COMPUTATIONAL

BIOLOGY AND APPLIED

BIOINFORMATICS
Edited by Heitor Silvério Lopes
and Leonardo Magalhães Cruz
Computational Biology and Applied Bioinformatics
Edited by Heitor Silvério Lopes and Leonardo Magalhães Cruz

Published by InTech
Janeza Trdine 9, 51000 Rijeka, Croatia

Copyright © 2011 InTech

All chapters are Open Access articles distributed under the Creative Commons
Non Commercial Share Alike Attribution 3.0 license, which permits to copy,
distribute, transmit, and adapt the work in any medium, so long as the original
work is properly cited. After this work has been published by InTech, authors
have the right to republish it, in whole or part, in any publication of which they
are the author, and to make other personal use of the work. Any republication,
referencing or personal use of the work must explicitly identify the original source.

Statements and opinions expressed in the chapters are these of the individual contributors
and not necessarily those of the editors or publisher. No responsibility is accepted
for the accuracy of information contained in the published articles. The publisher
assumes no responsibility for any damage or injury to persons or property arising out
of the use of any materials, instructions, methods or ideas contained in the book.

Publishing Process Manager Davor Vidic

Technical Editor Teodora Smiljanic
Cover Designer Jan Hyrat
Image Copyright Gencay M. Emin, 2010. Used under license from Shutterstock.com

First published August, 2011

Printed in Croatia

A free online edition of this book is available at www.intechopen.com

Additional hard copies can be obtained from [email protected]

Computational Biology and Applied Bioinformatics,

Edited by Heitor Silvério Lopes and Leonardo Magalhães Cruz
p. cm.
ISBN 978-953-307-629-4
free online editions of InTech
Books and Journals can be found at
www.intechopen.com
Contents

Preface IX

Part 1 Reviews 1

Chapter 1 Molecular Evolution & Phylogeny:

What, When, Why & How? 3
Pandurang Kolekar, Mohan Kale and Urmila Kulkarni-Kale

Chapter 2 Understanding Protein Function - The Disparity

Between Bioinformatics and Molecular Methods 29
Katarzyna Hupert-Kocurek and Jon M. Kaguni

Chapter 3 In Silico Identification

of Regulatory Elements in Promoters 47
Vikrant Nain, Shakti Sahi and Polumetla Ananda Kumar

Chapter 4 In Silico Analysis of Golgi Glycosyltransferases:

A Case Study on the LARGE-Like Protein Family 67
Kuo-Yuan Hwa, Wan-Man Lin and Boopathi Subramani

Chapter 5 MicroArray Technology - Expression Profiling

of MRNA and MicroRNA in Breast Cancer 87
Aoife Lowery, Christophe Lemetre, Graham Ball and Michael Kerin

Chapter 6 Computational Tools for Identification

of microRNAs in Deep Sequencing Data Sets 121
Manuel A. S. Santos and Ana Raquel Soares

Chapter 7 Computational Methods in Mass

Spectrometry-Based Protein 3D Studies 133
Rosa M. Vitale, Giovanni Renzone,
Andrea Scaloni and Pietro Amodeo

Chapter 8 Synthetic Biology & Bioinformatics

Prospects in the Cancer Arena 159
Lígia R. Rodrigues and Leon D. Kluskens
VI Contents

Chapter 9 An Overview of Hardware-Based

Acceleration of Biological Sequence Alignment 187
Laiq Hasan and Zaid Al-Ars

Part 2 Case Studies 203

Chapter 10 Retrieving and Categorizing Bioinformatics

Publications through a MultiAgent System 205
Andrea Addis, Giuliano Armano,
Eloisa Vargiu and Andrea Manconi

Chapter 11 GRID Computing and Computational Immunology 223

Ferdinando Chiacchio and Francesco Pappalardo

Chapter 12 A Comparative Study of Machine Learning

and Evolutionary Computation Approaches
for Protein Secondary Structure Classification 239
César Manuel Vargas Benítez, Chidambaram Chidambaram,
Fernanda Hembecker and Heitor Silvério Lopes

Chapter 13 Functional Analysis of the Cervical Carcinoma Transcriptome:

Networks and New Genes Associated to Cancer 259
Mauricio Salcedo, Sergio Juarez-Mendez,
Vanessa Villegas-Ruiz, Hugo Arreola, Oscar Perez,
Guillermo Gómez, Edgar Roman-Bassaure,
Pablo Romero, Raúl Peralta

Chapter 14 Number Distribution of Transmembrane

Helices in Prokaryote Genomes 279
Ryusuke Sawada and Shigeki Mitaku

Chapter 15 Classifying TIM Barrel Protein Domain Structure

by an Alignment Approach
Using Best Hit Strategy and PSI-BLAST 287
Chia-Han Chu, Chun Yuan Lin,
Cheng-Wen Chang, Chihan Lee and Chuan Yi Tang

Chapter 16 Identification of Functional Diversity

in the Enolase Superfamily Proteins 311
Kaiser Jamil and M. Sabeena

Chapter 17 Contributions of Structure Comparison Methods

to the Protein Structure Prediction Field 329
David Piedra, Marco d'Abramo and Xavier de la Cruz

Chapter 18 Functional Analysis of Intergenic

Regions for Gene Discovery 345
Li M. Fu
 Contents VII

Chapter 19 Prediction of Transcriptional Regulatory

Networks for Retinal Development 357
Ying Li, Haiyan Huang and Li Cai

Chapter 20 The Use of Functional Genomics in

Synthetic Promoter Design 375
Michael L. Roberts

Chapter 21 Analysis of Transcriptomic

and Proteomic Data in Immune-Mediated Diseases 397
Sergey Bruskin, Alex Ishkin, Yuri Nikolsky,
Tatiana Nikolskaya and Eleonora Piruzian

Chapter 22 Emergence of the Diversified Short ORFeome

by Mass Spectrometry-Based Proteomics 417
Hiroko Ao-Kondo, Hiroko Kozuka-Hata and Masaaki Oyama

Chapter 23 Acrylamide Binding to Its Cellular Targets:

Insights from Computational Studies 431
Emmanuela Ferreira de Lima and Paolo Carloni
Preface

Nowadays it is difficult to imagine an area of knowledge that can continue developing

without the use of computers and informatics. It is not different with biology, that has
seen an unpredictable growth in recent decades, with the rise of a new discipline,
bioinformatics, bringing together molecular biology, biotechnology and information
technology. More recently, the development of high throughput techniques, such as
microarray, mass spectrometry and DNA sequencing, has increased the need of
computational support to collect, store, retrieve, analyze, and correlate huge data sets
of complex information. On the other hand, the growth of the computational power
for processing and storage has also increased the necessity for deeper knowledge in
the field. The development of bioinformatics has allowed now the emergence of
systems biology, the study of the interactions between the components of a biological
system, and how these interactions give rise to the function and behavior of a living
being.

Bioinformatics is a cross-disciplinary field and its birth in the sixties and seventies
depended on discoveries and developments in different fields, such as: the proposed
double helix model of DNA by Watson and Crick from X-ray data obtained by
Franklin and Wilkins in 1953; the development of a method to solve the phase problem
in protein crystallography by Perutz's group in 1954; the sequencing of the first protein
by Sanger in 1955; the creation of the ARPANET in 1969 at Stanford UCLA; the
publishing of the Needleman-Wunsch algorithm for sequence comparison in 1970; the
first recombinant DNA molecule created by Paul Berg and his group in 1972; the
announcement of the Brookhaven Protein DataBank in 1973; the establishment of the
Ethernet by Robert Metcalfe in the same year; the concept of computers network and
the development of the Transmission Control Protocol (TCP) by Vint Cerf and Robert
Khan in 1974, just to cite some of the landmarks that allowed the rise of bioinformatics.
Later, the Human Genome Project (HGP), started in 1990, was also very important for
pushing the development of bioinformatics and related methods of analysis of large
amount of data.

This book presents some theoretical issues, reviews, and a variety of bioinformatics
applications. For better understanding, the chapters were grouped in two parts. It was
not an easy task to select chapters for these parts, since most chapters provide a mix of
review and case study. From another point of view, all chapters also have extensive
X Preface

biological and computational information. Therefore, the book is divided into two
parts. In Part I, the chapters are more oriented towards literature review and
theoretical issues. Part II consists of application-oriented chapters that report case
studies in which a specific biological problem is treated with bioinformatics tools.

Molecular phylogeny analysis has become a routine technique not only to understand
the sequence-structure-function relationship of biomolecules but also to assist in their
classification. The first chapter of Part I, by Kolekar et al., presents the theoretical basis,
discusses the fundamental of phylogenetic analysis, and a particular view of steps and
methods used in the analysis.

Methods for protein function and gene expression are briefly reviewed in Hupert-
Kocurek and Kaguni’s chapter, and contrasted with the traditional approach of
mapping a gene via the phenotype of a mutation and deducing the function of the
gene product, based on its biochemical analysis in concert with physiological studies.
An example of experimental approach is provided that expands the current
understanding of the role of ATP binding and its hydrolysis by DnaC during the
initiation of DNA replication. This is contrasted with approaches that yield large sets
of data, providing a different perspective on understanding the functions of sets of
genes or proteins and how they act in a network of biochemical pathways of the cell.

Due to the importance of transcriptional regulation, one of the main goals in the post-
genomic era is to predict how the expression of a given gene is regulated based on the
presence of transcription factor binding sites in the adjacent genomic regions. Nain et
al. review different computational approaches for modeling and identification of
regulatory elements, as well as recent advances and the current challenges.

In Hwa et al., an approach is proposed to group proteins into putative functional

groups by designing a workflow with appropriate bioinformatics analysis tools, to
search for sequences with biological characteristics belonging to the selected protein
family. To illustrate the approach, the workflow was applied to LARGE-like protein
family.

Microarray technology has become one of the most important technologies for
unveiling gene expression profiles, thus fostering the development of new
bioinformatics methods and tools. In the chapter by Lowery et al. a thorough review of
microarray technology is provided, with special focus on MRNA and microRNA
profiling of breast cancer.

MicroRNAs are a class of small RNAs of approximately 22 nucleotides in length that

regulate eukaryotic gene expression at the post-transcriptional level. Santos and Soares
present several tools and computational pipelines for miRNA identification, discovery
and expression from sequencing data.

Currently, the mass spectroscopy-based methods represent very important and

flexible tools for studying the dynamic features of proteins and their complexes. Such
 Preface XI

high-resolution methods are especially used for characterizing critical regions of the
systems under investigation. Vitale et al. present a thorough review of mass
spectrometry and the related computational methods for studying the three-
dimensional structure of proteins.

Rodrigues and Kluskens review synthetic biology approaches for the development of
alternatives for cancer diagnosis and drug development, providing several application
examples and pointing challenging directions of research.

Biological sequence alignment is an important and widely used task in bioinformatics.

It is essential to provide valuable and accurate information in the basic research, as
well as in daily use of the molecular biologist. The well-known Smith and Waterman
algorithm is an optimal sequence alignment method, but it is computationally
expensive for large instances. This fact fostered the research and development of
specialized hardware platforms to accelerate biological data analysis that use that
algorithm. Hasan and Al-Ars provide a thorough discussion and comparison of
available methods and hardware implementations for sequence alignment on different
platforms.

Exciting and updated issues are presented in Part II, where theoretical bases are
complemented with case studies, showing how bioinformatics analysis pipelines were
applied to answer a variety of biological issues.

During the last years we have witnessed an exponential growth of the biological data
and scientific articles. Consequently, retrieving and categorizing documents has
become a challenging task. The second part of the book starts with the chapter by
Addis et al. that propose a multiagent system for retrieving and categorizing
bioinformatics publications, with special focus on the information extraction task and
adopted hierarchical text categorization technique.

Computational immunology is a field of science that encompasses high-throughput

genomic and bioinformatic approaches to immunology. On the other hand, grid
computing is a powerful alternative for solving problems that are computationally
intensive. Pappalardo and Chiachio present two different studies of using
computational immunology approaches implemented in a grid infrastructure:
modeling atherosclerosis and optimal protocol searching for vaccine against
mammary carcinoma.

Despite the growing number of proteins discovered as sub-product of the many

genome sequencing projects, only a very few number of them have a known three-
dimensional structure. A possible way to infer the full structure of an unknown
protein is to identify potential secondary structures in it. Chidambaram et al. compare
the performance of several machine learning and evolutionary computing methods for
the classification of secondary structure of proteins, starting from their primary
structure.
XII Preface

Cancer is one of the most important public health problems worldwide. Breast and
cervical cancer are the most frequent in female population. Salcedo et al. present a
study about the functional analysis of the cervical carcinoma transcriptome, with focus
on the methods for unveiling networks and finding new genes associated to cervical
cancer.

In Sawada and Mitaku, the number distribution of transmembrane helices is

investigated to show that it is a feature under natural selection in prokaryotes and
how membrane proteins with high number of transmembrane helices disappeared in
random mutations by simulation data.

In Chu et al., an alignment approach using the pure best hit strategy is proposed to
classify TIM barrel protein domain structures in terms of the superfamily and family
categories with high accuracy.

Jamil and Sabeena use classic bioinformatic tools, such as ClustalW for Multiple
Sequence Alignment, SCI-PHY server for superfamily determination, ExPASy tools for
pattern matching, and visualization softwares for residue recognition and functional
elucidation to determine the functional diversity of the enolase enzyme superfamily.

Quality assessment of structure predictions is an important problem in bioinformatics

because quality determines the application range of predictions. Piedra et al. briefly
review some applications used in protein structure prediction field, were they are used
to evaluate overall prediction quality, and show how structure comparison methods
can also be used to identify the more reliable parts in “de novo” analysis and how this
information can help to refine/improve these models.

In Fu, a new method is presented that explores potential genes in intergenic regions of
an annotated genome on the basis of their gene expression activity. The method was
applied to the M. tuberculosis genome where potential protein-coding genes were
found, based on bioinformatics analysis in conjunction with transcriptional evidence
obtained using the Affymetrix GeneChip. The study revealed potential genes in the
intergenic regions, such as DNA-binding protein in the CopG family and a nickel
binding GTPase, as well as hypothetical proteins.

Cai et al. present a new method for developmental studies. It combines experimental
studies and computational analysis to predict the trans-acting factors and
transcriptional regulatory networks for mouse embryonic retinal development.

The chapter by Roberts shows how advances in bioinformatics can be applied to the
development of improved therapeutic strategies. The chapter describes how functional
genomics experimentation and bioinformatics tools could be applied to the design of
synthetic promoters for therapeutic and diagnostic applications or adapted across the
biotech industry. Designed synthetic gene promoters can than be incorporated in
novel gene transfer vectors to promote safer and more efficient expression of
therapeutic genes for the treatment of various pathological conditions. Tools used to
 Preface XIII

analyze data obtained from large-scale gene expression analyses, which are
subsequently used in the smart design of synthetic promoters are also presented.

Bruskin et al. describe how candidate genes commonly involved in psoriasis and
Crohn's disease were detected using lists of differentially expressed genes from
microarrays experiments with different numbers of probes. These gene codes for
proteins are particular targets for elaborating new approaches to treating these
pathologies. A comprehensive meta-analysis of proteomics and transcriptomics of
psoriatic lesions from independent studies is performed. Network-based analysis
revealed similarities in regulation at both proteomics and transcriptomics level.

Some eukaryotic mRNAs have multiple ORFs, which are recognized as polycistronic
mRNAs. One of the well-known extra ORFs is the upstream ORF (uORF), that
functions as a regulator of mRNA translation. In Ao-Kondo et al., this issue is
addressed and an introduction to the mechanism of translation initiation and
functional roles of uORF in translational regulation is given, followed by a review of
how the authors identified novel small proteins with Mass Spectrometry and a
discussion on the progress of bioinformatics analyses for elucidating the
diversification of short coding regions defined by the transcriptome.

Acrylamide might feature toxic properties, including neurotoxicity and

carcinogenicity in both mice and rats, but no consistent effect on cancer incidence in
humans could be identified. In the chapter written by Lima and Carloni, the authors
report the use of bioinformatics tools, by means of molecular docking and molecular
simulation procedures, to predict and explore the structural determinants of
acrylamide and its derivative in complex with all of their known cellular target
proteins in human and mice.

Professor Heitor Silvério Lopes

Bioinformatics Laboratory, Federal University of Technology – Paraná,
Brazil

Professor Leonardo Magalhães Cruz

Biochemistry Department, Federal University of Paraná,
Brazil
 Part 1

Reviews
 1

Molecular Evolution & Phylogeny:

What, When, Why & How?
Pandurang Kolekar1, Mohan Kale2 and Urmila Kulkarni-Kale1
1Bioinformatics Centre, University of Pune
2Department of Statistics, University of Pune
India

1. Introduction
The endeavour for the classification and study of evolution of organisms, pioneered by
Linneaus and Darwin on the basis of morphological and behavioural features of organisms,
is now being propelled by the availability of molecular data. The field of evolutionary
biology has experienced a paradigm shift with the advent of sequencing technologies and
availability of molecular sequence data in the public domain databases. The post-genomic
era provides unprecedented opportunities to study the process of molecular evolution,
which is marked with the changes organisms acquire and inherit. The species are
continuously subjected to evolutionary pressures and evolve suitably. These changes are
observed in terms of variations in the sequence data that are collected over a period of time.
Thus, the molecular sequence data archived in various databases are the snapshots of the
evolutionary process and help to decipher the evolutionary relationships of genes/proteins
and genomes/proteomes for a group of organisms. It is known that the individual genes
may evolve with varying rates and the evolutionary history of a gene may or may not
coincide with the evolution of the species as a whole. One should always refrain from
discussing the evolutionary relationship between organisms when analyses are performed
using limited/partial data. Thorough understanding of the principles and methods of
phylogeny help the users not only to use the available software packages in an efficient
manner, but also to make appropriate choices of methods of analysis and parameters so that
attempts can be made to maximize the gain on huge amount of available sequence data.
As compared to classical phylogeny based on morphological data, molecular phylogeny has
distinct advantages, for instance, it is based on sequences (as descrete characters) unlike the
morphological data, which is qualitative in nature. While the tree of life is depicted to have
three major branches as bacteria, archaea and eukaryotes (it excludes viruses), the trees
based on molecular data accounts for the process of evolution of bio-macromolecules (DNA,
RNA and protein). The trees generated using molecular data are thus referred to as ‘inferred
trees’, which present a hypothesized version of what might have happened in the process of
evolution using the available data and a model. Therefore, many trees can be generated
using a dataset and each tree conveys a story of evolution. The two main types of
information inherent in any phylogenetic tree are the topology (branching pattern) and the
branch lengths.
4 Computational Biology and Applied Bioinformatics

Before getting into the actual process of molecular phylogeny analysis (MPA), it will be
helpful to get familiar with the concepts and terminologies frequently used in MPA.
Phylogenetic tree: A two-dimensional graph depicting nodes and branches that illustrates
evolutionary relationships between molecules or organisms.
Nodes: The points that connect branches and usually represent the taxonomic units.
Branches: A branch (also called an edge) connects any two nodes. It is an evolutionary
lineage between or at the end of nodes. Branch length represents the number of
evolutionary changes that have occurred in between or at the end of nodes. Trees with
uniform branch length (cladograms), branch lengths proportional to the changes or distance
(phylograms) are derived based on the purpose of analysis.
Operational taxonomic units (OTUs): The known external/terminal nodes in the
phylogenetic tree are termed as OTU.
Hypothetical taxonomic units (HTUs): The internal nodes in the phylogenetic tree that are
treated as common ancestors to OTUs. An internal node is said to be bifurcating if it has
only two immediate descendant lineages or branches. Such trees are also called binary or
dichotomous as any dividing branch splits into two daughter branches. A tree is called a
‘multifurcating’ or ‘polytomous’ if any of its nodes splits into more than two immediate
descendants.
Monophyletic: A group of OTUs that are derived from a single common ancestor
containing all the descendents of single common ancestor.
Polyphyletic: A group of OTUs that are derived from more than one common ancestor.
Paraphyletic: A group of OTUs that are derived from a common ancestor but the group
doesn’t include all the descendents of the most recent common ancestor.
Clade: A monophyletic group of related OTUs containing all the descendants of the
common ancestor along with the ancestor itself.
Ingroup: A monophyletic group of all the OTUs that are of primary interest in the
phylogenetic study.
Outgroup: One or more OTUs that are phylogenetically outside the ingroup and known to
have branched off prior to the taxa included in a study.
Cladogram: The phylogenetic tree with branches having uniform lengths. It only depicts the
relationship between OTUs and does not help estimate the extent of divergence.
Phylogram: The phylogenetic tree with branches having variable lengths that are
proportional to evolutionary changes.
Species tree: The phylogenetic tree representing the evolutionary pathways of species.
Gene tree: The phylogenetic tree reconstructed using a single gene from each species. The
topology of the gene tree may differ from ‘species tree’ and it may be difficult to reconstruct
a species tree from a gene tree.
Unrooted tree: It illustrates the network of relationship of OTUs without the assumption of
common ancestry. Most trees generated using molecular data are unrooted and they can be
rooted subsequently by identifying an outgroup. Total number of bifurcating unrooted trees
can be derived using the equation: Nu= (2n-5)!/2n-3 (n-3)!
Rooted tree: An unrooted phylogenetic tree can be rooted with outgroup species, as a
common ancestor of all ingroup species. It has a defined origin with a unique path to each
ingroup species from the root. The total number of bifurcating rooted trees can be calculated
using the formula, Nr= (2n-3)!/2n-2 (n-2)! (Cavalli-Sforza & Edwards, 1967). Concept of
unrooted and rooted trees is illustrated in Fig. 1.
Molecular Evolution & Phylogeny: What, When, Why & How? 5

Fig. 1. Sample rooted and unrooted phylogenetic trees drawn using 5 OTUs . The external
and internal nodes are labelled with alphabets and Arabic numbers respectively. Note that
the rooted and unrooted trees shown here are one of the many possible trees (105 rooted
and 15 unrooted) that can be obtained for 5 OTUs.
The MPA typically involves following steps
• Definition of problem and motivation to carry out MPA
• Compilation and curation of homologous sequences of nucleic acids or proteins
• Multiple sequence alignments (MSA)
• Selection of suitable model(s) of evolution
• Reconstruction of phylogenetic tree(s)
• Evaluation of tree topology
A brief account of each of these steps is provided below.

2. Definition of problem and motivation to carry out MPA

Just like any scientific experiment, it is necessary to define the objective of MPA to be carried
out using a set of molecular sequences. MPA has found diverse applications, which include
classification of organisms, DNA barcoding, subtyping of viruses, study the co-evolution of
genes and proteins, estimation of divergence time of species, study of the development of
pandemics and pattern of disease transmission, parasite-vector-host relationships etc. The
biological investigations where MPA constitute a major part of analyses are listed here. A
virus is isolated during an epidemic. Is it a new virus or an isolate of a known one? Can a
genotype/serotype be assigned to this isolate just by using the molecular sequence data? A
few strains of a bacterium are resistant to a drug and a few are sensitive. What and where
are the changes that are responsible for such a property? How do I choose the attenuated
strains, amongst available, such that protection will be offered against most of the wild type
strains of a given virus? Thus, in short, the objective of the MPA plays a vital role in
deciding the strategy for the selection of candidate sequences and adoption of the
appropriate phylogenetic methods.

3. Compilation and curation of homologous sequences

The compilation of nucleic acid or protein sequences, appropriate to undertake validation of
hypothesis using MPA, from the available resources of sequences is the next step in MPA.
6 Computational Biology and Applied Bioinformatics

At this stage, it is necessary to collate the dataset consisting of homologous sequences with the
appropriate coverage of OTUs and outgroup sequences, if needed. Care should be taken to
select the equivalent regions of sequences having comparable lengths (± 30 bases or amino
acids) to avoid the subsequent errors associated with incorrect alignments leading to incorrect
sampling of dataset, which may result in erroneous tree topology. Length differences of >30
might result in insertion of gaps by the alignment programs, unless the gap opening penalty is
suitably modified. Many comprehensive primary and derived databases of nucleic acid and
protein sequences are available in public domain, some of which are listed in Table 1. The
database issue published by the journal ‘Nucleic Acids research’ (NAR) in the month of
January every year is a useful resource for existing as well as upcoming databases. These
databases can be queried using the ‘text-based’ or ‘sequence-based’ database searches.

Database URL Reference

Nucleotide
GenBank http://www.ncbi.nlm.nih.gov/genbank/ Benson et al., 2011
EMBL http://www.ebi.ac.uk/embl/ Leinonen et al., 2011
DDBJ http://www.ddbj.nig.ac.jp/ Kaminuma et al., 2011
Protein
GenPept http://www.ncbi.nlm.nih.gov/protein Sayers et al., 2011
Swiss-Prot http://expasy.org/sprot/ The UniProt Consortium (2011)
UniProt http://www.uniprot.org/ The UniProt Consortium (2011)
Derived
RDP http://rdp.cme.msu.edu/ Cole et al., 2009
HIV http://www.hiv.lanl.gov/content/index Kuiken et al., 2009
HCV http://www.hcvdb.org/ http://www.hcvdb.org/

Table 1. List of some of the commonly used nucleotide, protein and molecule-/species-
specific databases.
Text-based queries are supported using search engines viz., Entrez and SRS, which are
available at NCBI and EBI respectively. The list of hits returned after the searches needs to
be curated very carefully to ensure that the data corresponds to the gene/protein of interest
and is devoid of partial sequences. It is advisable to refer to the feature-table section of every
entry to ensure that the data is extracted correctly and corresponds to the region of interest.
The sequence-based searches involve querying the databases using sequence as a probe and
are routinely used to compile a set of homologous sequences. Once the sequences are
compiled in FASTA or another format, as per the input requirements of MPA software, the
sequences are usually assigned with unique identifiers to facilitate their identification and
comparison in the phylogenetic trees. If the sequences posses any ambiguous characters or
low complexity regions, they could be carefully removed from sequences as they don’t
contribute to evolutionary analysis. The presence of such regions might create problems in
alignment, as it could lead to equiprobable alternate solutions to ‘local alignment’ as part of
Molecular Evolution & Phylogeny: What, When, Why & How? 7

a global alignment. Such regions possess ‘low’ information content to favour a tree topology
over the other. The inferiority of input dataset interferes with the analysis and interpretation
of the MPA. Thus, compilation of well-curated sequences, for the problem at hand, plays a
crucial role in MPA.
The concept of homology is central to MPA. Sequences are said to be homologous if they share
a common ancestor and are evolutionarily related. Thus, homology is a qualitative description
of the relationship and the term %homology has no meaning. However, supporting data for
deducing homology comes from the extent of sequence identity and similarity, both of which
are quantitative terms and are expressed in terms of percentage.
The homologous sequences are grouped into three types, viz., orthologs (same gene in
different species), paralogs (the genes that originated from duplication of an ancestral gene
within a species) and xenologs (the genes that have horizontally transferred between the
species). The orthologous protein sequences are known to fold into similar three-dimensional
shapes and are known to carry out similar functions. For example, haemoglobin alpha in horse
and human. The paralogous sequences are copies of the ancestral genes evolving within the
species such that nature can implement a modified function. For example haemoglobin alpha
and beta in horse. The xenologs and horizontal transfer events are extremely difficult to be
proved only on the basis of sequence comparison and additional experimental evidence to
support and validate the hypothesis is needed. The concepts of sequence alignments,
similarity and homology are extensively reviewed by Phillips (2006).

4. Multiple sequence alignments (MSA)

MSA is one of the most common and critical steps of classical MPA. The objective of MSA is to
juxtapose the nucleotide or amino acid residues in the selected dataset of homologous
sequences such that residues in the column of MSA could be used to derive the sequence of
the common ancestor. The MSA algorithms try to maximize the matching residues in the given
set of sequences with a pre-defined scoring scheme. The MSA produces a matrix of characters
with species in the rows and character sites in columns. It also introduces the gaps, simulating
the events of insertions and deletions (also called as indels). Insertion of gaps also helps in
making the lengths of all sequences same for the sake of comparison. All the MSA algorithms
are guaranteed to produce optimal alignment above a threshold value of detectable sequence
similarity. The alignment accuracy is observed to decrease when sequence similarity drops
below 35% towards the twilight (<35% but > 25%) and moonlight zones (<25%) of similarity.
The character matrix obtained in MSA reveals the pattern of conservation and variability
across the species, which in turn reveals the motifs and the signature sequences shared by
species to retain the fold and function. The analysis of variations can be gainfully used to
identify the changes that explain functional and phenotypic variability, if any, across OTUs.
Many algorithms have been specially developed for MSA and subsequently improved to
achieve higher accuracy. One of the popular heuristics-based MSA approach follows
progressive alignment procedure, in which sequences are compared in a pair wise fashion to
build a distance matrix containing percent identity values. A clustering algorithm is then
applied to distance matrix to generate a guide tree. The algorithm then follows a guide tree
to add the pair wise alignments together starting from the leaf to root. This ensures the
sequences with higher similarity are aligned initially and distantly related sequences are
progressively added to the alignment of aligned sequences. Thus, the gaps inserted are
always retained. A suitable scoring function, sum-of-pairs, consensus, consistency-based etc.
8 Computational Biology and Applied Bioinformatics

is employed to derive the optimum MSA (Nicholas et al., 2002; Batzoglou, 2005). Most of the
MSA packages use Needleman and Wunsch (1970) algorithm to compute pair wise sequence
similarity. The ClustalW is the widely used MSA package (Thompson et al., 1994). Recently
many alternative MSA algorithms are also being developed, which are enlisted in Table 2.
The standard benchmark datasets are used for comparative assessment of the alternative
approaches (Aniba et al., 2010; Thompson et al., 2011). Irrespective of the proven
performance of MSA methods for individual genes and proteins, some of the challenges and
issues regarding computational aspects involved in handling genomic data are still the
causes of concern (Kemena & Notredame, 2009).

Alignment
Algorithm description Available at / Reference
programs
http://www.ebi.ac.uk/Tools/msa/clustalw2/ ;
ClustalW Progressive
Thompson et al., 1994
http://www.ebi.ac.uk/Tools/msa/muscle/ ;
MUSCLE Progressive/iterative
Edgar, 2004
http://www.ebi.ac.uk/Tools/msa/tcoffee/ ;
T-COFFEE Progressive
Notredame et al., 2000
http://bibiserv.techfak.uni-bielefeld.de/dialign/ ;
DIALIGN2 Segment-based
Morgenstern et al., 1998
http://mafft.cbrc.jp/alignment/software/ ;
MAFFT Progressive/iterative
Katoh et al., 2005
Alignment visualization programs
http://www.mbio.ncsu.edu/bioedit/bioedit.html
*BioEdit
; Hall, 1999
http://www.megasoftware.net/ ;
MEGA5
Kumar et al., 2008
http://dambe.bio.uottawa.ca/dambe.asp ;
DAMBE
Xia & Xie, 2001
http://aig.cs.man.ac.uk/research/utopia/cinema ;
CINEMA5
Parry-Smith et al., 1998
*: Not updated since 2008, but the last version is available for use.
Table 2. List of commonly used multiple sequence alignment programs and visualization
tools.
The MSA output can also be visualized and edited, if required, with the software like
BioEdit, DAMBE etc. Multiple alignment output shows the conserved and variable sites,
usually residues are colour coded for the ease of visualisation, identification and analysis.
The character sites in MSA can be divided as conserved (all the sequences have same
residue or base), variable-non-informative (singleton site) and variable-informative sites.
The sites containing gaps in all or majority of the species are of no importance from the
evolutionary point of view and are usually removed from MSA while converting MSA data
to input data for MPA. A sample MSA is shown in Fig. 2. The sequences of surface
hydrophobic (SH) protein from various genotypes (A to M) of Mumps virus, are aligned. A
careful visual inspection of MSA allows us to locate the patterns and motifs (LLLXIL) in a
given set of sequences. Apart from MPA, the MSA data in turn can be used for the
construction of position specific scoring matrix (PSSM), generation of consensus sequence,
Molecular Evolution & Phylogeny: What, When, Why & How? 9

sequence logos, identification and prioritisation of potential B- and T-cell epitopes etc.
Nowadays the databases of curated, pre-computed alignments of reference species are also
being made available, which can be used for the benchmark comparison, evaluation purpose
(Thompson et al., 2011) and it also helps to keep the track of changes that get accumulated in
the species over a period of time. For example, in case of viruses, observed changes are
correlated with emergence of new genotypes (Kulkarni-Kale et al., 2004; Kuiken et al., 2005).

Fig. 2. The complete multiple sequence alignment of the surface hydrophobic (SH) proteins
of Mumps virus genotypes (A to M) carried out using ClustalW. The MSA is viewed using
BioEdit. The species labels in the leftmost column begin with genotype letter (A-M) followed
by GenBank accession numbers. The scale for the position in alignment is given at the top of
the alignment. The columns with conserved residues are marked with an “*” in the last row.

5. Selection of a suitable model of evolution

The existing MPA methods utilize the mathematical models to describe the evolution of
sequence by incorporating the biological, biochemical and evolutionary considerations.
These mathematical models are used to compute genetic distances between sequences. The
use of appropriate model of evolution and statistical tests help us to infer maximum
evolutionary information out of sequence data. Thus, the selection of the right model of
10 Computational Biology and Applied Bioinformatics

sequence evolution becomes important as a part of effective MPA. Two types of approaches
are adapted for the building of models, first one is empirical i.e. using the properties
revealed through comparative studies of large datasets of observed sequences, and the other
is parametrical, which uses biological and biochemical knowledge about the nucleic acid
and protein sequences, for example the favoured substitution patterns of residues.
Parametric models obtain the parameters from the MSA dataset under study. Both types of
approaches result in the models based on the Markov process, in the form of matrix
representing the rate of all possible transitions between the types of residues (4 nucleotides
in nucleic acids and 20 amino acids in proteins). According to the type of sequence (nucleic
acid or protein), two categories of models have been developed.

5.1 Models of nucleotide substitution

The nucleotide substitution models are based on the parametric approach with the use of
mainly three parameters i) nucleotides frequencies, ii) rate of nucleotide substitutions and
iii) rate heterogeneity. Nucleotide frequencies, account for the compositional sequence
constraints such as GC content. These are subsequently used in a model to allow the
substitutions of a certain type to occur more likely than others. The nucleotide substitution
parameter is used to represent a measure of biochemical similarity. Higher the similarity
between the nucleotide bases, the more is the rate of substitution between them, for
example, the transitions are more frequent than transversions. A parameter of rate
heterogeneity accounts for the unequal rates of substitution across the variable sites, which
can be correlated with the constraints of genetic code, selection for the gene function etc. The
site variability is modelled by gamma distribution of rates across sites. The shape parameter
of gamma distribution determines amount of heterogeneity among sites, larger values of
shape parameter gives a bell shaped distribution suggesting little or no rate variation across
the sites whereas small values of it gives J-shaped distribution indicating high rate variation
among sites along with low rates of evolution at many sites.
Varieties of nucleotide substitution models have been developed with a set of assumptions
and parameters described as above. Some of the well-known models of nucleotide
substitutions include Jukes-Cantor (JC) one-parameter model (Jukes & Cantor, 1969),
Kimura two-parameter model (K2P) (Kimura, 1980), Tamura’s model (Tamura, 1992),
Tamura and Nei model (Tamura & Nei, 1993) etc. These models make use of different
biological properties such as, transitions, transversions, G+C content etc. to compute
distances between nucleotide sequences. The substitution patterns of nucleotides for some
of these models are shown in Fig. 3.

5.2 Models of amino acid replacement

In contrast to nucleotide substitution models, amino acid replacement models are developed
using empirical approach. Schwarz and Dayhoff (1979) developed the most widely used
model of protein evolution in which, the replacement matrix was obtained from the alignment
of globular protein sequences with 15% divergence. The Dayhoff matrices, known as PAM
matrices, are also used by database searching methods. The similar methodology was adopted
by other model developers but with specialized databases. Jones et al., (1994) have derived a
replacement matrix specifically for membrane proteins, which has values significantly
different from Dayhoff matrix suggesting the remarkably different pattern of amino acid
replacements observed in the membrane proteins. Thus, such a matrix will be more
Molecular Evolution & Phylogeny: What, When, Why & How? 11

appropriate for the phylogenetic study of membrane proteins. On the other hand, Adachi and
Hasegawa (1996) obtained a replacement matrix using mitochondrial proteins across 20
vertebrate species and can be effectively used for mitochondrial protein phylogeny. Henikoff
and Henikoff (1992) derived the series of BLOSUM matrices using local, ungapped alignments
of distantly related sequences. The BLOSUM matrices are widely used in similarity searches
against databases than for phylogenetic analyses.

Fig. 3. The types of substitutions in nucleotides. α denotes the rate of transitions and β
denotes the rate of transversions. For example, in the case of JC model α=β while in the case
of K2P model α>β.
Recently, structural constraints of the nucleic acids and proteins are also being incorporated in
the building of models of evolution. For example, Rzhetsky (1995) contributed a model to
estimate the substitution patterns in ribosomal RNA genes with the account of secondary
structure elements like stem-loops in ribosomal RNAs. Another approach introduced a model
with the combination of protein secondary structures and amino acid replacement (Lio &
Goldman, 1998; Thorne et al., 1996). The overview of different models of evolution and the
criteria for the selection of models is also provided by Lio & Goldman (1998); Luo et al. (2010).

6. Reconstruction of a phylogenetic tree

The phylogeny reconstruction methods result in a phylogenetic tree, which may or may not
corroborate with the true phylogenetic tree. There are various methods of phylogeny
reconstruction that are divided into two major groups viz. character-based and distance-
based.
Character-based methods use a set of discrete characters, for example, in case of MSA data
of nucleotide sequences, each position in alignment is reffered as “character” and nucleotide
(A, T, G or C) present at that position is called as the “state” of that “character”. All such
characters are assumed to evolve independent of each other and analysed separately.
Distance-based methods on other hand use some form of distance measure to compute the
dissimilarity between pairs of OTUs, which subsequently results in derivation of distance
matrix that is given as an input to clustering methods like Neighbor-Joining (N-J) and
Unweighted Pair Group Method with Arithmetic mean (UPGMA) to infer phylogenetic tree.
The character-based and distance-based methods follow exhaustive search and/or stepwise
clustering approach to arrive at an optimum phylogenetic tree, which explains the
12 Computational Biology and Applied Bioinformatics

evolutionary pattern of the OTUs under study. The exhaustive search method examines
theoretically all possible tree topologies for a chosen number of species and derives the best
tree topology using a set of certain criteria. Table 3 shows the possible number of rooted and
unrooted trees for n number of species/OTUs.

Number of Number of Number of

OTUs unrooted trees rooted trees
2 1 1
3 1 3
4 3 15
5 15 105
6 105 945
10 2027025 34459425
Table 3. The number of possible rooted and unrooted trees for a given number of OTUs. The
number of possible unrooted trees for n OTUs is given by (2n-5)!/[2n-3(n-3)!]; and rooted
trees is given by (2n-3)!/[2n-2(n-2)!]
Whereas, stepwise clustering methods employ an algorithm, which begins with the
clustering of highly similar OTUs. It then combines the clustered OTUs such that it can be
treated as a single OTU representing the ancestor of combined OTUs. This step reduces the
complexity of data by one OTU. This process is repeated and in a stepwise manner adding
the remaining OTUs until all OTUs are clustered together. The stepwise clustering approach
is faster and computationally less intensive than the exhaustive search method.
The most widely used distance-based methods include N-J & UPGMA and character-based
methods include Maximum Parsimony (MP) and Maximum Likelihood (ML) methods
(Felsenstein, 1996). All of these methods make particular assumptions regarding
evolutionary process, which may or may not be applicable to the actual data. Thus, before
selection of a phylogeny reconstruction method, it is recommended to take into account the
assumptions made by the method to infer the best phylogenetic tree. The list of widely used
phylogeny inference packages is given in Table 4.

Package Available from / Reference

http://evolution.genetics.washington.edu/phylip.html ;
PHYLIP
Felsenstein, 1989
http://paup.csit.fsu.edu/ ;
PAUP
Wilgenbusch & Swofford, 2003
http://www.megasoftware.net/ ;
MEGA5
Kumar et al., 2008
http://mrbayes.csit.fsu.edu/ ;
MrBayes
Ronquist & Huelsenbeck, 2003
http://www.tree-puzzle.de/ ;
TREE-PUZZLE
Schmidt et al., 2002
Table 4. The list of widely used packages for molecular phylogeny.
Molecular Evolution & Phylogeny: What, When, Why & How? 13

6.1 Distance-based methods of phylogeny reconstruction

The distance-based phylogeny reconstruction begins with the computation of pair wise genetic
distances between molecular sequences with the use of appropriate substitution model, which
is built on the basis of evolutionary assumptions, discussed in section 4. This step results in
derivation of a distance matrix, which is subsequently used to infer a tree topology using the
clustering method. Fig. 4 shows the distance matrix computed for a sample sequence dataset
of 5 OTUs with 6 sites using Jukes-Cantor distance measure. A distance measure possesses
three properties, (a) a distance of OTU from itself is zero, D(i, i) = 0; (b) the distance of OTU i
from another OTU j must be equal to the distance of OTU j from OTU i, D(i, j) = D(j, i); and (c)
the distance measure should follow the triangle inequality rule i.e. D(i, j) ≤ D(i, k) + D(k, j). The
accurate estimation of genetic distances is a crucial requirement for the inference of correct
phylogenetic tree, thus choice of the right model of evolution is as important as the choice of
clustering method. The popular methods used for clustering are UPGMA and N-J.

Fig. 4. The distance matrix obtained for a sample nucleotide sequence dataset using Jukes-
Cantor model. Dataset contains 5 OTUs (A-E) and 6 sites shown in Phylip format. Dnadist
program in PHYLIP package is used to compute distance matrix.

6.1.1 UPGMA method for tree building

The UPGMA method was developed by Sokal and Michener (1958) and is the most widely
used clustering methodology. The method is based on the assumptions that the rate of
substitution for all branches in the tree is constant (which may not hold true for all data) and
branch lengths are additive. It employs hierarchical agglomerative clustering algorithm,
which produces ultrametric tree in such a way that every OTU is equidistant from the root.
The clustering process begins with the identification of the highly similar pair of OTUs (i & j)
as decided from the distance value D(i, j) in distance matrix. The OTUs i and j are clustered
together and combined to form a composite OTU ij. This gives rise to new distance matrix
shorter by one row and column than initial distance matrix. The distances of un-clustered
OTUs remain unchanged. The distances of remaining OTUs (for e.g. k) from composite OTUs
are represented as the average of the initial distances of that OTU from the individual
members of composite OTU (i.e. D(ij, k) = [D(i, k) + D(j, k)]/2). In this way a new distance
matrix is calculated and in the next round, the OTUs with least dissimilarity are clustered
together to form another composite OTU. The remaining steps are same as discussed in the
first round. This process of clustering is repeated until all the OTUs are clustered.
The sample calculations and steps involved in UPGMA clustering algorithm using distance
matrix shown in Fig. 4 are given below.
14 Computational Biology and Applied Bioinformatics

Iteration 1: OTU A is minimally equidistant from OTUs B and C. Randomly we select the
OTUs A and B to form one composite OTU (AB). A and B are clustered together. Compute
new distances of OTUs C, D and E from composite OTU (AB). The distances between
unclustered OTUs will be retained. See Fig. 4 for initial distance matrix and Fig. 5 for
updated matrix after first iteration of UPGMA.
d(AB,C) = [d(A,C) + d(B,C)]/2 = [0.188486 + 0.440840]/2 = 0.314633
d(AB,D) = [d(A,D) + d(B,D)]/2 = [0.823959 + 0.440840]/2 = 0.632399
d(AB,E) = [d(A,E) + d(B,E)]/2 = [1.647918 + 0.823959]/2 = 1.235938

Fig. 5. The updated distance matrix and clustering of A and B after the 1st iteration of
UPGMA.
Iteration 2: OTUs (AB) and C are minimally distant. We select these OTUs to form one
composite OTU (ABC). AB and C are clustered together. We then compute new distances of
OTUs D and E from composite OTU (ABC). See Fig. 5 for distance matrix obtained in
iteration 1 and Fig. 6 for updated matrix after the second iteration of UPGMA.
d(ABC,D) = [d(AB,D) + d(C,D)]/2 = [0.632399 + 1.647918]/2 = 1.140158
d(ABC,E) = [d(AB,E) + d(C,E)]/2 = [1.235938 + 0.823959]/2 = 1.029948

Fig. 6. The updated distance matrix and clustering of A, B and C after the 2nd iteration of
UPGMA.
Iteration 3: OTUs D and E are minimally distant. We select these OTUs to form one
composite OTU (DE). D and E are clustered together. Compute new distances of OTUs
(ABC) and (DE) from each other. Finally the remaining two OTUs are clustered together. See
Fig. 6 for distance matrix obtained in iteration 2 and Fig. 7 for updated matrix after third
iteration of UPGMA.
d(ABC,DE) = [d(ABC,D) + d(ABC,E)]/2 = [1.140158 + 1.029948]/2 = 1.085053
Molecular Evolution & Phylogeny: What, When, Why & How? 15

Fig. 7. The updated distance matrix and clustering of OTUs after the 3rd iteration of
UPGMA. Numbers on the branches indicate branch lengths, which are additive.

6.1.2 N-J method for tree building

The N-J method for clustering was developed by Saitou and Nei (1987). It reconstructs the
unrooted phylogenetic tree with branch lengths using minimum evolution criterion that
minimizes the lengths of tree. It does not assume the constancy of substitution rates across
sites and does not require the data to be ultrametric, unlike UPGMA. Hence, this method is
more appropriate for the sites with variable rates of evolution.
N-J method is known to be a special case of the star decomposition method. The initial tree
topology is a star. The input distance matrix is modified such that the distance between
every pair of OTUs is adjusted using their average divergence from all remaining OTUs. The
least dissimilar pair of OTUs is identified from the modified distance matrix and is
combined together to form single composite OTU. The branch lengths of individual
members, clustered in composite OTU, are computed from internal node of composite OTU.
Now the distances of remaining OTUs from composite OTU are redefined to give a new
distance matrix shorter by one OTU than the initial matrix. This process is repeated till all
the OTUs are grouped together, while keeping track of nodes, which results in a final
unrooted tree topology with minimized branch lengths. The unrooted phylogenetic tree,
thus obtained can be rooted using an outgroup species. The BIONJ (Gascuel 1997),
generalized N-J (Pearson et al., 1999) and Weighbor (Bruno et al., 2000) are some of the
recently proposed alternative versions of N-J algorithm. The sample calculation and steps
involved in N-J clustering algorithm, using distance matrix shown in Fig. 4, are given below.
Iteration 1: Before starting the actual process of clustering the vector r is calculated as
following with N=5, refer to the initial distance matrix given in Fig. 4 for reference values.
r(A) = [d(A,B)+ d(A,C)+ d(A,D)+ d(A,E)]/(N-2) = 0.949616
r(B) = [d(B,A)+ d(B,C)+ d(B,D)+ d(B,E)]/(N-2) = 0.631375
r(C) = [d(C,A)+ d(C,B)+ d(C,D)+ d(C,E)]/(N-2) = 1.033755
r(D) = [d(D,A)+ d(D,B)+ d(D,C)+ d(D,E)]/(N-2) = 1.245558
r(E) = [d(E,A)+ d(E,B)+ d(E,C)+ d(E,D)]/(N-2) = 1.373265
Using these r values, we construct a modified distance matrix, Md, such that
MD(i,j) = d(i,j) – (ri + rj).
See Fig. 8 for Md.
16 Computational Biology and Applied Bioinformatics

Fig. 8. The modified distance matrix Md and clustering for iteration 1 of N-J.
As can be seen from Md in Fig. 8, OTUs A and C are minimally distant. We select the OTUs
A and C to form one composite OTU (AC). A and C are clustered together.
Iteration 2: Compute new distances of OTUs B, D and E from composite OTU (AC).
Distances between unclustered OTUs will be retained from the previous step.
d(AC,B) = [d(A,B) + d(C,B)-d(A,C)]/2 = 0.22042
d(AC,D) = [d(A,D) + d(C,D) -d(A,C)]/2 = 1.141695
d(AC,E) = [d(A,E) + d(C,E) -d(A,C)]/2 = 1.141695
Compute r as in the previous step with N=4. See Fig. 9 for new distance matrix and r vector.

Fig. 9. The new distance matrix D and vector r obtained for NJ algorithm iteration 2.
Now, we compute the modified distance matrix, Md as in the previous step and cluster the
minimally distant OTUs. See Fig. 10

Fig. 10. The modified distance matrix Md, obtained during N-J algorithm iteration 2.
In this step, AC & B and D & E are minimally distant, so we cluster AC with B and D with E.
Repeating the above steps we will finally get the following phylogenetic tree, Fig. 11.
Both the distance-based methods, UPGMA and N-J, are computationally faster and hence
suited for the phylogeny of large datasets. N-J is the most widely used distance-based
method for phylogenetic analysis. The results of these methods are highly dependent on the
model of evolution selected a priori.
Molecular Evolution & Phylogeny: What, When, Why & How? 17

Fig. 11. The phylogenetic tree obtained using N-J algorithm for distance matrix in Fig 4.
Numbers on the branches indicate branch length.

6.2 Character-based methods of phylogeny reconstruction

The most commonly used character-based methods in molecular phylogenetics are
Maximum parsimony and Maximum likelihood. Unlike the distance-based MPA, character-
based methods use character information in alignment data as an input for tree building.
The aligned data is in the form of character-state matrix where the nucleotide or amino acid
symbols represent the states of characters. These character-based methods employ
optimality criterion with the explicit definition of objective function to score the tree
topology in order to infer the optimum tree. Hence, these methods are comparatively slower
than distance-based clustering algorithms, which are simply based on a set of rules and
operations for clustering. But character based methods are advantageous in the sense that
they provide a precise mathematical background to prefer one tree over another unlike in
distance-based clustering algorithms.

6.2.1 Maximum parsimony

The Maximum parsimony (MP) method is based on the simple principle of searching the
tree or collection of trees that minimizes the number of evolutionary changes in the form of
change of one character state into other, which are able to describe observed differences in
the informative sites of OTUs. There are two problems under the parsimony criterion, a)
determining the length of the tree i.e. estimating the number of changes in character states,
b) searching overall possible tree topologies to find the tree that involves minimum number
of changes. Finally all the trees with minimum number of changes are identified for each of
the informative sites. Fitch’s algorithm is used for the calculation of changes for a fixed tree
topology (Fitch, 1971). If the number of OTUs, N is moderate, this algorithm can be used to
calculate the changes for all possible tree topologies and then the most parsimonious rooted
tree with minimum number of changes is inferred. However, if N is very large it becomes
computationally expensive to calculate the changes for the large number of possible rooted
trees. In such cases, a branch and bound algorithm is used to restrict the search space of tree
topologies in accordance with Fitch’s algorithm to arrive at parsimonious tree (Hendy &
Penny, 1982). However, this approach may miss some parsimonious topologies in order to
reduce the search space.
An illustrative example of phylogeny analysis using Maximum parsimony is shown in Table
5 and Fig. 12. Table 5 shows a snapshot of MSA of 4 sequences where 5 columns show the
18 Computational Biology and Applied Bioinformatics

aligned nucleotides. Since there are four taxa (A, B, C & D), three possible unrooted trees
can be obtained for each site. Out of 5 character sites, only two sites, viz., 4 & 5 are
informative i.e. sites having at least two different types of characters (nucleotides/amino
acids) with a minimum frequency 2. In the Maximum parsimony method, only informative
sites are analysed. Fig. 12 shows the Maximum parsimony phylogenetic analysis of site 5
shown in Table 5. Three possible unrooted trees are shown for site 5 and the tree length is
calculated in terms of number of substitutions. Tree II is favoured over trees I and III as it
can explain the observed changes in the sequences just with a single substitution. In the
same way unrooted trees can be obtained for other informative sites such as site 4. The most
parsimonious tree among them will be selected as the final phylogenetic tree. If two or more
trees are found and no unique tree can be inferred, trees are said to be equally
parsimonious.

Table 5. Example of phylogenetic analysis from 5 aligned character sites in 4 OTUs using
Maximum parsimony method.

Fig. 12. Example showing various tree topologies based on site 5 in Table 5 using the
Maximum parsimony method.
This method is suitable for a small number of sequences with higher similarity and was
originally developed for protein sequences. Since this method examines the number of
evolutionary changes in all possible trees it is computationally intensive and time
consuming. Thus, it is not the method of choice for large sized genome sequences with high
variation. The unequal rates of variation in different sites can lead to erroneous parsimony
tree with some branches having longer lengths than others as parsimony method assumes
the rate of change across all sites to be equal.
Molecular Evolution & Phylogeny: What, When, Why & How? 19

6.2.2 Maximum likelihood

As mentioned in the beginning, another character based method for the MPA is the
Maximum likelihood method. This method is based on probabilistic approach to phylogeny.
This approach is different from the methods discussed earlier. In this method probabilistic
models for phylogeny are developed and the tree would be reconstructed using Maximum
likelihood method or by sampling method for the given set of sequences. The main
difference between this method and some of the available methods discussed before is that
it ranks various possible tree topologies according to their likelihood. The same can be
obtained by either using the frequentist approach (using the probability (data|tree)) or by
using the Baysian approach (likelihood based on the posterior probabilities i.e. by using
probability (tree|data)). This method also facilitates computing the likelihood of a sub-tree
topology along the branch.
To make the method operative, one must know how to compute P(x*|T,t*) probability of set
of data given tree topology T and set of branch length t*. The tree having maximum
probability or the one, which maximizes the likelihood would be chosen as the best tree. The
maximization can also be based on the posterior probability P(tree|data) and can be carried
out by obtaining required probability using P(x*|T,t*)=P(data|tree) and by applying the
Baye’s theorem.
The exercise of maximization involves two steps:
a. A search over all possible tree topologies with order of assignment of sequences at the
leaves specified.
b. For each topology, a search over all possible lengths of edges in t*
As mentioned in the chapter earlier, the number of rooted trees for given number of
sequences (N) grows very rapidly even as N increases to 10. An efficient search procedure
for these tasks is required, which was proposed by Felsenstein (1981) and is extensively
being used in the MPA. The maximization of likelihood of edge lengths can be carried out
using various optimization techniques.
An alternative method is to search stochastically over trees by sampling from posterior
distribution P(T,t*|x*). This method uses techniques such as Monte Carlo method, Gibb’s
sampling etc. The results of this method are very promising and are often recommended.
Having briefly reviewed the principles, merits and limitations of various methods available
for reconstruction of phylogenetic trees using molecular data, it becomes evident that the
choice of method for MPA is very crucial. The flowchart shown in Fig. 13 is intended to
serve as a guideline to choose a method based on extent of similarity between the sequences.
However, it is recommended that one uses multiple methods (at least two) to derive the
trees. A few programs have also been developed to superimpose trees to find out
similarities in the branching pattern and tree topologies.

7. Assessing the reliability of phylogenetic tree

The assessment of the reliability of phylogenetic tree is an important part of MPA as it helps
to decide the relationships of OTUs with a certain degree of confidence assigned by
statistical measures. Bootstrap and Jackknife analyses are the major statistical procedures to
evaluate the topology of phylogenetic tree (Efron, 1979; Felsenstein, 1985).
In bootstrap technique, the original aligned dataset of sequences is used to generate the
finite population of pseudo-datasets by “sampling with replacement” protocol. Each
pseudo-dataset is generated by sampling n character sites (columns in the alignment)
20 Computational Biology and Applied Bioinformatics

Fig. 13. Flowchart showing the analysis steps involved in phylogenetic reconstruction.

Fig. 14. The procedure to generate pseudo-replicate datasets of original dataset using
bootstrap is shown above. The character sites are shown in colour codes at the bottom of
datasets to visualize “sampling with replacement protocol”.
randomly from original dataset with a possibility of sampling the same site repeatedly, in
the process of regular bootstrap. This leads to generation of population of datasets, which
are given as an input to tree building methods thus giving rise to population of phylogenetic
Molecular Evolution & Phylogeny: What, When, Why & How? 21

trees. The consensus phylogenetic tree is then inferred by the majority rule that groups those
OTUs, which are found to cluster most of the times in the population of trees. The branches
in consensus phylogenetic tree are labelled with bootstrap support values enabling the
significance of the relationship of OTUs as depicted using a branching pattern. The
procedure for regular bootstrap is illustrated in the Fig. 14. It shows the original dataset
along with four pseudo-replicate datasets.
The sites in the original dataset are colour coded to visualize the “sampling with
replacement protocol” used in generation of pseudo-replicate datasets 1-4. Seqboot program
in PHYLIP package was used for this purpose with choice of regular bootstrap. For
example, pseudo-replicate dataset 1 contains the site 1 (red) from original dataset sampled 3
times. In the general practice, usually 100 to 1000 datasets are generated and for each of the
datasets phylogenetic tree is obtained. The consensus phylogenetic tree is then obtained by
majority rule. The reliability of the consensus tree is assessed from the “branch times” value
displayed along the branches of tree.
In Jackknife procedure, the pseudo-datasets are generated by “sampling without replacement”
protocol. In this process, sampling (<n) character sites randomly from original dataset
generates each pseudo dataset. This leads to generation of population of datasets, which are
given as an input to tree building methods thus giving rise to population of phylogenetic trees.
The consensus phylogenetic tree is inferred by the majority rule that groups those OTUs,
which are found to be clustered most of the times in the population of trees.

8. The case study of Mumps virus phylogeny

We have chosen a case study of Mumps virus (MuV) phylogeny using the amino acid
sequences of surface hydrophobic (SH) proteins. There are 12 different known genotypes of
MuV, which are designated through A to L, based on the sequence similarity of SH gene
sequences. Recently a new genotype of MuV, designated as M, has been identified during
parotitis epidemic 2006-2007 in the state of São Paulo, Brazil (Santos et al., 2008). Extensive
phylogenetic analysis of newly discovered genotype with existing genotypes of reference
strains (A-L) has been used for the confirmation of new genotype using character-based
Maximum likelihood method (Santos et al., 2008). In the case study to be presented here, we
have used distance-based Neighbor-Joining method with an objective to re-confirm the
presence of new MuV genotype M. The dataset reported in Santos et al., (2008) is used for
the re-confirmation analysis. The steps followed in the MPA are listed below.
a. Compilation and curation of sequences: The sequences of SH protein of the strains of
reference genotypes (A to L) as well as newly discovered genotype (M) of MuV were
retrieved using GenBank accession numbers as given in Santos et al., (2008). Sequences
were saved in Fasta format.
b. Multiple sequence alignment (MSA): SH proteins were aligned using ClustalW (See Fig.
2). MSA was saved in Phylip or interleaved (.phy) format.
c. Bootstrap analysis: 100 pseudo-replicate datasets of the original MSA data (obtained in
step b) were generated using regular bootstrap methods in Seqboot program of PHYLIP
package.
d. Derivation of distance: The distances between sequences in each dataset were calculated
using Dayhoff PAM model assuming uniform rate of variation at all sites. The ‘outfile’
generated by Seqboot program was used as an input to Protdist program in PHYLIP
package.
22 Computational Biology and Applied Bioinformatics

Fig. 15. The unrooted consensus phylogenetic tree obtained for Mumps virus genotypes
using Neighbor-Joining method. The first letter in OTU labels indicates the genotype (A-M),
which is followed by the GenBank accession numbers for the sequences. The OTUs are also
colour coded according to genotypes as following, A: red; B: light blue; C: yellow; D: light
green; E: dark blue; F: magenta; G: cyan; H: brick; I: pink; J: orange; K: black; L: dark green;
M: purple. All of the genotypes have formed monophyletic clades with high bootstrap
support values shown along the branches. The monophyletic clade of M genotypes (with 98
bootstrap support at its base) separated from the individual monophyletic clades of other
genotypes (A-L) re-confirms the detection of new genotype M.
e. Building phylogenetic tree: The distance matrices obtained in the previous step were
given as an input to N-J method to build phylogenetic trees. The ‘outfile’ generated by
Protdist program containing distance matrices was given as an input to Neighbor
program in PHYLIP package.
f. The consensus phylogenetic tree was then obtained using Consense program. For this
purpose the ‘outtree’ file (in Newick format) generated by Neighbor program was given
as an input to Consense program.
g. The consensus phylogenetic tree was visualized using FigTree software (available from
http://tree.bio.ed.ac.uk/software/figtree/). The consensus unrooted phylogenetic tree
is shown in Fig. 15.
The phylogenetic tree for the same dataset was also obtained by using Maximum parsimony
method, implemented as the Protpars program in PHYLIP by carrying out MSA and
bootstrap as detailed above. The consensus phylogenetic tree is shown in Fig. 16.
Comparison of the trees shown in Fig. 15 & Fig. 16 with that of the published tree re-
confirms the emergence of new MuV genotype M during the epidemic in São Paulo, Brazil
(Santos et al., 2008), as the members of genotype M have formed a distinct monophyletic
clade similar to the known genotypes (A-L). But, a keen observer would note the differences
in ordering of clades in the two phylograms obtained using two different methods viz., N-J
Molecular Evolution & Phylogeny: What, When, Why & How? 23

and MP. For example, the clade of genotype J is close to the clade of genotype I in the N-J
phylogram (see Fig. 15) whereas in the MP phylogram (Fig. 16) the clade of genotype J is
shown to cluster with the clade of genotype F. Such differences in the ordering of clades are
observed some times as these methods (N-J & MP) employ different assumptions and
models of evolution. The user can interpret the results with reasonable confidence where the
similar clustering pattern of clades is observed in trees drawn using multiple methods. The
user, on the other hand, should refrain from over interpretation of sub-tree topolologies,
where branching order doesn’t match in the trees drawn using different methods. Similarly,
a lot of case studies pertaining to the emergence of new species as well as evolution of
individual genes/proteins have been published. It is advisable to re-run through a few case
studies, which are published, to understand the way in which the respective authors have
interpreted the results on the basis of phylogenetic analyses.

Fig. 16. The unrooted consensus phylogenetic tree obtained for Mumps virus genotypes
using Maximum parsimony method. The labelling of OTUs and colour coding is same as in
Fig. 15.

9. Challenges and opportunities in phylogenomics

The introduction of next-generation sequencing technology has totally revived the pace of
genome sequencing. It has inevitably posed challenges on traditional ways of molecular
phylogeny analysis based on single gene, set of genes or markers. Currently the phylogeny
based on molecular markers such as 16S rRNA, mitochondrial, nuclear genes etc. provide
the taxonomic backbone for Tree of Life (http://tolweb.org/tree/phylogeny.html). But the
single gene based phylogeny does not necessarily reflect the phylogenetic history among the
genomes of organisms from which these genes are derived. Also the types of evolutionary
events such as lateral gene transfer, recombination etc. may not be revealed through the
phylogeny of single gene. Thus whole genome based phylogeny analyses become important
for deeper understanding of the evolutionary pattern in the organisms (Konstantinidis &
24 Computational Biology and Applied Bioinformatics

Tiedje, 2007). But whole genome based phylogeny poses many challenges to the traditional
methods of MPA, major concerns of them being the size, memory and computational
complexity involved in alignment of genomes (Liu et al., 2010).
The methods of MSA developed so far are adequate to handle the requirements of limited
amount of data viz. individual gene or protein sequences from various organisms. The
increased size of data in terms of the whole genome sequences, however, poses constrains
on use and applicability of currently available methods of MSA as they become
computationally intensive with requirement of higher memory. The uncertainty associated
with alignment procedures, which leads to variations in the inferred phylogeny, has also
been pointed out to be the cause of concern (Wong et al., 2008). The benchmark datasets are
made available to validate performance of multiple sequence alignment methods (Kemena
& Notredame, 2009). These challenges have opened up opportunities for development of
alternative approaches for MPA with emergence of alignment-free methods for the same
(Kolekar et al., 2010; Sims et al., 2009; Vinga & Almeida, 2003). The field of MPA is also
evolving with attempts to develop novel methods based on various data mining techniques
viz. Hidden Markov Model (HMM) (Snir & Tuller, 2009), Chaos game theory (Deschavanne
et al., 1999), Return Time Distributions (Kolekar et al., 2010) etc. The recent approaches are
undergoing refinement and will have to be evaluated with the benchmark datasets before
they are routinely used. However, sheer dimensionality of genomic data demands their
application. These approaches along with the conventional approaches are extensively
reviewed elsewhere (Blair & Murphy, 2010; Wong & Nielsen, 2007).

10. Conclusion
The chapter provides excursion of molecular phylogeny analyses for potential users. It gives
an account of available resources and tools. The fundamental principles and salient features
of various methods viz. distance-based and character-based are explained with worked out
examples. The purpose of the chapter will be served if it enables the reader to develop
overall understanding, which is critical to perform such analyses involving real data.

11. Acknowledgment
PSK acknowledges the DBT-BINC junior research fellowship awarded by Department of
Biotechnology (DBT), Government of India. UKK acknowledges infrastructural facilities and
financial support under the Centre of Excellence (CoE) grant of DBT, Government of India.

12. References
Adachi, J. & Hasegawa, M. (1996) Model of amino acid substitution in proteins encoded by
mitochondrial DNA. Journal of Molecular Evolution 42(4):459-468.
Aniba, M.; Poch, O. & Thompson J. (2010) Issues in bioinformatics benchmarking: the case
study of multiple sequence alignment. Nucleic Acids Research 38(21):7353-7363.
Batzoglou, S. (2005) The many faces of sequence alignment. Briefings in Bioinformatics 6(1):6-
22.
Benson, D.; Karsch-Mizrachi, I.; Lipman, D.; Ostell, J. & Sayers, E. (2011) GenBank. Nucleic
Acids Research 39(suppl 1):D32-D37.
Molecular Evolution & Phylogeny: What, When, Why & How? 25

Blair, C. & Murphy, R. (2010) Recent Trends in Molecular Phylogenetic Analysis: Where to
Next? Journal of Heredity 102(1):130.
Bruno, W.; Socci, N. & Halpern A. (2000) Weighted neighbor joining: a likelihood-based
approach to distance-based phylogeny reconstruction. Molecular Biology and
Evolution 17(1):189.
Cavalli-Sforza, L. & Edwards, A. (1967) Phylogenetic analysis. Models and estimation
procedures. American Journal of Human Genetics 19(3 Pt 1):233.
Cole, J.; Wang, Q.; Cardenas, E.; Fish, J.; Chai, B.; Farris, R.; Kulam-Syed-Mohideen, A.;
McGarrell, D.; Marsh, T.; Garrity, G. & others. (2009) The Ribosomal Database
Project: improved alignments and new tools for rRNA analysis. Nucleic Acids
Research 37(suppl 1):D141-D145.
Deschavanne, P.; Giron, A.; Vilain, J.; Fagot, G. & Fertil, B. (1999) Genomic signature:
characterization and classification of species assessed by chaos game representation
of sequences. Mol Biol Evol 16(10):1391-9.
Edgar, R.(2004). MUSCLE: a multiple sequence alignment method with reduced time and
space complexity. BMC Bioinformatics 5:113
Efron, B. (1979) Bootstrap Methods: Another Look at the Jackknife. Ann. Statist. 7:1-26.
Felsenstein, J. (1981) Evolutionary trees from DNA sequences: a maximum likelihood
approach. J. Mol Evol 17:368-376.
Felsenstein, J. (1985) Confidence Limits on Phylogenies: An Approach Using the Bootstrap.
Evolution 39 783-791.
Felsenstein, J.(1989). PHYLIP-phylogeny inference package (version 3.2). Cladistics 5:164-166
Felsenstein, J. (1996) Inferring phylogenies from protein sequences by parsimony, distance,
and likelihood methods. Methods Enzymol 266:418-27.
Fitch, W. (1971) Toward Defining the Course of Evolution: Minimum Change for a Specific
Tree Topology. Systematic Zoology 20(4):406-416.
Gascuel, O. (1997) BIONJ: an improved version of the NJ algorithm based on a simple model
of sequence data. Molecular Biology and Evolution 14(7):685-695.
Hall, T. (1999). BioEdit: A user-friendly biological sequence alignment editor and analysis
program for Windows 95/98/NT. Nucleic Acids Symp Ser 41:95-98.
Hendy, M. & Penny, D. (1982) Branch and bound algorithms to determine minimal
evolutionary trees. Mathematical Biosciences 59(2):277-290.
Henikoff, S. & Henikoff, J. (1992) Amino acid substitution matrices from protein blocks.
Proceedings of the National Academy of Sciences of the United States of America
89(22):10915.
Jones, D.; Taylor, W. & Thornton, J. (1994) A mutation data matrix for transmembrane
proteins. FEBS Letters 339(3):269-275.
Jukes, T. & Cantor, C. (1969) Evolution of protein molecules. In “Mammalian Protein
Metabolism”(HN Munro, Ed.). Academic Press, New York.
Kaminuma, E.; Kosuge, T.; Kodama, Y.; Aono, H.; Mashima, J.; Gojobori, T.; Sugawara, H.;
Ogasawara, O; Takagi, T.; Okubo, K. & others. (2011). DDBJ progress report.
Nucleic Acids Research 39(suppl 1):D22-D27
Katoh, K.; Kuma, K.; Toh, H. & Miyata T (2005) MAFFT version 5: improvement in accuracy
of multiple sequence alignment. Nucleic Acids Res 33:511 - 518
Kemena, C. & Notredame, C. (2009) Upcoming challenges for multiple sequence alignment
methods in the high-throughput era. Bioinformatics 25(19):2455-2465.
26 Computational Biology and Applied Bioinformatics

Kimura, M. (1980) A simple method for estimating evolutionary rates of base substitutions
through comparative studies of nucleotide sequences. Journal of Molecular Evolution
16(2):111-120.
Kolekar, P.; Kale, M. & Kulkarni-Kale, U. (2010) `Inter-Arrival Time' Inspired Algorithm and
its Application in Clustering and Molecular Phylogeny. AIP Conference Proceedings
1298(1):307-312.
Konstantinidis, K. & Tiedje, J. (2007) Prokaryotic taxonomy and phylogeny in the genomic
era: advancements and challenges ahead. Current Opinion in Microbiology 10(5):504-
509.
Kuiken, C.; Leitner, T.; Foley, B.; Hahn, B.; Marx, P.; McCutchan, F.; Wolinsky, S.; Korber, B.;
Bansal, G. & Abfalterer, W. (2009) HIV sequence compendium 2009. Document LA-
UR:06-0680
Kuiken, C.; Yusim, K.; Boykin, L. & Richardson, R. (2005) The Los Alamos hepatitis C
sequence database. Bioinformatics 21(3):379.
Kulkarni-Kale, U.; Bhosle, S.; Manjari, G. & Kolaskar, A. (2004) VirGen: a comprehensive
viral genome resource. Nucleic Acids Research 32(suppl 1):D289.
Kumar, S.; Nei, M.; Dudley, J. & Tamura, K. (2008) MEGA: a biologist-centric software for
evolutionary analysis of DNA and protein sequences. Brief Bioinform, 9(4):299-306.
Leinonen, R.; Akhtar, R.; Birney, E.; Bower, L.; Cerdeno-Tárraga, A.; Cheng, Y.; Cleland, I.;
Faruque, N.; Goodgame, N.; Gibson, R. & others. (2011) The European Nucleotide
Archive. Nucleic Acids Research 39(suppl 1):D28-D31.
Lio, P. & Goldman, N. (1998) Models of molecular evolution and phylogeny. Genome Res
8(12):1233-44.
Liu, K.; Linder, C. & Warnow, T. (2010) Multiple sequence alignment: a major challenge to
large-scale phylogenetics. PLoS Currents 2.
Luo, A.; Qiao, H.; Zhang, Y.; Shi, W.; Ho, S.; Xu, W.; Zhang, A. & Zhu, C. (2010) Performance
of criteria for selecting evolutionary models in phylogenetics: a comprehensive
study based on simulated datasets. BMC Evolutionary Biology 10(1):242.
Morgenstern, B.; French, K.; Dress, A. & Werner, T. (1998). DIALIGN: finding local
similarities by multiple sequence alignment. Bionformatics 14:290 - 294
Needleman, S. & Wunsch, C. (1970) A general method applicable to the search for
similarities in the amino acid sequence of two proteins. Journal of Molecular Biology
48(3):443-453.
Nicholas, H.; Ropelewski, A. & Deerfield DW. (2002) Strategies for multiple sequence
alignment. Biotechniques 32(3):572-591.
Notredame, C.; Higgins, D. & Heringa, J. (2000) T-Coffee: A novel method for fast and
accurate multiple sequence alignment. Journal of Molecular Biology 302:205 - 217
Parry-Smith, D.; Payne, A.; Michie, A.& Attwood, T. (1998). CINEMA--a novel colour
INteractive editor for multiple alignments. Gene 221(1):GC57-GC63
Pearson, W.; Robins, G. & Zhang, T. (1999) Generalized neighbor-joining: more reliable
phylogenetic tree reconstruction. Molecular Biology and Evolution 16(6):806.
Phillips, A. (2006) Homology assessment and molecular sequence alignment. Journal of
Biomedical Informatics 39(1):18-33.
Ronquist, F. & Huelsenbeck, J. (2003). MrBayes 3: Bayesian phylogenetic inference under
mixed models. Bioinformatics 19(12):1572-1574
Molecular Evolution & Phylogeny: What, When, Why & How? 27

Rzhetsky, A. (1995) Estimating substitution rates in ribosomal RNA genes. Genetics

141(2):771.
Saitou, N. & Nei, M. (1987) The neighbor-joining method: a new method for reconstructing
phylogenetic trees. Mol Biol Evol 4(4):406-25.
Santos, C.; Ishida, M.; Foster, P.; Sallum, M.; Benega, M.; Borges, D.; Corrêa, K.; Constantino,
C.; Afzal, M. & Paiva, T. (2008) Detection of a new mumps virus genotype during
parotitis epidemic of 2006–2007 in the State of São Paulo, Brazil. Journal of Medical
Virology 80(2):323-329.
Sayers, E.; Barrett, T.; Benson, D.; Bolton, E.; Bryant, S.; Canese, K.; Chetvernin, V.; Church,
D.; DiCuccio, M.; Federhen, S. & others. (2011). Database resources of the National
Center for Biotechnology Information. Nucleic Acids Research 39(suppl 1):D38-D51.
Schwartz, R. & Dayhoff, M. (1979) Matrices for detecting distant relationships. M. 0. Dayhoff
(ed.), Atlas of protein sequence and structure 5:353-358.
Schmidt, H.; Strimmer, K.; Vingron, M. & Von Haeseler A. (2002). TREE-PUZZLE:
maximum likelihood phylogenetic analysis using quartets and parallel computing.
Bioinformatics 18(3):502.
Sims, G.; Jun, S.; Wu, G. & Kim, S. (2009) Alignment-free genome comparison with feature
frequency profiles (FFP) and optimal resolutions. Proc Natl Acad Sci U S A
106(8):2677-82.
Snir, S. & Tuller, T. (2009) The net-hmm approach: phylogenetic network inference by
combining maximum likelihood and hidden Markov models. Journal of
bioinformatics and computational biology 7(4):625-644.
Sokal, R. & Michener, C. (1958) A statistical method for evaluating systematic relationships.
Univ. Kans. Sci. Bull. 38:1409-1438.
Tamura, K. (1992) Estimation of the number of nucleotide substitutions when there are
strong transition-transversion and G+C-content biases. Molecular Biology and
Evolution 9(4):678-687.
Tamura, K. & Nei, M. (1993) Estimation of the number of nucleotide substitutions in the
control region of mitochondrial DNA in humans and chimpanzees. Molecular
Biology and Evolution 10(3):512-526.
The UniProt Consortium. (2011). Ongoing and future developments at the Universal Protein
Resource. Nucleic Acids Research 39(suppl 1):D214-D219.
Thompson, J.; Higgins, D. & Gibson, T. (1994) CLUSTAL W: improving the sensitivity of
progressive multiple sequence alignment through sequence weighting, position-
specific gap penalties and weight matrix choice. Nucleic Acids Res 22(22):4673-80.
Thompson, J.; Linard, B.; Lecompte, O. & Poch, O. (2011) A Comprehensive Benchmark
Study of Multiple Sequence Alignment Methods: Current Challenges and Future
Perspectives. PLoS ONE 6(3):e18093.
Thorne, J.; Goldman, N. & Jones, D. (1996) Combining protein evolution and secondary
structure. Molecular Biology and Evolution 13(5):666-673.
Vinga, S. & Almeida, J. (2003) Alignment-free sequence comparison-a review. Bioinformatics
19(4):513-23.
Wilgenbusch, J. & Swofford, D. (2003). Inferring Evolutionary Trees with PAUP*. Current
Protocols in Bioinformatics. 6.4.1–6.4.28
Wong, K.; Suchard, M. & Huelsenbeck, J. (2008) Alignment Uncertainty and Genomic
Analysis. Science 319(5862):473-476.
28 Computational Biology and Applied Bioinformatics

Wong, W. & Nielsen, R. (2007) Finding cis-regulatory modules in Drosophila using

phylogenetic hidden Markov models. Bioinformatics 23(16):2031-2037.
Xia, X. & Xie, Z. (2001). DAMBE: software package for data analysis in molecular biology
and evolution. Journal of Heredity 92(4):371
 2

Understanding Protein Function - The Disparity

Between Bioinformatics and Molecular Methods
Katarzyna Hupert-Kocurek1 and Jon M. Kaguni2
1University of Silesia
2Michigan State University
1Poland
2United States of America

1. Introduction
Bioinformatics has its origins in the development of DNA sequencing methods by Alan
Maxam and Walter Gilbert (Maxam and Gilbert, 1977), and by Frederick Sanger and
coworkers (Sanger et al., 1977). By entirely different approaches, the first genomes
determined at the nucleotide sequence level were that of bacteriophage φX174, and the
recombinant plasmid named pBR322 composed of about 5,400 (Sanger et al., 1977), or 4,400
base pairs (Sutcliffe, 1979), respectively. In contrast, two articles that appeared in February
2001 reported on the preliminary DNA sequence of the human genome, which corresponds
to 3 billion nucleotides of DNA sequence information (Lander et al., 2001; Venter et al.,
2001). Only two years later, the GenBank sequence database contained more than 29.3
billion nucleotide bases in greater than 23 million sequences. With the development of new
technologies, experts predict that the cost to sequence an individual’s DNA will be about
$1000. This reduction in cost suggests that efforts in the area of comparative genomics will
increase substantially, leading to an enormous database that vastly exceeds the existing one.
By way of comparative genomics approaches, computational methods have led to the
identification of homologous genes shared among species, and their classification into
superfamilies based on amino acid sequence similarity. In combination with their
evolutionary relatedness, superfamily members have been clustered into clades. In addition,
high throughput sequencing of small RNAs and bioinformatics analyses have contributed to
the identification of regions between genes that can code small RNAs (siRNA, microRNA,
and long noncoding RNA), which act during the development of an organism to modulate
gene expression at the post-transcriptional level (Fire et al., 1998; Hamilton and Baulcombe,
1999) reviewed in Elbashir et al., 2001; Ghildiyal and Zamore, 2009; Christensen et al., 2010).
An emerging area is functional genomics whereby gene function is deduced using large-
scale methods by identifying the involvement of specific genes in metabolic pathways. More
recently, phenotype microarray methods have been used to correlate the functions of genes
of microbes with cell phenotypes under a variety of growth conditions (Bochner, 2009).
These methods contrast with the traditional approach of mapping a gene via the phenotype
of a mutation, and deducing the function of the gene product based on its biochemical
analysis in concert with physiological studies. Such studies have been performed to confirm
the functional importance of conserved residues shared by superfamily members, and also
30 Computational Biology and Applied Bioinformatics

to determine the role of specific residues for a given protein. In comparison, comparative
genomics methods are unable to distinguish if a nonconserved amino acid among
superfamily members is functionally important, or simply reflects sequence divergence due
to the absence of selection during evolution. Without functional information, it is not
possible to determine if a nonconserved amino acid is important.

2. Bioinformatics analysis of AAA+ proteins

On the basis of bioinformatics analysis, P-loop nucleotide hydrolases compose a very large
group of proteins that use an amino acid motif named the phosphate binding loop (P-loop)
to hydrolyze the phosphate ester bonds of nucleotides. A positively charged group in the
side chain of an amino acid (often lysine) in the P-loop promotes nucleotide hydrolysis by
interacting with the phosphate of the bound nucleotide. Additional bioinformatics analysis
of this group of proteins led to a category of nucleotidases containing the Walker A and B
motifs, as well as additional motifs shared by the AAA (ATPases Associated with diverse
cellular Activities) superfamily (Beyer, 1997; Swaffield and Purugganan, 1997). These
diverse activities include protein unfolding and degradation, vesicle transport and
membrane fusion, transcription and DNA replication. The additional motifs of the AAA
superfamily differentiate its members from the larger set of P-loop nucleotidases. Neuwald
et al., and Iyer et al. then integrated structural information with bioinformatics analysis to
classify members of the AAA+ superfamily into clades (Neuwald et al., 1999; Iyer et al.,
2004). These clades are the clamp loader clade, the DnaA/CDC6/ORC clade, the classical
AAA clade, the HslU/ClpX/Lon/ClpAB-C clade, and the Helix-2 insert clade. The last two
clades have been organized into the Pre-sensor 1 hairpin superclade.
Members of the superfamily of AAA+ ATPases carry a nucleotide-binding pocket called the
AAA+ domain that ranges from 200 to 250 amino acids, which is formed by an αβα-type
Rossmann fold followed by several α helices (Figure 1) (Lupas and Martin, 2002; Iyer et al.,
2004; Hanson and Whiteheart, 2005). Such proteins often assemble into ring-shaped or
helical oligomers (Davey et al., 2002; Iyer et al., 2004; Erzberger and Berger, 2006). Using the
nomenclature of Iyer et al., the Rossmann fold is formed by a β sheet of parallel strands
arranged in a β5-β1-β4-β3-β2 series. Its structure resembles a wedge. An α helix preceding
the β1 strand and a loop that is situated across the face of the β sheet is a distinguishing
feature of the AAA+ superfamily. Another characteristic is the position of several α helices
positioned above the wide end of the wedge. The P-loop or the Walker A motif (GX4GKT/S
where X is any amino acid) is located between the β1 strand and the following α helix. The
Walker B motif (φφφφDE where φ is a hydrophobic amino acid) coordinates a magnesium
ion complexed with the nucleoside triphosphate via the conserved aspartate residue. The
conserved glutamate is thought to interact with a water molecule to make it a better
nucleophile for nucleotide hydrolysis.
AAA+ proteins also share conserved motifs named the Sensor 1, Box VII, and Sensor 2
motifs that coordinate ATP hydrolysis with a change in conformation (Figure 1) (Lupas and
Martin, 2002; Iyer et al., 2004; Hanson and Whiteheart, 2005). Relative to the primary amino
acid sequence, these motifs are on the C-terminal side of the Walker B motif. The Sensor 1
motif contains a polar amino acid at the end of the β4 strand. On the basis of the X-ray
crystal structure of N-ethylmaleimide-sensitive factor, an ATPase involved in intracellular
vesicle fusion (Beyer, 1997; Swaffield and Purugganan, 1997), this amino acid together with
Understanding Protein Function -
The Disparity Between Bioinformatics and Molecular Methods 31

the acidic residue in the Walker B motif interacts with and aligns the activated water
molecule during nucleotide hydrolysis. The Box VII motif, which is also called the SRH
(Second Region of Homology) motif, contains an arginine named the arginine finger by its
analogous function with the corresponding arginine of GTPase activator proteins that
interacts with GTP bound to a small G protein partner to promote GTP hydrolysis. The
crystal structures of several AAA+ proteins have shown that the Box VII motif in an
individual molecule is located some distance away from the nucleotide binding pocket. In
AAA+ proteins that assemble into ring-shaped or helical oligomers, the Box VII motif of one
protomer directs an arginine residue responsible for interaction with the γ phosphate of
ATP toward the ATP binding pocket of the neighboring molecule. It is proposed that this
interaction or lack thereof coordinates ATP hydrolysis with a conformational change. The
Sensor 2 motif, which resides in one of the α helices that follow the Rossmann fold, also
contains a conserved arginine. For proteins whose structures contain the bound nucleoside
triphosphate or a nucleotide analogue, this amino acid interacts with the γ phosphate of the
nucleotide. As reviewed by Ogura (Ogura et al., 2004), this residue is involved in ATP
binding or its hydrolysis in some but not all AAA+ proteins. Like the arginine finger
residue, this arginine is thought to coordinate a change in protein conformation with
nucleotide hydrolysis.

Fig. 1. Structural organization of the AAA+ domain, and the locations of the Walker A/P-
loop, Walker B, Sensor 1, Box VII and Sensor 2 motifs are shown (adapted from ref.
(Erzberger and Berger, 2006)).
Because this chapter focuses on members of the DnaA/CDC6/ORC or initiator clade, the
following summarizes properties of this clade and not others. Like the clamp loader clade,
proteins in the initiator clade as represented by DnaA and DnaC have a structure
resembling an open spiral on the basis of X-ray crystallography (Erzberger et al., 2006; Mott
et al., 2008). In comparison, oligomeric proteins in the remaining clades form closed rings. A
characteristic feature of proteins in the initiator clade is the presence of two α helices
between the β2 and β3 strands (Figure 1). Compared with the function of DnaA in the
initiation of E. coli DNA replication, DnaC plays a separate role. Their functions are
32 Computational Biology and Applied Bioinformatics

described in more detail below. The ORC/CDC6 group of eukaryotic proteins in the
initiator clade, like DnaA and DnaC, act to recruit the replicative helicase to replication
origins at the stage of initiation of DNA replication (Lee and Bell, 2000; Liu et al., 2000). The
origin recognition complex (ORC) is composed of six related proteins named Orc1p through
Orc6p, and likely originated along with Cdc6p from a common ancestral gene.
Bioinformatics analysis of DnaC suggests that this protein is a paralog of DnaA, arising by
gene duplication and then diverging with time to perform a separate role from DnaA during
the initiation of DNA replication (Koonin, 1992). This notion leads to the question of what
specific amino acids are responsible for the different functions of DnaA and DnaC despite
the shared presence of the AAA+ amino acid sequence motifs. Presumably, specific amino
acids that are not conserved between these two proteins have critical roles in determining
their different functions, but how are these residues identified and distinguished from those
that are not functionally important? In addition, some amino acids that are conserved
among homologous DnaC proteins, which were identified by multiple sequence alignment
of twenty-eight homologues (Figure 2), are presumable responsible for the unique activities
of DnaC, but what are these unique activities? These issues underscore the limitation of
deducing the biological function of protein by relying only on bioinformatics analysis.

3. Reverse genetics as an approach to identify the function of an unknown

gene
Using various amino acid sequence alignment methods for a particular gene, the postulated
function for this gene remains unknown if amino acid sequence homology is not obtained
relative to a gene of known function. In such cases, the general approach is to employ
reverse genetics to attempt to correlate a phenotype with a mutation in the gene. By way of
comparison, forward genetics begins with a phenotype caused by a specific mutation at an
unknown site in the genome. The approximate position of the gene can be determined by
classical genetic methods that involve its linkage to another mutation that gives rise to a
separate phenotype. Refined linkage mapping can localize the gene of interest, followed by
PCR (polymerase chain reaction) amplification of the region and DNA sequence analysis to
determine the nature of the mutation. As a recent development, whole genome sequencing
has been performed to map mutations, dispensing with the classical method of genetic
linkage mapping (Lupski et al., 2010; Ng and Kirkness, 2010). The DNA sequence obtained
may reveal that the gene and the corresponding gene product have been characterized in the
same or different organism, and disclose its physiological function.
In a reverse genetics approach with a haploid organism, the standard strategy is to
inactivate the gene with the hope that a phenotype can be measured. Inactivation can be
achieved either by deleting the gene or by insertional mutagenesis, usually with a
transposon. As examples, transposon mutagenesis has been performed with numerous
microbial species, and with Caenorhabditis elegans (Vidan and Snyder, 2001; Moerman and
Barstead, 2008; Reznikoff and Winterberg, 2008). Using E. coli or S. cerevisiae as model
organisms for gene disruption, one method relies on replacing most of the gene with a drug
resistance cassette, or a gene that causes a detectable phenotype. The technique of gene
disruption relies on homologous recombination in which the drug resistance gene, for
example, has been joined to DNA sequences that are homologous to the ends of the target
gene (Figure 3). After introduction of this DNA into the cell, recombination between the
ends of the transfected DNA and the homologous regions in the chromosome leads to
Understanding Protein Function -
The Disparity Between Bioinformatics and Molecular Methods 33

Fig. 2. Multiple sequence alignment of dnaC homologues using the Constraint-based

Multiple Alignment Tool reveals amino acids that are highly conserved. The Walker A,
Walker B, Sensor I, Sensor II and Box VII motifs shared among the AAA+ family of ATPases
are shown below the alignment. These motifs are involved in ATP binding, ATP hydrolysis,
and coordinating a conformational change with ATP hydrolysis. The figure also shows the
region of DnaC near the N-terminus that is involved in interaction with DnaB helicase
(Ludlam et al., 2001; Galletto et al., 2003; Mott et al., 2008), and the ISM (Initiator/loader-
Specific Motif), which corresponds to two α helices between the β2 and β3 strands, that
distinguishes members of the DnaA/CDC6/ORC clade from other AAA+ clades. The ISM
causes DnaC assembled as oligomers to form a spiral structure (Mott et al., 2008).
34 Computational Biology and Applied Bioinformatics

replacement of the chromosomal copy of the gene with the drug resistance cassette, after
which the excised copy of the chromosomal gene is lost. In both E. coli and S. cerevisiae, this
approach has been used in seeking to correlate a phenotype with genes of unknown
function, and to identify those that are essential for viability (Winzeler et al., 1999; Baba et
al., 2006). By either gene disruption or transposon mutagenesis, genetic mapping of the
mutation can be performed by inverse PCR where primers complementary to a sequence
near the ends of the drug resistance cassette or the transposon are used. This approach first
involves partially digesting the chromosomal DNA with a restriction enzyme followed by
ligation of the resulting fragments to form a collection of circular DNAs. DNA sequence
analysis of the amplified DNA with the primers described above followed by comparison of
the nucleotide sequence with the genomic DNA sequence can identify the site of the
disrupted gene, or the site of insertion of the transposon.

Fig. 3. DNA recombination between the chromosomal gene and homologous DNA at ends
of the drug resistance gene leads to replacement of the chromosomal copy of the gene with
the drug resistance cassette. The chromosomal gene, and homologous DNA at the ends of
the drug resistance cassette are indicated by the lighter shaded rectangles. The drug
resistance gene is indicated by the darker rectangles. The thin line represents chromosomal
DNA flanking the gene.
With a multicellular organism, a similar strategy that relies on homologous recombination is
used to delete a gene. The type of cell to introduce the deletion is usually an embryonic stem
cell so that the effect of deletion can be measured in the whole organism. Many eukaryotic
organisms have two complete sets of chromosomes. Because the process of homologous
recombination introduces the deletion mutation in one of the two pairs of chromosomes,
yielding a heterozygote, the presence of the wild type copy on the sister chromosome may
conceal the biological effect of the deletion. Thus, the ideal objective is to delete both copies
of a gene in order to measure the associated phenotype. To attempt to obtain an organism in
which both copies of a gene have been “knocked out,” the standard strategy is to mate
heterozygous individuals. By Mendelian genetics, one-fourth of the progeny should carry
the deletion in both copies of the gene. The drawback with the approach of deleting a gene
Understanding Protein Function -
The Disparity Between Bioinformatics and Molecular Methods 35

is that it may be essential for viability as suggested if a homozygous knockout organism

cannot be obtained. Another pitfall is that it may not be possible to construct a heterozygous
knockout because the single wild type copy is insufficient to maintain viability. In either
case, no other hint of gene function is obtained except for the requirement for life.
Another complication with attempting to determine the role of a eukaryotic gene by
deleting it is the existence of gene families where a specific biochemical function is provided
by allelic variants. Hence, to correlate a phenotype by introducing a mutation into a specific
allelic variant requires inactivation of all other members of the family. A further
complication with eukaryotic organisms is that a product formed by an enzyme of interest
in one biochemical pathway may be synthesized via an alternate pathway that involves a
different set of proteins. In these circumstances, deletion of the gene does not yield a
measurable phenotype.
In the event that deletion of a gene is not possible, an alternate approach to characterize the
function of an unknown gene is by RNA interference (reviewed in Carthew and Sontheimer,
2009; Fischer, 2010; Krol et al., 2010). This strategy exploits a natural process that acts to
repress the expression of genes during development, or as cells progress through the cell
cycle (Fire et al., 1998; Ketting et al., 1999; Tabara et al., 1999). Small RNA molecules named
microRNA (miRNA) and small interfering RNA (siRNA) become incorporated into a large
complex called the RNA-inducing silencing complex (RISC), which reduces the expression
of target genes by facilitating the annealing of the RNA with the complementary sequence in
a messenger RNA (Liu et al., 2003). The duplex RNA is recognized by a component of the
RISC complex, followed by degradation of the messenger RNA to block its expression. The
RNA interference pathway has been adapted as a method to reduce or “knockdown” the
expression of a specific gene in order to explore its physiological function. Compared with
other genetic methods that examine the effect of a specific amino acid substitution on a
particular activity of a multifunctional protein, the knockout and knockdown approaches
are not as refined in that they measure the physiological effect of either the reduced
function, or the loss of function of the entire protein.

4. E. coli as a model organism for structure-function studies

Escherichia coli is a rod-shaped bacterium (0.5 micron x 2 microns in the nongrowing state)
that harbors a 4.4 x 106 base pair genome encoding more than 4,000 genes. By transposon-
based insertional mutagenesis and independently by systematic deletion of each open
reading frame, these genes have been separated into those that are essential for viability,
and those that are considered nonessential (Baba et al., 2006). Of the total, about 300 genes
are of undetermined function, including 37 genes that are essential. BLAST analysis
indicates that some of the genes of unknown function are conserved among bacteria,
suggesting their functional importance.
In comparison, many of the genes of known function have been studied extensively. Among
these are the genes required for duplication of the bacterial chromosome, including a subset
that acts at the initiation stage of DNA replication. The following section describes a specific
example that focuses on DnaC protein. Studies on this protein take advantage of
bioinformatics in combination with its X-ray crystallographic structure, molecular genetic
analysis, and the biochemical characterization of specific mutant DnaC proteins to obtain
new insight into its role in DNA replication.
36 Computational Biology and Applied Bioinformatics

5. Molecular analysis of E. coli DnaC, an essential protein involved in the

initiation of DNA replication, and replication fork restart
DNA replication is the basis for life. Occurring only once per cell cycle, DNA replication
must be tightly coordinated with other major cellular processes required for cell growth so
that each progeny cell receives an accurate copy of the genome at cell division (reviewed in
DePamphilis and Bell, 2010). Improper coordination of DNA replication with cell growth
leads to aberrant cell division that causes cell death in severe cases. In addition, the failure to
control the initiation process leads to excessive initiations, followed by the production of
double strand breaks that apparently arise due to head-to-tail fork collisions. In eukaryotes,
aneuploidy and chromosome fusions appear if the broken DNA is not fixed that can lead to
malignant growth.
In bacteria, chromosomal DNA replication starts at a specific locus called oriC (Figure 4).

Fig. 4. Organization of DNA sequence motifs in the E. coli replication origin (oriC). Near the
left border are 13-mer motifs that are unwound by DnaA complexed to ATP. Sites
recognized by DnaA are the DnaA boxes (arrow), I-sites (warped oval), and τ-sites (warped
circle). Sites recognized by IHF (shaded rectangle), Fis (filled rectangle), and DNA adenine
methyltransferase (shaded circle) are also indicated.
Recent reviews describe the independent mechanisms that control the frequency of
initiation from this site (Nielsen and Lobner-Olesen, 2008; Katayama et al., 2010). In
Escherichia coli, the minimal oriC sequence of 245 base pairs contains DNA-binding sites for
many different proteins that either act directly in DNA replication, or modulate the
frequency of this process (reviewed in Leonard and Grimwade, 2009). One of them is DnaA,
which is the initiator of DNA replication, and has been placed in one of the clades of the
AAA+ superfamily via bioinformatics analysis (Koonin, 1992; Erzberger and Berger, 2006).
DnaA binds to a consensus 9 base pair sequence known as the DnaA box. There are five
DnaA boxes individually named R1 through R5 within oriC that are similar in sequence and
are recognized by DnaA (Leonard and Grimwade, 2009). In addition to these sites, DnaA
complexed to ATP specifically recognizes three I- sites and τ-sites in oriC, which leads to the
unwinding of three AT-rich 13-mer repeats located in the left half of oriC. Binding sites are
also present for IHF protein (integration host factor) and FIS protein (factor for inversion
stimulation). As these proteins induce bends in DNA, their apparent ability to modulate the
binding of DnaA to the respective sites in oriC may involve DNA bending. Additionally,
oriC carries 11 GATC sequences recognized by DNA adenine methyltransferase, and sites
Understanding Protein Function -
The Disparity Between Bioinformatics and Molecular Methods 37

recognized by IciA, Rob, and SeqA. The influence of IHF, FIS, IciA, Rob and SeqA proteins
on the initiation process is described in more detail in a review (Leonard and Grimwade,
2005).
At the initiation stage of DNA replication, the first step requires the binding of DnaA
molecules, each complexed to ATP, to the five DnaA boxes, I- and τ- sites of oriC. After
binding, DnaA unwinds the duplex DNA in the AT-rich region to form an intermediate
named the open complex. HU or IHF stimulates the formation of the open complex. In the
next step, the replicative helicase named DnaB becomes stably bound to the separated
strands of the open complex to form an intermediate named the prepriming complex. At
this juncture, DnaC must be complexed to DnaB for a single DnaB hexamer to load onto
each of the separated strands. DnaC protein must then dissociate from the complex in order
for DnaB to be active as a helicase. Following the loading of DnaB, this helicase enlarges the
unwound region of oriC, and then interacts with DnaG primase (Tougu and Marians, 1996).
This interaction between DnaB and DnaG primase, which synthesize primer RNAs that are
extended by DNA polymerase III holoenzyme during semi-conservative DNA replication,
marks the transition between the process of initiation and elongation stage of DNA
replication (McHenry, 2003; Corn and Berger, 2006; Langston et al., 2009). Replication fork
movement that is supported by DnaB helicase and assisted by a second helicase named Rep
proceeds bidirectionally around the chromosome until it reaches the terminus region (Guy
et al., 2009). The two progeny DNAs then segregate near opposite poles of the cell before
septum formation and cell division.
DnaC protein (27 kDa) is essential for cell viability because it is required during the
initiation stage of DNA replication (reviewed in Kornberg and Baker, 1992; Davey and
O'Donnell, 2003). DnaC is also required for DNA replication of the single stranded DNA of
phage φX174, and for many plasmids (e.g. pSC101, P1, R1). DnaC additionally acts to
resurrect collapsed replication forks that appear when a replication fork encounters a nick,
gap, double-stranded break, or modified bases in the parental DNA (Sandler et al., 1996).
This process of restarting a replication fork involves assembly of the replication restart
primosome that contains PriA, PriB, PriC, DnaT, DnaB, DnaC, and Rep protein (Sandler,
2005; Gabbai and Marians, 2010). The major roles of DnaC at oriC, at the replication origins
of the plasmids and bacteriophage described above, or in restarting collapsed replication
forks is to form a complex with DnaB, which is required to escort the helicase onto the DNA,
and then to depart. Since the discovery of the dnaC gene over 40 years ago (Carl, 1970), its
ongoing study by various laboratories using a variety of approaches continue to reveal new
aspects of the molecular mechanisms of DnaC in DNA replication.
Biochemical analysis combined with the X-ray crystallographic structure of the majority of
Aquifex aeolicus DnaC (residues 43 to the C-terminal residue at position 235) reveals that
DnaC protein consists of a smaller N-terminal domain that is responsible for binding to the
C- terminal face of DnaB helicase, and larger ATP-binding region of 190 amino acids (Figure
2; (Ludlam et al., 2001; Galletto et al., 2003; Mott et al., 2008)). Sequence comparison of
homologues of the dnaC gene classifies DnaC as a member of the AAA+ family of ATPases
(Koonin, 1992; Davey et al., 2002; Mott et al., 2008). However unlike other AAA+ proteins,
DnaC contains two additional α helices named the ISM motif (Initiator/loader–Specific
Motif) that directs the oligomerization of this protein into a right-handed helical filament
(Mott et al., 2008). In contrast, the majority AAA+ proteins lacking these α helices assemble
into a closed-ring. Phylogenetic analysis of the AAA+ domain reveals that DnaC is most
38 Computational Biology and Applied Bioinformatics

closely related to DnaA, suggesting that both proteins arose from a common ancestor
(Koonin, 1992). In support, the X-ray crystallographic structures of the ATPase region of
DnaA and DnaC are very similar (Erzberger et al., 2006; Mott et al., 2008).
For DnaC, ATP increases its affinity for single-stranded DNA, which stimulates its ATPase
activity (Davey et al., 2002; Biswas et al., 2004). Other results suggest that ATP stabilizes the
interaction of DnaC with DnaB in the DnaB-DnaC complex (Wahle et al., 1989; Allen and
Kornberg, 1991), which contradicts studies that support the contrary conclusion that ATP is
not necessary for DnaC to form a stable complex with DnaB (Davey et al., 2002; Galletto et
al., 2003; Biswas and Biswas-Fiss, 2006). As mutant DnaC proteins bearing amino acid
substitutions in the Walker A box are both defective in ATP binding and apparently fail to
interact with DnaB, the consequence is that these mutants cannot escort DnaB to oriC
(Ludlam et al., 2001; Davey et al., 2002). Hence, despite the ability of DnaB by itself to bind
to single-stranded DNA in vitro (LeBowitz and McMacken, 1986), DnaC is essential for
DnaB to become stably bound to the unwound region of oriC (Kobori and Kornberg, 1982;
Ludlam et al., 2001). The observation that DnaC complexed to ATP interacts with DnaA
raises the possibility that both proteins act jointly in helicase loading (Mott et al., 2008).
Together, these observations indicate that the ability of DnaC to bind to ATP is essential for
its function in DNA replication, but the paradox about the role of ATP binding and its
hydrolysis on the activity of DnaC and about the mechanism that leads to the dissociation of
DnaC from DnaB have been long-standing issues.

Fig. 5. Plasmid shuffle method. With an E. coli strain lacking the chromosomal copy of dnaC,
a plasmid carrying the wild type dnaC gene residing in the bacterial cell can provide for
dnaC function. Using a plasmid that depends on IPTG for its maintenance, the strain will not
survive in the absence of IPTG. An introduced plasmid that does not depend on IPTG for its
propagation can sustain viability in the absence of IPTG only if it encodes a functional dnaC
allele.
Understanding Protein Function -
The Disparity Between Bioinformatics and Molecular Methods 39

As described above, one of the characteristics of AAA+ proteins is the presence of a

conserved motif named Box VII, which carries a conserved arginine called the “arginine
finger”. Structural studies of other AAA+ proteins have led to the proposal that this arginine
interacts with the γ phosphate of ATP to promote and coordinate ATP hydrolysis with a
conformational change. Recent experiments were performed to examine the role of the
arginine finger of DnaC and to attempt to clarify how ATP binding and its hydrolysis by
DnaC are involved in the process of initiation of DNA replication (Makowska-Grzyska and
Kaguni, 2010). Part of this study relied on an E. coli mutant lacking the chromosomal copy of
the dnaC gene (Hupert-Kocurek et al., 2007). Because the dnaC gene is essential, this
deficiency of the host strain can be complemented by a plasmid encoding the dnaC gene that
depends on IPTG (isopropyl β-D-1-thiogalactopyranoside) for plasmid maintenance (Figure
5). If another plasmid is introduced into the null dnaC mutant, it maintains viability of the
strain in the absence of IPTG only if it carries a functional dnaC allele. In contrast, if the
second plasmid carries an inactivating mutation in dnaC, the host strain cannot survive
without IPTG. This plasmid exchange method showed that an alanine substitution for the
arginine finger residue inactivated DnaC (Makowska-Grzyska and Kaguni, 2010).
Biochemical experiments performed in parallel showed that this conserved arginine plays a
role in the signal transduction process that involves ATP hydrolysis by DnaC that then leads
to the release of DnaC from DnaB. Finally, the interaction of primase with DnaB that is
coupled with primer formation is also apparently necessary for DnaC to dissociate from
DnaB.

6. Conclusions
In summary, the combination of various experimental approaches on the study of DnaC
have led to insightful experiments that expand our understanding of the role of ATP
binding and its hydrolysis by DnaC during the initiation of DNA replication. Evidence
suggests that ATP hydrolysis by DnaC that leads to the dissociation of DnaC from DnaB
helicase is coupled with primer formation that requires an interaction between DnaG
primase and DnaB. Hence, these critical steps are involved in the transition from the process
of initiation to the elongation phase of DNA replication in E. coli.
This example on the molecular mechanism of DnaC protein is a focused study of one
protein and its interaction with other required proteins during the process of initiation of
DNA replication. One may consider this a form of vertical thinking. It contrasts with
bioinformatics approaches that yield large sets of data for proteins based on the DNA
sequences of genomes, and with microarray approaches that, for example, survey the
expression of genes and their regulation at the genome level under different conditions, or
identify interacting partners for a specific protein. The vast wealth of data from these global
approaches provide a different perspective on understanding the functions of sets of genes
or proteins and how they act in a network of biochemical pathways of the cell.

7. Acknowledgements
We thank members of our labs for discussions on the content and organization of this
chapter. This work was supported by a grant GM090063 from the National Institutes of
Health, and the Michigan Agricultural Station to JMK.
40 Computational Biology and Applied Bioinformatics

8. References
Allen, G. J. and A. Kornberg (1991). Fine balance in the regulation of DnaB helicase by DnaC
protein in replication in Escherichia coli. J Biol Chem 266(33): 22096-22101
Baba, T., T. Ara, M. Hasegawa, Y. Takai, Y. Okumura, M. Baba, K. A. Datsenko, M. Tomita,
B. L. Wanner and H. Mori (2006). Construction of Escherichia coli K-12 in-frame,
single-gene knockout mutants: the Keio collection. Mol Syst Biol 2: 1-11
Beyer, A. (1997). Sequence analysis of the AAA protein family. Protein Sci 6(10): 2043-58,
0961-8368 (Print), 0961-8368 (Linking)
Biswas, S. B. and E. E. Biswas-Fiss (2006). Quantitative analysis of binding of single-stranded
DNA by Escherichia coli DnaB helicase and the DnaB-DnaC complex. Biochemistry
45(38): 11505-11513
Biswas, S. B., S. Flowers and E. E. Biswas-Fiss (2004). Quantitative analysis of nucleotide
modulation of DNA binding by the DnaC protein of Escherichia coli. Biochem J 379:
553-562
Bochner, B. R. (2009). Global phenotypic characterization of bacteria. FEMS Microbiol Rev
33(1): 191-205, 0168-6445 (Print), 0168-6445 (Linking)
Carl, P. L. (1970). Escherichia coli mutants with temperature-sensitive synthesis of DNA. Mol
Gen Genet 109(2): 107-122
Carthew, R. W. and E. J. Sontheimer (2009). Origins and Mechanisms of miRNAs and
siRNAs. Cell 136(4): 642-55, 1097-4172 (Electronic), 0092-8674 (Linking)
Christensen, N. M., K. J. Oparka and J. Tilsner (2010). Advances in imaging RNA in plants.
Trends Plant Sci 15(4): 196-203, 1878-4372 (Electronic), 1360-1385 (Linking)
Corn, J. E. and J. M. Berger (2006). Regulation of bacterial priming and daughter strand
synthesis through helicase-primase interactions. Nucleic Acids Res 34(15): 4082-
4088
Davey, M. J., L. Fang, P. McInerney, R. E. Georgescu and M. O'Donnell (2002). The DnaC
helicase loader is a dual ATP/ADP switch protein. EMBO J 21(12): 3148-3159
Davey, M. J., D. Jeruzalmi, J. Kuriyan and M. O'Donnell (2002). Motors and switches: AAA+
machines within the replisome. Nat Rev Mol Cell Biol 3(11): 826-835
Davey, M. J. and M. O'Donnell (2003). Replicative helicase loaders: ring breakers and ring
makers. Curr Biol 13(15): R594-596
DePamphilis, M. L. and S. D. Bell (2010). Genome duplication, Garland Science/Taylor &
Francis Group, 9780415442060, London
Elbashir, S. M., J. Harborth, W. Lendeckel, A. Yalcin, K. Weber and T. Tuschl (2001).
Duplexes of 21-nucleotide RNAs mediate RNA interference in cultured mammalian
cells. Nature 411(6836): 494-8, 0028-0836 (Print), 0028-0836 (Linking)
Erzberger, J. P. and J. M. Berger (2006). Evolutionary relationships and
structural mechanisms of AAA+ proteins. Annu Rev Biophys Biomol Struct 35: 93-
114
Erzberger, J. P., M. L. Mott and J. M. Berger (2006). Structural basis for ATP-dependent
DnaA assembly and replication-origin remodeling. Nat Struct Mol Biol 13(8): 676-
683
Understanding Protein Function -
The Disparity Between Bioinformatics and Molecular Methods 41

Fire, A., S. Xu, M. K. Montgomery, S. A. Kostas, S. E. Driver and C. C. Mello (1998). Potent
and specific genetic interference by double-stranded RNA in Caenorhabditis elegans.
Nature 391(6669): 806-11, 0028-0836 (Print), 0028-0836 (Linking)
Fischer, S. E. (2010). Small RNA-mediated gene silencing pathways in C. elegans. Int J
Biochem Cell Biol 42(8): 1306-15, 1878-5875 (Electronic), 1357-2725 (Linking)
Gabbai, C. B. and K. J. Marians (2010). Recruitment to stalled replication forks of the PriA
DNA helicase and replisome-loading activities is essential for survival. DNA
Repair (Amst) 9(3): 202-209, 1568-7856 (Electronic), 1568-7856 (Linking)
Galletto, R., M. J. Jezewska and W. Bujalowski (2003). Interactions of the Escherichia coli
DnaB Helicase Hexamer with the Replication Factor the DnaC Protein. Effect of
Nucleotide Cofactors and the ssDNA on Protein-Protein Interactions and the
Topology of the Complex. J Mol Biol 329(3): 441-465
Ghildiyal, M. and P. D. Zamore (2009). Small silencing RNAs: an expanding universe. Nat
Rev Genet 10(2): 94-108, 1471-0064 (Electronic), 1471-0056 (Linking)
Guy, C. P., J. Atkinson, M. K. Gupta, A. A. Mahdi, E. J. Gwynn, C. J. Rudolph, P. B. Moon, I.
C. van Knippenberg, C. J. Cadman, M. S. Dillingham, R. G. Lloyd and P. McGlynn
(2009). Rep provides a second motor at the replisome to promote duplication of
protein-bound DNA. Mol Cell 36(4): 654-666, 1097-4164 (Electronic), 1097-2765
(Linking)
Hamilton, A. J. and D. C. Baulcombe (1999). A species of small antisense RNA in
posttranscriptional gene silencing in plants. Science 286(5441): 950-952, 0036-8075
(Print), 0036-8075 (Linking)
Hanson, P. I. and S. W. Whiteheart (2005). AAA+ proteins: have engine, will work. Nat Rev
Mol Cell Biol 6(7): 519-529
Hupert-Kocurek, K., J. M. Sage, M. Makowska-Grzyska and J. M. Kaguni (2007). Genetic
method to analyze essential genes of Escherichia coli. Appl Environ Microbiol 73(21):
7075-7082
Iyer, L. M., D. D. Leipe, E. V. Koonin and L. Aravind (2004). Evolutionary history and higher
order classification of AAA+ ATPases. J Struct Biol 146(1-2): 11-31, 1047-8477
(Print), 1047-8477 (Linking)
Katayama, T., S. Ozaki, K. Keyamura and K. Fujimitsu (2010). Regulation of the replication
cycle: conserved and diverse regulatory systems for DnaA and oriC. Nat Rev
Microbiol 8(3): 163-170, 1740-1534 (Electronic), 1740-1526 (Linking)
Ketting, R. F., T. H. Haverkamp, H. G. van Luenen and R. H. Plasterk (1999). Mut-7 of C.
elegans, required for transposon silencing and RNA interference, is a homolog of
Werner syndrome helicase and RNaseD. Cell 99(2): 133-141, 0092-8674 (Print), 0092-
8674 (Linking)
Kobori, J. A. and A. Kornberg (1982). The Escherichia coli dnaC gene product. II.
Purification, physical properties, and role in replication. J Biol Chem 257(22):
13763-13769
Koonin, E. V. (1992). DnaC protein contains a modified ATP-binding motif and belongs
to a novel family of ATPases including also DnaA. Nucleic Acids Res 20(8):
1997
42 Computational Biology and Applied Bioinformatics

Kornberg, A. and T. A. Baker (1992). DNA Replication Second Edition, W.H. Freeman and
Company, 9781891389443, New York
Krol, J., I. Loedige and W. Filipowicz (2010). The widespread regulation of microRNA
biogenesis, function and decay. Nat Rev Genet 11(9): 597-610, 1471-0064
(Electronic), 1471-0056 (Linking)
Lander, E. S., L. M. Linton, B. Birren, C. Nusbaum, M. C. Zody, J. Baldwin, K. Devon, K.
Dewar, M. Doyle, W. FitzHugh, R. Funke, D. Gage, K. Harris, A. Heaford, J.
Howland, L. Kann, J. Lehoczky, R. LeVine, P. McEwan, K. McKernan, J. Meldrim, J.
P. Mesirov, C. Miranda, W. Morris, J. Naylor, C. Raymond, M. Rosetti, R. Santos, A.
Sheridan, C. Sougnez, N. Stange-Thomann, N. Stojanovic, A. Subramanian, D.
Wyman, J. Rogers, J. Sulston, R. Ainscough, S. Beck, D. Bentley, J. Burton, C. Clee,
N. Carter, A. Coulson, R. Deadman, P. Deloukas, A. Dunham, I. Dunham, R.
Durbin, L. French, D. Grafham, S. Gregory, T. Hubbard, S. Humphray, A. Hunt, M.
Jones, C. Lloyd, A. McMurray, L. Matthews, S. Mercer, S. Milne, J. C. Mullikin, A.
Mungall, R. Plumb, M. Ross, R. Shownkeen, S. Sims, R. H. Waterston, R. K. Wilson,
L. W. Hillier, J. D. McPherson, M. A. Marra, E. R. Mardis, L. A. Fulton, A. T.
Chinwalla, K. H. Pepin, W. R. Gish, S. L. Chissoe, M. C. Wendl, K. D. Delehaunty,
T. L. Miner, A. Delehaunty, J. B. Kramer, L. L. Cook, R. S. Fulton, D. L. Johnson, P. J.
Minx, S. W. Clifton, T. Hawkins, E. Branscomb, P. Predki, P. Richardson, S.
Wenning, T. Slezak, N. Doggett, J. F. Cheng, A. Olsen, S. Lucas, C. Elkin, E.
Uberbacher, M. Frazier, R. A. Gibbs, D. M. Muzny, S. E. Scherer, J. B. Bouck, E. J.
Sodergren, K. C. Worley, C. M. Rives, J. H. Gorrell, M. L. Metzker, S. L. Naylor, R.
S. Kucherlapati, D. L. Nelson, G. M. Weinstock, Y. Sakaki, A. Fujiyama, M. Hattori,
T. Yada, A. Toyoda, T. Itoh, C. Kawagoe, H. Watanabe, Y. Totoki, T. Taylor, J.
Weissenbach, R. Heilig, W. Saurin, F. Artiguenave, P. Brottier, T. Bruls, E. Pelletier,
C. Robert, P. Wincker, D. R. Smith, L. Doucette-Stamm, M. Rubenfield, K.
Weinstock, H. M. Lee, J. Dubois, A. Rosenthal, M. Platzer, G. Nyakatura, S.
Taudien, A. Rump, H. Yang, J. Yu, J. Wang, G. Huang, J. Gu, L. Hood, L. Rowen, A.
Madan, S. Qin, R. W. Davis, N. A. Federspiel, A. P. Abola, M. J. Proctor, R. M.
Myers, J. Schmutz, M. Dickson, J. Grimwood, D. R. Cox, M. V. Olson, R. Kaul, N.
Shimizu, K. Kawasaki, S. Minoshima, G. A. Evans, M. Athanasiou, R. Schultz, B. A.
Roe, F. Chen, H. Pan, J. Ramser, H. Lehrach, R. Reinhardt, W. R. McCombie, M. de
la Bastide, N. Dedhia, H. Blocker, K. Hornischer, G. Nordsiek, R. Agarwala, L.
Aravind, J. A. Bailey, A. Bateman, S. Batzoglou, E. Birney, P. Bork, D. G. Brown, C.
B. Burge, L. Cerutti, H. C. Chen, D. Church, M. Clamp, R. R. Copley, T. Doerks, S.
R. Eddy, E. E. Eichler, T. S. Furey, J. Galagan, J. G. Gilbert, C. Harmon, Y.
Hayashizaki, D. Haussler, H. Hermjakob, K. Hokamp, W. Jang, L. S. Johnson, T. A.
Jones, S. Kasif, A. Kaspryzk, S. Kennedy, W. J. Kent, P. Kitts, E. V. Koonin, I. Korf,
D. Kulp, D. Lancet, T. M. Lowe, A. McLysaght, T. Mikkelsen, J. V. Moran, N.
Mulder, V. J. Pollara, C. P. Ponting, G. Schuler, J. Schultz, G. Slater, A. F. Smit, E.
Stupka, J. Szustakowski, D. Thierry-Mieg, J. Thierry-Mieg, L. Wagner, J. Wallis, R.
Wheeler, A. Williams, Y. I. Wolf, K. H. Wolfe, S. P. Yang, R. F. Yeh, F. Collins, M. S.
Guyer, J. Peterson, A. Felsenfeld, K. A. Wetterstrand, A. Patrinos, M. J. Morgan, P.
de Jong, J. J. Catanese, K. Osoegawa, H. Shizuya, S. Choi and Y. J. Chen (2001).
Understanding Protein Function -
The Disparity Between Bioinformatics and Molecular Methods 43

Initial sequencing and analysis of the human genome. Nature 409(6822): 860-921,
0028-0836 (Print), 0028-0836 (Linking)
Langston, L. D., C. Indiani and M. O'Donnell (2009). Whither the replisome: emerging
perspectives on the dynamic nature of the DNA replication machinery. Cell Cycle
8(17): 2686-2691, 1551-4005 (Electronic), 1551-4005 (Linking)
LeBowitz, J. H. and R. McMacken (1986). The Escherichia coli dnaB replication protein is a
DNA helicase. J Biol Chem 261(10): 4738-48,
Lee, D. G. and S. P. Bell (2000). ATPase switches controlling DNA replication initiation. Curr
Opin Cell Biol 12(3): 280-285, 0955-0674 (Print), 0955-0674 (Linking)
Leonard, A. C. and J. E. Grimwade (2005). Building a bacterial orisome: emergence of
new regulatory features for replication origin unwinding. Mol Microbiol 55(4): 978-
985
Leonard, A. C. and J. E. Grimwade (2009). Initiating chromosome replication in E. coli: it
makes sense to recycle. Genes Dev 23(10): 1145-1150,
Liu, J., C. L. Smith, D. DeRyckere, K. DeAngelis, G. S. Martin and J. M. Berger (2000).
Structure and function of Cdc6/Cdc18: implications for origin recognition and
checkpoint control. Mol Cell 6(3): 637-648
Liu, Q., T. A. Rand, S. Kalidas, F. Du, H. E. Kim, D. P. Smith and X. Wang (2003). R2D2, a
bridge between the initiation and effector steps of the Drosophila RNAi pathway.
Science 301(5641): 1921-1925, 1095-9203 (Electronic), 0036-8075 (Linking)
Ludlam, A. V., M. W. McNatt, K. M. Carr and J. M. Kaguni (2001). Essential amino acids of
Escherichia coli DnaC protein in an N-terminal domain interact with DnaB helicase. J
Biol Chem 276(29): 27345-27353
Lupas, A. N. and J. Martin (2002). AAA proteins. Curr Opin Struct Biol 12(6): 746-53, 0959-
440X (Print), 0959-440X (Linking)
Lupski, J. R., J. G. Reid, C. Gonzaga-Jauregui, D. Rio Deiros, D. C. Chen, L. Nazareth, M.
Bainbridge, H. Dinh, C. Jing, D. A. Wheeler, A. L. McGuire, F. Zhang, P.
Stankiewicz, J. J. Halperin, C. Yang, C. Gehman, D. Guo, R. K. Irikat, W. Tom, N. J.
Fantin, D. M. Muzny and R. A. Gibbs (2010). Whole-genome sequencing in a
patient with Charcot-Marie-Tooth neuropathy. N Engl J Med 362(13): 1181-91, 1533-
4406 (Electronic), 0028-4793 (Linking)
Makowska-Grzyska, M. and J. M. Kaguni (2010). Primase Directs the Release of DnaC from
DnaB. Mol Cell 37(1): 90-101
Maxam, A. M. and W. Gilbert (1977). A new method for sequencing DNA. Proc Natl Acad
Sci U S A 74(2): 560-4, 0027-8424 (Print), 0027-8424 (Linking)
McHenry, C. S. (2003). Chromosomal replicases as asymmetric dimers: studies of subunit
arrangement and functional consequences. Mol Microbiol 49(5): 1157-1165
Moerman, D. G. and R. J. Barstead (2008). Towards a mutation in every gene in
Caenorhabditis elegans. Brief Funct Genomic Proteomic 7(3): 195-204, 1477-4062
(Electronic), 1473-9550 (Linking)
Mott, M. L., J. P. Erzberger, M. M. Coons and J. M. Berger (2008). Structural synergy and
molecular crosstalk between bacterial helicase loaders and replication initiators.
Cell 135(4): 623-634
44 Computational Biology and Applied Bioinformatics

Neuwald, A. F., L. Aravind, J. L. Spouge and E. V. Koonin (1999). AAA+: A class of

chaperone-like ATPases associated with the assembly, operation, and disassembly
of protein complexes. Genome Res 9(1): 27-43
Ng, P. C. and E. F. Kirkness (2010). Whole genome sequencing. Methods Mol Biol 628: 215-
226, 1940-6029 (Electronic), 1064-3745 (Linking)
Nielsen, O. and A. Lobner-Olesen (2008). Once in a lifetime: strategies for preventing re-
replication in prokaryotic and eukaryotic cells. EMBO Rep 9(2): 151-156
Ogura, T., S. W. Whiteheart and A. J. Wilkinson (2004). Conserved arginine residues
implicated in ATP hydrolysis, nucleotide-sensing, and inter-subunit interactions in
AAA and AAA+ ATPases. J Struct Biol 146(1-2): 106-112
Reznikoff, W. S. and K. M. Winterberg (2008). Transposon-based strategies for the
identification of essential bacterial genes. Methods Mol Biol 416: 13-26, 1064-3745
(Print), 1064-3745 (Linking)
Sandler, S. J. (2005). Requirements for replication restart proteins during constitutive stable
DNA replication in Escherichia coli K-12. Genetics 169(4): 1799-1806
Sandler, S. J., H. S. Samra and A. J. Clark (1996). Differential suppression of priA2::kan
phenotypes in Escherichia coli K- 12 by mutations in priA, lexA, and dnaC. Genetics
143(1): 5-13
Sanger, F., G. M. Air, B. G. Barrell, N. L. Brown, A. R. Coulson, C. A. Fiddes, C. A.
Hutchison, P. M. Slocombe and M. Smith (1977). Nucleotide sequence of
bacteriophage phi X174 DNA. Nature 265(5596): 687-95, 0028-0836 (Print), 0028-
0836 (Linking)
Sanger, F., S. Nicklen and A. R. Coulson (1977). DNA sequencing with chain-terminating
inhibitors. Proc Natl Acad Sci U S A 74(12): 5463-7, 0027-8424 (Print), 0027-8424
(Linking)
Sutcliffe, J. G. (1979). Complete nucleotide sequence of the Escherichia coli plasmid pBR322.
Cold Spring Harb Symp Quant Biol 43 Pt 1: 77-90, 0091-7451 (Print), 0091-7451
(Linking)
Swaffield, J. C. and M. D. Purugganan (1997). The evolution of the conserved ATPase
domain (CAD): reconstructing the history of an ancient protein module. J Mol Evol
45(5): 549-563, 0022-2844 (Print), 0022-2844 (Linking)
Tabara, H., M. Sarkissian, W. G. Kelly, J. Fleenor, A. Grishok, L. Timmons, A. Fire and C. C.
Mello (1999). The rde-1 gene, RNA interference, and transposon silencing in C.
elegans. Cell 99(2): 123-32, 0092-8674 (Print), 0092-8674 (Linking)
Tougu, K. and K. J. Marians (1996). The interaction between helicase and primase sets the
replication fork clock. J Biol Chem 271(35): 21398-405,
Venter, J. C., M. D. Adams, E. W. Myers, P. W. Li, R. J. Mural, G. G. Sutton, H. O. Smith, M.
Yandell, C. A. Evans, R. A. Holt, J. D. Gocayne, P. Amanatides, R. M. Ballew, D. H.
Huson, J. R. Wortman, Q. Zhang, C. D. Kodira, X. H. Zheng, L. Chen, M. Skupski,
G. Subramanian, P. D. Thomas, J. Zhang, G. L. Gabor Miklos, C. Nelson, S. Broder,
A. G. Clark, J. Nadeau, V. A. McKusick, N. Zinder, A. J. Levine, R. J. Roberts, M.
Simon, C. Slayman, M. Hunkapiller, R. Bolanos, A. Delcher, I. Dew, D. Fasulo, M.
Flanigan, L. Florea, A. Halpern, S. Hannenhalli, S. Kravitz, S. Levy, C. Mobarry, K.
Reinert, K. Remington, J. Abu-Threideh, E. Beasley, K. Biddick, V. Bonazzi, R.
Understanding Protein Function -
The Disparity Between Bioinformatics and Molecular Methods 45

Brandon, M. Cargill, I. Chandramouliswaran, R. Charlab, K. Chaturvedi, Z. Deng,

V. Di Francesco, P. Dunn, K. Eilbeck, C. Evangelista, A. E. Gabrielian, W. Gan, W.
Ge, F. Gong, Z. Gu, P. Guan, T. J. Heiman, M. E. Higgins, R. R. Ji, Z. Ke, K. A.
Ketchum, Z. Lai, Y. Lei, Z. Li, J. Li, Y. Liang, X. Lin, F. Lu, G. V. Merkulov, N.
Milshina, H. M. Moore, A. K. Naik, V. A. Narayan, B. Neelam, D. Nusskern, D. B.
Rusch, S. Salzberg, W. Shao, B. Shue, J. Sun, Z. Wang, A. Wang, X. Wang, J. Wang,
M. Wei, R. Wides, C. Xiao, C. Yan, A. Yao, J. Ye, M. Zhan, W. Zhang, H. Zhang, Q.
Zhao, L. Zheng, F. Zhong, W. Zhong, S. Zhu, S. Zhao, D. Gilbert, S. Baumhueter, G.
Spier, C. Carter, A. Cravchik, T. Woodage, F. Ali, H. An, A. Awe, D. Baldwin, H.
Baden, M. Barnstead, I. Barrow, K. Beeson, D. Busam, A. Carver, A. Center, M. L.
Cheng, L. Curry, S. Danaher, L. Davenport, R. Desilets, S. Dietz, K. Dodson, L.
Doup, S. Ferriera, N. Garg, A. Gluecksmann, B. Hart, J. Haynes, C. Haynes, C.
Heiner, S. Hladun, D. Hostin, J. Houck, T. Howland, C. Ibegwam, J. Johnson, F.
Kalush, L. Kline, S. Koduru, A. Love, F. Mann, D. May, S. McCawley, T. McIntosh,
I. McMullen, M. Moy, L. Moy, B. Murphy, K. Nelson, C. Pfannkoch, E. Pratts, V.
Puri, H. Qureshi, M. Reardon, R. Rodriguez, Y. H. Rogers, D. Romblad, B. Ruhfel,
R. Scott, C. Sitter, M. Smallwood, E. Stewart, R. Strong, E. Suh, R. Thomas, N. N.
Tint, S. Tse, C. Vech, G. Wang, J. Wetter, S. Williams, M. Williams, S. Windsor, E.
Winn-Deen, K. Wolfe, J. Zaveri, K. Zaveri, J. F. Abril, R. Guigo, M. J. Campbell, K.
V. Sjolander, B. Karlak, A. Kejariwal, H. Mi, B. Lazareva, T. Hatton, A. Narechania,
K. Diemer, A. Muruganujan, N. Guo, S. Sato, V. Bafna, S. Istrail, R. Lippert, R.
Schwartz, B. Walenz, S. Yooseph, D. Allen, A. Basu, J. Baxendale, L. Blick, M.
Caminha, J. Carnes-Stine, P. Caulk, Y. H. Chiang, M. Coyne, C. Dahlke, A. Mays,
M. Dombroski, M. Donnelly, D. Ely, S. Esparham, C. Fosler, H. Gire, S. Glanowski,
K. Glasser, A. Glodek, M. Gorokhov, K. Graham, B. Gropman, M. Harris, J. Heil, S.
Henderson, J. Hoover, D. Jennings, C. Jordan, J. Jordan, J. Kasha, L. Kagan, C. Kraft,
A. Levitsky, M. Lewis, X. Liu, J. Lopez, D. Ma, W. Majoros, J. McDaniel, S. Murphy,
M. Newman, T. Nguyen, N. Nguyen, M. Nodell, S. Pan, J. Peck, M. Peterson, W.
Rowe, R. Sanders, J. Scott, M. Simpson, T. Smith, A. Sprague, T. Stockwell, R.
Turner, E. Venter, M. Wang, M. Wen, D. Wu, M. Wu, A. Xia, A. Zandieh and X.
Zhu (2001). The sequence of the human genome. Science 291(5507): 1304-1351, 0036-
8075 (Print), 0036-8075 (Linking)
Vidan, S. and M. Snyder (2001). Large-scale mutagenesis: yeast genetics in the genome era.
Curr Opin Biotechnol 12(1): 28-34, 0958-1669 (Print), 0958-1669 (Linking)
Wahle, E., R. S. Lasken and A. Kornberg (1989). The dnaB-dnaC replication protein
complex of Escherichia coli. I. Formation and properties. J Biol Chem 264(5): 2463-
2468
Winzeler, E. A., D. D. Shoemaker, A. Astromoff, H. Liang, K. Anderson, B. Andre, R.
Bangham, R. Benito, J. D. Boeke, H. Bussey, A. M. Chu, C. Connelly, K. Davis, F.
Dietrich, S. W. Dow, M. El Bakkoury, F. Foury, S. H. Friend, E. Gentalen, G.
Giaever, J. H. Hegemann, T. Jones, M. Laub, H. Liao, N. Liebundguth, D. J.
Lockhart, A. Lucau-Danila, M. Lussier, N. M'Rabet, P. Menard, M. Mittmann, C.
Pai, C. Rebischung, J. L. Revuelta, L. Riles, C. J. Roberts, P. Ross-MacDonald, B.
Scherens, M. Snyder, S. Sookhai-Mahadeo, R. K. Storms, S. Veronneau, M. Voet, G.
46 Computational Biology and Applied Bioinformatics

Volckaert, T. R. Ward, R. Wysocki, G. S. Yen, K. Yu, K. Zimmermann, P. Philippsen,

M. Johnston and R. W. Davis (1999). Functional characterization of the S. cerevisiae
genome by gene deletion and parallel analysis. Science 285(5429): 901-906, 0036-
8075 (Print), 0036-8075 (Linking)
 3

In Silico Identification of Regulatory

Elements in Promoters
Vikrant Nain1, Shakti Sahi1 and Polumetla Ananda Kumar2
1Gautam Buddha University, Greater Noida
2National Research Centre on Plant Biotechnology, New Delhi
India

1. Introduction
In multi-cellular organisms development from zygote to adult and adaptation to different
environmental stresses occur as cells acquire specialized roles by synthesizing proteins
necessary for each task. In eukaryotes the most commonly used mechanism for maintaining
cellular protein environment is transcriptional regulation of gene expression, by recruiting
required transcription factors at promoter regions. Owing to the importance of
transcriptional regulation, one of the main goals in the post-genomic era is to predict gene
expression regulation on the basis of presence of transcription factor (TF) binding sites in the
promoter regions. Genome wide knowledge of TF binding sites would be useful to build
transcriptional regulatory networks model that result in cell specific differentiation. In
eukaryotic genomes only a fraction (< 5%) of total genome codes for functional proteins or
RNA, while remaining DNA sequences consist of non-coding regulatory sequences, other
regions and sequences still with unknown functions.
Since the discovery of trans-acting factors in gene regulation by Jacob and Monads in lac
operon of E. coli, scientists had an interest in finding new transcription factors, their specific
recognition and binding sequences. In DNAse footprinting (or DNase protection assay);
transcription factor bound regions are protected from DNAse digestion, creating a
"footprint" in a sequencing gel. This methodology has resulted in identification of hundreds
of regulatory sequences. However, limitation of this methodology is that it requires the TF
and promoter sequence (100-300 bp) in purified form. Our knowledge of known
transcription factors is limited and recognition and binding sites are scattered over the
complete genome. Therefore, in spite of high degree of accuracy in prediction of TF binding
site, this methodology is not suitable for genome wide or across the genomes scanning.
Detection of TF binding sites through phylogenetic footprinting is gradually becoming
popular. It is based on the fact that random mutations are not easily accepted in functional
sequences, while they continuously keep on tinkering non functional sequences. Many
comparative genomics studies have revealed that during course of evolution regulatory
elements remain conserved while the non-coding DNA sequences keep on mutating. With
an ever increasing number of complete genome sequence from multiple organisms and
mRNA profiling through microarray and deep sequencing technologies, wealth of gene
expression data is being generated. This data can be used for identification of regulatory
48 Computational Biology and Applied Bioinformatics

elements through intra and inter species comparative genomics. However, the identification
of TF binding sites in promoters still remains one of the major challenges in bioinformatics
due to following reasons:
1. Very short (5-15 nt) size of regulatory motifs that also differ in their number of
occurrence and position on DNA strands with respect to transcription start site. This
wide distribution of short TF binding sites makes their identification with commonly
used sequence alignment programmes challenging.
2. A very high degree of conservation between two closely related species generally
shows no clear signature of highly conserved motifs.
3. Absence of significant similarities between highly diverse species hinders the alignment
of functional sequences.
4. Sometimes, functional conservation of gene expression is not sufficient to assure the
evolutionary preservation of corresponding cis-regulatory elements (Pennacchio and
Rubin, 2001).
5. Transcription factors binding sites are often degenerate.
In order to overcome these challenges, in the last few years novel approaches have been
developed that integrate comparative, structural, and functional genomics with the
computational algorithms. Such interdisciplinary efforts have increased the sensitivity of
computational programs to find composite regulatory elements.
Here, we review different computational approaches for identification of regulatory
elements in promoter region with seed specific legumin gene promoter analysis as an
example. Based on the type of DNA sequence information the motif finding algorithms are
classified into three major classes: (1) methods that use promoter sequences from co
regulated genes from a single genome, (2) methods that use orthologous promoter
sequences of a single gene from multiple species, also known as phylogenetic footprinting
and (3) methods that use promoter sequences of co regulated genes as well as phylogenetic
footprinting (Das and Dai, 2007).

2. Representation of DNA motifs

In order to discover motifs of unknown transcription factors, models to represent motifs are
essential (Stormo, 2000).There are three models which are generally used to describe a motif
and its binding sites:
1. string representation (Buhler and Tompa, 2002)
2. matrix representation (Bailey and Elkan, 1994) and
3. representation with nucleotide dependency (Chin and Leung, 2008)

2.1 String representation

String representation is the basic representation using string of symbols or nucleotides A, C,
G and T of length-l to describe a motif. Wildcard symbols are introduced into the string to
represent choice from a subset of symbols at a particular position. The International Union
of Pure and Applied Chemistry (IUPAC) nucleic acid codes (Thakurta and Stormo, 2007) are
used to represent the information about degeneracy for example: W = A or T (‘Weak’ base
pairing); S= C or G (‘Strong’ base pairing); R= A or G (Purine); Y= C or T (Pyrimidine); K= G
or T (Keto group on base); M= A or C (Amino group on base); B= C, G, or T; D= A, G, or T ;
H= A, C, or T ; V= A, C, or G; N= A, C, G, or T.
In Silico Identification of Regulatory Elements in Promoters 49

2.2 Matrix representation

In matrix representation, motifs of length l are represented by position weight matrices
(PWMs) or position specific scoring matrices (PSSMs) of size 4x l. This gives the occurrence
probabilities of each of the four nucleotides at a position j. The score of any specific sequence
is the sum of the position scores from the weight matrix corresponding to that sequence.
Using this representation an entire genome can be scanned by a matrix and the score at
every position obtained (Stormo, 2000). Any sequence with score that is higher than the
predefined cut-off is a potential new binding site. A consensus sequence is deduced from a
multiple alignment of input sequences and then converted into a position weight matrix.
A PWM score is the sum of position-specific scores for each symbol in the substring. The
matrix has one row for each symbol of the alphabet, and one column for each position in the
( ) j =1 , is defined as ∑ Nj =1 msj , j
N
pattern. The score assigned by a PWM to a substring S = S j

where j represents position in the substring, sj is the symbol at position j in the substring,
and mα,j is the score in row α, column j of the matrix.
Although matrix representation appears superior, the solution space for PWMs and PSSMs,
which consists of 4l real numbers is infinite in size, and there are many local optimal
matrices, thus, algorithms generally either produce a suboptimal motif matrix or take too
long to run when the motif is longer than 10 bp (Francis and Henry, 2008).

2.3 Representation with nucleotide dependency

The interdependence between neighboring nucleotides with similar number of parameters
as string and matrix representations is described by Scored Position Specific Pattern (SPSP).
A set of length-l binding site patterns can be described by a SPSP representation P, which
contains c (c ≤ l) sets of patterns Pi, 1 ≤ i ≤ c, where each set of patterns Pi contains length-li
patterns Pi,j of symbols A, C, G, and T and ∑i li = l. Each length- li pattern Pi,j is associated
with a score si,j that represents the “closeness” of a pattern to be a binding site. The lower the
score, the pattern is more likely a binding site (Henry and Fracis, 2006).

3. Methods of finding TF binding sites in a DNA sequence

3.1 Searching known motifs
Development of databases of complete information on experimentally validated TF binding
site is indispensable for promoter sequence analysis. Information about TF binding sites
remain scattered in literature. In the last one and half decade phenomenal increase in
computational power, cheaper electronic storage with faster communication technologies,
have resulted in development of a range of web accessible databases having experimentally
validated TF binding sites. These TF binding site databases are not only highly useful for
identification of putative TF binding sites in new promoter sequences (Table1), but also are
valuable for providing positive dataset required for improvement and validation of new TF
binding site prediction algorithms.

3.1.1 TRANSFAC
TRANSFAC is the largest repository of transcription factors binding sites. TRANSFAC
(TRANSFAC 7.0, 2005) web accessible database consists of 6,133 factors with 7,915 sites,
while professional version (TRANSFAC 2008.3) consists of 11,683 factors with 30,227 sites.
TRANSFAC database is composed of six tables SITE, GENE, FACTOR, CELL, CLASS and
50 Computational Biology and Applied Bioinformatics

MATRIX. GENE table gives a short explanation of the gene where a site (or group of sites)
belongs to; FACTOR table describes the proteins binding to these sites. CELL gives brief
information about the cellular source of proteins that have been shown to interact with the
sites. CLASS contains some background information about the transcription factor classes,
while the MATRIX table gives nucleotide distribution matrices for the binding sites of
transcription factors. This database is most frequently used as reference for TFB sites as well
as for development of new algorithms. However, new users find it difficult to access the
database because it requires search terms to be entered manually. There is no criterion to
select the organism, desired gene or TF from a list, so web interface is not user friendly.
Other web tools such as TF search and Signal Scan overcome this limitation to certain extent.

3.1.2 Signal Scan

Signal Scan finds and lists homologies of published TF binding site signal sequences in the
input DNA sequence by using TRANSFAC, TFD and IMD databases. It also allows to select
from different classes viz mammal, bird, amphibian, insect, plant, other eukaryotes,
prokaryote, virus (TRANSFAC only), insect and yeast (TFD only).

3.1.3 TRRD
The transcription regulatory region database (TRRD) is a database of transcription
regulatory regions of the eukaryotic genome. The TRRD database contains three
interconnected tables: TRRDGENES (description of the genes as a whole), TRRDSITES
(description of the sites), and TRRDBIB (references). The current version, TRRD 3.5,
comprises of the description of 427 genes, 607 regulatory units (promoters, enhancers, and
silencers), and 2147 transcription factor binding sites. The TRRDGENES database compiles
the data on human (185 entries), mouse (126), rat (69), chicken (29), and other genes.

Transcription
Developmental/Environmental
factor binding Position
stimulus Sequence
site
TATA Box -33 tcccTATAaataa
Core promoter
Cat Box -49 gCCAAc
G Box -66 tgACGgtgt
Stress responsive ABRE -76 acaccttctttgACTGtccatccttc
ABI4 -245 CACCg
W Box -72 cttctTTGAcgtgtcca
Pathogen defense
TCA gAGAAgagaa
Light Response I box -302 gATATga
WUN -348 tAATTacac
Wound specific
TCA -646 gAGAAgagaa
Legumin -118 tccatacCCATgcaagctgaagaatgtc
Opaque-2 -348 TAATtacacatatttta
Seed Specific
Prolamine box -385 TTaaaTGTAAAAgtAa
AAGAA-motif -294 agaAAGAa
Table 1. In silico analysis of pigeonpea legumin gene promoter for identification of
regulatory elements. Database search reveals that it consist of regulatory elements that can
direct its activation under different envirnmental conditions and developmental stages.
In Silico Identification of Regulatory Elements in Promoters 51

3.1.4 PlantCARE
PlantCARE is database of plant specific cis-Acting regulatory elements in the promoter
regions (Lescot et al., 2002). It generates a sequence analysis output on a dynamic webpage,
on which TF binding sites are highlighted in the input sequence. The database can be
queried on names of transcription factor (TF) sites, motif sequence, function, species, cell
type, gene, TF and literature references. Information regarding TF site, organism, motif
position, strand, core similarity, matrix similarity, motif sequence and function are listed
whereas the potential sites are mapped on the query sequence.

3.1.5 PLACE
PLACE is another database of plant cis-acting regulatory elements extracted from published
reports (Higo et al., 1999). It also includes variations in the motifs in different genes or plant
species. PLACE also includes non-plant cis-elements data that may have homologues with
plant. PLACE database also provides brief description of each motif and links to
publications.

3.1.6 RegulonDB
RegulonDB is a comprehensive database of gene regulation and interaction in E. coli. It
consists of data on almost every aspect of gene regulation such as terminators, promoters,
TF binding sites, active and inactive transcription factor conformations, matrices alignments,
transcription units, operons, regulatory network interactions, ribosome binding sites (rbs),
growth conditions, gene product and small RNAs.

3.1.7 ABS
ABS is a database of known TF binding sites identified in promoters of orthologous
vertebrate genes. It has 650 annotated and experimental validated binding sites from 68
transcription factors and 100 orthologous target genes in human, mouse, rat and chicken
genome sequences. Although it’s a simple and easy-to-use web interface for data retrieval
but it does not facilitate either analysis of new promoter sequence or mapping user defined
motif in the promoter.

3.1.8 MatInspector
MatInspector identifies cis-acting regulatory elements in nucleotide sequences using
library of weight matrices (Cartharius et al., 2005). It is based on novel matrix family
concept, optimized thresholds, and comparative analysis that overcome the major limitation
of large number of redundant binding sites predicted by other programs. Thus it increases
the sensitivity of reducing false positive predictions. MatInspector also allows integration of
output with other sequence analysis programs e.g. DiAlignTF, FrameWorker,
SequenceShaper, for an in-depth promoter analysis and designing regulatory sequences.
MatInspector library contains 634 matrices representing one of the largest libraries available
for public searches.

3.1.9 JASPAR
JASPAR is the another open access database that compete with the commercial TF binding
site databases such as TRANSFAC (Portales-Casamar et al., 2009). The latest release has a
52 Computational Biology and Applied Bioinformatics

collection of 457 non-redundant, curated profiles. It is a collection of smaller databases, viz

JASPAR CORE, JASPAR FAM, JASPAR PHYLOFACTS, JASPAR POLII and others, among
which JASPAR CORE is most commonly used. The JASPAR CORE database contains a
curated, non-redundant set of profiles, derived from published collections of experimentally
determined transcription factor binding sites for multicellular eukaryotes (Portales-Casamar
et al., 2009). The JASPAR database can also be accessed remotely through external
application programming interface (API).

3.1.10 Cister: cis-element cluster finder

Cister is based on the technique of posterior decoding, with Hidden Markov model and
predicts regulatory regions in DNA sequences by searching for clusters of cis-elements
(Frith et al., 2001). The Cister input page consists of 16 common TF sites to define a cluster
and additional user defined PWM or TRANSFAC entries can also be entered. For web based
analysis maximum input sequence length is 100 kb, however, the program is downloadable
for standalone applications and analysis of longer sequences.

3.1.11 MAPPER
MAPPER stands for Multi-genome Analysis of Positions and Patterns of Elements of
Regulation It is a platform for the computational identification of TF binding sites in
multiple genomes (Marinescu et al., 2005). The MAPPER consists of three modules, the
MAPPER database, the Search Engine, and rSNPs and combines TRANSFAC and JASPAR
data. However, MAPPER database is limited to TFBSs found only in the promoter of genes
from the human, mouse and D.melanogaster genomes.

3.1.12 Stubb
Like Cister, Stubb also uses hidden Markov models (HMM) to obtain a statistically
significant score for modules (Sinha et al., 2006). STUBB is more suitable for finding
modules over genomic scales with small set of transcription factors whose binding sites are
known. Stubb differs from MAPPER in that the application of latter is limited to binding
sites of a single given motif in an input sequence.

3.1.13 Clover
Clover is another program for identifying functional sites in DNA sequences. It take a set of
DNA sequences that share a common function, compares them to a library of sequence
motifs (e.g. transcription factor binding patterns), and identifies which, if any, of the motifs
are statistically overrepresented in the sequence set (Frith et al., 2004). It requires two input
files one for sequences in fasta format and another for sequence motif. Clover provides
JASPAR core collection of TF binding sites that can be converted to clover format. Clover is
also available as standalone application for windows, Linux as well as Mac operating
systems.

3.1.14 RegSite
Regsite consists of plant specific largest repository of transcription factor binding sites.
Current RegSite release contains 1816 entries. It is used by transcription start site prediction
programs (Sinha et al., 2006).
In Silico Identification of Regulatory Elements in Promoters 53

3.1.15 JPREdictor
JPREdictor is a JAVA based cis-regulatory TF binding site prediction program (Fiedler and
Rehmsmeier, 2006). The JPREdictor can use different types of motifs: Sequence Motifs,
Regular Expression Motifs, PSPMs as well as PSSMs and the complex motif type
(MultiMotifs). This tool can be used for the prediction of cis-regulatory elements on a
genome-wide scale.

3.2 Motif finding programs

3.2.1 Phylogenetic footprinting
Comparative DNA sequence analysis shows local difference in mutation rates and reveals a
functional site by virtue of its conservation in a background of non-functional sequences. In
the phylogenetic equivalent, regulatory elements are protected from random drift across
evolutionary time by selection. Orthologous noncoding DNA sequences from multiple
species provide a strong base for identification of regulatory elements by Phylogenetic
footprinting (Fig. 1) (Rombauts et al., 2003).
The major advantage of phylogenetic footprinting over the single genome is that multigene
approach requires data of co regulated genes. While phylogenetic footprinting can
identifying regulatory elements present in single gene, that remain conserved during the
course of divergence of two species under investigation. With steep increase in available
complete genome sequences, across species comparisons for a wide variety of organisms has
become possible (Blanchette and Tompa, 2002; Das and Dai, 2007). A multiple sequence
alignment algorithm suited for phylogenetic footprinting should be able to indentify small
(5-15 bp) sequence in a background of highly diverse sequences.

Fig. 1. Identification of new regulatory elements (L-19) in legumin gene promoters by

phylogenetic footprinting.
3.2.1.1 Clustal W, LAGAN, AVID
In phylogenetic footprinting primary aim is to construct global multiple alignment of the
orthologous promoter sequences and then identify a region conserved across orthologous
sequences. Alignment algorithms, such as ClustalW (Thompson et al., 1994), LAGAN
(Brudno et al., 2003), AVID (Bray et al., 2003) and Bayes-Block Aligner (Zhu edt al., 1998),
have proven useful for phylogenetic footprinting, but the short length of the conserved
motif compared to the length of the non-conserved background sequence; and their variable
position in a promoter hampers the alignment of conserved motifs. Moreover multiple
sequence alignment does not reveal meaningful biological information if the species used
54 Computational Biology and Applied Bioinformatics

for comparison are too closely related. If the species are too distantly related, it is difficult to
find an accurate alignment. It requires computational tools that bypass the requirement of
sequence alignment completely and have the capabilities to identify short and scattered
conserved regions.
3.2.1.2 MEME, Consensus, Gibbs sampler, AlignAce
In cases where multiple alignment algorithms fails, motif finding algorithms such as MEME,
Consensus and Gibbs sampler have been used (Fig. 2). The feasibility of using comparative
DNA sequence analysis to identify functional sequences in the genome of S. cerevisiae, with
the goal of identifying regulatory sequences and sequences specifying nonprotein coding
RNAs was investigated (Cliften et al., 2001). It was found that most of the DNA sequences of
the closely related Saccharomyces species aligned to S.cerevisiae sequences and known
promoter regions were conserved in the alignments. Pattern search algorithms like
CONSENSUS (Hertz et al., 1990), Gibbs sampling (Lawrence et al., 1993) and AlignAce
(Roth et al., 1998) were useful for identifying known regulatory sequence elements in the
promoters, where they are conserved through the most diverged Saccharomyces species.
Gibbs sampler was used for motif finding using phylogenetic footprinting in proteobacterial
genomes (McCue et al., 2001). These programs employ two approaches for motif finding.
One approach is to employ a training set of transcription factor binding sites and a scoring
scheme to evaluate predictions. The scoring scheme is often based on information theory
and the training set is used to empirically determine a score threshold for reporting of the
predicted transcription factor binding sites. The second method relies on a rigorous
statistical analysis of the predictions, based upon modeled assumptions. The statistical
significance of a sequence match to a motif can be accessed through the determination of p-
value. P-value is the probability of observing a match with a score as good or better in a
randomly generated search space of identical size and nucleotide composition. The smaller
the p-value, the lesser the probability that the match is due to chance alone. Since the motif
finding algorithms assume the input sequences to be independent, therefore, they are
limited by the fact that the data sets containing a mixture of some closely related species will
have an unduly high weight in the results of motifs reported.
Multiple genome sequences were compared that are as optimally diverged as possible in
Saccharomyces genomes. Phylogenetic footprints were searched among the genome
sequences of six Saccharomyces species using the sequence alignment tool CLUSTAL W and
many statistically significant conserved sequence motifs (Cliften et al., 2003) were found.

Fig. 2. Combined Block diagram of an MEME output highlighting conserved motifs in

promoter regions of legumin seed storage protein genes of four different species.
In Silico Identification of Regulatory Elements in Promoters 55

3.2.1.3 Footprinter
This promising novel algorithm was developed to overcome the limitations imposed by
motif finding algorithms. This algorithm identifies the most conserved motifs among the
input sequences as measured by a parsimony score on the underlying phylogenetic tree
(Blanchette and Tompa, 2002). It uses dynamic programming to find most parsimonious k-
mer from each of the input sequences where k is the motif length. In general, the algorithm
selects motifs that are characterized by a minimal number of mismatches and are conserved
over long evolutionary distances. Furthermore, the motifs should not have undergone
independent losses in multiple branches. In other words, the motif should be present in the
sequences of subsequent taxa along a branch. The algorithm, based on dynamic
programming, proceeds from the leaves of the phylogenetic tree to its root and seeks for
motifs of a user-defined length with a minimum number of mismatches. Moreover, the
algorithm allows a higher number of mismatches for those sequences that span a greater
evolutionary distance. Motifs that are lost along a branch of the tree are assigned an
additional cost because it is assumed that multiple independent losses are unlikely in
evolution. To compensate for spurious hits, statistical significance is calculated based on a
random set of sequences in which no motifs occur.
3.2.1.4 CONREAL
CONREAL (Conserved Regulatory Elements Anchored Alignment Algorithm) is another
motif finding algorithm based on phylogenetic footprinting (Berezikov et al., 2005). This
algorithm uses potential motifs as represented by positional weight matrices (81 vertebrate
matrices form JASPAR database and 546 matrices from TRANSFAC database) to establish
anchors between orthologous sequences and to guide promoter sequence alignment.
Comparison of the performance of CONREAL with the global alignment programs LAGAN
and AVID using a reference data set, shows that CONREAL performs equally well for
closely related species like rodents and human, and has a clear added value for aligning
promoter elements of more divergent species like human and fish, as it identifies conserved
transcription-factor binding sites that are not found by other methods.
3.2.1.5 PHYLONET
The PHYLONET computational approach identifies conserved regulatory motifs directly
from whole genome sequences of related species without reliance on additional information
was developed by (Wang and Stormo, 2005). The major steps involved are: i) construction of
phylogenetic profiles for each promoter , ii) searching through the entire profile space of all
the promoters in the genome to identify conserved motifs and the promoters that contain
them using algorithm like BLAST, iii) determination of statistical significance of motifs
(Karlin and Altschul, 1990). By comparing promoters using phylogenetic profiles (multiple
sequence alignments of orthologous promoters) rather than individual sequences, together
with the application of modified Karlin– Altschul statistics, they readily distinguished
biologically relevant motifs from background noise. When applied to 3524 Saccharomyces
cerevisiae promoters with Saccharomyces mikatae, Saccharomyces kudriavzevii, and Saccharomyces
bayanus sequences as references PHYLONET identified 296 statistically significant motifs
with a sensitivity of >90% for known transcription factor binding sites. The specificity of the
predictions appears very high because most predicted gene clusters have additional
supporting evidence, such as enrichment for a specific function, in vivo binding by a known
TF, or similar expression patterns.
56 Computational Biology and Applied Bioinformatics

However, the prediction of additional transcription factor binding sites by comparison of a

motif to the promoter regions of an entire genome has its own problems due to the large
database size and the relatively small width of a typical transcription factor binding site.
There is an increased chance of identification of many sites that match the motif and the
variability among the transcription factor binding sites permits differences in the level of
regulation, due to the altered intrinsic affinities for the transcription factor (Carmack et al.,
2007).
3.2.1.6 Phyloscan
PhyloScan combines evidence from matching sites found in orthologous data from several
related species with evidence from multiple sites within an intergenic region.The
orthologous sequence data may be multiply aligned, unaligned, or a combination of aligned
and unaligned. In aligned data, PhyloScan statistically accounts for the phylogenetic
dependence of the species contributing data to the alignment and, in unaligned data; the
evidence for sites is combined assuming phylogenetic independence of the species. The
statistical significance of the gene predictions is calculated directly, without employing
training sets (Carmack et al., 2007). The application of the algorithm to real sequence data
from seven Enterobacteriales species identifies novel Crp and PurR transcription factor
binding sites, thus providing several new potential sites for these transcription factors.

3.3 Software suites for motif discovery

3.3.1 BEST
BEST is a suite of motif-finding programs that include four motif-finding programs:
AlignACE (Roth et al., 1998), BioProspector(Liu et al., 2001), Consensus(Hertz and Stormo,
1999), MEME (Bailey et al., 2006) and the optimization program BioOptimizer (Jensen and
Liu, 2004). BEST was compiled on Linux, and thus it can only be run on Linux machines
(Che et al., 2005).

3.3.2 Seqmotifs
Seqmotifs is a suite of web based programs to find regulatory motifs in coregulated genes of
both prokaryotes and eukaryotes. In this suite BioProspector (Liu et al., 2001) is used for
finding regulatory motifs in prokaryote or lower eukaryote sequences while
CompareProspector(Liu et al., 2002) is used for higher eukaryotes. Another program Mdscan
(Liu et al., 2002) is used for finding protein-DNA interaction sites from ChIP-on-chip targets.
These programs analyze a group of sequences of coregulated genes so they may share
common regulatory motifs and output a list of putative motifs as position-specific probability
matrices, the individual sites used to construct the motifs, and the location of each site on the
input sequences. CompareProspector has been used for identification of transcription factors
Mef2, My, Srf, and Sp1 motifs from a human-muscle-specific co regulated genes. Additionally
in a C. elegans–C briggsae comparison, CompareProspector found the PHA-4 motif and the
UNC-86 motif.(Liu et al., 2004) Another C. Elegans CompareProspector analysis showed that
intestine genes have GATA transcription factor binding motif that was latter experimentally
validated (Pauli et al., 2006).

3.3.3 RSAT
The Regulatory Sequence Analysis Tools (RSAT) is an integrated online tool to analyze
regulatory sequences in co regulated genes (Thomas-Chollier et al., 2008). The only input
In Silico Identification of Regulatory Elements in Promoters 57

required is a list of genes of interest; subsequently upstream sequences of desired distance

can be retrieved and appended to the list. Further tasks of putative regulatory signals
detection, matching positions for the detected signals in the original dataset can then be
performed. The suite includes programs for sequence retrieval, pattern discovery,
phylogenetic footprint detection, pattern matching, genome scanning and feature map
drawing. Random controls can be performed with random gene selections or by generating
random sequences according to a variety of background models (Bernoulli, Markov). The
results can be displayed graphically highlighting the desired features. As RSAT web
services is implemented using SOAP and WSDL, so either Perl, Python or Java scripts can
used for developing custom workflows by combining different tools.

3.3.4 TFM
TFM (Transcription Factor Matrices) is a software suite for identifying and analyzing
transcription factor binding sites in DNA sequences. It consists of TFM-Explorer, TFM-Scan,
TFM-Pvalue and TFM-Cluster.
TFM-Explorer (Transcription Factor Matrix Explorer) proceeds in two steps: scans sequences
for detecting all potential TF binding sites, using JASPAR or TRANSFAC and extracts
significant transcription factor.
TFM-Scan is a program dedicated to the location of large sets of putative transcription factor
binding sites on a DNA sequence. It uses Position Weight Matrices such as those available in
the Transfac or JASPAR databases. The program takes as input a set of matrices and a DNA
sequence. It computes all occurrences of matrices on the sequence for a given P-value
threshold. The algorithm is very fast and allows for large-scale analysis. TFM-Scan is also
able to cluster similar matrices and similar occurrences
TFM-Pvalue is a software suite providing tools for computing the score threshold associated
to a given P-value and the P-value associated to a given score threshold. It uses Position
Weight Matrices such as those available in the Transfac or JASPAR databases. The program
takes as input a matrix, the background probabilities for the letters of the DNA alphabet and
a score or a P-value.

3.3.5 rVISTA
rVISTA (regulatory VISTA) combines searching the major transcription binding site
database TRANSFAC Professional from Biobase with a comparative sequence analysis and
this procedure reduced the number of predicted transcription factor binding sites by several
orders of magnitude (Loots and Ovcharenko, 2004). It can be used directly or through links
in mVISTA, Genome VISTA, or VISTA Browser. Human and mouse sequences are aligned
using the global alignment program AVID (Bray et al., 2003).

3.3.6 Mulan
Mulan brings together several novel algorithms: the TBA multi-aligner program for rapid
identification of local sequence conservation and the multiTF program for detecting
evolutionarily conserved transcription factor binding sites in multiple alignments. In
addition, Mulan supports two-way communication with the GALA database; alignments of
multiple species dynamically generated in GALA can be viewed in Mulan, and conserved
transcription factor binding sites identified with Mulan/multiTF can be integrated and
overlaid with extensive genome annotation data using GALA. Local multiple alignments
58 Computational Biology and Applied Bioinformatics

computed by Mulan ensure reliable representation of short- and large-scale genomic

rearrangements in distant organisms. Mulan allows for interactive modification of critical
conservation parameters to differentially predict conserved regions in comparisons of both
closely and distantly related species. The uses and applications of the Mulan tool through
multispecies comparisons of the GATA3 gene locus and the identification of elements that
are conserved in a different way in avians than in other genomes, allowing speculation on
the evolution of birds.

3.3.7 MotifVoter
MotifVoter is a variance based ensemble method for discovery of binding sites. It uses 10
most commonly used individual basic motif finders as its component (Wijaya et al., 2008).
AlignACE (Hughes et al., 2000), MEME (Bailey and Elkan, 1994; Bailey et al., 2006),
ANNSpec, Mitra, BioProspector, MotifSampler, Improbizer, SPACE, MDScan and
Weeder. All programs can be selected individually or collectively. Though the existing
ensemble methods overall perform better than stand-alone motif finders, the improvement
gained is not substantial. These methods do not fully exploit the information obtained from
the results of individual finders, resulting in minor improvement in sensitivity and poor
precision.

3.3.8 ConSite
ConSite is a, web-based tool for finding cis-regulatory elements in genomic sequences
(Sandelin et al., 2004). Two genomic sequences submitted for analysis are aligned by ORCA
method. Alternatively, prealigned sequences can be submitted in ClustalW, MSF (GCG),
Fasta or Pair wise BLAST format. For analysis Transcription factors can be selected on the
basis of species, name, domain or user defined matrix (raw counts matrix or position weight
matrix). Predictions are based on the integration of binding site prediction generated with
high-quality transcription factor models and cross-species comparison filtering
(phylogenetic footprinting). ConSite (Sandelin et al., 2004) is based on the JASPAR database
(Portales-Casamar et al., 2009). By incorporating evolutionary constraints, selectivity is
increased by an order of magnitude as compared to single sequence analysis. ConSite offers
several unique features, including an interactive expert system for retrieving orthologous
regulatory sequences.

3.3.9 OPOSSUM
OPOSSUM identifies statistically over-represented, conserved TFBSs in the promoters of co-
expressed genes (Ho Sui et al., 2005). OPOSSUM integrates a precomputed database of
predicted, conserved TFBSs, derived from phylogenetic footprinting and TFBS detection
algorithms, with statistical methods for calculating overrepresentation. The background
data set was compiled by identifying all strict one-to-one human/mouse orthologues from
the Ensemble database. These orthologues were then aligned using ORCA, a pair-wise DNA
alignment program. The conserved non-coding regions were identified. The conserved
regions which fell within 5000 nucleotides upstream and downstream of the transcription
start site (TSS) were then scanned for TF sites using the position weight matrices (PWMs)
from the JASPAR database (Portales-Casamar et al., 2009). These TF sites were stored in the
OPOSSUM database and comprise the background set.
In Silico Identification of Regulatory Elements in Promoters 59

3.3.10 TOUCAN2
TOUCAN 2 is an operating system independent, open source, JAVA based workbench for
regulatory sequence analysis (Aerts et al., 2004). It can be used for detection of significant
transcription factor binding sites from comparative genomics data or for detection of
combinations of binding sites in sets of co expressed/co regulated genes. It tightly integrates
with Ensemble and EMBL for retrieval of sequences data. TOUCAN provides options to
align sequences with different specialized algorithms viz AVID (Bray et al., 2003), LAGAN
(Brudno et al., 2003), or BLASTZ. MotifScanner algorithm is used to search occurrence of
sites of transcription factors by using libraries of position weight matrices from TRANSFAC
6 (Matys et al., 2003), JASPAR, PLANTCARE (Lescot et al., 2002), SCPD and others. Motif
Sampler can be used for detection of over-represented motifs. More significantly TOUCAN
provides an option to select cis-regulatory modules using the ModuleSearch In essence
TOUCAN 2 provides one of the best integration of different algorithms for identification of
cis regulatory elements.

3.3.11 WebMOTIFS
WebMOTIFS web server combines TAMO and THEME tools for identification of conserved
motifs in co regulated genes (Romer et al., 2007). TAMO combines results from four motif
discovery programs viz AlignACE, MDscan, MEME, and Weeder, followed by clustering of
results (Gordon et al., 2005). Subsequently Bayesian motif analysis of known motifs is done
by THEME. Thus it integrates de novo motif discovery programs with Bayesian approaches
to identify the most significant motifs. However, current version of Web MOTIFS supports
motif discovery only for S. cerevisiae, M. musculus, and H. sapiens genomes.

3.3.12 Pscan
Pscan is a software tool that scans a set of sequences (e.g. promoters) from co-regulated or
co-expressed genes with motifs describing the binding specificity of known transcription
factors (Zambelli et al, 2009). It assesses which motifs are significantly over or
underrepresented, providing thus hints on which transcription factors could be common
regulators of the genes studied, together with the location of their candidate binding sites in
the sequences. Pscan does not resort to comparisons with orthologous sequences and
experimental results show that it compares favorably to other tools for the same task in
terms of false positive predictions and computation time.

4. Identification of regulatory elements in legumin promoter

The plant seeds are rich source of almost all essential supplements of diet viz. proteins,
carbohydrates, and lipids. Seeds not only provide nutrients to germinating seeding but are
major source of energy and other cellular building blocks for human and other heterotrophs
as well. Consequently, there is a plethora of pathogens attacking plant seeds. Certain pests
such as coleopteran insects of the family Bruchidae, have evolved with leguminous plants
(Sales et al., 2000). It is believed that these seeds, most of which are not essential for the
establishment of the new plant following germination, contribute to the protection and
defense of seeds against pathogens and predators.
Genes encoding seed storage proteins, like zein, phaseolin and legumin, were among the
first plant genes studied at gene expression level. Hetrologous expression of reporter genes
60 Computational Biology and Applied Bioinformatics

confirmed that there promoters direct the seed specific expression and subsequently these
seed storage gene promoters were used for developing transgenic plants for expressing
different genes of interest. Earlier we isolated legumin gene promoter from pigeon pea
(Cajanus cajan) and identified regulatory elements in its promoter region(Jaiswal et al., 2007).
Sequence analysis with PLACE, PLANTCARE, MATINSPECTOR and TRANSFC shows
that legumin promoter not only consist of regulatory elements for seed specific expression
but also have elements that are present in promoters of other genes involved in pathogen
defense, abiotic stress, light response and wounding (Table 1). Our pervious study also
confirmed that legumin promoter in expressed in the developing seedling as well (Jaiswal et
al., 2007). Recent studies have shown that these promoters are expressed in non seed tissues
as well (Zakharov et al., 2004) and play a role in bruchid (insect) resistance (Sales et al.,
2000).
In such a scenario where seed storage protein performs an additional task of pathogen
defense its prompter must have responsive elements to such stresses. In fact legumin
promoter consists of transcription factor binding site for wounding, a signal of insect attack
and pathogen defense (Table 1). Since promoter sequences are available for legumin
promoter from different species it becomes a good system for identification novel regulatory
elements in these promoter. Phylogenetic footprinting analysis reveals presentence of
another conserved motif 19 base pair downstream to legumin box (Fig. 1), named L-19 box
(Jaiswal et al., 2007). Further MEME analysis shows that in addition to four conserved
blocks present for TATA box, G box, Legumin box and L-19 box, there are other conserved,
non overlapping sequence blocks are present that were not revealed by multiple sequence
alignment based phylogenetic footprinting (Fig. 2).

5. Conclusion
With the critical role of cis-regulatory elements in differentiation of specific cells leading to
growth, development and survival of an organism, scientists have a great interest in their
identification and characterization. However, due to the limited availability of known
transcription factors identification of the corresponding regulatory elements through
conventional DNA-protein interaction techniques is challenging. Rapid increase in number
of complete genome sequences, identification of co-regulated genes through microarray
technology with available electronic storage and computational power has put before the
scientific community a challenge to integrate these advancing technology and develop
computational program to identify these short but functionally important regulatory
sequences. Although some progress has been made in this direction and databases like
TRANSFC and others store libraries of transcription factor binding sites. However, there are
limitations primarily because publicly available libraries are very small and complete
datasets are not freely available. Secondly, because of their very small size there is certain
degree of redundancy in binding and therefore the chances of false prediction are very high.
These limitations have been overcome to some extent by databases like JASPAR that are
freely available and have a collection of regulatory elements large enough to compete with
the commercially available datasets. Another concern with cis-acting regulatory elements is
that the data pool of these functional non coding transcription factor binding sites is very
small (a few thousands), compared with the fact that thousand of genes are expressed in a
cell at any point of time and every gene is transcribed by a combination of minimum 5-10
transcription factors. Phylogenetic footprinting, has enormous potential in identifying new
In Silico Identification of Regulatory Elements in Promoters 61

regulatory elements and completing the gene network map. Although routine sequence
alignment programs such as clustalW fail to align short conserved sequences in a
background of hyper variable sequences, more sophisticated sequence alignment programs
have been specially developed for identification of conserved regulatory elements. These
programs such as CONREAL uses available transcription factor binding site data to align
the two sequences that decreases the chances of missing a regulatory site considerably.
Moreover, other approaches such as MEME altogether abandons the presence of sequence
alignment and directly identifies the rare conserved blocks even if they have jumbled up to
the complementary strand. With the increasing sophistication and accuracy of motif finding
programs and available genome sequences it can be assumed that knowledge of these
regulatory sequences will definitely increase (Table 2). Once we have sufficient data it can
be used for development of synthetic promoters with desired expression patterns (Fig. 3).

Fruit
Specific

Wound
Specific

Seed Specific

CAMV 35S Promotor

Assemble
Specific Motifs
s

Construction of Synthetic
Promotor
Gene

Gene

Fig. 3. Future prospects: development of synthetic promoters for expression of gene of

interest in desired tissue at defined time and developmental stage. Example: Regulatory
elements can be combined from wound, fruit and seed specific promoters and combined
with strong CaMV 35S promoter for high level expression of desired gene in all these
tissues.
62 Computational Biology and Applied Bioinformatics

Tools Web site

Sequence Alignment
Blast-Z http://www.bx.psu.edu/miller_lab/
Dialign http://dialign.gobics.de/chaos-dialign-submission
AVID (Mvista) http://genome.lbl.gov/vista/mvista/submit.shtml
Lagan http://lagan.stanford.edu/lagan_web/index.shtml
Clustal W http://www.ebi.ac.uk/Tools/msa/clustalw2/
TF Binding Site search
Consite http://www.phylofoot.org/consite
CONREAL http://conreal.niob.knaw.nl/
http://www.softberry.com/berry.phtml?topic=promhg&group=progra
PromH
ms&subgroup=promoter
Trafac http://trafac.cchmc.org/trafac/index.jsp
http://wingless.cs.washington.edu/htbin-
Footprinter
post/unrestricted/FootPrinterWeb/FootPrinterInput2.pl
rVISTA http://rvista.dcode.org/
TFBIND http://tfbind.hgc.jp/
TESS http://www.cbil.upenn.edu/cgi-bin/tess/tess
TFSearch http://www.cbrc.jp/research/db/TFSEARCH.html
Toucan http://homes.esat.kuleuven.be/~saerts/software/toucan.php
Phyloscan http://bayesweb.wadsworth.org/cgi-bin/phylo_web.pl
OFTBS http://www.bioinfo.tsinghua.edu.cn/~zhengjsh/OTFBS/
http://alggen.lsi.upc.es/cgi-
PROMO
bin/promo_v3/promo/promoinit.cgi?dirDB=TF_8.3
R-Motif http://bioportal.weizmann.ac.il/~lapidotm/rMotif/html/
Motif Finding
MEME http://meme.sdsc.edu/meme4_6_1/intro.html
AlignAce http://atlas.med.harvard.edu/cgi-bin/alignace.pl
MotifVoter http://compbio.ddns.comp.nus.edu.sg/~edward/MotifVoter2/
RSAT http://rsat.scmbb.ulb.ac.be/rsat/
Gibbs Sampler http://bayesweb.wadsworth.org/gibbs/gibbs.html
BioProspector http://ai.stanford.edu/~xsliu/BioProspector/
MatInspector http://www.genomatix.de/
Improbizer http://users.soe.ucsc.edu/~kent/improbizer/improbizer.html
WebMOTIFS http://fraenkel.mit.edu/webmotifs-tryit.html
Psacn http://159.149.109.9/pscan/
http://wingless.cs.washington.edu/htbin-
FootPrinter
post/unrestricted/FootPrinterWeb/FootPrinterInput2.pl
Table 2. Regulatory sequences identification programs.

6. References
Aerts, S., Van Loo, P., Thijs, G., Mayer, H., de Martin, R., Moreau, Y., and De Moor, B.
(2004). TOUCAN 2: the all-inclusive open source workbench for regulatory
sequence analysis. Nucleic Acids Research 33, W393-W396.
In Silico Identification of Regulatory Elements in Promoters 63

Bailey, T.L., and Elkan, C. (1994). Fitting a mixture model by expectation maximization to
discover motifs in biopolymers. Proc Int Conf Intell Syst Mol Biol 2, 28-36.
Bailey, T.L., Williams, N., Misleh, C., and Li, W.W. (2006). MEME: discovering and
analyzing DNA and protein sequence motifs. Nucleic Acids Research 34, W369-
W373.
Berezikov, E., Guryev, V., and Cuppen, E. (2005). CONREAL web server: identification and
visualization of conserved transcription factor binding sites. Nucleic Acids
Research 33, W447-W450.
Blanchette, M., and Tompa, M. (2002). Discovery of Regulatory Elements by a
Computational Method for Phylogenetic Footprinting. Genome Research 12, 739-
748.
Bray, N., Dubchak, I., and Pachter, L. (2003). AVID: A Global Alignment Program. Genome
Research 13, 97-102.
Brudno, M., Do, C.B., Cooper, G.M., Kim, M.F., Davydov, E., Program, N.C.S., Green, E.D.,
Sidow, A., and Batzoglou, S. (2003). LAGAN and Multi-LAGAN: Efficient Tools for
Large-Scale Multiple Alignment of Genomic DNA. Genome Research 13, 721-731.
Buhler, J., and Tompa, M. (2002). Finding motifs using random projections. J Comput Biol 9,
225-242.
Carmack, C.S., McCue, L., Newberg, L., and Lawrence, C. (2007). PhyloScan: identification
of transcription factor binding sites using cross-species evidence. Algorithms for
Molecular Biology 2, 1.
Cartharius, K., Frech, K., Grote, K., Klocke, B., Haltmeier, M., Klingenhoff, A., Frisch, M.,
Bayerlein, M., and Werner, T. (2005). MatInspector and beyond: promoter analysis
based on transcription factor binding sites. Bioinformatics 21, 2933-2942.
Che, D., Jensen, S., Cai, L., and Liu, J.S. (2005). BEST: Binding-site Estimation Suite of Tools.
Bioinformatics 21, 2909-2911.
Cliften, P., Sudarsanam, P., Desikan, A., Fulton, L., Fulton, B., Majors, J., Waterston, R.,
Cohen, B.A., and Johnston, M. (2003). Finding Functional Features in
Saccharomyces Genomes by Phylogenetic Footprinting. Science.
Cliften, P.F., Hillier, L.W., Fulton, L., Graves, T., Miner, T., Gish, W.R., Waterston, R.H., and
Johnston, M. (2001). Surveying Saccharomyces Genomes to Identify Functional
Elements by Comparative DNA Sequence Analysis. Genome Research 11, 1175-
1186.
Das, M., and Dai, H.-K. (2007). A survey of DNA motif finding algorithms. BMC
Bioinformatics 8, S21.
Fiedler, T., and Rehmsmeier, M. (2006). jPREdictor: a versatile tool for the prediction of cis-
regulatory elements. Nucleic Acids Research 34, W546-W550.
Francis, C., and Henry, C.M.L. (2008). DNA Motif Representation with Nucleotide
Dependency. IEEE/ACM Trans. Comput. Biol. Bioinformatics 5, 110-119.
Frith, M.C., Hansen, U., and Weng, Z. (2001). Detection of cis -element clusters in higher
eukaryotic DNA. Bioinformatics 17, 878-889.
Frith, M.C., Fu, Y., Yu, L., Chen, J.F., Hansen, U., and Weng, Z. (2004). Detection of
functional DNA motifs via statistical over†representation. Nucleic Acids
Research 32, 1372-1381.
64 Computational Biology and Applied Bioinformatics

Gordon, D.B., Nekludova, L., McCallum, S., and Fraenkel, E. (2005). TAMO: a flexible,
object-oriented framework for analyzing transcriptional regulation using DNA-
sequence motifs. Bioinformatics 21, 3164-3165.
Henry, C.M.L., and Fracis, Y.L.C. (2006). Discovering DNA Motifs with Nucleotide
Dependency, Y.L.C. Francis, ed, pp. 70-80.
Hertz, G.Z., and Stormo, G.D. (1999). Identifying DNA and protein patterns with
statistically significant alignments of multiple sequences. Bioinformatics 15, 563-
577.
Hertz, G.Z., Hartzell, G.W., and Stormo, G.D. (1990). Identification of Consensus Patterns in
Unaligned DNA Sequences Known to be Functionally Related. Comput Appl Biosci
6, 81 - 92.
Higo, K., Ugawa, Y., Iwamoto, M., and Korenaga, T. (1999). Plant cis-acting regulatory DNA
elements (PLACE) database: 1999. Nucleic Acids Research 27, 297-300.
Ho Sui, S.J., Mortimer, J.R., Arenillas, D.J., Brumm, J., Walsh, C.J., Kennedy, B.P., and
Wasserman, W.W. (2005). oPOSSUM: identification of over-represented
transcription factor binding sites in co-expressed genes. Nucleic Acids Research 33,
3154-3164.
Hughes, J.D., Estep, P.W., Tavazoie, S., and Church, G.M. (2000). Computational
identification of Cis-regulatory elements associated with groups of functionally
related genes in Saccharomyces cerevisiae. Journal of Molecular Biology 296, 1205-
1214.
Jaiswal, R., Nain, V., Abdin, M.Z., and Kumar, P.A. (2007). Isolation of pigeon pea (Cajanus
cajan L.) legumin gene promoter and identification of conserved regulatory
elements using tools of bioinformatics. Indian Journal of experimental Biology 6,
495-503.
Jensen, S.T., and Liu, J.S. (2004). BioOptimizer: a Bayesian scoring function approach to
motif discovery. Bioinformatics 20, 1557-1564.
Karlin, S., and Altschul, S.F. (1990). Methods for assessing the statistical significance of
molecular sequence features by using general scoring schemes. Proceedings of the
National Academy of Sciences 87, 2264-2268.
Lawrence, C.E., Altschul, S.F., Boguski, M.S., Liu, J.S., Neuwald, A.F., and Wootton, J.C.
(1993). Detecting subtle sequence signals: a Gibbs sampling strategy for multiple
alignment. Science 262, 208-214.
Lescot, M., Déhais, P., Thijs, G., Marchal, K., Moreau, Y., Van de Peer, Y., RouzÃ, P., and
Rombauts, S. (2002). PlantCARE, a database of plant cis-acting regulatory elements
and a portal to tools for in silico analysis of promoter sequences. Nucleic Acids
Research 30, 325-327.
Liu, X., Brutlag, D.L., and Liu, J.S. (2001). BioProspector: discovering conserved DNA motifs
in upstream regulatory regions of co-expressed genes. Pac Symp Biocomput, 127-
138.
Liu, X.S., Brutlag, D.L., and Liu, J.S. (2002). An algorithm for finding protein-DNA binding
sites with applications to chromatin-immunoprecipitation microarray experiments.
Nat Biotech 20, 835-839.
Liu, Y., Liu, X.S., Wei, L., Altman, R.B., and Batzoglou, S. (2004). Eukaryotic Regulatory
Element Conservation Analysis and Identification Using Comparative Genomics.
Genome Research 14, 451-458.
In Silico Identification of Regulatory Elements in Promoters 65

Loots, G.G., and Ovcharenko, I. (2004). rVISTA 2.0: evolutionary analysis of transcription
factor binding sites. Nucleic Acids Research 32, W217-W221.
Loots, G.G., Locksley, R.M., Blankespoor, C.M., Wang, Z.E., Miller, W., Rubin, E.M., and
Frazer, K.A. (2000). Identification of a Coordinate Regulator of Interleukins 4, 13,
and 5 by Cross-Species Sequence Comparisons. Science 288, 136-140.
Marinescu, V., Kohane, I., and Riva, A. (2005). MAPPER: a search engine for the
computational identification of putative transcription factor binding sites in
multiple genomes. BMC Bioinformatics 6, 79.
Matys, V., Fricke, E., Geffers, R., Gößling, E., Haubrock, M., Hehl, R., Hornischer, K.,
Karas, D., Kel, A.E., Kel-Margoulis, O.V., Kloos, D.U., Land, S., Lewicki-Potapov,
B., Michael, H., Münch, R., Reuter, I., Rotert, S., Saxel, H., Scheer, M., Thiele, S.,
and Wingender, E. (2003). TRANSFAC®: transcriptional regulation, from patterns
to profiles. Nucleic Acids Research 31, 374-378.
Pauli, F., Liu, Y., Kim, Y.A., Chen, P.-J., and Kim, S.K. (2006). Chromosomal clustering and
GATA transcriptional regulation of intestine-expressed genes in C. elegans.
Development 133, 287-295.
Pennacchio, L.A., and Rubin, E.M. (2001). Genomic strategies to identify mammalian
regulatory sequences. Nat Rev Genet 2, 100-109.
Portales-Casamar, E., Thongjuea, S., Kwon, A.T., Arenillas, D., Zhao, X., Valen, E., Yusuf, D.,
Lenhard, B., Wasserman, W.W., and Sandelin, A. (2009). JASPAR 2010: the greatly
expanded open-access database of transcription factor binding profiles. Nucleic
Acids Research.
Rombauts, S., Florquin, K., Lescot, M., Marchal, K., Rouzé, P., and Van de Peer, Y. (2003).
Computational Approaches to Identify Promoters and cis-Regulatory Elements in
Plant Genomes. Plant Physiology 132, 1162-1176.
Romer, K.A., Kayombya, G.-R., and Fraenkel, E. (2007). WebMOTIFS: automated discovery,
filtering and scoring of DNA sequence motifs using multiple programs and
Bayesian approaches. Nucleic Acids Research 35, W217-W220.
Roth, F.P., Hughes, J.D., Estep, P.W., and Church, G.M. (1998). Finding DNA regulatory
motifs within unaligned noncoding sequences clustered by whole-genome mRNA
quantitation. Nat Biotech 16, 939-945.
Sales, M.Ì.c.P., Gerhardt, I.R., Grossi-de-Sá, M.F.t., and Xavier-Filho, J. (2000). Do Legume
Storage Proteins Play a Role in Defending Seeds against Bruchids? Plant
Physiology 124, 515-522.
Sandelin, A., Wasserman, W.W., and Lenhard, B. (2004). ConSite: web-based prediction of
regulatory elements using cross-species comparison. Nucleic Acids Research 32,
W249-W252.
Sinha, S., Liang, Y., and Siggia, E. (2006). Stubb: a program for discovery and analysis of cis-
regulatory modules. Nucleic Acids Research 34, W555-W559.
Stormo, G.D. (2000). DNA Binding Sites: Representation and Discovery. Bioinformatics 16,
16 - 23.
Thomas-Chollier, M., Sand, O., Turatsinze, J.-V., Janky, R.s., Defrance, M., Vervisch, E.,
Brohee, S., and van Helden, J. (2008). RSAT: regulatory sequence analysis tools.
Nucleic Acids Research 36, W119-W127.
Thompson, J.D., Higgins, D.G., and Gibson, T.J. (1994). CLUSTAL W: improving the
sensitivity of progressive multiple sequence alignment through sequence
66 Computational Biology and Applied Bioinformatics

weighting, position-specific gap penalties and weight matrix choice. Nucleic Acids
Research 22, 4673-4680.
Wang, T., and Stormo, G.D. (2005). Identifying the conserved network of cis-regulatory sites
of a eukaryotic genome. Proceedings of the National Academy of Sciences of the
United States of America 102, 17400-17405.
Wijaya, E., Yiu, S.-M., Son, N.T., Kanagasabai, R., and Sung, W.-K. (2008). MotifVoter: a
novel ensemble method for fine-grained integration of generic motif finders.
Bioinformatics 24, 2288-2295.
Zakharov, A., Giersberg, M., Hosein, F., Melzer, M., Müntz, K., and Saalbach, I. (2004).
Seed-specific promoters direct gene expression in non-seed tissue. Journal of
Experimental Botany 55, 1463-1471.
Zhu, J., Liu, J.S., and Lawrence, C.E. (1998). Bayesian adaptive sequence alignment
algorithms. Bioinformatics 14, 25-39.
 4

In Silico Analysis of Golgi Glycosyltransferases:

A Case Study on the LARGE-Like Protein Family
Kuo-Yuan Hwa1,2, Wan-Man Lin and Boopathi Subramani
Institute of Organic & Polymeric Materials
1Department of Molecular Science and Engineering
2Centre for Biomedical Industries

National Taipei University of Technology, Taipei,

Taiwan, ROC

1. Introduction
Glycosylation is one of the major post-translational modification processes essential for
expression and function of many proteins. It has been estimated that 1% of the open reading
frames of a genome is dedicated to glycosylation. Many different enzymes are involved in
glycosylation, such as glycosyltransferases and glycosidases.
Traditionally, glycosyltransferases are classified based on their enzymatic activities by
Enzyme Commission (http://www.chem.qmul.ac.uk/iubmb/enzyme/). Based on the
activated donor type, glycosyltransferases are named, for example glucosyltransferase,
mannosyltransferase and N-acetylglucosaminyltransferases. However, classification of
glycosyltransferases based on the biochemical evidence is a difficult task since most of the
enzymes are membrane proteins. Reconstruction of enzymatic assay for the membrane
proteins are intrinsically more difficult than soluble proteins. Thus the purification of
membrane-bound glycosyltransferase is a difficult task. On the other hand, with the recent
advancement of genome projects, DNA sequences of an organism are readily available.
Furthermore, bioinformatics annotation tools are now commonly used by life science
researchers to identify the putative function of a gene. Hence, new approaches based on in
silico analysis for classifying glycosyltransferase have been used successfully. The best
known database for classification of glycosyltransferase by in silico approach is the CAZy
(Carbohydrate- Active enZymes) database (http://afmb.cnrs-mrs.fr/CAZy/) (Cantarel et al.,
2009).
Glycosyltransferases are enzymes involved in synthesizing sugar moieties by transferring
activated saccharide donors into various macro-molecules such as DNA, proteins, lipids and
glycans. More than 100 glycosyltransferases are localized in the endoplasmic reticulum (ER)
and Golgi apparatus and are involved in the glycan synthesis (Narimatsu, H., 2006). The
structural studies on the ER and golgi glycosyltransferases has revealed several common
domains and motifs present between them. The glycosyltransferases are grouped into
functional subfamilies based on similarities of sequence, their enzyme characteristics, donor
specificity, acceptor specificity and the specific donor and acceptor linkages (Ishida et al.,
2005). The glycosyltransferase sequences comprise of 330-560 amino acids long and share
the same type II transmembrane protein structure with four functional domains: a short
68 Computational Biology and Applied Bioinformatics

cytoplasmic domain, a targeting / membrane anchoring domain, a stem region and a

catalytic domain (Fukuda et al., 1994). Mammals utilize only 9 sugar nucleotide donors for
glycosyltransferases such as UDP-glucose, UDP-galactose, UDP-GlcNAc, UDP-GalNAc,
UDP-xylose, UDP-glucuronic acid, GDP-mannose, GDP-fucose, and CMP-sialic acid. Other
organisms have an extensive range of nucleotide sugar donors (Varki et al., 2008). Based on
the structural studies, we have designed an intelligent platform for the LARGE protein, a
golgi glycosyltransferase. The LARGE is a member of glycosyltransferase which has been
studied in protein glycosylation (Fukuda & Hindsgaul, 2000). It was originally isolated from
a region in chromosome 22 of the human genome which was frequently deleted in human
meningiomas with alteration in glycosphingolipid composition. This led to a suggestion that
the LARGE may have possible role in complex lipid glycosylation (Dumanski et al., 1987;
Peyrard et al., 1999).

2. LARGE
LARGE is one of the largest genes present in the human genome and it is comprised of 660
kb of genomic DNA and contains 16 exons encoding a 756-amino-acid protein. It showed
98% amino acid identity to the mouse homologue and similar genomic organization. The
expression of LARGE is ubiquitous but the highest levels of LARGE mRNA are present in
heart, brain and skeletal muscle (Peyrard et al., 1999).
LARGE encodes a protein which has an N-terminal transmembrane anchor, coiled coil motif
and two putative catalytic domains with a conserved DXD (Asp-any-Asp) motif typical of
many glycosyltransferases that uses nucleoside diphosphate sugars as donors (Longman et
al., 2003 & Peyrard et al., 1999). The proximal catalytic domain in the LARGE was most
homologous to the bacterial glycosyltransferase family 8 (GT8 in CAZy database) members
(Coutinho et al., 2003). The members of this family are mainly involved in the synthesis of
bacterial outer membrane lipopolysaccharide. The distal domain resembled the human β1,3-
N-acetytglucosaminyltransferase (iGnT), a member of GT49 family. The iGnT enzyme is
required for the synthesis of the poly-N-acetyllactosamine backbone which is part of the
erythrocyte i antigen (Sasaki et al., 1997). The presence of two catalytic domains in the
LARGE is extremely unusual among the glycosyltransferase enzymes.

2.1 Functions of LARGE

2.1.1 Dystroglycan glycosylation
The Dystroglycan (DG) is an important constituent of the dystrophin-glycoprotein complex
(DGC). This complex plays an essential role in the maintaining the stability of the muscle
membrane and for the correct localization and/or ligand-binding activity, the glycosylation
of some of these components are required (Durbeej et al., 1998). The DG comprises of two
subunits, the extracellular α-DG and the transmembrane β-DG (Barresi, 2004). Various
components present in the extracellular matrix including laminin (Smalheiser & Schwartz
1987), agrin (Gee et al., 1994), neurexin, (Sugita et al., 2001), and perlecan (Peng et al., 1998)
interacts with α-DG. The carbohydrate moieties present in the α-DG are essential to bind
with laminin and other ligands. The α-DG is modified by three different types of glycans
such as: mucin type O-glycosylation, O-mannosylation, and N-glycosylation. The
glycosylated α-DG is essential for the protein’s ability to bind the laminin globular domain-
containing proteins of the Extracellular Matrix (Kanagawa, 2005). LARGE is required for the
generation of functional, properly glycosylated forms of α-DG (Barresi, 2004).
In Silico Analysis of Golgi Glycosyltransferases: A Case Study on the LARGE-Like Protein Family 69

2.1.2 Human LARGE and α-Dystroglycan

The α-DG functional glycosylation by LARGE is likely to be involved in the generation of a
glycan polymer which gives rise to the broad molecular weight range observed for α-DG
detected by VIA4-1 and IIH6 antibodies. Both the human and mouse LARGE C-terminal
glycosyltransferase domain is similar to β3GnT6, which adds GlcNAc to Gal to generate
linear polylactosamine chains (Sasaki et al., 1997), the chain formed by LARGE might also be
composed of GlcNAc and Glc.
In 1963, Myodystrophy, myd, was first described (Lane et al., 1976) as a recessive myopathy
mapping to chromosome (Chr) 8, was identified as an intragenic deletion within the
glycosyltransferase gene, LARGE. In Largemyd and enr mice, the hypoglycosylation of α-DG
in DGC was due to the mutation in LARGE (Grewal et al., 2001). The α-DG function was
restored in Largemyd skeletal muscle and ameliorates muscular dystrophy when LARGE
gene was transferred, which indicated that adjustment in the glycosylation status of α-DG
can improve the muscle phenotype.
The patients with clinical spectrum ranging from severe congenital muscular dystrophy
(CMD), structural brain and eye abnormalities [Walker-Warburg syndrome (WWS), MIM
236670] to a relative mild form of limb-girdle muscular dystrophy (LGMD2I, MIM 607155)
are linked to the abnormal O-linked glycosylation of α-DG (van Reeuwijk et al., 2005). A
study made by Barresi R. et al. (2004) revealed the existence of dual and concentration
dependent functions of LARGE. In physiological concentration, LARGE may be involved in
regulating the α-DG O-mannosylation pathway. But when the LARGE is expressed by force,
it may trigger some other alternative pathways for the O-glycosylation of α-DG which can
generate a type of repeating polymer of variable lengths, such as glycosaminoglycan-like or
core 1 or core 2 structures. This alternative glycan mimics the O-mannose glycan in its
ability to bind α-DG ligands and can compensate for the defective tetrasaccharide. The
functional LARGE protein is also required for neuronal migration during CNS development
and it rescues α-DG in MEB fibroblasts and WWS cells (Barresi R. et al., 2004).

2.1.3 LARGE in visual signal processing

The role of LARGE in proper visual signal processing was studied from the retina retinal
pathology in Largemyd mice. The functional abnormalities of the retina was investigated by a
sensitive tool called Electroretinogram (ERG). In Largemyd mice, the normal a-wave indicated
that the mutant glycosyltransferase does not have any effect on its photoreceptor function.
But the alteration in b-wave may have resulted in downstream retinal circuitry with altered
signal processing (Newman & Frishman, 1991). The DGC may also have a possible role in
this aspect of the phenotype. The abnormal b-wave was responsible for the loss of retinal
isoforms of dystrophin in humans and mice similar to the Largemyd mice.

2.2 LARGE homologues

A homologous gene to LARGE was identified and named as LARGE2. It is found to be
involved in α-DG maturation as like LARGE, according to Fujimura et al., (2005). It is still not
well understood whether these two proteins are compensatory or cooperative. The co-
expression of LARGE and LARGE2 did not increase the maturation of α-DG in comparison
with either one of them alone and it proved that for the maturation of α-DG, the function of
LARGE2 is compensatory and not cooperative. Gene therapy for muscular dystrophy using
the LARGE gene is a current topic of research (Barresi R. et al., 2004; Braun, 2004). When
compared to LARGE, LARGE2 gene may be more effective because it can glycosylate heavily
than LARGE and it also prevents the harmful and immature α-DG production.
70 Computational Biology and Applied Bioinformatics

The closely related homologues of LARGE are found in the human genome,
(glycosyltransferase-like 1B; GYLTL1B), mouse genome (Glylt1b; also called LARGE-Like or
LargeL) and in some other vertebrate species (Grewal & Hewitt, 2002). The homologue gene
is positioned on the chromosome 11p11.2 of the human genome and it encodes 721 amino
acid protein which has 67% identity with LARGE, suggests that the two genes may have
risen by gene duplication. Like LARGE, it is also predicted to have two catalytic domains,
though it lacks the coiled-coiled motif present in the former protein. The hyperglycosylation
of α-dystroglycan by the overexpression of GYLTL1B increased its ability to bind laminin
and both the genes showed the same level of increase in laminin binding ability
(Brockington, et al., 2005).

3. Bioinformatics workflow and platform design

Many public databases and bioinformatics tools have been developed and are currently
available for use (Ding & Berleant, 2002). The primary goal of bioinformaticians is to
develop reliable databases and effective analysis tools capable of handling bulk amount of
biological data. But the objective of laboratory researchers is to study specific areas within
the life sciences, which requires only a limited set of databases and analysis tools. Thus the
existing free bioinformatics tools are sometimes too complicated for the biologists to choose.
One solution is to have an expert team who are familiar with both bioinformatics databases
and to know the needs of a research group in a particular field. The expert team will
recommend a workflow by using selected bioinformatics tools and databanks and also helps
the scientists with the complicated issue of tools and databases. Moreover, such a team
could organize large number of heterogeneous sources of biological information into a
specific, expertly annotated databank.
The team can also regularly and systematically update the information essential to help
biologists overcome the problems of integrating and keeping up-to-date with heterogeneous
biological information (Gerstein, 2000).
We have built a novel information management platform, LGTBase (Hyperlink).This
composite knowledge management platform includes the “LARGE-like GlcNAc Transferase
Database” by integrating specific public databases like CAZy database, and the workflow
analysis combined the usage of specific, public & designed bioinformatics tools to identify
the members of the LARGE-like protein family.

4. Tools and database selection

To analyze a novel protein family, biologists need to understand many different types of
information. Moreover, the speed of discovery in biology has been expanding exponentially
in recent years. So the biologists have to pick the right information available from the vast
resources available. To overcome these obstacles, a bioinformatics workflow can be
designed for analysing a specific protein family. In our study, a workflow was designed
based on the structure and characteristics of LARGE protein as shown in Figure 1 (Hwa et
al., 2007). The unknown DNA/protein sequences will be first identified as members of the
known gene families by using the Basic Local Alignment Search Tool (BLAST). The blastp
search tool is used to look for new LARGE-like proteins present in different organisms. The
researchers who wish to use our platform can obtain the protein sequences either from the
experimental data or through the blastp results. The search results were then analyzed with
In Silico Analysis of Golgi Glycosyltransferases: A Case Study on the LARGE-Like Protein Family 71

the following tools. To begin with, the sequences are searched for the aspartate-any residue-
aspartate (DXD) motif. The DXD motifs present in some glycosyltransferase families are
essential for its enzyme activity.

Fig. 1. Bioinformatics workflow of LGTBase.

The DXD motif prediction was then followed by the transmembrane domain prediction by
using the TMHMM program (version 2.0; Center for Biological Sequence Analysis, Technical
University of Denmark [http://www.cbs.dtu.dk/services/TMHMM-2.0/]). The
transmembrane domain is a characteristic feature of the Golgi enzymes.
The sequence motifs are then identified by MEME (Multiple Expectation-maximization for
Motif Elicitation) program (version 3.5.4; San Diego Supercomputer Center, UCSD
[http://meme.sdsc.edu/meme/]).
This program finds the motif-homology between the target sequence and other known
glycosyltransferases. In addition to all the above mentioned tools, the Pfam search (Sanger
Institute [http://www.sanger.ac.uk/Software/Pfam/search.shtml]) can also be used to find
the multiple sequence alignments and hidden Markov models in many existing protein
domains and families. The Pfam results will indicate what kind of protein family the peptide
belongs to. If it is a desired protein, investigators can then identify the evolutionary
relationships by using phylogenetic analysis.

4.1 LARGE-like GlcNAc transferase database

The specific annotation entries used in the LGTBase are currently being used in a
configuration that uses the information retrieved from several databases.
In CAZy database (Carbohydrate- Active enZymes) database ([http://afmb.cnrs-
mrs.fr/CAZY/]), the glycosyltransferases are classified as families, clans, and folds based on
their structural and sequence similarities, and also on their mechanistic investigation. The
other databases used in this platform were listed in Table 1.
72 Computational Biology and Applied Bioinformatics

Database Description Website

EntrezGene NCBI's repository for gene-specific http://www.ncbi.nlm.nih.gov/sites
information /entrez?db=gene
GenBank NIH genetic sequence database, an http://www.ncbi.nlm.nih.gov/sites
annotated collection of all publicly /entrez?db=nucleotide
available DNA sequences
Dictybase Database for model organism http://dictybase.org/
Dictyostelium discoideum
UniProtKB/S High-quality, manually annotated, http://www.uniprot.org/
wiss-Prot non-redundant protein sequence
database
InterPro Database of protein families, http://www.ebi.ac.uk/interpro/
domains and functional sites
MGI Database provides integrated genetic, http://www.informatics.jax.org/
genomic, and biological data of the
laboratory mouse
Ensembl It provides genome- annotation http://www.ensembl.org/index.htm
information l
HGMD Human Gene Mutation Database http://www.hgmd.cf.ac.uk/ac/inde
(HGMD) provides comprehensive x.php
data on human inherited disease
mutations
UniGene NCBI database of the transcriptome http://www.ncbi.nlm.nih.gov/unige
ne
GeneWiki The database transfers information http://en.wikipedia.org/wiki/Gene
on human genes to Wikipedia article _Wiki
TGDB Database with information about the http://www.tumor-
genes involved in cancers gene.org/TGDB/tgdb.html
HUGE The database provides the results of http://zearth.kazusa.or.jp/huge/
the Human cDNA project at the
Kazusa DNA Research Institute
RGD Database with collection of genetic http://rgd.mcw.edu/
and genomic information on the rat
OMIM Database provides information on http://www.ncbi.nlm.nih.gov/sites
human genes and genetic disorders. /entrez?db=omim
CGAP Information of gene expression http://cgap.nci.nih.gov/
profiles of normal, precancer, and
cancer cells.
PubMed Database with 20 million citations for http://www.ncbi.nlm.nih.gov/Pub
biomedical literature from medical Med/
journals, life science journals, related
books.
GO Representation of gene and gene http://www.geneontology.org/
product attributes across all species
Table 1. The information sources of LARGE-like GlcNAc Transferase Database
In Silico Analysis of Golgi Glycosyltransferases: A Case Study on the LARGE-Like Protein Family 73

All the information related to the LARGE-like protein family was retrieved from the
different biological databases. In order to confirm that the information obtained was
reliable, the data was scrutinized at two levels. First the information was selected from the
above mentioned biological databases with customized programs (using the perl compatible
regular expressions). Then the obtained information was annotated and validated by experts
in glycobiology and bioinformatics.
The annotated data in the LGTBase database was divided into nine categories (Figure 2).
The first category is related to genomic location, displays the chromosome, the cytogenetic
band and the map location of the gene. The second is related to aliases and descriptions,
displays synonyms and aliases for the relevant gene, and descriptions of its function,
cellular localization and effect on phenotype. The third category on proteins provides
annotated information about the proteins encoded by the relevant genes. The fourth is about
protein domains and families, provides annotated information about protein domains and
families and the fifth on protein function which provides annotated information about gene
function. The sixth category is related to pathways and interactions, provides links to
pathways and interactions followed by the seventh on disorders and mutations which
draws its information from OMIM and UniProt. The eighth category is on expression in
specific tissues, shows that the tissue expression values are available for a given gene. The
last category is about research articles, lists the references related to the proteins which are
studied. In addition, the investigator can also use DNA or protein sequences to assemble the
dataset for the analysis using this workflow.

Fig. 2. The contents of LGTBase database

4.2 LARGE-like GlcNAc transferase workflow

4.2.1 Reference sequences search
The unknown DNA/protein sequences are identified as members of the known gene
families using the Basic Local Alignment Search Tool (BLAST). BlastP is one of the BLAST
programs and it searches protein databases using a protein query. We used BlastP to look
for new LARGE-like proteins from different species and gathered the protein sequences of
74 Computational Biology and Applied Bioinformatics

LARGE like GlcNAc Transferases and built a protein database of ‘LARGE-like protein’. This
database would assist in search for more reference sequences of LARGE-like protein.

4.2.2 DXD motif search

In several glycosyltransferase families, the DXD motif is essential for the enzymatic activity
(Busch et al. 1998). So we first searched for aspartate-any residue-aspartate (DXD) motif,
commonly found in glycosyltransferase. Therefore, the ‘DXD Motif Search’ tool was
designed. The input protein sequences are loaded or pasted in this tool and the results
indicate the presence or absence of DXD motif.

4.2.3 Transmembrane helices search

The LARGE protein is a member of the N-acetylglucosaminyltransferase family. The
presence of transmembrane domain is a characteristics feature of this family. TMHMM
program is used to predict the transmembrane helices based on the hidden Markov model.
The prediction gives the most probable location and orientation of transmembrane helices in
the sequence. TMHMM can predict the location of transmembrane alpha helices and the
location of intervening loop regions. This program also predicts the location of the loops
that are present between the helices either inside or outside of the cell or organelle. The
program is designed based on a 20 amino acids long alpha helix which contains
hydrophobic amino acids that can span through a cell membrane.

4.2.4 MEME analysis

A motif is a sequence pattern that occurs repeatedly in a group of related protein or DNA
sequences. MEME (Multiple Expectation-maximization for Motif Elicitation) represents
motifs as position-dependent letter-probability matrices which describe the probability of
each possible letter at each position in the pattern. The program can search for homologous
sequences among the input protein sequences.

4.2.5 Protein families search

The Pfam HMM search was used to identify the protein family to which the input protein
sequences belong. The Pfam database contains the information about most of the protein
domains and families. The results from the Pfam HMM search will show the relation of
input protein sequences with the existing protein families and domains.

4.2.6 Phylogenetic analysis

The phylogenetic analysis was performed to find any significant evolutionary relationship
between the new protein sequences and the LARGE protein family and to support our
previous findings. ClustalW, a multiple alignment program which aligns two or more
sequences to determine any significant consensus sequences between them (Thompson et
al., 1994). This approach can also be used for searching patterns in the sequence. The
phylogenetic tree was constructed by using PHYLIP program (v.3.6.9) and viewed by
Treeview software (v.1.6.6). In GlcNAc-transferase phylogenetic analysis, once the multiple
alignment of all GlcNAc-transferase has been done, it can be used to construct the
phylogenetic tree. About 25 protein sequences were identified as the LARGE-like protein
family. By using the neighbor joining distance method, the phylogenetic tree showed that
these proteins can be divided into 6 groups (Figure 3). The evolutionary history inferred
In Silico Analysis of Golgi Glycosyltransferases: A Case Study on the LARGE-Like Protein Family 75

from phylogenetic analysis is usually depicted as branching, tree-like diagrams which

represents an estimated pedigree of the inherited relationships among the protein sequences
from different species. These evolutionary relationships can be viewed either as Cladograms
(Chenna et al., 2003) or Phylograms (Radomski & Slonimski, 2001).

Fig. 3. Phylogenetic tree of LARGE-like Protein Family

4.3 Organization of the LGTBase platform

The data obtained from the analyses were stored in a MySQL relational database and the
web interface was built by using PHP and CGI/Java scripts. According to the characteristics
of LARGE-like GlcNAc transferase proteins, the workflow was designed and developed by
using Java language and several open source bioinformatics programs. Tools with different
languages, C, perl, java were integrated by using Java language (Figure 4). Adjustable
parameters of the tools were reserved to fulfill the needs in future.

5. Application with LARGE protein family

A protein sequence (fasta format) can be entered into the BlastP assistant interface, enabling
the other known proteins with similar sequences to be identified (Figure 5). The investigator
can select all the resulting sequences or use only some of them. The data can then be
transferred to the DXD analysis page (Figure 6). The rationale behind choosing the DXD
analysis was since they are represented in many families of glycosyltransferases and it will
be easy to narrow down the analysis of putative protein sequences to particular protein
families or domains. There were many online tools available for the identification and
76 Computational Biology and Applied Bioinformatics

characterization of unknown protein sequences. So depending upon the target protein of

study, one can pick the tools to characterize it.

Fig. 4. Database selected for construction of the knowledge management platform

The sequences are analyzed with the DXD motif search tool (Figure 6), which selects those
sequences containing the DXD motif for the TMHMM analysis. The transmembrane helices
can be predicted with TMHMM analysis (Figure 7). The transmembrane domains are
predicted by the hydrophobic nature of the proteins and mainly used to identify the cellular
location of the proteins. Similar to transmembrane domain prediction, there were several
other domains that can be predicted based on the protein’s characters like hydrophobic,
hydrophilic etc., The dataset containing DXD motifs and transmembrane helices are then
selected for MEME (Figure 8) and Pfam analysis (Figure 9). Some sequence motifs occur
repeatedly in the data set and are conjectured to have a biological significance are predicted
by MEME analysis. This application plays a significant role in characterization of the
putative protein sequences after the initial studies with the DXD motif, transmembrane
domain, and other tools. This tool can be used for all kind of protein sequences since its
prediction is based on the pattern of sequences present in the study. The protein sequences
in the dataset can be identified to the known protein families by Pfam analysis. The pfam
classification can also be used for almost all the putative protein sequences because of its
large collection of protein domain families represented by multiple sequence alignments
and Hidden Markov Models. After the MEME and Pfam analysis were done, ClustalW and
Phylip programs were used for Phylogenetic Analysis (Figure 9) to see the evolutionary
relationship among the data sets (Figure 10). Finally, these results can be used to design
experiments to be performed in the laboratory.
In Silico Analysis of Golgi Glycosyltransferases: A Case Study on the LARGE-Like Protein Family 77

Fig. 5. The BlastP tool of the LGTBase platform to find similar sequences to LARGE
78 Computational Biology and Applied Bioinformatics

Fig. 6. DXD motif search tool of the LGTBase platform for DXD motif prediction

Fig. 7. TMHMM analysis tool of the LGTBase platform for Transmembrane domain
prediction
In Silico Analysis of Golgi Glycosyltransferases: A Case Study on the LARGE-Like Protein Family 79

Fig. 8. MEME analysis tool of the LGTBase platform to predict the sequence motifs

Fig. 9. Pfam analysis tool of the LGTBase platform to identify the known protein family of
the target protein which is studied
80 Computational Biology and Applied Bioinformatics

Fig. 10. Phylogenetic analysis tool of the LGTBase platform to study the evolutionary
relationship of the target protein

6. Future direction
We have described how to construct a computational platform to analyze the LARGE
protein family. Since the platform was built based on several commonly shared protein
domains and motifs, it can also be modified for analyzing other golgi glycosyltransferases.
Furthermore, the phylogenetic analysis (Figure 3) revealed that LARGE protein family is
related to β-1,3-N-acetylglucosaminyltransferase 1 (β3GnT). β3GnT (EC 2.4.1.149) is a group
of enzymes belong to glycosyltransferases family. Some β3GnT enzymes catalyze the
transfer of GlcNAc from UDP-GlcNAc to Gal in the Galβ1-4 Glc(NAc) structure with β-1,3
linkage. These enzymes were grouped into GT family 31, 49 in the CAZy database. The
enzyme uses 2 substrates namely, UDP-N-acetyl-D-glucosamine and D-galactosyl-β-1,4-N-
acetyl-D-glucosaminyl-R and the products are formed as UDP, N-acetyl-β-D- glucosaminyl-
In Silico Analysis of Golgi Glycosyltransferases: A Case Study on the LARGE-Like Protein Family 81

β-1,3 -D-galactosamine. These enzymes participate in the formation of keratan sulfate,

glycosphingolipid biosynthesis, neo-lacto series and N-linked glycans. There are currently 9
members known from the β3GnT family.
The β3GnT1 (iGnT) was the first enzyme to be isolated when cDNA of a human β-1,3-N-
acetylglucosaminyltransferase essential for poly-N-acetyllactosamine synthesis was studied
(Zhou et al., 1999). The poly-N-acetyllactosamine synthesized by iGnT provides critical
backbone structure for the addition of functional oligosaccharides such as Sialyl Lewis X. It
has been reported recently that β3GnT1 is involved in attenuating prostate cancer cell
locomotion by regulating the synthesis of laminin-binding glycans on α-DG (Bao et al.,
2009). Since there are several common shared domains similar to the LARGE protein, the
new platform for β3GnT protein family can be constructed based on the original platform.
Apart from β3GnT1, β3GnT2 enzyme is responsible for elongation of poly-lactosamine
chains. This enzyme was isolated based on structural similarity with the β3GalT family.
Studies showed that on a panel of invasive and noninvasive fresh transitional cell
carcinomas (TCCs) showed strong down regulation of β3GnT2 in the invasive lesions,
suggesting that a decline in the expression levels of some members of the
glycosyltransferase (Gromova et al., 2001).
The β3GnT3 and β3GnT4 enzymes were subsequently isolated based on the structural
similarity with β3GalT family. β3GnT3 is a type II transmembrane protein and contains a
signal anchor that is not cleaved. It prefers the substrates of lacto-N-tetraose and lacto-N-
neotetraose, and it is also involved in the biosynthesis of poly-N-acetyllactosamine chains
and the biosynthesis of the backbone structure of dimeric sialyl Lewis A. It plays dominant
role in the L-selectin ligand biosynthesis, lymphocyte homing and lymphocyte trafficking.
The β3GnT3 enzyme is highly expressed in the non-invasive colon cancer cells. β3GnT4 is
involved in the biosynthesis of poly-N-acetyllactosamine chains and prefers lacto-N-
neotetraose as the substrate. It is a type II transmembrane protein and it is expressed more
in bladder cancer cells (Shiraishi et al., 2001). β3GnT5 is responsible for
lactosyltriaosylceramide synthesis, an essential component of lacto/neolacto series
glycolipids (Togayachi et al., 2001 ). The expression of the HNK-1 and Lewis x antigens on
the lacto/neo-lacto-series of glycolipids is developmentally and tissue-specifically regulated
by β3GnT5. The overexpression of β3GnT5 in human gastric carcinoma cell lines led to
increased sialyl-Lewis X expression and increased H.pylori adhesion (Marcos et al., 2008).
The β3GnT6 synthesizes the core 3 O-glycan structure and speculates that this enzyme plays
an important role in the synthesis and function of mucin O-glycan in the digestive organs. In
addition, the expression of β3GnT6 was markedly down regulated in gastric and colorectal
carcinomas (Iwai et al., 2005). Expression of β3GnT7 has been reported to be down-
regulated upon malignant transformation (Kataoka et al., 2002). Elongation of the
carbohydrate backbone of keratan sulfate proteoglycan is catalyzed by β3GnT7 and β1,4-
galactosyltransferase 4 (Hayatsu et al., 2008). β3GnT7 can transfer GlcNAc to Gal to
synthesize a polylactosamine chain with each enzyme differing in its acceptor molecule
preference. The polylactosamine and related structures plays crucial role in cell-cell
interaction, cell-extracellular matrix interaction, immune response and determining
metastatic capacity. The β3GnT8 enzyme extends a polylactosamine chain specifically on a
tetraantennary N-glycans. β3GnT8 transfers GlcNAc to the non-reducing terminus of the
82 Computational Biology and Applied Bioinformatics

Galβ1-4GlcNAc of tetra antennary N-glycan in vitro. Intriguingly, β3GnT8 is significantly

upregulated in colon cancer tissues than in normal tissue (Ishida et al., 2005). The co-
transfection of β3GnT8 and β3GnT2 resulted in synergistic enhancement of the activity of
the polylactosamine synthesis. This indicates that these two enzymes interact and
complement each other’s function in the cell. As a summary, the members of the β3GnT
protein family are important in human cancer biology.
Our initial motif analysis showed that there are 3 important functional domains predicted
are commonly found among the β3GnT enzymes. The first motif is a structural motif
necessary for maintaining the protein fold. The second, DXD motif represented in many
glycosyltransferases is involved in the binding of the nucleotide-sugar donor substrate, both
directly and indirectly through coordination of metal ions such as magnesium or manganese
in the active site. A glycine-rich loop is the third motif found at the bottom of the active site
cleft. This loop is likely to play a role in the recognition of both the GlcNAc portion of the
donor and the substrate. Since the three common domains of β3GnT are similar to the
LARGE protein family, it is feasible to modify the current LARGE platform to analyze other
golgi glycosyltransferases such as β3GnT.

7. References
Bao, X., Kobayashi, M., Hatakeyama, S., Angata, K., Gullberg, D., Nakayama, J., Fukuda,
M.N. & Fukuda, M. (2009). Tumor suppressor function of laminin-binding
α-dystroglycan requires a distinct β-3-N-acetylglucosaminyltransferase. Proceedings
of the National Academy of Sciences USA, Vol.106, No.29, (July 2009),
pp. 12109-12114
Barresi, R., Michele, D.E., Kanagawa, M., Harper, H.A., Dovico, S.A., Satz, J.S., Moore, S.A.,
Zhang, W., Schachter, H., Dumanski, J.P., Cohn, R.D., Nishino, I. & Campbell, K.P.
(2004). LARGE can functionally bypass alpha-dystroglycan glycosylation defects in
distinct congenital muscular dystrophies. Nature Medicine, Vol.10, No.7, (July 2004),
pp. 696-703.
Braun, S. (2004). Naked plasmid DNA for the treatment of muscular dystrophy. Current
Opinion in Molecular Therapeutics, Vol.6, (October 2004), pp. 499-505.
Brockington, M., Torelli, S., Prandini, P., Boito, C., Dolatshad, N.F., Longman, C., Brown,
S.C., Muntoni, F. (2005). Localization and functional analysis of the LARGE family
of glycosyltransferases: significance for muscular dystrophy. Human Molecular
Genetics, Vol.14, No.5, (March 2005), pp. 657-665.
Busch, C., Hofmann, F., Selzer, J., Munro, S., Jeckel, D. & Aktories, K. (1998). A common
motif of eukaryotic glycosyltransferases is essential for the enzyme activity of large
clostridial cytotoxins. Journal of Biological Chemistry, Vol.273, No.31, (July 1998),
pp.19566-19572.
Cantarel, B.L., Coutinho, P.M., Rancurel, C., Bernard, T., Lombard, V. & Henrissat, B.
(2009) The Carbohydrate-Active EnZymes database (CAZy): an expert
resource for Glycogenomics. Nucleic Acids Research, Vol. 37, (January 2009), pp.
D233-238
In Silico Analysis of Golgi Glycosyltransferases: A Case Study on the LARGE-Like Protein Family 83

Chenna, R., Sugawara, H., Koike, T., Lopez, R., Gibson, T.J. & Higgins, D.G. Thompson JD
(2003). Multiple sequence alignment with the Clustal series of programs. Nucleic
Acids Res. Vol.31, No.13, (July 2003), pp. 3497-3500.
Coutinho, P.M., Deleury, E., Davies, G.J. & Henrissat, B. (2003). An evolving hierarchical
family classification for glycosyltransferases. Journal of Molecular Biology, Vol.328,
(April 2003), pp. 307-317.
Ding, J., Berleant, D., Nettleton, D. & Wurtele, E. (2002). Mining MEDLINE: abstracts,
sentences, or phrases? Pacific Symposium on Biocomputing, Vol.7, pp. 326-337.
Dumanski, J.P., Carlbom, E., Collins, V.P., Nordenskjold, M. (1987). Deletion mapping of a
locus on human chromosome 22 involved in the oncogenesis of meningioma.
Proceedings of the National Academy of Sciences USA, Vol.84, (December 1987), pp.
9275-9279.
Durbeej, M., Henry, M.D. & Campbell, K.P. (1998). Dystroglycan in development and
disease. Current Opinions in Cell Biology Vol. 10, (October 1998), pp. 594-601.
Fujimura, K., Sawaki, H., Sakai, T., Hiruma, T., Nakanishi, N., Sato, T., Ohkura, T.,
Narimatsu, H. (2005). LARGE2 facilitates the maturation of a-dystroglycan more
effectively than LARGE. Biochemical and Biophysical Research Communications,
Vol.329, No.3, (April 2005), pp. 1162-1171
Fukuda, M., Hindsgaul, O., Hames, B.D. & Glover, D.M. (1994). In Molecular Glycobiology,
Oxford Univ. Press, Oxford.
Fukuda, M., & Hindsgaul, O. (2000). Molecular and Cellular Glycobiology (2nd ed.), Oxford
Univ. Press, Oxford.
Gee, S.H., Montanaro, F., Lindenbaum, M.H., and Carbonetto, S. (1994). Dystroglycan-α, a
dystrophin-associated glycoprotein, is a functional agrin receptor. Cell, Vol.77,
(June 1994), pp. 675-686.
Gerstein, M. (2000). Integrative database analysis in structural genomics. Nature Structural
Biology, Vol.7, (November 2000), Suppl: 960-963.
Grewal, K., Holzfeind, P.J., Bittner, R.E. & Hewitt, J.E., (2001). Mutant glycosyltransferase
and altered glycosylation of alpha-dystroglycan in the myodystrophy mouse.
Nature Genetics, Vol.28, (June 2001), pp.151-154.
Grewal, P.K. & Hewitt, J.E. (2002). Mutation of Large, which encodes a putative
glycosyltransferase, in an animal model of muscular dystrophy. Biochimica et
Biophysica Acta, Vol.1573, (December 2002), pp. 216-224.
Gromova, I., Gromov, P. & Celis J.E. (2001). A Novel Member of the Glycosyltransferase
Family, β3GnT2, highly down regulated in invasive human bladder Transitional
Cell Carcinomas. Molecular Carcinogenesis, Vol. 32, No. 2, (October 2001), pp.
61-72
Hayatsu, N., Ogasawara, S., Kaneko, M.K., Kato, Y. & Narimatsu, H. (2008). Expression of
highly sulfated keratan sulfate synthesized in human glioblastoma cells. Biochemical
and Biophysical Research Communications, Vol. 368, No. 2, (April 2008), pp. 217-222
Hwa, K.Y., Pang, T.L. & Chen, M.Y. (2007). Classification of LARGE-like GlcNAc-
Transferases of Dictyostelium discoideum by Phylogenetic Analysis. Frontiers in the
Convergence of Bioscience and Information Technologies, pp. 289-293.
84 Computational Biology and Applied Bioinformatics

Ishida, H., Togayachi, A., Sakai, T., Iwai, T., Hiruma, T., Sato, T., Okubo, R., Inaba, N., Kudo,
T., Gotoh, M., Shoda, J., Tanaka, N., & Narimatsu, H. A novel beta1,3-N-
acetylglucosaminyltransferase (beta3Gn-T8), which synthesizes poly-N-
acetyllactosamine, is dramatically upregulated in colon cancer. FEBS Letters.
(January 2005), Vol. 579, No.1, pp. 71-78.
Ishida, H., Togayachi, A., Sakai, T., Iwai, T., Hiruma, T., Sato, T., Okubo, R., Inaba, N., Kudo,
T., Gotoh, M., Shoda, J., Tanaka, N. & Narimatsu, H. (2005). A novel beta1,3-N-
acetylglucosaminyltransferase (beta3Gn-T8), which synthesizes poly-N-
acetyllactosamine, is dramatically upregulated in colon cancer. FEBS Letters,
Vol.579, No.1, (January 2005), pp. 71-8.
Iwai, T., Kudo, T., Kawamoto, R., Kubota, T., Togayachi, A., Hiruma, T., Okada, T.,
Kawamoto, T., Morozumi, K. & Narimatsu, H. (2005). Core 3 synthase is down-
regulated in colon carcinoma and profoundly suppresses the metastatic potential of
carcinoma cells. Proceedings of the National Academy of Sciences USA, Vol.102, No.12,
(March 2005), pp. 4572-4577
Kanagawa, M., Michele, D.E., Satz, J.S., Barresi, R., Kusano, H., Sasaki, T., Timpl, R., Henry,
M. D., and Campbell, K.P. (2005). Disruption of Perlecan Binding and Matrix
Assembly by Post-Translational or Genetic Disruption of Dystroglycan Function.
FEBS Letters, Vol.579, No.21, (August 2005), pp. 4792-4796.
Kataoka, K. & Huh, N.H. (2002). A novel β1,3-N-acetylglucosaminyltransferase involved in
invasion of cancer cells as assayed in vitro. Biochemical and Biophysical Research
Communications, Vol. 294, No.4, (June 2002), pp. 843-848
Lane, P.W., Beamer, T.C. & Myers, D.D. (1976). Myodystrophy, a new myopathy on
chromosome 8 of the mouse. Journal of Heredity, Vol. 67, No.3 (May-June 1976), pp.
135-138.
Longman, C., Brockington, M., Torelli, S., Jimenez-Mallebrera, C., Kennedy, C., Khalil, N.,
Feng, L., Saran, R.K., Voit, T., Merlini, L., Sewry, C.A., Brown, S.C. & Muntoni F.
(2003). Mutations in the human LARGE gene cause MDC1D, a novel form of
congenital muscular dystrophy with severe mental retardation and abnormal
glycosylation of alpha dystroglycan. Human Molecular Genetics, Vol.12, No.21,
(November 2003), pp. 2853-2861.
Marcos, N.T., Magalhães, A., Ferreira, B., Oliveira, M.J., Carvalho, A.S., Mendes, N.,
Gilmartin, T., Head, S.R., Figueiredo, C., David, L., Santos-Silva, F. & Reis, C.A.
(2008). Helicobacter pylori induces β3GnT5 in human gastric cell lines, modulating
expression of the SabA ligand Sialyl-Lewis X. Journal of Clinical Investigation, Vol.
118, No. 6, (June 2008), pp.2325-2336
Narimatsu, H. (2006). Human glycogene cloning: focus on beta 3-glycosyltransferase and
beta 4-glycosyltransferase families. Current Opinions in Structural Biology. Vol.16,
No.5, (October 2006), pp. 567-575.
Newman, E.A. & Frishman, L.J. (1991). The b-wave. In Arden, G.B. (ed.), Principles and
Practice of Clinical Electrophysiology of Vision, Mosby-Year Book, St Louis, MO.
Peng, H.B., Ali, A.A., Daggett, D.F., Rauvala, H., Hassell, J.R., & Smalheiser, N.R. (1998). The
relationship between perlecan and dystroglycan and its implication in the
In Silico Analysis of Golgi Glycosyltransferases: A Case Study on the LARGE-Like Protein Family 85

formation of the neuromuscular junction. Cell Adhesion and Communication, Vol.5,

No.6, (September 1998), pp. 475-489
Peyrard, M., Seroussi, E., Sandberg-Nordqvist, A.C., Xie, Y.G., Han, F.Y.,Fransson, I.,
Collins, J., Dunham, I., Kost-Alimova, M., Imreh, S.,Dumanski, J.P., (1999). The
human LARGE gene from 22q12.3-q13.1 is a new, distinct member of the
glycosyltransferase gene family. Proceedings of the National Academy of Sciences USA,
Vol.96, No.2, (January 1999), pp. 589-603.
Radomski, J.P. & Slonimski, P.P. (2001). Genomic style of proteins: concepts, methods and
analyses of ribosomal proteins from 16 microbial species. FEMS Microbiol Reviews,
Vol.25, No.4, (August 2001), pp. 425-435.
Sasaki, K., Kurata-Miura, K., Ujita, M., Angata, K., Nakagawa, S., Sekine, S., Nishi, T. &
Fukuda, M. (1997). Expression cloning of cDNA encoding a human beta-1,3-N-
acetylglucosaminyl transferase that is essential for poly-N-acetyllactosamine
synthesis. Proceedings of the National Academy of Sciences USA, Vol.94, No.26,
(December 1997), pp. 14294-14299.
Shiraishi, N., Natsume, A., Togayachi, A., Endo, T., Akashima, T., Yamada, Y., Imai, N.,
Nakagawa, S., Koizumi, S., Sekine, S., Narimatsu, H. & Sasaki K. (2001).
Identification and characterization of 3 novel β1,3-N-
Acetylglucosaminyltransferases. Structurally Related to the β1,3-
Galactosyltransferase family. The Journal of Biological Chemistry, Vol. 276, No.5,
(February 2001), pp. 3498-3507
Smalheiser, N. R., and Schwartz, N. B. (1987) Cranin: a laminin-binding protein of cell
membranes. Proceedings of the National Academy of Sciences USA, Vol.84, No.18,
(September 1987), pp. 6457-6461.
Sugita, S., Saito, F., Tang, J., Satz, J., Campbell, K., & Sudhof, T.C. (2001). A stoichiometric
complex of neurexins and dystroglycan in brain. Journal of Cell Biology, Vol.154,
No.2, (July 2001), pp. 435-445
Thompson, J.D., Higgins, D.G. & Gibson, T.J. (1994). CLUSTAL W: improving the sensitivity
of progressive multiple sequence alignment through sequence weighting, position-
specific gap penalties and weight matrix choice. Nucleic Acids Research, Vol.22,
No.22, (November 1994), pp. 4673-4680.
Togayachi, A., Akashima, T., Ookubo, R., Kudo, T., Nishihara, S., Iwasaki, H., Natsume, A.,
Mio, H. Inokuchi J. and T. Irimura et al., Molecular cloning and characterization of
UDP-GlcNAc: Lactosylceramide β1,3-N-acetylglucosaminyltransferase (β3Gn-T5),
an essential enzyme for the expression of HNK-1 and Lewis X epitopes on
glycolipids, Journal of Biological Chemistry, Vol. 276, No.5, (March 2001), pp. 22032–
22040
van Reeuwijk, J., Brunner, H.G., van Bokhoven, H. (2005). Glyc-O-genetics of
Walker-Warburg syndrome. Clinical Genetics, Vol. 67, No.4, (April 2005), pp. 281-
289.
Varki, A., Cummings, R.D., Esko, J.D., Freeze, H.H., Stanley, P., Bertozzi, C.R., Hart, G.W. &
Etzler, M.E. (2008). Essentials of Glycobiology, (2nd ed.) Plainview (NY): Cold Spring
Harbor Laboratory Press
86 Computational Biology and Applied Bioinformatics

Zhou, D., Dinter, A., Gutiérrez Gallego, R., Kamerling, J.P., Vliegenthart, J.F., Berger, E.G. &
Hennet, T. (1999). A β-1,3-N-acetylglucosaminyltransferase with poly-N-
acetyllactosamine synthase activity is structurally related to β-1,3-
galactosyltransferases. Proceedings of the National Academy of Sciences USA, Vol. 96,
No. 2, (January 1999), pp. 406-411
 5

MicroArray Technology - Expression Profiling of

MRNA and MicroRNA in Breast Cancer
Aoife Lowery, Christophe Lemetre, Graham Ball and Michael Kerin
1Department of Surgery, National University of Ireland Galway,
2John Van Geest Cancer Research Centre, School of Science & Technology,
Nottingham Trent University, Nottingham
1Ireland
2UK

1. Introduction
Breast cancer is the most common form of cancer among women. In 2009, an estimated
194,280 new cases of breast cancer were diagnosed in the United States; breast cancer was
estimated to account for 27% of all new cancer cases and 15% of cancer-related mortality in
women (Jemal et al, 2009). Similarly, in Europe in 2008, the disease accounted for some 28%
and 17% of new cancer cases and cancer-related mortality in women respectively (Ferlay et
al, 2008). The increasing incidence of breast cancer worldwide will result in an increased
social and economic burden; for this reason there is a pressing need from a health and
economics perspective to develop and provide appropriate, patient specific treatment to
reduce the morbidity and mortality of the disease. Understanding the aetiology, biology and
pathology of breast cancer is hugely important in diagnosis, prognostication and selection of
primary and adjuvant therapy. Breast tumour behaviour and outcome can vary
considerably according to factors such as age of onset, clinical features, histological
characteristics, stage of disease, degree of differentiation, genetic content and molecular
aberrations. It is increasingly recognised that breast cancer is not a single disease but a
continuum of several biologically distinct diseases that differ in their prognosis and
response to therapy (Marchionni et al, 2008; Sorlie et al, 2001). The past twenty years has
seen significant advances in breast cancer management. Targeted therapies such as
hormonal therapy for estrogen receptor (ER) positive breast tumours and trastuzumab for
inhibition of HER2/neu signalling have become an important component of adjuvant
therapy and contributed to improved outcomes (Fisher et al, 2004; Goldhirsch et al, 2007;
Smith et al, 2007). However, our understanding of the molecular basis underlying breast
cancer heterogeneity remains incomplete. It is likely that there are significant differences
between breast cancers that reach far beyond the presence or absence of ER or HER2/neu
amplification. Patients with similar morphology and molecular phenotype based on ER, PR
and HER2/neu receptor status can have different clinical courses and responses to therapy.
There are small ER positive tumours that behave aggressively while some large high grade
ER negative, HER2/neu receptor positive tumours have an indolent course. ER-positive
tumours are typically associated with better clinical outcomes and a good response to
88 Computational Biology and Applied Bioinformatics

hormonal therapies such as tamoxifen (Osborne et al, 1998). However, a subset of these
patients recur and up to 40% develop resistance to hormonal therapy (Clarke et al, 2003).
Furthermore, clinical studies have shown that adding adjuvant chemotherapy to tamoxifen
in the treatment of node negative, ER positive breast cancer improves disease outcome
(Fisher et al, 2004). Indeed, treatment with tamoxifen alone is only associated with a 15%
risk of distant recurrence, indicating that 85% of these patients would do well without, and
could be spared the cytotoxic side-effects of adjuvant chemotherapy.
The heterogeneity of outcome and response to adjuvant therapy has driven the discovery of
further molecular predictors. Particular attention has focused on those with prognostic
significance which may help target cancer treatment to the group of patients who are likely
to derive benefit from a particular therapy. There has been a huge interest in defining the
gene expression profiles of breast tumours to further understand the aetiology and progression
of the disease in order to identify novel prognostic and therapeutic markers. The sequencing
of the human genome and the advent of high throughput molecular profiling has facilitated
comprehensive analysis of transcriptional variation at the genomic level. This has resulted in
an exponential increase in our understanding of breast cancer molecular biology. Gene
expression profiling using microarray technology was first introduced in 1995 (Schena et al,
1995). This technology enables the measurement of expression of tens of thousands of
mRNA sequences simultaneously and can be used to compare gene expression within a
sample or across a number of samples. Microarray technology has been productively
applied to breast cancer research, contributing enormously to our understanding of the
molecular basis of breast cancer and helping to achieve the goal of individualised breast
cancer treatment. However as the use of this technology becomes more widespread, our
understanding of the inherent limitations and sources of error increases. The large amount
of data produced from such high throughput systems has necessitated the use of complex
computational tools for management and analysis of this data; leading to rapid
developments in bioinformatics.
This chapter provides an overview of current gene expression profiling techniques, their
application to breast cancer prognostics and the bioinformatic challenges that must be
overcome to generate meaningful results that will be translatable to the clinical setting. A
literature search was performed using the PubMed database to identify publications
relevant to this review. Citations from these articles were also examined to yield further
relevant publications.

2. Microarray technology – principles & technical considerations

2.1 High throughput genomic technology
There are a multitude of high throughput genomic approaches which have been developed
to simultaneously measure variation in thousands of DNA sequences, mRNA transcripts,
peptides or metabolites:
• DNA microarray measures gene expression
• Microarray comparative genomic hybridisation (CGH) measures genomic gains and
losses or identifies differences in copy number for genes involved in pathological states
(Oosterlander et al, 2004)
• Single nucleotide polymorphism (SNP) microarray technology (Huang et al, 2001) has
been developed to test for genetic aberrations that may predispose an individual to
disease development.
MicroArray Technology - Expression Profiling of MRNA and MicroRNA in Breast Cancer 89

• CpG arrays (Yan et al , 2000) can be used to determine whether patterns of specific
epigenetic alterations correlate with pathological parameters.
• Protein microarrays (Stoll et al, 2005) consisting of antibodies, proteins, protein
fragments, peptides or carbohydrate elements, are used to detect patterns of protein
expression in diseased states.
• ChIP-on-chip (Oberley et al, 2004) combines chromatin immunoprecipitation (ChIP)
with glass slide microarrays (chip) to detect how regulatory proteins interact with the
genome.
All of these approaches offer unique insights into the genetic and molecular basis of disease
development and progression.
This chapter focuses primarily on gene expression profiling and cDNA microarrays,
however many of the issues raised, particularly in relation to bioinformatics are also
applicable to the other “-omic” technologies.
Gene expression which is a measurement of gene “activity” can be determined by the
abundance of its messenger RNA (mRNA) transcripts or by the expression of the protein
which it encodes. ER, PR and HER2/neu receptor status are determined in clinical practice
using immunohistochemistry (IHC) to quantitate protein expression or fluorescence in situ
hybridisation (FISH) to determine copy number. These techniques are semi-quantitative and
are optimal when determining the expression of individual or a small number of genes.
Microarray technology is capable of simultaneously measuring the expression levels of
thousands of genes in a biological sample at the mRNA level. The abundance of individual
mRNA transcripts in a sample is a reflection of the expression levels of corresponding genes.
When a complementary DNA (cDNA) mixture reverse transcribed from the mRNA is
labelled and hybridised to a microarray, the strength of the signal produced at each address
shows the relative expression levels of the corresponding gene.
cDNA microarrays are miniature platforms containing thousands of DNA sequences which
act as gene specific probes, immobilised on a solid support (nylon, glass, silicon) in a parallel
format. They are reliant on the complementarity of the DNA duplex i.e. reassembly of
strands with base pairing A to T and C to G which occurs with high specificity. There are
microarray platforms available containing bound librarys of oligonucleotides representing
literally all known human genes e.g. Affymetrix GeneChip (Santa Clara, CA), Agilent array
(Santa Clara, CA), Illumina bead array (San Diego, CA). When fluorescence-labelled cDNA
is hybridised to these arrays, expression levels of each gene in the human genome can be
quantified using laser scanning microscopes. These microscopes measure the intensity of the
signal generated by each bound probe; abundant sequences generate strong signals and rare
sequences generate weaker signals. Despite differences in microarray construction and
hybridization methodologies according to manufacturing, microarray-based measurements
of gene expression appear to be reproducible across a range of different platforms when the
same starting material is used, as demonstrated by the MicroArray Quality Control project
(Shi et al, 2006).

2.2 Experimental approach

There are experimental design and quality control issues that must be considered when
undertaking a microarray experiment. The experiment should be designed appropriately to
answer a specific question and samples must be acquired from either patients or cultured
cells which are appropriate to the experimental setup. If the aim of a microarray experiment
90 Computational Biology and Applied Bioinformatics

is to identify differentially expressed genes between two groups of samples i.e.

“experiment” and “control”, it is critical that the largest source of variation results from the
phenotype under investigation (e.g. patient characteristic or treatment). The risk of
confounding factors influencing the results can be minimised by ensuring that the groups of
samples being compared are matched in every respect other than the phenotype under
investigation. Alternatively, large sample numbers can be used to increase the likelihood
that the experimental variable is the only consistent difference between the groups.
For a microarray experiment, fresh frozen tissue samples are required which have been
snap-frozen in liquid nitrogen or collected in an RNARetain™ or RNA LaterTM solution to
preserve the quality of the RNA. Formalin-fixed and paraffin embedded tissue samples are
generally unsuitable for microarray studies as the RNA in the sample suffers degradation
during tissue processing (Cronin et al, 2004; Masuda et al, 1999, Paik et al, 2005).
Due to the omnipresence of ribonucleases and the inherent instability of RNA, it is essential to
measure the integrity of RNA after extraction. Only samples of the highest integrity should be
considered for reverse transcription to cDNA and hybridisation to the microarray platform
(figure 1). Once obtained, intensity readings must be background adjusted and transformed;
this data is then normalised and analysed and results are generally interpreted according to
biological knowledge. The success of microarray experiments is highly dependent on
replication. Technical replication refers to the repeated assaying of the same biological sample
to facilitate quality assessment. Even more important is biological replication on larger sample
sets. The accuracy of microarray expression measurements must be confirmed using a reliable
independent technology, such as real-time quantitative PCR, and validated on a larger set of
independent biological samples. It is independent validation studies that determine the
strength or clinical relevance of a gene expression profile.

Fig. 1. The steps involved in a cDNA microarray experiment

MicroArray Technology - Expression Profiling of MRNA and MicroRNA in Breast Cancer 91

3. Molecular profiling – unlocking the heterogeneity of breast cancer

Breast cancer researchers were quick to adopt high throughput microarray technology,
which is unsurprising considering the opportunity it provides to analyse thousands of
genes simultaneously.

3.1 Class discovery

Microarray studies can be used in three different manners;
• class comparison
• class prediction
• class discovery (Simon et al, 2003)
All of these approaches have been applied to the study of breast cancer.
Class discovery involves analyzing a given set of gene expression profiles with the goal of
discovering subgroups that share common features. The early gene expression profiling
studies of breast cancer (Perou et al, 2000; Sorlie et al, 2001) were class discovery studies.
Researchers used an unsupervised method of analysis, in which tumours were clustered
into subgroups by a 496-gene “intrinsic” gene set that reflects differences in gene expression
between tumours without using selection criteria. The tumour subtype groupings consist of
luminal like subtypes which are predominantly ER and PR positive, basal-like subtypes
which are predominantly triple negative for ER, PR and HER2/neu, HER2/neu-like
subtypes which have increased expression of the HER2/neu amplicon and a normal-like
subtype (Perou et al, 2000). Subsequent studies from the same authors, on a larger cohort of
patients with follow-up data showed that the luminal subgroup could be further subdivided
into at least two groups, and that these molecular subtypes were actually associated with
distinct clinical outcomes (Sorlie et al 2001). These molecular subtypes of breast cancer have
been confirmed and added to in subsequent microarray datasets (Hu et al, 2006; Sorlie et al,
2003; Sotiriou et al, 2003). Given the importance of the ER in breast cancer biology, it is not
surprising that the most striking molecular differences were identified between the ER-
positive (luminal) and ER-negative subtypes. These differences have been repeatedly
identified and validated with different technologies and across different platforms (Fan et al,
2006; Farmer et al, 2005; Sorlie et al, 2006). The luminal subgroup has been subdivided into
two subgroups of prognostic significance:
• luminal A tumours which have high expression of ER –activated genes, and low
expression of proliferation related genes
• luminal B tumours which have higher expression of proliferation related genes and a
poorer prognosis than luminal A tumours (Geyer et al, 2009; Paik et al, 2000; Parker et
al, 2009; Sorlie et al, 2001, 2003).
The ER negative tumours are even more heterogeneous and comprise the:
• basal-like subgroup which lack ER and HER2/neu expression and feature more frequent
overexpression of basal cytokeratins, epidermal growth factor receptor and c-Kit
(Nielsen et al, 2004)
• HER2/neu subgroup which overexpress HER2/neu and genes associated with the
HER2/neu pathway and/or the HER2/neu amplicon on chromosome 17.
The HER2/neu and basal-like subtypes have in common an aggressive clinical behaviour
but appear to be more responsive to neoadjuvant chemotherapy than the luminal subtypes
(Carey et al, 2007; Rouzier et al, 2005). Also clustering with the ER negative tumours are the
normal-like breast cancers; these are as yet poorly characterised and have been shown to
92 Computational Biology and Applied Bioinformatics

cluster with fibroadenoma and normal breast tissue samples (Peppercorn et al, 2008). It is
important at this point to acknowledge the limitations of this molecular taxonomy;
intrasubtype heterogeneity has been noted despite the broad similarities defined by these
large subtypes (Parker et al, 2009). In particular the basal-like subgroup can be divided into
multiple additional subgroups (Kreike et al, 2007; Nielsen et al, 2004). Additionally,
although the luminal tumours have been separated into subgroups of prognostic
significance, meta-analysis of published expression data has suggested that these luminal
tumours actually form a continuum and their separation based on expression of
proliferation genes may be subjective (Shak et al, 2006; Wirapati et al, 2008). Furthermore,
the clinical significance of the normal-like subtype is yet to be determined; it has been
proposed that this subgroup may in fact represent an artefact of sample contamination with
a high content of normal breast tissue (Parker et al, 2009; Peppercorn et al, 2008). Due to
these limitations and the subjective nature of how the molecular subtypes were identified,
the translation of this taxonomy to the clinical setting as a definitive classification has been
difficult (Pustzai et al, 2006). The development of a prognostic test based on the intrinsic
subtypes has not been feasible to date. However, the seminal work by Sorlie and Perou
(Perou et al, 2000; Sorlie et al, 2001) recognized for the first time the scale of biological
heterogeneity within breast cancer and led to a paradigm shift in the way breast cancer is
perceived.

3.2 Class comparison

A number of investigators undertaking microarray expression profiling studies in breast
cancer have since adopted class comparison studies. These studies employ supervised
analysis approaches to determine gene expression differences between samples which
already have a predefined classification. The “null hypothesis” is that a given gene on the
array is not differentially expressed between the two conditions or classes under study. The
alternative hypothesis is that the expression level of that gene is different between the two
conditions. An example of this approach is the microarray experiments that have been
undertaken to define differences between invasive ductal and invasive lobular carcinomas
(Korkola, 2003; Weigelt, 2009; Zhao, 2004), between hereditary and sporadic breast cancer
(Berns, 2001; Hedenfalk, 2001) and between different disease stages of breast cancer
(Pedraza, 2010).

3.3 Class prediction

Perhaps the most clinically relevant use of this technology, however, are the microarray
class prediction studies which have been designed to answer specific questions regarding
gene expression in relation to clinical outcome and response to treatment. The latter
approach attempts to identify predictive markers, as opposed to the prognostic markers
which were identified in the “intrinsic gene-set”. There is frequently some degree of
confusion regarding the terms of “prognostic” and “predictive biomarkers”. This is partially
due to the fact that many prognostic markers also predict response to adjuvant therapy. This
is particularly true in breast cancer where, for example, the ER is prognostic, and predictive
of response to hormonal therapy, but also predictive of a poorer response to chemotherapy
(Carey 2007; Kim, 2009; Rouzier 2005,).
One of the first microarray studies designed to identify a gene-set predictive of prognosis in
breast cancer was that undertaken by van’t Veer and colleagues (van’t Veer et al, 2002). They
MicroArray Technology - Expression Profiling of MRNA and MicroRNA in Breast Cancer 93

developed a 70-gene set capable of predicting the development of metastatic disease in a

group of 98 patients made up of 34 who had developed metastasis within 5-years of follow-
up, 40 patients who remained disease-free at 5-years, 18 patients with a BRCA-1 mutation,
and 2 patients with a BRCA-2 mutation. The 70-gene signature was subsequently validated
in a set of 295 breast cancers, including the group used to train the model, and shown to be
more accurate than standard histopathological parameters at predicting outcome in these
breast cancer patients (van de Vijver et al, 2002). The signature includes many genes
involved in proliferation, and genes associated with invasion, metastasis, stromal integrity
and angiogenesis are also represented. This 70-gene prognostic signature classifies patients
based on correlation with a “good-prognosis” gene expression profile; a coefficient of
greater than 0.4 is classified as good prognosis. The signature was initially criticised for the
inclusion of some patients in both the discovery and validation stages (van de Vijver et al,
2002). However, it has been subsequently validated in multiple cohorts of node-positive and
node-negative patients and has been shown to outperform traditional clinical and
histological parameters at predicting prognosis (Buyse et al, 2006; Mook et al, 2009).

3.3.1 Mammaprint assay

The 70-gene signature was approved by the FDA to become the MammaPrint Assay
(Agendia BV, Amsterdam, The Netherlands); the first fully commercialized microarray
based multigene assay for breast cancer. This prognostic tool is now available and can be
offered to women under the age of 61 years with lymph node negative breast cancer. The
MammaPrint test results are dichotomous, indicating either a high or low risk of disease
recurrence, and the test performs best at the extremes of the spectrum of disease outcome
i.e. identifying patients with a very good or a very poor prognosis.
The MammaPrint signature is a purely prognostic tool, and its role as a predictive marker
for response to therapy was not examined at the time it was developed. Its’ clinical utility is
currently being assessed, however, in a prospective clinical trial called microarray in node
negative and 1 to 3 positive lymph node disease may avoid chemotherapy (MINDACT) trial
(Cardoso et al, 2008). The trial aims to recruit 6000 patients, all of whom will be assessed by
standard clinicopathologic prognostic factors and by the MammaPrint assay. In cases where
there is concordance between the standard prognostic factors and the molecular assay,
patients will be treated accordingly with adjuvant chemotherapy with or without endocrine
therapy for poor prognosis patients. If both assays predict a good prognosis, no adjuvant
chemotherapy is given, and adjuvant hormonal therapy is given alone where indicated. In
cases where there is disconcordance between the standard clinicopathological prognostic
factors and the MammaPrint assays’ prediction of prognosis the patients are randomised to
receive adjuvant systemic therapy based on either the clinicopathological or the
MammaPrint prognostic prediction results. The expected outcome is that there will be a
reduction of 10-15% in the number of patients requiring adjuvant chemotherapy based on
the MammaPrint assay prediction. It is envisaged that this trial will answer the questions of
what patients can be spared chemotherapy and still have a good prognosis, thus
accelerating progress towards the goal of more tailored therapy for breast cancer patients.

3.3.2 Oncotype Dx assay

While MammaPrint was developed as a prognostic assay, the other most widely established
commercialized multigene assay Oncotype Dx was developed in a more context specific
94 Computational Biology and Applied Bioinformatics

manner as a prognostic and predictive test to determine the benefit of chemotherapy in

women with node-negative, ER-positive breast cancer treated with tamoxifen (Paik et al,
2004). The authors used published microarray datasets, including those that identified the
intrinsic breast cancer subtypes and the 70-gene prognostic signature identified by the
Netherlands group to develop real time quantitative polymerase chain reaction (RQ-PCR)
tests for 250 genes. Research undertaken by the National Surgical Adjuvant Breast and
Bowel Project (NSABP) B14 protocol using three independent clinical series, resulted in the
development of an optimised 21-gene predictive assay (Paik et al, 2004). The assay has been
commercialised as Oncotype® DX by Genomic Health Inc1 and consists of a panel of 16
discriminator genes and 5 endogenous control genes which are detected by RQ-PCR using
formalin-fixed paraffin embedded (FFPE) sections from standard histopathology blocks. The
ability to use FFPE tissue facilitates clinical translation and has allowed retrospective
analysis of archived tissue in large cohorts with appropriate follow up data. The assay has
been used to generate Recurrence Scores (RS) by differentially weighting the constituent
genes which are involved in:
• proliferation (MKI67, STK15, BIRC5/Survivin, CCNB1, MYBL2)
• estrogen response (ER, PGR, SCUBE2)
• HER2/neu amplicon (HER2/neu/ERBB2, GRB7),
• invasion (MMP11, CTSL2)
• apoptosis (BCL2, BAG1)
• drug metabolism (GSTM1)
• macrophage response (CD68).
The assay was evaluated in 651 ER positive lymph node negative breast cancer patients who
were treated with either tamoxifen or tamoxifen and chemotherapy as part of the NSABP
B20 protocol (Paik et al, 2006). It was found that patients with high recurrence scores had a
large benefit from chemotherapy, with a 27.6% mean decreased in 10 year distance
recurrence rates, while those with a low recurrence score derived virtually no benefit from
chemotherapy. The RS generated by the expression of the 21 genes is a continuous variable
ranging from 1-100, but has been divided into three groups for clinical decision making; low
(<18), intermediate (18-31) and high (>31). It has been shown in a number of independent
datasets that ER positive breast cancer patients with a low RS have a low risk of recurrence
and derive little benefit from chemotherapy. Conversely, ER positive patients with high RS
have a high risk of recurrence but do benefit from chemotherapy (Goldstein, 2006; Habel,
2006; Mina, 2007; Paik, 2006). The ability of the 21-gene signature to so accurately predict
prognosis has led to the inclusion of the Oncotype Dx assay in American Society of Clinical
Oncology (ASCO) guidelines on the use of tumour markers in breast cancer as a predictor of
recurrence in ER-positive, node-negative patients. However, despite the accurate
performance of the assay for high and low risk patients, there remains uncertainty regarding
the management of patients with intermediate RS (18-31). This issue is being addressed in a
prospective randomized trial assigning individual options for treatment (TAILORx)
sponsored by the National Cancer Institute (Lo et al, 2007). This multicentre trial aims to
recruit 10,000 patients with ER –positive, lymph node negative breast cancer who are
assigned to one of three groups based on their RS; low<11, intermediate 11-25 and high >25.
Notably, the RS criteria have been changed for the TAILORx trial, with the intermediate

1 http://www.genomichealth.com/OncotypeDX
MicroArray Technology - Expression Profiling of MRNA and MicroRNA in Breast Cancer 95

range being changed from RS 18-30 to RS 11-25 to avoid exluding patients who may derive a
small benefit from chemotherapy (Sparano et al, 2006). Patients in the intermediate RS
group are randomly assigned to receive either adjuvant chemotherapy and hormonal
therapy, or hormonal therapy alone. The primary aim of the trial is to determine if ER
positive patients with an intermediate RS benefit from adjuvant chemotherapy or not.
The MammaPrint and Oncotype Dx gene signatures both predict breast cancer behaviour,
however there are fundamental differences between them (outlined in table 1). This chapter
has focused on these signatures as they were the first to be developed, have been extensively
validated, and are commercially available. However it is important to note that there are
other multi-gene based assays that have been developed and commercialized but are not
discussed in detail as they are not yet as widely utilized (Loi et al, 2007; Ma et al, 2008; Ross
et al, 2008; Wang et al, 2005 ).

Assay MammaPrint Oncotype Dx

Manufacturer Agendia BV Genomic Health, Inc.
From candidate set of 25,000 From candidate set of 250
Development of Signature
genes in 98 patients genes in 447 patients
Gene signature 70 genes 21 genes
Stage I & II breast cancer
Stage I & II breast cancer
Lymph node negative
Patient cohort Lymph node negative
ER positive
<55yrs
Receiving Tamoxifen
Platform cDNA Microarray RQ-PCR
Fresh frozen tissue or
Sample requirements collected in RNA FFPE tissue
preservative
5-year distant relapse free 10-year distant relapse free
Outcome
survival survival
Dichotomous correlation Continuous recurrence score
coefficient <18 = low risk
Test Results
>4.0 = good prognosis 18-31= intermediate risk
<4.0 = poor prognosis >31 = high risk
Predictive No; purely prognostic Yes
Prospective Trial MINDACT TAILORx
FDA approved Yes No
ASCO Guidelines No Yes
Table 1. Comparison of commercially available prognostic assays MammaPrint and
Oncotype Dx

4. Microarray data integration

4.1 Setting standards for microarray experiments
It must be acknowledged that despite the multitude of breast cancer prognostic signatures
available, the overlap between the gene lists is minimal (Ahmed, 2005; Brenton, 2005; Fan et
96 Computational Biology and Applied Bioinformatics

al, 2006; Michiels et al, 2005). This lack of concordance has called into question the
applicability of microarray analysis across the entire breast cancer population. In order to
facilitate external validation of signatures and meta-analysis in an attempt to devise more
robust signatures, it is important that published microarray data be publicly accessible to
the scientific community. In 2001 the Microarray Gene Expression Data Society proposed
experimental annotation standards known as minimum information about a microarray
experiment (MIAME), stating that raw data supporting published studies should be made
publicly available in one of a number of online repositories (table 2), these standards are
now upheld by leading scientific journals and facilitating in depth interrogation of multiple
datasets simultaneously.

Public Database
for Microarray URL Organization Description
Data
Array Express http://www.ebi.ac.uk/arrayexpress/ European Public data
Bioinformatics deposition and
Institute (EBI) queries
GEO Gene http://www.ncbi.nlm.nih.gov/geo/ National Centre for Public data
Expression Biotechnology deposition and
Omnibus Information (NCBI) queries
CIBEX Center http://cibex.nig.ac.jp/index.jsp National Institute Public data
for Information of Genetics deposition and
Biology Gene queries
Expression
Database
ONCOMINE http://www.oncomine.org/main/ University of Public queries
Cancer Profiling index.jsp Michigan
Database
PUMAdb http://puma.princeton.edu/ Princeton Public queries
Princeton University
University
MicroArray
database
SMD Stanford http://genome- Stanford Univeristy Public queries
Microarray www5.stanford.edu/
Database
UNC Chapel https://genome.unc.edu/ University of North Public queries
Hill Microarray Carolina at Chapel
database Hill
Table 2. List of Databases with Publicly Available Microarray Data

4.2 Gene ontology

The volume of data generated by high throughput techniques such as microarray poses the
challenge of how to integrate the genetic information obtained from large scale experiments
with information about specific biological processes, and how genetic profiles relate to
functional pathways. The development of the Gene Ontology (GO) as a resource for
MicroArray Technology - Expression Profiling of MRNA and MicroRNA in Breast Cancer 97

experimentalists and bioinformaticians has contributed significantly to overcoming this

challenge (Ashburner et al, 2000). The GO Consortium was established with the aim of
producing a structured, precisely defined, common, controlled vocabulary for describing
the roles of genes and gene products in any organism. Initially a collaboration between three
organism databases: Flybase (The Flybase Consortium, 1999), Mouse Genome Informatics
(Blake et al, 2000) and the Saccharomyces Genome Database (Ball et al, 2000), the GO
Consortium has grown to include several of the world’s major repositories for plant, animal
and microbial genomes.
The Gene Ontology provides a structure that organizes genes into biologically related
groups according to three criteria. Genes and gene products are classified according to:
• Molecular Function: biochemical activity of gene products at the molecular level
• Biological Process: biological function of a gene product
• Cellular Component: location in the cell or extracellular environment where molecular
events occur
Every gene is described by a finite, uniform vocabulary. Each GO entry is defined by a
numeric ID in the format GO#######. These GO identifiers are fixed to the textual
definition of the term, which remains constant. A GO annotation is the specific association
between a GO identifier and a gene or protein and has a distinct evidence source that
supports the association. A gene product can take part in one or more biological process and
perform one or more molecular functions. Thus, a well characterized gene product can be
annotated to multiple GO terms in the three GO categories outlined above. GO terms are
related to each-other such that each term is placed in the context of all of the other terms in a
node-directed acyclic graph (DAC). The relationships used by the GO are: “is_a”, “part_of”,
“regulates”, “positively_regulates”, “negatively_regulates” and “disjoint_from”. Each term
in the DAC may have one or more parent terms and possibly one or more child nodes, and
the DAC gives a graphical representation of how GO terms relate to each other in a
hierarchical manner.
The development of Gene Ontology has facilitated analysis of microarray gene sets in the
context of the molecular functions and pathways in which they are involved (Blake &
Harris, 2002). GO-term analysis can be used to determine whether genetic “hits” show
enrichment for a particular group of biological processes, functions or cellular
compartments. One approach uses statistical analysis to determine whether a particular GO
is over or under-represented in the list of differentially expressed genes from a microarray
experiment. The statistical tests used for such analysis include hypergeometric, binomial or
Chi-square tests (Khatri et al, 2005).
An alternative approach known as “gene-set testing” has been described which involves
beginning with a known set of genes and testing whether this set as a whole is differentially
expressed in a microarray experiment (Lamb et al, 2003; Mootha et al, 2003). The results of
such analyses inform hypotheses regarding the biological significance of microarray
analyses.
Several tools have been developed to facilitate analysis of microarray data using GO, and a
list of these can be found at: http://www.geneontology.org/GO.tools.microarray.shtml
Analysing microarray datasets in combination with biological knowledge provided by GO
makes microarray data more accessible to the molecular biologist and can be a valuable
strategy for the selection of biomarkers and the determination of drug treatment effect in
breast cancer (Arciero et al, 2003; Cunliffe et al, 2003).
98 Computational Biology and Applied Bioinformatics

4.3 Microarray meta-analysis – combining datasets

Meta-analyses have confirmed that different prognostic signatures identify similar
biological subgroups of breast cancer patients (Fan et al, 2006) and have also shown that the
designation of tumours to a “good prognosis”/”low risk” group or a “poor
prognosis”/”high risk” group is largely dependent on the expression patterns of
proliferative genes. In fact, some of these signatures have been shown to have improved
performance when only the proliferative genes are used (Wirapati, 2008). Metanalyses of the
signatures have also proposed that the prognostic ability of the signatures is optimal in the
ER positive and HER2-negative subset of breast tumours (Desmedt, 2008; Wirapati, 2008),
the prognosis of this group of tumours being governed by proliferative activity.
Despite obvious clinical application, none of these prognostic assays are perfect, and they all
carry a false classification rate. The precise clinical value for these gene expression profiles
remains to be established by the MINDACT and TAILORx trials. In the interim the
performance of these assays is likely to be optimised by combining them with data from
traditional clinicopathological features, an approach which has been shown to increase
prognostic power (Sun et al, 2007).
Microarray technology has undoubtedly enhanced our understanding of the molecular
mechanisms underlying breast carcinogenesis; profiling studies have provided a myriad of
candidate genes that may be implicated in the cancer process and are potentially useful as
prognostic and predictive biomarkers or as therapeutic targets. However, as yet there is
little knowledge regarding the precise regulation of these genes and receptors, and further
molecular categories are likely to exist in addition to and within the molecular subtypes
already delineated. Accumulating data reveal the incredible and somewhat foreboding
complexity and variety of breast cancers and while mRNA expression profiling studies are
ongoing, a new player in breast cancer biology has come to the fore in recent years; a
recently discovered RNA species termed MiRNA (miRNA) which many scientists believe
may represent a crucial link in the cancer biology picture.

5. MicroRNA - a recently discovered layer of molecular complexity

It has been proposed that the discovery of miRNAs as regulators of gene expression represents
a paradigm changing event in biology and medicine. This discovery was made in 1993 by
researchers at the Ambros laboratory in Dartmouth Medical School, USA at which time it was
thought to be a biological entity specific to the nematode C. Elegans (Lee et al, 1993). In the
years following this discovery, hundreds of miRNAs were identified in animals and plants.
However it is only in the past 5 years that the field of miRNA research has really exploded
with the realisation that miRNAs are critical to the development of multicellular organisms
and the basic functions of cells (Bartel, 2004). MiRNAs are fundamental to genetic regulation,
and their aberrant expression and function have been linked to numerous diseases and
disorders (Bartel, 2004; Esquela-Kerscher & Slack, 2006). Importantly, miRNA have been
critically implicated in the pathogenesis of most human cancers, thus uncovering an entirely
new repertoire of molecular factors upstream of gene expression.

5.1 MicroRNA - novel cancer biomarkers

The first discovery of a link between miRNAs and malignancy was the identification of a
translocation-induced deletion at chromosome 13q14.3 in B-cell Chronic Lymphocytic
MicroArray Technology - Expression Profiling of MRNA and MicroRNA in Breast Cancer 99

Leukaemia (Calin et al, 2002). Loss of miR-15a and miR-16-1 from this locus results in
increased expression of the anti-apoptotic gene BCL2. Intensifying research in this field,
using a range of techniques including miRNA cloning, quantitative PCR, microarrays and
bead-based flow cytometric miRNA expression profiling has resulted in the identification
and confirmation of abnormal miRNA expression in a number of human malignancies
including breast cancer (Heneghan et al, 2010; Lowery et al, 2007). MiRNA expression has
been observed to be upregulated or downregulated in tumours compared with normal
tissue, supporting their dual role in carcinogenesis as either oncogenic miRNAs or tumour
suppressors respectively (Lu et al, 2005). The ability to profile miRNA expression in human
tumours has led to remarkable insight and knowledge regarding the developmental lineage
and differentiation states of tumours. It has been shown that distinct patterns of miRNA
expression are observed within a single developmental lineage, which reflect mechanisms of
transformation, and support the idea that miRNA expression patterns encode the
developmental history of human cancers. In contrast to mRNA profiles it is possible also to
successfully classify poorly differentiated tumours using miRNA expression profiles
(Volinia et al, 2006). In this manner, miRNA expression could potentially be used to
accurately diagnose poorly differentiated tissue samples of uncertain histological origin, e.g.
metastasis with an unknown primary tumour, thus facilitating treatment planning.
MicroRNAs exhibit unique, inherent characteristics which make them particularly attractive
for biomarker development. They are known to be dysregulated in cancer, with
pathognomonic or tissue specific expression profiles and even a modest number of miRNAs
is sufficient to classify human tumours, which is in contrast to the relatively large mRNA
signatures generated by microarray studies (Lu et al, 2005). Importantly, miRNA are
remarkably stable molecules. They undergo very little degradation even after processing
such as formalin fixation and remain largely intact in FFPE clinical tissues, lending
themselves well to the study of large archival cohorts with appropriate follow-up data (Li et
al, 2007; Xi et al, 2007). The exceptional stability of miRNAs in visceral tissue has stimulated
investigation into their possible preservation in the circulation and other bodily fluids
(urine, saliva etc.). The hypothesis is that circulating miRNAs, if detectable and quantifiable
would be the ideal biomarker accessible by minimally invasive approaches such as simple
phlebotomy (Cortez et al, 2009; Gilad et al, 2008; Mitchell et al, 2008).

5.2 MicroRNA microarray

The unique size and structure of miRNAs has necessitated the modification of existing
laboratory techniques, to facilitate their analysis. Due to the requirement for high quality
large RNA molecules, primarily for gene expression profiling, many laboratories adopted
column-based approaches to selectively isolate large RNA molecules, discarding small RNA
fractions which were believed to contain degradation products. Modifications to capture
miRNA have been made to existing protocols to facilitate analysis of the miRNA fraction.
Microarray technology has also been modified to facilitate miRNA expression profiling.
Labelling and probe design were initially problematic due to the small size of miRNA
molecules. Reduced specificity was also an issue due to the potential of pre-miRNA and pri-
miRNAs to produce signals in addition to active mature miRNA. Castoldi et al described a
novel miRNA microarray platform using locked nucleic acid (LNA)-modified capture
probes (Castoldi et al, 2006). LNA modification improved probe thermostability and
increased specificity, enabling miRNAs with single nucleotide differences to be
100 Computational Biology and Applied Bioinformatics

discriminated - an important consideration as sequence-related family members may be

involved in different physiological functions (Abbott et al, 2005). An alternative high
throughput miRNA profiling technique is the bead-based flow cytometric approach
developed by Lu et al.; individual polystyrene beads coupled to miRNA complementary
probes are marked with fluorescent tags (Lu et al, 2005). After hybridization with size-
fractioned RNAs and streptavidin-phycoerythrin staining, the beads are analysed using a
flow-cytometer to measure bead colour and pycoerythrin, denoting miRNA identity and
abundance respectively. This method offered high specificity for closely related miRNAs
because hybridization occurs in solution. The high-throughput capability of array-based
platforms make them an attractive option for miRNA studies compared to lower
throughput techniques such as northern blotting and cloning; which remain essential for the
validation of microarray data.

5.2.1 MicroRNA microarray - application to breast cancer

Microarray analysis of miRNA expression in breast cancer is in its’ infancy relative to
expression profiling of mRNA. However, there is increasing evidence to support the
potential for miRNAs as class predictors in breast cancer. The seminal report of aberrant
miRNA expression in breast cancer by Iorio et al. in 2005 identified 29 miRNAs that were
differentially expressed in breast cancer tissue compared to normal, a subset of which could
correctly discriminate between tumour and normal with 100% accuracy (Iorio et al, 2005).
Among the leading miRNAs differentially expressed; miR-10b, miR-125b and mR-145 were
downregulated whilst miR-21 and miR-155 were consistently over-expressed in breast
tumours. In addition, miRNA expression correlated with biopathological features such as
ER and PR expression (miR-30) and tumour stage (miR-213 and miR-203). Mattie et al.
subsequently identified unique sets of miRNAs associated with breast tumors defined by
their HER2/neu or ER/PR status (Mattie et al, 2006). We have described 3 miRNA
signatures predictive of ER, PR and Her2/neu receptor status, respectively, which were
identified by applying artificial neural network analysis to miRNA microarray expression
data (Lowery et al, 2009). Blenkiron et al used an integrated approach of both miRNA and
mRNA microarray expression profiling to classify tumours according to “intrinsic subtype”.
This approach identified a number of miRNAs that are differentially expressed according to
intrinsic breast cancer subtype and associated with clinicopathological factors including ER
status and tumour grade. Importantly, there was overlap between the differentially
expressed miRNAs identified in these studies.
There has been interest in assessing the prognostic value of miRNAs, and expression studies
in this regard have focused on detecting differences in miRNA expression between primary
breast tumours and metastatic lymph nodes. This approach has identified numerous
miRNA that are dysregulated in primary breast tumours compared to metastatic lymph
nodes (Baffa et al 2009; Huang et al, 2008). MiRNA have also been identified that are
differentially expressed in patients who had a “poor prognosis” or a short time to
development of distant metastasis (Foekens et al, 2008); miR-516-3p, miR-128a, miR-210, and
miR-7 were linked to aggressiveness of lymph node-negative, ER-positive human breast
cancer.
The potential predictive value of miRNA is also under investigation. Preclinical studies have
reported associations between miRNA expression and sensitivity to adjuvant breast cancer
therapy including chemotherapy, hormonal therapy and HER2/neu targeted therapy (Ma
MicroArray Technology - Expression Profiling of MRNA and MicroRNA in Breast Cancer 101

et al, 2010; Tessel et al, 2010; Wang et al, 2010), prompting analysis of tumour response in
clinical samples. Rodriguez-Gonzalez et al attempted to identify miRNAs related to
response to tamoxifen therapy by exploiting the Foekens dataset (Foekens, 2008) which
comprised miRNA expression levels of 249 miRNAs in 38 ER positive breast cancer patients.
Fifteen of these patients were hormone naive and experienced relapse, which was treated
with tamoxifen. Ten patients responded and five did not, progressing within 6 months. Five
miRNAs (miR-4221, miR-30a-3p, miR-187, miR-30c and miR-182) were the most
differentially expressed between patients who benefitted from tamoxifen and those who
failed therapy. The predictive value for these miRNAs was further assessed in 246 ER
positive primary tumours of hormone naive breast cancer patients who received tamoxifen
as monotherapy for metastatic disease. MiR-30a-3p, miR-30c and miR-182 were significantly
associated with response to tamoxifen, but only miR-30c remained an independent predictor
on multivariate analysis (Rodriguez-Gonzalez, 2010).
Microarray-based expression profiling has also been used to identify circulating miRNAs
which are differentially expressed in breast cancer patients and matched healthy controls.
Zhao et al profiled 1145 miRNAs in the plasma of 20 breast cancer patients and 20 controls,
identifying 26 miRNAs with at least two-fold differential expression which reasonably
separated the 20 cases from the 20 controls (Zhao et al, 2010). This is the first example of
genome-wide miRNA expression profiling in the circulation of breast cancer patients and
indicates potential for development of a signature of circulating miRNAs that may function
as a diagnostic biomarker of breast cancer.
At present diagnostic, prognostic and predictive miRNA signatures and markers remain
hypothesis generating. They require validation in larger, independent clinical cohorts prior
to any consideration for clinical application. Furthermore as additional short non-coding
RNAs are continuously identified through biomarker discovery programmes, the available
profiling technologies must adapt their platforms to incorporate newer potentially relevant
targets. MicroRNAs possess the additional attraction of potential for development as
therapeutic targets due to their ability to regulate gene expression. It is likely that future
microarray studies will adopt and integrated approach of miRNA and mRNA expression
analysis in an attempt to decipher regulatory pathways in addition to expression patterns.

6. Limitations of microarray technology & bioinformatic challenges

In addition to the great promises and opportunities held by microarray technologies, several
issues need to be borne in mind and appropriately addressed in order to perform reliable
and non-questionable experiments. As a result, several steps need to be addressed in order
to identify and validate reliable biomarkers in the scope of potential future clinical
application. This is one of the reasons why, despite the promises of using powerful high-
throughput technologies as such as microarray, only very few useful biomarkers have been
identified so far and/or have been translated to useful clinical assay or companion
diagnostics (Mammaprint®, Oncotype DX®). There still remains a lack of clinically relevant
biomarkers (Rifai et al, 2006). Amongst the limitations and pitfalls around the technology
and the use of microarrays, some of the most important are the reported lack of
reproducibility, as well as the massive amount of data generated, often extremely noisy and
with an increasing complexity. As for example, in the recent Affymetrix GeneChip 1.0 ST
microarray platform (designed to target all known and predicted exons in human, mouse
and rat genomes), where there is approximately 1.2 million exon clusters corresponding to
102 Computational Biology and Applied Bioinformatics

over 1.4 million probesets (Lancashire et al, 2009). As a result, it appears clearly that
extracting any relevant key component from such datasets requires robust mathematical
and/or statistical models running on efficient hardware to perform the appropriate
analyses.
With this in mind, it is clear that the identification of new biomarkers still requires a
concerted, multidisciplinary effort. It requires the expertise of the biologist or pathologist, to
extract the samples, the scientist to perform the analysis on the platform and then the
bioinformatician/biostatistician to analyse and interpret the output. The data-mining
required to cope with these types of data needs careful consideration and specific
computational tools, and as such remains a major challenge in bioinformatics.

6.1 Problems with the analysis of microarray data

6.1.1 Dimensionality and false discovery
The statistical analysis of mRNA or miRNA array data poses a number of challenges. This
type of data is of extremely high dimensionality i.e. has a large number of variables. Each of
these variables represents the relative expression of a mRNA or miRNA in a sample. Each of
these components contain noise, are non-linear may not follow a normal distribution
through a population and may be strongly correlated with other probes in the profile. These
characteristics mean that the data may violate many of the assumptions of conventional
statistical techniques, particularly with parametric tests.
The dimensionality of the data poses a significant problem, and remains as one of the most
critical when analysing microarray data. When one analyses this type of data, one has to
consider what is referred to as the curse of dimensionality, firstly described by Bellman in 1961
as the “exponential growth of the search space as a function of dimensionality” (Bellman, 1961;
Bishop, 1995). This occurs in highly dimensional systems where the number of dimensions
masks the true importance of an individual single dimension (variable). It is particularly
true in a microarray experiment when the number of probes representing the number of
miRNA/mRNA studied far exceeds the number of available samples. So there is the
potential for a probe that is in reality of high importance to be missed when considered with
a large number of other probes. This problem is overcome by breaking down the analysis
into single or small groups of variables and repeating the analysis rather than considering
the whole profile in one single analysis. Other methods consists of using pre-processing
methods and feature extraction algorithms in order to only analyse a subset of the data
supposed to hold the most relevant features (Bishop, 1995), as determined by the pre-
processing steps.
High dimensionality also creates problems due to false discovery. The false discovery rate
(FDR) introduced by Benjamini and Hochberg (Benjamini and Hochberg, 1995) is a measure
of the number of features incorrectly identified as “differential” and various approaches
have been suggested to accurately control the FDR. In this case if one has a high number of
dimensions and analyses each singly (as above) a proportion can appear to be of high
importance due to random chance considering the distribution, even when they are not. To
overcome this one has to examine a rank order of importance and when testing for
significance one has to correct the threshold for significance by dividing it by the number of
dimensions. So for example when analysing the significance of single probes from a profile
with 4,000 probes in it the threshold becomes P < 0.05 divided by 4,000 i.e. P < 0.0000125.
MicroArray Technology - Expression Profiling of MRNA and MicroRNA in Breast Cancer 103

6.1.2 Quality and noise

Noise also poses a problem in the analysis of mRNA or miRNA data. The inherent technical
and biological variability necessarily induces noise within the data, eventually leading to
biased results. The noise may lead to misinterpretation of sample groups that may actually
have no biological relevance. As a consequence extreme care needs to be taken to address
the problem of noise.
Noise may be random where it is applied to all parts of the miRNA equally or systematic
where particular probes inherently have more noise than others because of the nature of the
component miRNA or genomic code that they represent.
It is now widely acknowledged that the reported high level of noise found in microarray
data is the most critical pull-back of microarray-based studies, as it is pointed by the MAQC
Consortium (Shi et al, 2006; Klebanov and Yakovlev, 2007).

6.1.3 Complexity and non-normality

Because of the complex nature of the profile a particular mRNA or miRNA may be non-
normally distributed through a population. Such non-normality will immediately invalidate
any statistical test that uses parametric statistics i.e. depends on the assumption of a normal
distribution. Invalidated tests would include ANOVA and t-test. To overcome this, the data
would have to be transformed mathematically to follow a normal distribution or an
alternative non parametric test would have to be employed. Examples of non-parametric
tests include Kruskal-Wallis and Mann Whitney U which are ANOVA and unpaired T-Test
alternatives respectively. Generally non-parametric tests lack power compared to their
parametric alternatives and this may prove to be a problem in high dimensional space due
to the reasons described previously.

6.1.4 Reproducibility
Reproducibility has a marked effect on the accuracy of any analysis conducted. Furthermore
reproducibility has a profound effect on the impact of other issues such as dimensionality
and false detection. Robust scientific procedures requires that the results have to be
reproducible in order to reduce the within sample variability, the variability between
sample runs and the variability across multiple reading instruments. Aspects of variability
can be addressed using technical and experimental replicates. The averaging of samples
profiles can be used to increase the confidence in the profiles for comparison (Lancashire et
al., 2009). Technical replicates provide information on the variability associated with
instrumental variability whilst experimental (or biological) replicates give a measure of the
natural sample to sample variation. Problems in data analysis occur when the technical
variability is high. In this situation the problem in part can be resolved by increasing the
number of replicates. If however the technical variation is higher than the biological
variation then the sample cannot be analysed.

6.1.5 Auto-correlation or co-correlation

Auto correlation exists when two components within a system are strongly linearly
correlated with one another. In any complicated system there are likely to be a number of
components that are auto correlated. This is especially true in array profiling of biological
samples. Firstly due to biological processes one protein in a set of samples is likely to
interact or correlate with another through a population.
104 Computational Biology and Applied Bioinformatics

Auto correlation becomes a problem when using linear based regression approaches. This is
because one of the assumptions of regression using multiple components is that the
components are not auto correlated. If intensity for multiple miRNA probes are to be added
into a regression to develop a classifier these components should not be auto correlated.
Auto correlation can be tested for using the Durbin Watson test.

6.1.6 Generality
The whole purpose of biomarker (or set of biomarkers) identification, using high-
throughput technologies or any other, is to provide the clinicians with an accurate model in
order to assess a particular aspect. However, a model is only as good as its ability to
generalize to unseen real world data. A model only able to explain the population on which
it was developed would be purely useless for any application.
As a result, if one is to develop classifiers from mRNA or miRNA array data the features
identified should be generalised. That is they will predict for new cases in the general
population of cases. When analysing high dimensional data there is an increased risk of over
fitting, particularly when the analysis methods imply supervised training on a subset of the
population. So for example, when a large number of mRNA or miRNA are analysed there is
the potential for false detection to arise. If a random element identified through false
detection is included as a component of a classifier (model) then the generality of that
classifier will be reduce; i.e. it is not a feature that relates to the broader population but is a
feature specific to the primary set of data used to develop the classifier. Standards of
validation required to determine generality have been defined by Michiels et al, 2007.
Generality of classifiers can be increased by the application of bootstrapping or cross
validation approaches.
Some algorithms and approaches, that usually involve supervised training, suffer from
over-fitting (sometimes called memorisation). This is a process where a classifier is
developed for a primary dataset but models the noise within the data as well as the relevant
features. This means that the classifier will not accurately classify for new cases i.e. it does
not represent a general solution to the problem which is applicable to all cases. This is
analogous, for example, to one developing a classifier that predicts well the risk of
metastasis for breast cancer patients from Nottingham but will not predict well for a set of
cases from Denmark. Over fitted classifiers seldom represent the biology of the system being
investigated and the features identified are often falsely detected.
One of the most common solutions to avoid over-fitting is to apply a Cross Validation
technique in combination with the supervised training. Random sample cross validation is a
process of mixing data. Firstly the data are divided into two or three parts (figure 2); the first
part is used to develop the classifier and the second or second and third parts are used to
test the classifier. These parts are sometimes termed training, test and validation data sets
respectively. In certain classifiers such as Artificial Neural Network based classifiers the
second blind set is used for optimisation and to prevent over fitting. In random sample cross
validation the random selection and training process is repeated a number of times to create
a number of models each looking at the global dataset in a number of different ways (figure
2). Often the mean performance of these models is considered.
Leave one out cross validation is an approach also used to validate findings. In this case one
sample is left out of the analysis. Once training is complete the sample left out is tested. This
process is repeated a number of times to determine the ability of a classifier to predict
MicroArray Technology - Expression Profiling of MRNA and MicroRNA in Breast Cancer 105

unseen cases. This approach of random sample cross validation drives the classifier solution
to a generalised one by stopping the classifier from training too much on a seen dataset and
stopping the training earlier based on a blind dataset.

7. Methods used to analyse microarray data and their limitations

With the advent of cutting edge new technologies such as microarrays, the analysis tools for
the data produced need to be appropriately applied. Although expression arrays have
brought high hopes and expectations, they have brought tremendous challenges with them.
They have been proven to suffer from different limitations as previously discussed.
However, innovative computational analysis solutions have been developed and have been
proven efficient and successful at identifying markers of interest regarding particular
questions. This section presents some of the most common methods employed to overcome
the limitations discuss above, and to analyse expression array data.

7.1 Application of ordination techniques

If we are to utilise the mRNA or miRNA profile we have to identify robust features despite
its high dimensionality that are statistically valid for the general population not just for a
subset. Ordination techniques are used to map the variation in data. They are not directly
predictive and cannot classify directly unless combined with another classification
technique.

Fig. 2. Illustration of Cross Validation technique, here with three subsets: the training subset
used to train the classifier, the test subset used to stop the training when it has reached an
optimal performance on this subset, and a validation subset to evaluate the performance
(generalization ability) of the trained classifier.
106 Computational Biology and Applied Bioinformatics

7.1.1 Principal components analysis

PCA is usually a method of choice for dimensionality reduction. It is a multivariate
exploratory technique used to simplify complex data space (Raychaudhuri et al, 2000) by
translating the data space into a new space defined by the principal components. It works
by identifying the main (principal) components that explain best the shape (variance) of a
data set. Each principal component is a vector (line) through the data set that explains a
proportion of the variance, it is the expression of a linear combination of the data. In PCA
the first component that is added is the one that explains the most variance the second
component added is then orthogonal to the first. Subsequent orthogonal components are
added until all of the variation is explained. The addition of vectors through a
multidimensional data set is difficult to visualise in print, we have tried to illustrate it with 3
dimensions in figure 3. In mRNA/miRNA profile data where thousands of dimensions
exist, PCA is a useful technique as it reduces the dimensionality to a manageable number of
principal components. If the majority of the variance is explained in 2 or 3 principal
components these can be used to visualise the structure of the population using 2 or 3
dimensional plots. A limited parameterisation can also be conducted to determine the
contribution of each parameter (miRNA) to each of the principal components. This however
suffers from the curse of dimensionality in high dimensional systems. Thus the main
limitation of using PCA for gene expression data is the inability to verify the association of a
principal component vector with the known experimental variables (Marengo et al, 2004).
This often makes it difficult to accurately identify the importance of the mRNA or miRNA in
the system, and make it a valuable tool only for data reduction.

Fig. 3. Example of a 3 dimension PCA with the 3 orthogonal PCs.

MicroArray Technology - Expression Profiling of MRNA and MicroRNA in Breast Cancer 107

7.1.2 Hierarchical clustering

Although several clustering techniques exist, the most used in the context of microarray
data analysis is hierarchical clustering. Hierarchical clustering is used to identify the
structure of a given population of cases or a given set of markers such as proteins. Every
case is considered to have a given position in multidimensional space. Hierarchical
clustering determines the similarity of cases in this space based on the distance between
points. There are various linkage methods used for calculating distance, such as single
linkage, complete linkage and average linkage. Single linkage computes the distance as the
distance between the two nearest points in the clusters being compared. Complete linkage
computes the distance between the two farthest points, whilst average linkage averages all
distances across all the points in the clusters being compared. One commonly used distance
measure is Euclidian distance which is the direct angular distance between two points. In
fact it considers the distance in multidimensional space between each point and every other
point. In this way a hierarchy of distances is determined. This hierarchy is plotted in the
form of a dendrogram (figure 4). From this dendrogram we can identify clusters of cases or
markers that are similar at a given distance.
The one major problem concerning clustering is that it suffers from the curse of
dimensionality when analysing complex datasets. In a high dimensional space, it is likely
that for any given pair of points within a cluster there will exist dimensions on which these
points are far apart from one another. Therefore distance functions using all input features
equally may not be truly effective (Domeniconi et al, 2004). Furthermore, clustering methods
will often fail to identify coherent clusters due to the presence of many irrelevant and
redundant features (Greene et al, 2005). Additionally, the important number of different
distance measure may add an additional bias: it has been reported that the choice of a
distance measure can greatly affect the results and produce different outcomes after the
analysis (Quackenbush, 2001). Dimensionality is also of importance when one is examining
the structure of a population through ordination techniques. This is particularly the case
when utilising hierarchical cluster analysis. This approach is of limited suitability for high
dimensional data as in a high dimensional space the distance between individual cases
reaches convergence making all cases appear the same (Domeniconi et al, 2004). This makes
it difficult to identify the real structure in the data or clusters of similar cases.

7.2 Application of modelling techniques

This second part of the section focusing on analysis tools considers more evolved techniques
with what is known as machine learning. There are however a number of other techniques
that can be employed in a predictive or classification capacity. Others include hidden
Markov and Bayesian methods. These are widely described in the literature.

7.2.1 Decision tree based methodologies

Decision tree methodologies include, boosted decision trees, classification and regression
trees, random forest methodologies. This approach is based on splitting a population into
groups based on a hierarchy of rules (figure 5). Thus a given case is split into a given class
based on a series of rules. This approach has been modified in a number of ways. Generally,
a decision is made based on a feature that separates classes (one branch of the cluster
dendrogram from another) within the population. This decision is based on a logical or
numerical rule. Although their use in the analysis of miRNA data has been limited, decision
108 Computational Biology and Applied Bioinformatics

trees have been used in the analysis of miRNA data derived to classify cancer patients (Xu,
et al, 2009).

Fig. 4. Example of a hierarchical clustering analysis result aiming to find clusters of similar
cases.

Fig. 5. Schematic example of the basic principle of Decision Trees

MicroArray Technology - Expression Profiling of MRNA and MicroRNA in Breast Cancer 109

Boosted decision trees take the primary decision tree algorithm and boost it. Boosting is a
process where classifiers are derived to allow prediction of those not correctly predicted by
earlier steps. This means that a supervised classification is run where the actual class is
known. A decision tree is created that classifies correctly as many cases as possible. Those
cases that are incorrectly classified are given more weighting. A new tree is then created
with these boosted weights. This process is similar to the iterative leaning that is conducted
with the Artificial Neural Network back propagation algorithm.
Random forest approaches take the basic decision tree algorithm and couple it with random
sample cross validation. In this way a forest of trees is created. Integration of a number of
decision trees identifies a combined decision tree which, as it is developed on blind cases,
represents what approaches a generalised solution for the problem being modelled
(Breiman et al, 2001). This approach has been shown to be very good at making generalised
classifications. The approach essentially derives each tree from a random vector with
equivalent distribution from within the data set, essentially an extensive form of cross
validation. Yousef et al, (2010) have used random forest as one method for the identification
of gene targets for miRNAs. Segura et al (2010) have used random forests as a part of an
analysis to define post recurrence survival in melanoma patients.

7.2.2 Artificial Neural Networks

Artificial Neural Networks are a non linear predictive system that may be used as a
classifier. A popular form of ANN is the multi-layer perceptron (MLP) and is used to solve
many types of problems such as pattern recognition and classification, function
approximation, and prediction. The approach is a form of artificial intelligence in that it
“learns” a solution to a problem from a preliminary set of samples. This is achieved by
comparing predicted versus actual values for a seen data set (the training data set described
earlier) and using the error of the predicted values from the ANN to iteratively develop a
solution that is better able to classify. In MLP ANNs, learning is achieved by updating the
weights that exist between the processing elements that constitute the network topology
(figure 6). The algorithm fits multiple activation functions to the data to define a given class
in an iterative fashion, essentially an extension of logistic regression. Once trained, ANNs
can be used to predict the class of an unknown sample of interest. Additionally, the
variables of the trained ANN model may be extracted to assess their importance in the system
of interest. ANNs can be coupled with Random sample cross validation or any other cross
validation method (LOO or MCCV) in order to ensure that the mode developed is not over
fitted. One of the advantages of ANNs is that the process generates a mathematical model that
can be interrogated and explored in order to elucidate further biological details and validate
the model developed on a wide range of cases. A review of their use is in a clinical setting
presented in Lisboa and Taktak (2006). Back propagation MLP ANNs have been proposed for
use in the identification of biomarkers from miRNA data by Lowery et al, 2009.

7.2.3 Linear Discriminant Analysis (LDA)

Linear discriminant analysis attempts to separate the data into two subgroups by calculating
the optimal linear line that best splits the population. Calculation of this discriminating line
is conducted by taking into account sample variation within similar classes, and minimizing
it between classes. As a result, any additional sample has its class determined by the side of
the discriminating line it falls.
110 Computational Biology and Applied Bioinformatics

LDA can outperform other linear classification methods as LDA tries to consider the
variation within the sample population. Nevertheless, LDA still suffers from its linear
characteristic, and often fails to accurately classify non-linear problems, which is mostly the
case in biomedical sciences (Stekel et al, 2003). This is the reason why non-linear classifiers
are recommended.

Fig. 6. Example of a classical MLP ANN topology with the details of a node (or neurone)

7.2.4 Support Vector Machines

Support Vector Machines (SVMs) are another popular form of machine learning algorithms
in the field of analyzing MA data for non-linear modeling (Vapnik and Lerner, 1963). They
are an evolution of LDA in the sense that they work by separating the data into 2 sub-
groups. They work by separating the data into two regions by constructing a straight line or
hyper plane that best separates between classes (figure 7). In the common example of a two-
class classification problem, SVMs attempt to find a linear “maximal margin hyperplane”
MicroArray Technology - Expression Profiling of MRNA and MicroRNA in Breast Cancer 111

able to accurately discriminate the classes (Dreiseitlet al, 2001), similarly to what does Linear
Discriminant Analysis. If no such linear hyperplane can be found, usually due to the
inherent non-linearity of the dataset, the data are mapped into a high-dimensional feature
space using a kernel function (for example polynomial or radial basis functions) in which
the two classes can now be separated by a hyperplane which corresponds to a non-linear
classifier (Furey et al, 2000). The class of the unknown sample is then determined by the side
of the “maximal marginal hyper plane” on which it lies. SVMs have been used to analyse
miRNA data by Xue et al, 2005.

Fig. 7. Schematic representation of the principle of SVM. SVM tries to maximise the margin
from the hyperplane in order to best separate the two classes (red positives from blue
negatives).

8. Conclusion
The capability of microarray to simultaneously analyse expression patterns of thousands of
DNA sequences, mRNA or miRNA transcripts has the potential to provide a unique insight
into the molecular biology of malignancy. However, the clinical relevance and value of
microarray data is highly dependent on a number of crucial factors including appropriate
experimental design and suitable bioinformatic analysis. Breast cancer is a heterogeneous
disease with many biological variables which need to be considered to generate meaningful
results. Cohort selection is critical and sufficient biological and technical replicates must be
included as part of microarray study design. Experimental protocols should be appropriate
to the research question. The research community have enthusiastically applied high
throughput technologies to the study of breast cancer. Class prediction, class comparison
and class discovery studies have been undertaken in an attempt to unlock the heterogeneity
of breast cancer and identify novel biomarkers. Molecular signatures have been generated
which attempt to outperform current histopathological parameters at prognostication and
112 Computational Biology and Applied Bioinformatics

prediction of response to therapy. Two clinical tests based on gene expression profiling
(Oncotype DX and Mammaprint) are already in clinical use and being evaluated in
multicentre international trials. It is essential that the potential of microarray signatures is
carefully validated before they are adopted as prognostic tools in the clinical setting.
Standards have been set for the reporting of microarray data (MIAME) and such data is
publically available to facilitate external validation and meta-analysis. It is imperative that
the data is integrated with knowledge normally processed in the clinical setting if we are to
overcome the difficulties in reproducibility, standardization and lack of proof of significance
beyond traditional clinicopathological tools that are limiting the incorporation of microarray
based tools into today’s standard of care.
Deriving biologically and clinically relevant results from microarray data is highly
dependent on bioinformatic analysis. Microarray data is limited by inherent characteristics
that render traditional statistical approaches less effective. These include high
dimensionality, false discovery rates, noise, complexity, non-normality and limited
reproducibility. High dimensionality remains one of the most critical challenges in the
analysis of microarray data. Hierarchical clustering approaches, which have been widely
used in the analysis of breast cancer microarray data, do not cope well with dimensionality.
In overcoming this challenge supervised machine learning techniques have been adapted to
the clinical setting to complement the existing statistical methods. The majority of machine
learning techniques originated in weak-theory domains such as business and marketing.
However, these approaches including Artificial Neural Networks and Support Vector
Machines have been successfully applied to the analysis of miRNA microarray data in the
context of clinical prognostication and prediction.
It is clear that the goal of translating microarray technology to the clinical setting requires
close collaboration between the involved scientific disciplines.If the current momentum in
microarray-based miRNA and mRNA translational research can be maintained this will add
an exciting new dimension to the field of diagnostics and prognostics and will bring us
closer to the ideal of individualized care for breast cancer patients.

9. References
Abbott AL, Alvarez-Saavedra E, Miska EA et al( 2005) The let-7 MiRNA family members
mir-48, mir-84, and mir-241 function together to regulate developmental timing in
Caenorhabditis elegans. Dev Cell. 9(3):403-14.
Adam BL, Qu Y, Davis JW, Ward MD et al (2002) Serum protein fingerprinting coupled with
a pattern-matching algorithm distinguishes prostate cancer from benign prostate
hyperplasia and healthy men. Cancer Research 62:3609-3614.
Ahmed AA, Brenton JD (2005) Microarrays and breast cancer clinical studies: forgetting
what we have not yet learnt. Breast Cancer Res 7:96–99.
Arciero C, Somiari SB, Shriver CD, et al. (2003). Functional relationship and gene ontology
classification of breast cancer biomarkers. Int. J. Biol. Markers 18: 241-272.
Ashburner M, Ball CA, Blake JA et al (2000). Gene Ontology: tool for the unification of
biology. Nat Genet. 25(1): 25–29.
Baffa R, Fassan M, Volinia S et al.(2009) MicroRNA expression profiling of human metastatic
cancers identifies cancer gene targets. J Pathol, 219(2), 214‐221
MicroArray Technology - Expression Profiling of MRNA and MicroRNA in Breast Cancer 113

Ball CA, Dolinski K, Dwight SS, et al (2000). Integrating functional genomic information into
the Saccharomyces Genome Database. Nucleic Acids Res;28:77–80
Ball G, Mian S, Holding F, et al (2002) An integrated approach utilizing artificial neural
networks and SELDI mass spectrometry for the classification of human tumours
and rapid identification of potential biomarkers. Bioinformatics 18:3395-3404.
Bartel DP. (2004) MiRNAs: genomics, biogenesis, mechanism and function. Cell; 116:281-97.
Bellman RE (1961) Adaptive Control Processes. Princeton University Press, Princeton, NJ
Benjamini Y, Hochberg Y (1995) Controlling the False Discovery Rate: A Practical and
Powerful Approach to Multiple Testing. Journal of the Royal Statistical Society.
Series B (Methodological);57:289-300.
Berns EM, van Staveren IL, Verhoog L et al (2001) Molecular profiles of BRCA1-mutated and
matched sporadic breast tumours: relation with clinico-pathological features.British
journal of cancer;85(4):538-45.
Bishop C (1995) Neural networks for pattern recognition. Oxford University Press.
Blake JA, Eppig JT, Richardson JE, Davisson MT (2000). The Mouse Genome Database
(MGD): expanding genetic and genomic resources for the laboratory mouse.
Nucleic Acids Res.;28:108–111
Blake JA, Harris MA (2002) The Gene Ontology (GO) project: structured vocabularies for
molecular biology and their application to genome and expression analysis. Curr
Protoc Bioinformatics. Chapter 7:Unit 7.2.
Blenkiron C, Goldstein LD, Thorne NP, et al (2007) MicroRNA expression profiling of
human breast cancer identifies new markers of tumor subtype. Genome Biol
8(10):R214
Brenton JD, Carey LA, Ahmed AA, Caldas C (2005). Molecular classification and molecular
forecasting of breast cancer: ready for clinical application? J Clin Oncol 23:7350–
7360.
Breiman L. Random Forests (2001) Machine Learning 45:5-32.
Buyse M, Loi S, van’t Veer L, et al. (2006) Validation and clinical utility of a 70-gene
prognostic signature for women with node-negative breast cancer. J Natl Cancer
Inst. 98(17):1183-1192.
Calin GA, Dumitru CD, Shimizu M, et al (2002) Frequent deletions and down-regulation of
micro- RNA genes miR15 and miR16 at 13q14 in chronic lymphocytic leukemia
PNAS.;99(24):15524-9.
Cardoso F, Van’t Veer L, Rutgers E, et al. (2008) Clinical application of the 70-gene profile:
the MINDACT trial. J Clin Oncol; 26:729–735.
Carey LA, Dees EC, Sawyer L et al (2007). The triple negative paradox: Primary tumor
chemosensitivity of breast cancer subtypes. Clin Cancer Res 2007; 13:2329 –2334.
Castoldi M, Schmidt S, Benes V, et al (2006) A sensitive array for MiRNA expression
profiling (miChip) based on locked nucleic acids (LNA). RNA;12(5):913-20.
Clarke R, Liu MC, Bouker KB, et al (2003). Antiestrogen resistance in breast cancer and the
role of estrogen receptor signaling. Oncogene;22(47):7316-39.
Cortez MA, Calin GA (2009). MiRNA identification in plasma and serum: a new tool to
diagnose and monitor diseases. Expert Opin Biol Ther;9(6):703-711.
114 Computational Biology and Applied Bioinformatics

Cronin, M, Pho M, Dutta D et al (2004). Measurement of gene expression in archival

paraffin-embedded tissues: development and performance of a 92-gene reverse
transcriptase-polymerase chain reaction assay. Am J Pathol. 164:35–42
Cunliffe HE, Ringner M, Bilke S, et al. (2003). The gene expression response of breast cancer
to growth regulators: patterns and correlation with tumor expression profiles.
Cancer Res. 63:7158-7166.
Desmedt C, Haibe-Kains B, Wirapati P, et al (2008). Biological processes associated with
breast cancer clinical outcome depend on the molecular subtypes. Clin Cancer Res
;14:5158-65.
Domeniconi C, Papadopoulos D, Gunopulos D, et al (2004). Subspace clustering of high
dimensional data. Proceedings 4th SIAM International Conference on Data
Mining, pp. 517-521. Lake Buena Vista, FL, SIAM, 3600 UNIV CITY SCIENCE
CENTER, PHILADELPHIA, PA 19104-2688 USA.
Dreiseitl S, Ohno-Machado L, Kittler H, et al (2001). A Comparison of Machine Learning
Methods for the Diagnosis of Pigmented Skin Lesions. Journal of Biomedical
Informatics;34:28-36.
Esquela-Kerscher A, Slack FJ.(2006) Oncomirs - MiRNAs with a role in cancer. Nature
reviews;6(4):259-69.
Fan C, Oh DS, Wessels L, et al (2006) Concordance among gene-expression-based predictors
for breast cancer. N Engl J Med;355:560–9.
Farmer P, Bonnefoi H, Becette V, et al (2005) Identification of molecular apocrine breast
tumours by microarray analysis.Oncogene;24:4660–71.
Ferlay J, Parkin DM, Steliarova-Foucher E (2010) Estimates of cancer incidence and mortality
in Europe in 2008. Eur J Cancer; 46:765-781.
Fisher B, Jeong JH, Bryant J, et al (2004). Treatment of lymph-node-negative, oestrogen
receptor-positive breast cancer: Long-term findings from National Surgical
Adjuvant Breast and Bowel Project randomised clinical trials. Lancet;364:858–868
Foekens JA, Sieuwerts AM, Smid M et al (2008) Four miRNAs associated with
aggressiveness of lymph node negative, estrogen receptor-positive human breast
cancer.PNAS;105(35):13021-6.
Furey T S, Cristianini N, Duffy N, et al (2000). Support vector machine classification and
validation of cancer tissue samples using microarray expression data.
Bioinformatics16:906-914.
Geyer FC, Lopez-Garcia MA, Lambros MB, Reis-Filho JS (2009) Genetic Characterisation of
Breast Cancer and Implications for Clinical Management. J Cell Mol Med (10):4090
103.
Gilad S, Meiri E, Yogev Y, et al (2008). Serum MiRNAs are promising novel biomarkers.
PLoS ONE. ;3(9):e3148.
Goldhirsch A, Wood WC, Gelber RD,et al (2007). Progress and promise: highlights of the
international expert consensus on the primary therapy of early breast cancer 2007.
Ann Oncol;18(7):1133-44.
Goldstein LJ, Gray R, Badve S, et al (2008) Prognostic utility of the 21-gene assay in hormone
receptor-positive operable breast cancer compared with classical clinicopathologic
features. J Clin Oncol;26:4063–4071
MicroArray Technology - Expression Profiling of MRNA and MicroRNA in Breast Cancer 115

Greene D, Cunningham P, Jorge A, et al. (2005). Producing accurate interpretable clusters

from high-dimensional data, Proceedings 9th European Conference on Principles
and Practice of Knowledge Discovery in Databases (PKDD), pp. 486-494, Porto
Portugal.
Habel LA, Shak S, Jacobs MK, et al (2006). A population-based study of tumor gene
expression and risk of breast cancer death among lymph node-negative patients.
Breast Cancer Res;8:R25.
Hedenfalk I, Duggan D, Chen Y, et al (2001) Gene expression profiles in hereditary breast
cancer. N Engl J Med.;344(8):539-48.
Heneghan HM, Miller N, Kerin MJ. (2010)MiRNAs as biomarkers and therapeutic targets in
cancer. Curr Opin Pharmacol;10(5):543-50.
Hu Z, Fan C, Oh DS,, et al. (2006) The molecularportraits of breast tumors are conserved
across microarray platforms. BMC Genom ;7:96.
Huang JX, Mehrens D, Wiese R,, et al. 2001. High-throughput genomic and Proteomic
analysis using microarray technology. Clinical Chem, 47: 1912-16.
Huang Q, Gumireddy K, Schrier M et al.(2008) The microRNAs miR‐373 and miR‐520c
promote tumour invasion and metastasis. Nat Cell Biol;10(2):202‐210
Izmirlian G (2004) Application of the random forest classification algorithm to a SELDI-TOF
proteomics study in the setting of a cancer prevention trial. Annals of the New
York Academy of Sciences; 1020:154-174
Iorio MV, Ferracin M, Liu CG, et al (2005). MicroRNA gene expression deregulation in
human breast cancer. Cancer research;65: 7065-70.
Jemal A, Siegel R, Ward E, et al. (2009) Cancer statistics, 2009. CA Cancer J Clin;59:225-249.
Khatri, P., Draghici, S. (2005), Ontological analysis of gene expression data: current tools,
limitations, and open problems. Bioinformatics; 21: 3587–3595.
Kim C, Taniyama Y, Paik S (2009). Gene-expression-based prognostic and predictive
markers for breast cancer- A primer for practicing pathologists Crit Rev Oncol
Hematol.;70(1):1-11.
Klebanov L, Yakovlev A (2007) How high is the level of technical noise in microarray data?
Biology Direct;2:9.
Kreike B, van Kouwenhove M, Horlings H et al (2007). Gene expression profiling and
histopathological characterization of triple-negative/basal-like breast carcinomas.
Breast Cancer Res;9:R65.
Korkola JE, DeVries S, Fridlyand J, et al (2003). Differentiation of lobular versus ductal
breast carcinomas by expression microarray analysis. Cancer Res;63:7167–7175.
Lamb J, Ramaswamy S, Ford HL, et al (2003). A mechanism of cyclin D1 action encoded in
the patterns of gene expression in human cancer. Cell; 114(3):323-34.
Lancashire LJ, Lemetre C, Ball GR (2009). An introduction to artificial neural networks in
bioinformatics--application to complex microarray and mass spectrometry datasets
in cancer studies. Briefings in Bioinformatics;10:315-329.
Lee RC, Feinbaum RL, Ambros V. (1993) The C. elegans heterochronic gene lin-4 encodes
small RNAs with antisense complementarity to lin-14. Cell.;75(5):843-54.
Leopold E, Kindermann J (2006). Content Classification of Multimedia Documents using
Partitions of Low-Level Features. Journal of Virtual Reality and Broadcasting 3(6).
116 Computational Biology and Applied Bioinformatics

Li J, Smyth P, Flavin R, et al. (2007) Comparison of miRNA expression patterns using total
RNA extracted from matched samples of formalin- fixed paraffin-embedded (FFPE)
cells and snap frozen cells. BMC biotechnology;7:36
Lisboa PJ, Taktak AF(2006). The use of artificial neural networks in decision support in
cancer: A systematic review. Neural Networks;19:408-415.
Lo SS, Norton J, Mumby PB et al(2007). Prospective multicenter study of the impact of the
21-gene recurrence score (RS) assay on medical oncologist (MO) and patient (pt)
adjuvant breast cancer (BC) treatment selection. J Clin Oncol;25(18 suppl):577
Loi S, Haibe-Kains B, Desmedt C, et al (2007). Definition of clinically distinct molecular
subtypes in estrogen receptor-positive breast carcinomas through genomic grade. J.
Clin. Oncol. 25, 1239–1246
Lowery AJ, Miller N, McNeill RE, Kerin MJ (2008). MicroRNAs as prognostic indicators and
therapeutic targets: potential effect on breast cancer management. Clin Cancer Res.
;14(2):360-5.
Lowery AJ, Miller N, Devaney A, et al (2009) . MicroRNA signatures predict estrogen
receptor, progesterone receptor and Her2/neu receptor status in breast cancer.
Breast Cancer Res.;11(3):R27.
Lu J, Getz G, Miska EA, et al.(2005) MiRNA expression profiles classify human cancers.
Nature. 2005;435(7043):834-8
Ma XJ, Hilsenbeck SG, Wang W et al (2006). The HOXB13:IL17BR expression index is a
prognostic factor in early-stage breast cancer. J Clin Oncol; 24:4611– 4619.
Ma J, Dong C, Ji C (2010). MicroRNA and drug resistance. Cancer Gene Ther, 17(8), 523‐531
Manning AT, Garvin JT, Shahbazi RI, et al (2007). Molecular profiling techniques and
bioinformatics in cancer research Eur J Surg Oncol;33(3):255-65.
Marchionni L, Wilson RF, Wolff AC, et al (2008). Systematic review: gene expression
profiling assays in early-stage breast cancer. Ann Intern Med.;148(5):358-369.
Marengo E, Robotti E, Righetti PG, et al (2004). Study of proteomic changes associated with
healthy and tumoral murine samples in neuroblastoma by principal component
analysis and classification methods. Clinica Chimica Acta;345:55-67.
Masuda N, Ohnishi T, KawamotoS , et al (1999) Analysis of chemical modification of RNA
from formalin-fixed samples and optimization of molecular biology applications
for such samples. Nucleic Acids Res. 27, 4436–4443
Matharoo-Ball B, Ratcliffe L, Lancashire L, et al (2007). Diagnostic biomarkers differentiating
metastatic melanoma patients from healthy controls identified by an integrated
MALDI-TOF mass spectrometry/bioinformatic approach. Proteomics Clinical
Applications; 1:605-620
Mattie MD, Benz CC, Bowers J, et al (2006). Optimized high-throughput MiRNA expression
profiling provides novel biomarker assessment of clinical prostate and breast
cancer biopsies. Molecular cancer;5:24
Michiels S, Koscielny S, Hill C (2005). Prediction of cancer outcome with microarrays: a
multiple random validation strategy. Lancet;365: 488-92.
Michiels S, Koscielny S, Hill C (2007). Interpretation of microarray data in cancer. British
Journal of Cancer;96:1155–1158.
MicroArray Technology - Expression Profiling of MRNA and MicroRNA in Breast Cancer 117

Mina L, Soule SE, Badve S, et al. (2007) Predicting response to primary chemotherapy: gene
expression profiling of paraffin-embedded core biopsy tissue. Breast Cancer Res
Treat ;103:197–208.
Mitchell PS, Parkin RK, Kroh EM, et al (2008).Circulating MiRNAs as stable blood- based
markers for cancer detection.PNAS;105(30):10513-8
Mook S, Schmidt MK, Viale G, et al (2009). The 70-gene prognosis signature predicts disease
outcome in breast cancer patients with 1–3 positive lymph nodes in an
independent validation study. Breast Cancer Res Treat;116:295–302.
Mootha VK, Lindgren CM, Eriksson KF, et al (2003). PGC-1alpha Responsive Genes
Involved in Oxidative Phosphorylation are Coordinately Downregulated in Human
Diabetes, Nature Genetics 34(3):267-73
Nielsen TO, Hsu FD, Jensen K et al (2004). Immunohistochemical and clinical
characterization of the basal-like subtype of invasive breast carcinoma. Clin Cancer
Res ;10:5367-74.
Oberley MJ, Tsao J, Yau P, Farnham PJ (2004). Highthroughput screening of chromatin
immunoprecipitates using CpG-island microarrays. Methods Enzymol;376: 315-
34.
Oostlander AE, Meijer GA, Ylstra B (2004). Microarraybased comparative genomic
hybridization and its applications in human genetics. Clin Genet, 66: 488-495.
Osborne CK(1998) Tamoxifen in the treatment of breast cancer. N Engl J Med;339(22):1609-
18.
Paik S, Shak S, Tang G, et al (2004). A multigene assay to predict recurrence of tamoxifen-
treated, node-negative breast cancer. N Engl J Med;351(27):2817-26.
Paik, S. Kim, C. Y, Song, Y. K. & Kim, W. S. (2005) Technology insight: application of
molecular techniques to formalin-fixed paraffin-embedded tissues from breast
cancer. Nat. Clin. Pract. Oncol;2:246–254
Paik S, Tang G, Shak S, , et al(2006). Gene expression and benefit of chemotherapy in women
with node-negative, estrogen receptor-positive breastcancer. J Clin Oncol; 24 (23) :
3726-34.
Parker JS, Mullins M, Cheang MC, et al (2009). Supervised risk predictor of breast cancer
based on intrinsic subtypes. J Clin Oncol;27:1160–1167.
Pedraza V, Gomez-Capilla JA, Escaramis G, et al (2010) Gene expression signatures in breast
cancer distinguish phenotype characteristics, histologic subtypes, and tumor
invasiveness. Cancer.;116(2):486-96.
Peppercorn J, Perou CM, Carey LA. (2008) Molecular subtypes in breast cancer evaluation
and management: divide and conquer. Cancer Invest;26:1–10.
Perou CM, Sorlie T, Eisen MB, et al (2000). Molecular portraits of human breast tumours.
Nature;406: 747-52.
Pusztai L, Mazouni C, Anderson K, et al (2006). Molecular classification of breast cancer:
limitations and potential. Oncologist;11:868–877.
Quackenbush J (2001). Computational analysis of microarray data. Nature Reviews Genetics
;2:418-27.
118 Computational Biology and Applied Bioinformatics

Raychaudhuri S, Stuart JM, Altman RB (2000). Principal components analysis to summarize

microarray experiments: application to sporulation time series. In Pacific
Symposium on Biocomputing, pp. 455–466.
Rifai N, Gillette MA, Carr SA (2006). Protein biomarker discovery and validation: the long
and uncertain path to clinical utility. Nature biotechnology;24:971-983
Rouzier R, Perou CM, Symmans WF et al (2005). Breast cancer molecular subtypes
respond differently to preoperative chemotherapy. Clin Cancer Res;11:5678 –
5685.
Schena M, Shalon D, Davis RW, Brown PO (1995). Quantitative monitoring of
gene expression patterns with a complementary DNA microarray. Science;270:467-
70.
Segura MF, Belitskaya-Lévy I, Rose A, et al (2010) Melanoma MicroRNA Signature Predicts
Post-Recurrence Survival. Clinical Cancer Research;16:1577.
Shak S, Baehner FL, Palmer G, et al (2006) Subtypes of breast cancer defined by
standardized quantitative RT–PCR analysis of 10 618 tumors. Breast Cancer Res
Treat 2006;100:S295–295.
Shi L, Reid LH, Jones WD, et al (2006)The microarray quality control (MAQC) project shows
inter- and intraplatform reproducibility of gene expression measurements. Nature
Biotechnologie;24:1151–1161.
Simon RM, Korn EL, McShane LM, et al (2003). Design and analysis of DNA microarray
investigations. Springer New York
Smith I, Procter M, Gelber RD, et al (2007). 2-year follow up of trastuzumab after adjuvant
chemotherapy in HER2-positive breast cancer: a randomised controlled trial.
Lancet. ;369(9555):29-36.
Sorlie T, Perou CM, Tibshirani R, et al (2001) Gene expression patterns of breast
carcinomas distinguish tumor subclasses with clinical implications.PNAS;98: 10869-
74.
Sorlie T, Tibshirani R, Parker J, et al (2003). Repeated observation of breast tumor subtypes
in independent gene expression data sets. PNAS;100:8418–8423.
Sorlie T, Perou CM, Fan C, et al (2006) Gene expression profiles do not consistently predict
the clinical treatment response in locally advanced breast cancer. Mol Cancer
Ther;5:2914–8.
Sotiriou C, Neo SY, McShane LM, et al (2003) Breast cancer classification and prognosis
based on gene expression profiles from a population based study.PNAS;100:10393–
10398
Sparano JA. (2006). TAILORx: Trial assigning individualized options for treatment (Rx). Clin
Breast Cancer;7:347–350.
Stekel D. (2003). Microarray bioinformatics. Cambrigde University Press,
Stoll D, Templin MF, Bachmann J, Joos TO (2005). Protein microarrays: applications and
future challenges. Curr Opin Drug Discov Devel, 8: 239-252.
Sun Y, Goodison S. Li J, Liu L., Farmerie W (2007). Improved breast cancer prognosis
through the combination of clinical and genetic markers, Bioinformatics;23:30–
37
MicroArray Technology - Expression Profiling of MRNA and MicroRNA in Breast Cancer 119

Tessel MA, Krett NL, Rosen ST (2010).Steroid receptor and microRNA regulation in cancer.
Curr Opin Oncol;22(6):592‐597
The FlyBase Consortium (1999). The FlyBase database of the Drosophila Genome Projects
and community literature. Nucleic Acids Res;27:85–88.
van de Vijver M, He Y, van’t Veer L, et al (2002). A gene-expression signature as a predictor
of survival in breast cancer. N Engl J Med;347:1999–2009.
van’t Veer L, Dai H, van de Vijver M, et al (2002). Gene expression profiling predicts clinical
outcome of breast cancer. Nature;415:530–6.
Vapnik V, Lerner A (1963). Pattern recognition using generalized portrait method.
Automation and Remote Control 1963;24:774-780.
Volinia S, Calin GA, Liu CG, et al (2006). A MiRNA expression signature of human solid
tumors defines cancer gene targets. PNAS.;103(7):2257-61.
Wadsworth JT, Somers KD, Cazares LH, et al (2004) Serum protein profiles to identify head
and neck cancer. Clinical Cancer Research;10:1625-1632.
Wang Y, Klijn JG, Zhang Y et al (2005). Gene-expression profiles to predict
distant metastasis of lymph-node-negative primary breast cancer. Lancet;365:671–
679.
Warnat P, Eils R, Brors B (2005). Cross-platform analysis of cancer microarray data
improves gene expression based classification of phenotypes. BMC
bioinformatics 2;6: 265.
Weigelt B, Geyer FC, Natrajan R, et al (2010) The molecular underpinning of lobular
histological growth pattern: a genome-wide transcriptomic analysis of invasive
lobular carcinomas and grade- and molecular subtype-matched invasive ductal
carcinomas of no special type. J Pathol;220(1):45-57
Wirapati P, Sotiriou C, Kunkel S, et al (2008). Meta-analysis of gene expression profiles in
breast cancer: toward a unified understanding of breast cancer subtyping and
prognosis signatures. Breast Cancer Res;10:R65.
Wong JWH, Cagney G, Cartwright HM (2005). SpecAlign—processing and alignment of
mass spectra datasets. Bioinformatics;21:2088-2090
Xi Y, Nakajima G, Gavin E, et al (2007). Systematic analysis of MiRNA expression of RNA
extracted from fresh frozen and formalin-fixed paraffin-embedded samples.
RNA;13(10):1668-74.
Xue C, Li F, He T, Liu GP, Li Y, Xuegong Z (2005). Classification of real and pseudo
microRNA precursors using local structure-sequence features and support vector
machine. BMC Bioinformatics;6:310.
Xu R, Xu J, Wunsch DC (2009). Using default ARTMAP for cancer classification with
MicroRNA expression signatures, International Joint Conference on Neural Networks,
pp.3398-3404,
Yan PS, Perry MR, Laux DE, et al (2000). CpG island arrays: an application toward
deciphering epigenetic signatures of breast cancer. Clinical Cancer Research; 6:
1432-38.
Yousef M, Najami N, Khalifa W (2010). A comparison study between one-class and two-
class machine learning for MicroRNA target detection. Journal of Biomedical
Science and Engineering ;3:247-252.
120 Computational Biology and Applied Bioinformatics

Zhao H, Langerod A, Ji Y, et al (2004) Different gene expression patterns in invasive lobular

and ductal carcinomas of the breast. Mol Biol Cell;15:2523–2536.
Zhao H, Shen J, Medico L, et al (2010). A Pilot Study of Circulating miRNAs as Potential
Biomarkers of Early Stage Breast Cancer. PLoS One;5(10),e137:5 , 2010
Zheng T, Wang J, Chen X, Liu L (2010) Role of microRNA in anticancer drug resistance. Int J
Cancer;126(1):2-10.
 6

Computational Tools for Identification of

microRNAs in Deep Sequencing Data Sets
Manuel A. S. Santos and Ana Raquel Soares
University of Aveiro
Portugal

1. Introduction
MicroRNAs (miRNAs) are a class of small RNAs of approximately 22 nucleotides in length
that regulate eukaryotic gene expression at the post-transcriptional level (Ambros 2004;
Bartel 2004; Filipowicz et al. 2008). They are transcribed as long precursor RNA molecules
(pri-miRNAs) and are successively processed by two key RNAses, namely Drosha and
Dicer, into their mature forms of ~22 nucleotides (Kim 2005; Kim et al. 2009). These small
RNAs regulate gene expression by binding to target sites in the 3’ untranslated region of
mRNAs (3’UTR). Recognition of the 3’UTR by miRNAs is mediated through complementary
hybridization at least between nucleotides 2-8, numbered from the 5’ end (seed sequences)
of the small RNAs, and complementary sequences present in the 3’UTRs of mRNAs
(Ambros 2004; Bartel 2004; Zamore and Haley 2005). Perfect or nearly perfect
complementarities between miRNAs and their 3’UTRs induce mRNA cleavage by the RNA-
induced silencing complex (RISC), whereas imperfect base pair matching may induce
translational silencing through various molecular mechanisms, namely inhibition of
translation initiation and activation of mRNA storage in P-bodies and/or stress granules
(Pillai et al. 2007).
This class of small RNAs is well conserved between eukaryotic organisms, suggesting that
they appeared early in eukaryotic evolution and play fundamental roles in gene expression
control. Each miRNA may repress hundreds of mRNAs and regulate a wide variety of
biological processes, namely developmental timing (Feinbaum and Ambros 1999; Lau et al.
2001), cell differentiation (Tay et al. 2008), immune response (Ceppi et al. 2009) and infection
(Chang et al. 2008). For this reason, their identification is essential to understand eukaryotic
biology. Their small size, low abundance and high instability complicated early
identification, but these obstacles have been overcome by next generation sequencing
approaches, namely the Genome SequencerTM FLX from Roche, the Solexa/Illumina
Genome Analyzer and the Applied Biosystems SOLiDTM Sequencer which are currently
being routinely used for rapid miRNA identification and quantification in many eukaryotes
(Burnside et al. 2008; Morin et al. 2008; Schulte et al. 2010).
As in other vertebrates, miRNAs control gene expression in zebrafish, since defective
miRNA processing arrest development (Wienholds et al. 2003). Also, a specific subset of
miRNAs is required for brain morphogenesis in zebrafish embryos, but not for cell fate
determination or axis formation (Giraldez et al. 2005). In other words, miRNAs play an
122 Computational Biology and Applied Bioinformatics

important role in zebrafish organogenesis and their expression at specific time points is
relevant to organ formation and differentiation. Since identification of the complete set of
miRNAs is fundamental to fully understand biological processes, we have used high
throughput 454 DNA pyrosequencing technologies to fully characterize the zebrafish
miRNA population (Soares et al. 2009). For this, a series of cDNA libraries were prepared
from miRNAs isolated at different embryonic time points and from fully developed organs
sequenced using the Genome SequencerTM FLX. This platform yields reads of up to 200
bases each and can generate up to 1 million high quality reads per run, which provides
sufficient sequencing coverage for miRNA identification and quantification in most
organisms. However, deep sequencing of small RNAs may pose some problems that need to
be taken into consideration to avoid sequencing biases. For example, library preparation and
computational methodologies for miRNA identification from large pool of reads need to be
optimized. There are many variables to consider, namely biases in handling large sets of
data, sequencing errors and RNA editing or splicing. If used properly, deep sequencing
technologies have enormous analytical power and have been proven to be very robust in
retrieving novel small RNA molecules. One of the major challenges when analyzing deep
sequencing data is to differentiate miRNAs from other small RNAs and RNA degradation
products.
Different research groups are developing dedicated computational methods for the
identification of miRNAs from large sets of sequencing data generated by next
generation sequencing experiments. miRDeep (http://www.mdc-
berlin.de/en/research/research_teams/systems_biology_of_gene_regulatory_elements/pr
ojects/miRDeep/index.html) (Friedlander et al. 2008) and miRanalizer
(http://web.bioinformatics.cicbiogune.es/microRNA/miRanalyser.php) (Hackenberg et al.
2009) can both detect known miRNAs annotated in miRBase and predict new miRNAs
(although using different prediction algorithms) from small RNA datasets generated by
deep sequencing. Although these online algorithms are extremely useful for miRNA
identification, custom-made pipeline analysis of deep sequencing data may be performed in
parallel to uncover the maximum number of small non-coding RNA molecules present in
the RNA datasets.
In this chapter, we discuss the tools and computational pipelines used for miRNA
identification, discovery and expression from sequencing data, based on our own experience
of deep sequencing of zebrafish miRNAs, using the Genome SequencerTM FLX from Roche.
We show how a combination of a public available, user-friendly algorithm, such as
miRDeep, with custom-built analysis pipelines can be used to identify non-coding RNAs
and uncover novel miRNAs. We also demonstrate that population statistics can be applied
to statistical analysis of miRNA populations identified during sequencing and we
demonstrate that robust computational analysis of the data is crucial for extracting the
maximum information from sequencing datasets.

2. miRNA identification by next-generation sequencing

2.1 Extraction of next-generation sequencing data
Next generation sequencing methods have been successfully applied in the last years to
miRNA identification in a variety of organisms. However, the enormous amount of data
generated represents bioinformatics challenges that researchers have to overcome in order
to extract relevant data from the datasets.
Computational Tools for Identification of microRNAs in Deep Sequencing Data Sets 123

We have used the Genome SequencerTM FLX system (454 sequencing) to identify zebrafish
miRNAs from different developmental stages and from different tissues. For this, cDNA
libraries are prepared following commonly used protocols (Droege M and Hill B. 2008;
Soares et al. 2009). These libraries contain specific adaptors for the small RNA molecules
containing specific priming sites for sequencing. After sequencing, raw data filtration and
extraction is performed using specialist software incorporated into the Genome SequencerTM
FLX system (Droege M and Hill B. 2008). Raw images are processed to remove background
noise and the data is normalized. Quality of raw sequencing reads is based on complete read
through of the adaptors incorporated into the cDNA libraries. The 200 base pair of 454
sequencing reads provide enough sequencing data for complete read through of the
adaptors and miRNAs. During quality control, the adaptors are trimmed and the resulting
sequences are used for further analysis. Sequences ≥ 15 nucleotides are kept for miRNA
identification, and constitute the small RNA sequencing data.
Other sequencing platforms, such as Illumina/Solexa and SOLiDTM, also have specialist
software for raw data filtration. DSAP, for example, is an automated multiple-task web
service designed to analyze small RNA datasets generated by the Solexa platform (Huang et
al. 2010). This software filters raw data by removing sequencing adaptors and poly-
A/T/C/G/N nucleotides. In addition, it performs non-coding RNA matching by sequence
homology mapping against the non-coding RNA database Rfam (rfam.sanger.ac.uk/) and
detects known miRNAs in miRBase (Griffiths-Jones et al. 2008), based on sequence
homology.
The SOLiDTM platform has its own SOLiD™ System Small RNA Analysis Pipeline Tool
(RNA2MAP), which is available online (http://solidsoftwaretools.com/gf/project/
rna2map). This software is similar to DSAP, as it filters raw data and identifies known
miRNAs in the sequencing dataset by matching reads against miRBase sequences and
against a reference genome. Although these specialist software packages are oriented for
miRNA identification in sequencing datasets they are not able to identify novel miRNAs.
For this, datasets generated from any of the sequencing platforms available have to be
analyzed using tools that include algorithms to identify novel miRNAs.

2.2 miRNA identification from next generation sequencing databases

miRNA identification (of both known and novel molecules) from datasets generated by
deep-sequencing has been facilitated by the development of public user friendly algorithms,
such as miRDeep (Friedlander et al. 2008), miRanalyzer (Hackenberg et al. 2009) and
miRTools (Zhu et al. 2010).
We used miRDeep to identify miRNAs in our sequencing datasets (Figure 1). miRDeep was
the first public tool available for the analysis of deep-sequencing miRNA data. This software
was developed to extract putative precursor structures and predict secondary structures
using RNAfold (Hofacker 2003) after genome alignment of the sequences retrieved by next-
generation sequencing. This algorithm relies on the miRNA biogenesis model. Pre-miRNAs
are processed by DICER, which originates three different fragments, namely the mature
miRNA, the star and the hairpin loop sequences (Kim et al. 2009). miRDeep scores the
compatibility of the position and frequency of the sequenced RNA with the secondary
structures of the miRNA precursors and identifies new, conserved and non-conserved
miRNAs with high confidence. It distinguishes between novel and known miRNAs, by
evaluating the presence or absence of alignments of a given sequence with the stem loop
124 Computational Biology and Applied Bioinformatics

sequences deposited in miRBase. The sequence with the highest expression is always
considered as the mature miRNA sequence by the miRDeep algorithm. All hairpins that are
not processed by DICER will not match a typical secondary miRNA structure and are
filtered out.
After aligning the sequences against the desired genome using megaBlast, the blast output is
parsed for miRDeep uploading. As sequencing errors, RNA editing and RNA splicing may
alter the original miRNA sequence, one can re-align reads that do not match the genome
using SHRiMP (http://compbio.cs.toronto.edu/shrimp/). The retrieved alignments are also
parsed for miRDeep for miRNA prediction. miRDeep itself allows up to 2 mismatches in the
3’ end of each sequence, which already accounts with some degree of sequencing errors that
might have occurred.
Reads matching more than 10 different genome loci are generally discarded, as they likely
constitute false positives. The remaining alignments are used as guidelines for excision of
the potential precursors from the genome. After secondary structure prediction of putative
precursors, signatures are created by retaining reads that align perfectly with those putative
precursors to generate the signature format. miRNAs are predicted by discarding non-
plausible DICER products and scoring plausible ones. The latter are blasted against mature
miRNAs deposited in miRBase, to extract known and conserved miRNAs. The remaining
reads are considered novel miRNAs.
In order to evaluate the sensitivity of the prediction and data quality, miRDeep calculates
the false positive rate, which should be below 10%. For this, the signature and the structure-
pairings in the input dataset are randomly permutated, to test the hypothesis that the
structure (hairpin) of true miRNAs is recognized by DICER and causes the signature.
miRanalizer (Hackenberg et al. 2009) is a recently developed web server tool that detects
both known miRNAs annotated in miRBase and other non-coding RNAs by mapping
sequences to non-coding RNA libraries, such as Rfam. This feature is important, as more
classes of small non coding RNAs are being unravelled and their identification can provide
clues about their functions. At the same time, by removing reads that match other non
coding RNA classes, it reduces the false positive rate in the prediction of novel miRNAs, as
these small non coding RNAs can be confused with miRNAs. For novel miRNA prediction,
miRanalizer implements a machine learning approach based on the random forest method,
with the number of trees set to 100 (Breiman 2001). miRanalyzer can be applied to miRNA
discovery in different models, namely human, mouse, rat, fruit-fly, round-worm, zebrafish
and dog, and uses datasets from different models to build the final prediction model. In
comparison to miRDeep, this is disadvantageous as the latter can predict novel miRNAs
from any model. All pre-miRNAs candidates that match known miRNAs are extracted from
the experimental dataset and labelled as positive instances. Next, an equal amount of pre-
miRNA candidates from the same dataset are selected by random selection with the known
miRNAs removed and labelled as negative. Pre-processing of reads corresponding to
putative new miRNAs includes clustering of all reads that overlap with the genome, testing
whether the start of the current read overlaps less than 3 nucleotides with the end position
of previous reads. This avoids DICER products grouping together and be considered non-
miRNAs products, which would increase false negatives. Besides, clusters of more than 25
base pairs in length are discarded and the secondary structure of the miRNA is predicted
via RNAfold (Hofacker 2003). Structures where the cluster sequence is not fully included
and where part of the stem cannot be identified as a DICER product are discarded.
Computational Tools for Identification of microRNAs in Deep Sequencing Data Sets 125

Fig. 1. Data pipeline analysis using miRDeep.

126 Computational Biology and Applied Bioinformatics

miRTools is a comprehensive web server that can be used for characterization of the small
RNA transcriptome (Zhu et al. 2010). It offers some advantages relative to miRDeep and
miRanalyzer, since it integrates multiple computational approaches including tools for raw
data filtration, identification of novel miRNAs and miRNA expression profile generation. In
order to detect novel miRNAs, miRTools analyze all sequences that are not annotated to
known miRNAs, other small non-coding RNAs and genomic repeats or mRNA that match the
reference genome. These sequences are extracted and their RNA secondary structures are
predicted using RNAfold (Hofacker 2003) and novel miRNAs are identified using miRDeep.

2.3 Analysis of discarded reads by miRNA identification algorithms can identify new
miRNAs
Since miRDeep and miRanalyzer are highly stringent algorithms, some miRNAs may escape
detection. The false negative discovery rate can, however be calculated by simply
performing a megaBlast search of the sequencing data against the miRNAs deposited in
miRBase. Perfect alignments are considered true positives. The list of known miRNAs
identified by this method is compared to the list of known miRNAs identified by miRDeep
or miRanalyzer. False negatives are those miRNAs present in the blast analysis, but which
were missed by the miRNA prediction algorithms. This is, in our opinion, an essential
control, as it gives information about the percentage of miRNAs that may have escaped
miRDeep or miRanalyzer analysis. We have detected ~19% of false negatives, which
prompted us to develop a parallel pipeline to analyze reads that may have been incorrectly
discarded by the original algorithm (Figure 2). This analysis can and should be performed
independently of the algorithm used to retrieve miRNAs from deep sequencing data.
To overcome the lack of sensitivity of miRDeep, our parallel bioinformatics pipeline
includes a megaBlast alignment between the dataset of discarded reads by miRDeep and
mature sequences deposited in miRBase. Besides, novel transcripts encoding miRNAs
predicted by computational tools can be retrieved from the latest Ensembl version using
BioMart and also from literature predictions. These sequences are then used to perform a
megaBlast search against the sequencing data. The transcripts with perfect matches and
alignment length > 18 nucleotides are kept for further processing. These transcripts are then
compared with the mature miRNAs deposited in miRBase and those that produce imperfect
alignments or do not produce alignments are considered novel miRNAs. Imperfect alignments
may identify conserved miRNAs if there is a perfect alignment in the seed region.
Complementary alignments of our dataset reads against the zebrafish genome with SHRiMP
alignments and complementary miRDeep analysis with an analysis of the reads discarded
by this algorithm, allowed us to identify 90% of the 192 zebrafish miRNAs previously
identified, plus 107 miRNA star sequences and 25 novel miRNAs.

2.4 Generation of miRNA profiles from deep sequencing data

Deep sequencing of miRNAs can also be used to generate miRNA expression profiles as the
absolute number of sequencing reads of each miRNA is directly proportional to their
relative abundance. miRNA profiles can be generated based on the number of reads of each
particular miRNA. However, a normalization step is essential to compare miRNA
expression levels between different samples. The variation in the total number of reads
between samples leads to erroneous interpretation of miRNA expression patterns by direct
Computational Tools for Identification of microRNAs in Deep Sequencing Data Sets 127

comparison of read numbers (Chen et al. 2005). Normalization assumes that the small RNA
population is constant and is represented by an arbitrary value (e.g. 1000), and can be
calculated as indicated below:

miRNA relative expression = 1000 x (NRmiRNAXY)

TNRmiRNAsY
where NRmiRNAXY is the number of reads of miRNAX (X = any miRNA) in sample Y, and
TNRmiRNAsY is the total number of miRNAs in sample Y. 1000 is an arbitrary number of
reads that allows for data normalization across different samples. This calculates the relative
expression of a specific miRNA in a given sample, relative to all miRNAs expressed.

Fig. 2. Bioinformatics pipeline of reads discarded by miRDeep (-i and –d stand for query file
and database respectively).
128 Computational Biology and Applied Bioinformatics

Using this formula it is possible to generate miRNA profiles for each sample sequenced.
These profiles provide valuable information about relative miRNA expression, which is
essential to understand miRNA function in different tissues. In order to compare miRNA
profiles of two deep sequencing samples (e.g. condition vs control), a two-side t-test can be
applied to determine miRNA levels. Sequence count values should be log-transformed to
stabilize variance (Creighton et al. 2009). miRTools already include a computational
approach to identify significantly differentially expressed miRNAs (Zhu et al. 2010). It
compares differentially expressed miRNAs in multiple samples after normalization of the
read count of each miRNA with the total number of miRNA read counts which are matched
to the reference genome. The algorithm calculates statistical significance (P-value) based on
a Bayesian method (Audic and Claverie 1997), which accounts for sampling variability of
tags with low counts. Significantly differentially expressed miRNAs are those that show P-
values <0.01 and at least 2-fold change in normalized sequence counts.

2.5 Statistical analysis of miRNA population

The platforms available for miRNA sequencing offer different sequencing coverage, ranging
from thousands to millions of reads. In principle, higher sequencing coverage will enable
discovery of more miRNA molecules in a sequencing run. However, technical problems
during sample preparation can interfere with good quality sequencing of small RNAs. One
of the most common problems is the generation of primer dimmers during PCR
amplification of cDNA libraries. This may indicate an excess of primers during
amplification, when compared to the miRNA levels in a given cDNA library or low
annealing temperature. This problem is often only detected after sequencing. When this
happens, a large number of reads do not pass quality control filters and the number of reads
corresponding to small RNAs is considerably lower than the initial sequencing coverage.
Besides this, quality control filters do not consider reads with sequencing errors in the
adaptors or without recognizable adaptors. For these reasons, a tool that verifies if the
sequencing coverage is sufficient to retrieve most miRNAs in a given sample is important.
A useful approach to assess the representativeness of miRNA reads in a sequencing
experiment is to apply population statistics to the overall miRNA population. We have
developed a statistical tool to calculate how many miRNAs are expected in a given
sequencing experiment and how many reads are needed to identify them. Rarefaction
curves of the total number of reads obtained versus the total number of miRNA species
identified are plotted and the total richness of the miRNA population is determined. Chao1,
a non-parametric richness estimator (Chao 1987), can be used to determine the total richness
of the miRNA population, as a function of the observed richness (Sobs), and the number of
total sequences obtained by sequencing. The value obtained represents the number of
different miRNAs that can be identified in a specific sequencing experiment. The rarefaction
curve estimates the number of reads needed to identify the different miRNAs that may be
present in a sequencing run. For example, 206 miRNAs are expected to be present in a
sequencing experiment that retrieves approximately 40000 reads (Figure 3). The steep curve
levels off towards an asymptote, indicating the point (~20000 reads) where additional
sampling will not yield extra miRNAs. As that critical point is below the total number of
reads obtained, we can conclude that the sequencing coverage is sufficient to identify all
miRNAs predicted in the particular sample. Rarefaction curves and the Chao1 statistical
estimator are computed using EstimateS8.0 (Colwell and Coddington 1994).
Computational Tools for Identification of microRNAs in Deep Sequencing Data Sets 129

Fig. 3. Statistical analysis of miRNA population. A) A rarefaction curve of the total number
of reads generated by deep sequencing versus the total number of miRNA species identified
is shown. The steep curve levels off towards an asymptote, which indicates the point where
additional sampling will not yield new miRNAs. B) Homogeneity of the miRNA population
was assessed using population statistics and by determining the Chao1 diversity estimator.
The Chao1 reached a mean stable value of 207, with lower and upper limits of 200.37 to
229.66, respectively, for a level of confidence of 95%.

3. Conclusion
Small non-coding RNAs are a class of molecules that regulate several biological processes.
Identification of such molecules is crucial to understand the molecular mechanisms that
they regulate. There are already several deep sequencing approaches to identify these
molecules. However, correct interpretation of sequencing data depends largely on the
bioinformatics and statistical tools available. There are online algorithms that facilitate
identification of miRNAs and other small non-coding RNAs from large datasets. However,
there are no tools to predict novel small non-coding RNAs beyond miRNAs. As those
additional RNA classes, namely piRNAs, snRNAs and snoRNAs are processed differently,
the development of algorithms based solely on their biogenesis is challenging. Moreover,
the available algorithms have some limitations and additional data analysis should be
performed with the discarded reads that can potentially hold non-conventional miRNA
molecules. Analysis of deep sequencing data is a powerful methodology to identify novel
miRNAs in any organism and determine their expression profiles. The challenge is to deal
with increasing dataset size and to integrate the information generated by small RNA
sequencing experiments. This will be essential to understand how different RNA classes are
related. Computational tools to integrate small non-coding RNA data with gene expression
data and target predictions are pivotal to understand the biological processes regulated by
miRNAs and other small non-coding RNA classes.
130 Computational Biology and Applied Bioinformatics

4. References
Ambros, V. (2004). The functions of animal microRNAs. Nature 431(7006), 350-355.
Audic, S., and Claverie, J. M. (1997). The significance of digital gene expression profiles.
Genome Res. 7(10), 986-995.
Bartel, D. P. (2004). MicroRNAs: genomics, biogenesis, mechanism, and function. Cell 116(2),
281-297.
Breiman, L. Random forests. Machine Learning (2001). 45:28.
Burnside, J., Ouyang, M., Anderson, A., Bernberg, E., Lu, C., Meyers, B. C., Green, P. J.,
Markis, M., Isaacs, G., Huang, E., and Morgan, R. W. (2008). Deep sequencing of
chicken microRNAs. Bmc Genomics 9.
Ceppi, M., Pereira, P. M., Dunand-Sauthier, I., Barras, E., Reith, W., Santos, M. A., and
Pierre, P. (2009). MicroRNA-155 modulates the interleukin-1 signaling pathway in
activated human monocyte-derived dendritic cells. Proc. Natl. Acad. Sci. U. S. A
106(8), 2735-2740.
Chang, J. H., Cruo, J. T., Jiang, D., Guo, H. T., Taylor, J. M., and Block, T. M. (2008). Liver-
specific MicroRNA miR-122 enhances the replication of hepatitis C virus in
nonhepatic cells. Journal of Virology 82(16), 8215-8223.
Chao, A. (1987). Estimating the Population-Size for Capture Recapture Data with Unequal
Catchability. Biometrics 43(4), 783-791.
Chen, P. Y., Manninga, H., Slanchev, K., Chien, M. C., Russo, J. J., Ju, J. Y., Sheridan, R., John,
B., Marks, D. S., Gaidatzis, D., Sander, C., Zavolan, M., and Tuschl, T. (2005). The
developmental miRNA profiles of zebrafish as determined by small RNA cloning.
Genes & Development 19(11), 1288-1293.
Colwell, R. K., and Coddington, J. A. (1994). Estimating Terrestrial Biodiversity Through
Extrapolation. Philosophical Transactions of the Royal Society of London Series B-
Biological Sciences 345(1311), 101-118.
Creighton, C. J., Reid, J. G., and Gunaratne, P. H. (2009). Expression profiling of microRNAs
by deep sequencing. Brief. Bioinform. 10(5), 490-497.
Droege M, and Hill B. The Genome Sequencer FLX trade mark System-Longer reads, more
applications, straight forward bioinformatics and more complete data sets.
J.Biotechnol. 136 (1-2): 3-10. 2008.
Feinbaum, R., and Ambros, V. (1999). The timing of lin-4 RNA accumulation controls the
timing of postembryonic developmental events in Caenorhabditis elegans. Dev.
Biol. 210(1), 87-95.
Filipowicz, W., Bhattacharyya, S. N., and Sonenberg, N. (2008). Mechanisms of post-
transcriptional regulation by microRNAs: are the answers in sight? Nat. Rev. Genet.
9(2), 102-114.
Friedlander, M. R., Chen, W., Adamidi, C., Maaskola, J., Einspanier, R., Knespel, S., and
Rajewsky, N. (2008). Discovering microRNAs from deep sequencing data using
miRDeep. Nature Biotechnology 26(4), 407-415.
Giraldez, A. J., Cinalli, R. M., Glasner, M. E., Enright, A. J., Thomson, J. M., Baskerville, S.,
Hammond, S. M., Bartel, D. P., and Schier, A. F. (2005). MicroRNAs regulate brain
morphogenesis in zebrafish. Science 308(5723), 833-838.
Computational Tools for Identification of microRNAs in Deep Sequencing Data Sets 131

Griffiths-Jones, S., Saini, H. K., van, D. S., and Enright, A. J. (2008). miRBase: tools for
microRNA genomics. Nucleic Acids Res. 36(Database issue), D154-D158.
Hackenberg, M., Sturm, M., Langenberger, D., Falcon-Perez, J. M., and Aransay, A. M.
(2009). miRanalyzer: a microRNA detection and analysis tool for next-
generation sequencing experiments. Nucleic Acids Res. 37(Web Server issue),
W68-W76.
Hofacker, I. L. (2003). Vienna RNA secondary structure server. Nucleic Acids Res. 31(13),
3429-3431.
Huang, P. J., Liu, Y. C., Lee, C. C., Lin, W. C., Gan, R. R., Lyu, P. C., and Tang, P. (2010).
DSAP: deep-sequencing small RNA analysis pipeline. Nucleic Acids Res. 38(Web
Server issue), W385-W391.
Kim, V. N. (2005). MicroRNA biogenesis: coordinated cropping and dicing. Nat. Rev. Mol.
Cell Biol. 6(5), 376-385.
Kim, V. N., Han, J., and Siomi, M. C. (2009). Biogenesis of small RNAs in animals. Nat. Rev.
Mol. Cell Biol. 10(2), 126-139.
Lau, N. C., Lim, L. P., Weinstein, E. G., and Bartel, D. P. (2001). An abundant class of tiny
RNAs with probable regulatory roles in Caenorhabditis elegans. Science 294(5543),
858-862.
Morin, R. D., O'Connor, M. D., Griffith, M., Kuchenbauer, F., Delaney, A., Prabhu, A. L.,
Zhao, Y., McDonald, H., Zeng, T., Hirst, M., Eaves, C. J., and Marra, M. A. (2008).
Application of massively parallel sequencing to microRNA profiling and discovery
in human embryonic stem cells. Genome Research 18(4), 610-621.
Pillai, R. S., Bhattacharyya, S. N., and Filipowicz, W. (2007). Repression of protein synthesis
by miRNAs: how many mechanisms? Trends Cell Biol. 17(3), 118-126.
Schulte, J. H., Marschall, T., Martin, M., Rosenstiel, P., Mestdagh, P., Schlierf, S., Thor,
T., Vandesompele, J., Eggert, A., Schreiber, S., Rahmann, S., and Schramm, A.
(2010). Deep sequencing reveals differential expression of microRNAs in
favorable versus unfavorable neuroblastoma. Nucleic Acids Res. 38(17), 5919-
5928.
Soares, A. R., Pereira, P. M., Santos, B., Egas, C., Gomes, A. C., Arrais, J., Oliveira, J. L.,
Moura, G. R., and Santos, M. A. S. (2009). Parallel DNA pyrosequencing unveils
new zebrafish microRNAs. Bmc Genomics 10.
Tay, Y. M. S., Tam, W. L., Ang, Y. S., Gaughwin, P. M., Yang, H., Wang, W. J., Liu, R. B.,
George, J., Ng, H. H., Perera, R. J., Lufkin, T., Rigoutsos, I., Thomson, A. M., and
Lim, B. (2008). MicroRNA-134 modulates the differentiation of mouse embryonic
stem cells, where it causes post-transcriptional attenuation of Nanog and LRH1.
Stem Cells 26(1), 17-29.
Wienholds, E., Koudijs, M. J., van Eeden, F. J., Cuppen, E., and Plasterk, R. H. (2003). The
microRNA-producing enzyme Dicer1 is essential for zebrafish development. Nat.
Genet. 35(3), 217-218.
Zamore, P. D., and Haley, B. (2005). Ribo-gnome: the big world of small RNAs. Science
309(5740), 1519-1524.
132 Computational Biology and Applied Bioinformatics

Zhu, E., Zhao, F., Xu, G., Hou, H., Zhou, L., Li, X., Sun, Z., and Wu, J. (2010). mirTools:
microRNA profiling and discovery based on high-throughput sequencing. Nucleic
Acids Res. 38(Web Server issue), W392-W397.
 7

Computational Methods in Mass

Spectrometry-Based Protein 3D Studies
Rosa M. Vitale1, Giovanni Renzone2, Andrea Scaloni2 and Pietro Amodeo1
1Istitutodi Chimica Biomolecolare, CNR, Pozzuoli
2Laboratorio di Proteomica e Spettrometria di Massa, ISPAAM, CNR, Naples
Italy

1. Introduction
Mass Spectrometry (MS)-based strategies featuring chemical or biochemical probing
represent powerful and versatile tools for studying structural and dynamic features of
proteins and their complexes. In fact, they can be used both as an alternative for systems
intractable by other established high-resolution techniques, and as a complementary
approach to these latter, providing different information on poorly characterized or very
critical regions of the systems under investigation (Russell et al., 2004). The versatility of
these MS-based methods depends on the wide range of usable probing techniques and
reagents, which makes them suitable for virtually any class of biomolecules and complexes
(Aebersold et al., 2003). Furthermore, versatility is still increased by the possibility of
operating at very different levels of accuracy, ranging from qualitative high-throughput fold
recognition or complex identification (Young et al., 2000), to the fine detail of structural
rearrangements in biomolecules after environmental changes, point mutations or complex
formations (Nikolova et al.,1998; Millevoi et al., 2001; Zheng et al., 2007). However, these
techniques heavily rely upon the availability of powerful computational approaches to
achieve a full exploitation of the information content associated with the experimental data.
The determination of three-dimensional (3D) structures or models by MS-based techniques
(MS3D) involves four main activity areas: 1) preparation of the sample and its derivatives
labelled with chemical probes; 2) generation of derivatives/fragments of these molecules for
further MS analysis; 3) interpretation of MS data to identify those residues that have reacted
with probes; 4) derivation of 3D structures consistent with information from previous steps.
Ideally, this procedure should be considered the core of an iterative process, where the final
model possibly prompts for new validating experiments or helps the assignment of
ambiguous information from the mass spectra interpretation step.
Both the overall MS3D procedure and its different steps have been the subject of several
accurate review and perspective articles (Sinz, 2006; Back et al., 2003; Young et al., 2000;
Friedhoff, 2005, Renzone, et al., 2007a). However, with the partial exception of a few recent
papers (Van Dijk et al., 2005; Fabris et al., 2010; Leitner et al., 2010), the full computational
detail behind 3D model building (step 4) has generally received less attention than the
former three steps. Structural derivation in MS3D, in fact, is considered a special case of
structural determination from sparse/indirect constraints (SD-SIC). Nevertheless,
information for modelling derivable from MS-based experiments exhibits some peculiar
134 Computational Biology and Applied Bioinformatics

features that differentiate it from the data types associated with other experimental
techniques involved in SD-SIC procedures, such as nuclear magnetic resonance (NMR),
electron microscopy, small-angle X-ray scattering (SAXS), Förster resonance energy transfer
(FRET) and other fluorescence spectroscopy techniques, for which most of the currently
available SD-SIC methods have been developed and tailored (Förster et al., 2008; Lin et al.,
2008; Nilges et al., 1988a; Aszodi et al., 1995).
In this view, this study will illustrate possible approaches to model building in MS3D,
underlining the main issues related to this specific field and outlining some of the possible
solutions to these problems. Whenever possible, alternative methods employing either
different programs selected among most popular applications in homology modelling,
threading, docking and molecular dynamics (MD), or different strategies to exploit the
information contained in MS data will be described. Discussion will be limited to packages
either freely available, or costing less than 1,000 US$ for academic users. For programs, the
home web address has been reported, rather than references that are very often partial
and/or outdated. Some examples, derived from the literature available in this field, or
developed ad hoc to illustrate some critical features of the computational methods in MS3D,
should clarify potentiality and current limitations of this approach.

2. General MS3D modelling procedures

2.1 Possible computational protocols for MS3D approaches
MS3D can be fruitfully applied to many structure-related problems; thus, it requires the
(possibly combined) use of different modelling procedures. However, a very general scheme
for a MS3D approach can still be sketched (Fig. 1). It includes:
• an initial generation of possible structures for the investigated system by some
sampling algorithms (S1 or S2 stages);
• followed by classification, clustering and selection steps of the best sampled structures
based on one or more criteria (F1 or F2a-F2b-F2c);
• an optional narrowing of the ensemble by a refinement of the selected models (R);
• followed by new classification, clustering and selection stages for the identification of
the most representative models (FF).
Selection criteria are very often represented by more or less sophisticated combinations of
different scoring (i.e. the higher, the better), penalty (i.e. the lower, the better) or target (i.e.
the closer to its reference value, the better) functions. For the sake of brevity, from here
onwards the term “scoring” will be indiscriminately used for either true scoring, or penalty,
or target function, when their discrimination is not necessary.
The features characterizing a specific approach are: a) combination of sampling (and
optimization) algorithms, b) scoring functions in sampling/optimization and classification/
clustering/selection stages, c) strategies to introduce MS-based experimental information.
A first major branching in this scheme already occurs in the earliest modelling stages (box
A), depending if MS-based information is, at least in part, integrated in the structure
generation stage (path S1-F1), or rather deferred to a subsequent model classification/
selection step (path S2-F2a-F2b-F2c).
Depending on information types, programs and strategies used in modelling (see next
sections for theory and examples), MS-based data can be either all introduced during
sampling (S1), or all used in the filtering stage (F2a), or subdivided between the two steps
(S1+F1). The main advantage of the inclusion of MS-based information into sampling (path
Computational Methods in Mass Spectrometry-Based Protein 3D Studies 135

S1-F1) is an increase in model generation efficiency by limitation of the conformational or

configurational subspace to be explored. In several potentially problematic cases, i.e. large
molecules with very limited additional information available, this reduction can transform a
potentially insoluble problem into a reliable model generation, capable of correlating
structural and functional features of the investigated system. However, for the very same
reason, if information is introduced too abruptly or tightly during structural sampling, it can
artificially freeze the models into a wrong, or at least incomplete, set of solutions (Latek et
al., 2007; Bowers et al., 2000). Also the weight of erroneous restraints will be considerably
amplified by the impossibility of a comparison with solutions characterized by some
restraint violations, but considerably more favourable scoring function values, which are
often diagnostic of inadequate sampling and/or errors in the experimental restraint set.

Fig. 1. Flowchart of a generic MS3D modelling approach. Magenta, violet and pink represent
steps in which MS-based information is applied. Triangular arrows indicate use of MS-based
data. Dotted lines and borders are used for optional refinement stages. Blue codes in white
circles/ellipses label the corresponding stages within the text.
136 Computational Biology and Applied Bioinformatics

Accordingly, both the protocol used to implement MS-based information into modelling
procedures and the MS-based data themselves generally represent very critical features,
which require the maximum attention during computational setup and final analyses. In
addition, implementation of restraints in the sampling procedure either requires some
purposely programming activity, or severely limits the choice of modelling tools to
programs already including suitable user-defined restraints.
Use of MS-based information in post-sampling analyses (path S2-F2a-F2b-F2c) to help
classifying and selecting the final models exhibits a mostly complementary profile of
advantages-disadvantages. In fact, it decreases the sampling efficiency of the modelling
methods (S2), by leading to a potentially very large number of models to be subsequently
discarded on the mere basis of their violations of MS-derived restraints (F2a), and by
providing no ab initio limitations to the available conformational/configurational space of
the system. Furthermore, it may still require programming activity if available restraint
analysis tools (F2a) are lacking or inefficient in the case of the implemented information.
However, this approach warrants the maximum freedom to the user in the choice of the
sampling program; this may result very useful in those cases where the peculiar features of
a specific program are strongly required to model the investigated system. In addition, a
compared analysis of both structural features and scoring function values between models
accepted and rejected on the basis of MS-based data may allow the identification of potential
issues in the selected models and the corresponding data sets (steps F2c-X).

2.2 Integration of MS-based data into modelling procedures

Although an ever-increasing number of MS-based strategies has been developed, they
provide essentially two information classes for model building: i) surface accessible
residues, from chemical/isotopic labelling or limited proteolysis experiments (Renzone et
al., 2007a); ii) couples of residues whose relative distances span a prefixed range, from
crosslinking experiments (Sinz, 2006; Renzone et al., 2007a). Details on the nature of the
combined biochemical and MS approaches used to generate these data and the experimental
procedures adopted in these cases is provided in the exhaustive reviews reported above.

2.2.1 Surface-related information (selective proteolysis and chemical labelling)

Although many structural generation approaches include surface-dependant terms, usually
they are not exposed to the user; thus, direct implementation of accessibility information is
always indirect and ranges from very difficult to impossible. In some docking programs,
surface residue patches can be excluded from the exploration, thus restricting the region of
space to be sampled (Section 3.2). This information is generally exploited through programs
that build and evaluate different kinds of molecular surfaces, applied during the model
validation stages. In this view, the main available programs and their usage will be
described in the section dedicated to model validation (Section 3.3.2).
In the case of modelling procedures based on sequence alignment with templates of known
3D structure, surface-dependent data can be employed both to validate alignments before
modelling (early steps in S1 stage), and to filter the structures resulting from the different
steps of a traditional model building procedure (stages F1 or F2a, and FF).

2.2.2 Crosslinks
Cross-linking information often directly contribute to the model building procedure (under
the form of distance restraints or direct linker addition to the simulated systems) (stage S1 in
Fig.1), in addition to their model validation/interpretation role (stages F1, F2a, FF).
Computational Methods in Mass Spectrometry-Based Protein 3D Studies 137

Whenever information from crosslinking experiments is integrated within the modelling

procedure, the most common approach recurring in literature is its translation into distance
constraints (i.e. “hard”, fixed distances) or restraints (i.e. variable within an interval and/or
around a fixed distance with a given tolerance) involving atoms, in a full-atomistic
representation, or higher-order units, such as residues, secondary structure (SS) elements, or
domains, in coarse-grained models. A less common approach consists in the explicit
inclusion of the crosslinker atoms in the simulation.
2.2.2.1 Distance restraints
Distance restraints (DRs) are usually implemented by adding a penalty term to the scoring
function used to generate, classify or select the models, whenever the distance between
specified atom pairs exceeds a threshold value. In this way, associated experimental
information can be introduced rather easily and with moderate computational overheads in
all the molecular modelling and simulation approaches based on scoring functions.
However, since crosslinking agents are molecules endowed with well-defined and specific
conformational and interaction properties, both internal and with crosslinked molecules,
accurate theoretical and experimental estimates of distance ranges associated with the
corresponding cross-link agents only qualitatively correspond to experimentally-detected
distances between pairs of cross-linked residues (Green et al., 2001; Leitner et al., 2010).
Steric bumps, specific favourable or unfavourable electrostatic interactions, presence of
functional groups capable of promoting/hampering the crosslinking reaction and changes
in crosslinker conformational population under the effects of macromolecule are all possible
causes for observed discrepancies.
2.2.2.2 Explicit linkers
Explicit inclusion of crosslinkers in the systems, although potentially allowing to overcome
the limits of DRs, presently suffers from several drawbacks that limit its usage to either final
selection/validation stages, or to cases where a limited number of totally independent and
simultaneously holding crosslinks are observed. In fact, when many crosslinks are detected
in a system by MS analysis, they very often correspond to mixtures of different patterns,
because crosslinks can interfere each other either by direct steric hindrance, or by
competition for one of the macromolecule reacting groups, or by inducing deformation in
the linked system, thus preventing further reactions. However, the added information from
explicit crosslinkers may: i) allow disambiguation between alternative predicted binding
modes, ii) provide more realistic and strict estimates of the linker length to be used in
further stages of DR-based calculations, iii) help modelling convergence, iv) substantially
contribute to model validation.
An attempt to reproduce by an implicit approach at least the geometrical constraints
associated with a physical linker has been performed by developing algorithms to identify
minimum-length paths on protein surfaces (Potluri et al., 2004). This approach provides
upper/lower bounds to possible crosslinking distances on static structures but it only
worked on static structures as a post-modelling validation tool, and no further applications
have been reported so far.

3. Available computational approaches in MS3D

MS-based data can be used to obtain structural information on different classes of problems:
a. single conformational states (e.g. the overall fold);
138 Computational Biology and Applied Bioinformatics

b. conformational changes upon mutations/environmental modifications;

c. macromolecular aggregation (multimerization);
d. binding of small ligands to macromolecules.
Sampling efficiency and physical soundness of the scoring functions used during sampling
(stages S1/S2 of Fig. 1) and to select computed structures (stages F1/F2b and FF) generally
represent the main current limitations of 3D structure prediction and simulation methods. In
this view, introduction of experimental data represents a powerful approach to reduce the
geometrical space to be explored during sampling, and also an independent criterion to
evaluate the quality of selected models.
From a computational point of view, structural problems a)-d) translate into system-
dependent proper combinations of:
A. fold identification and characterization;
B. docking;
C. structural refinement and characterization of dynamic properties and of changes under
the effects of local or environmental perturbations.
Since the optimal combination of methods for a given problem depends upon a large
number of system- and data-dependent parameters, and the number of programs developed
for biomolecular simulations is huge, an exhaustive description and compared analysis of
methods for biomolecular structure generation/refinement is practically impossible.
However, we will try to offer a general overview of the main approaches to generate, refine
and select 3D structures in MS3D applications, with a special attention to possible ways of
introducing MS-based data and exploiting their full information content.

3.1 Fold identification and characterization

The last CASP (Critical Assessment of techniques for protein Structure Prediction)
experiment call (CASP9, 2010) classified modelling methods in two main categories:
“Template Based Modelling” (TBM) and “Template Free Modelling” (TFM), depending if
meaningful homology can be identified or not before modelling between the target sequence
and those of proteins/domains whose 3D structures are known (templates).
TFM represents the most challenging task because it requires the exploration of the widest
conformational space and heavily relies on scoring methods inspired by those principles of
physics governing protein folding (de novo or ab initio methods), eventually integrated by
statistical predictions, such as probabilities of interresidue contacts, surface accessibility of
single residues or local patches and SS occurrence. When number and quality of these
information increase, together with the extent of target sequence for which they are
available, “folding recognition” and “threading” techniques can be used, including a broad
range of methods at the interface between TFM and TBM. In these approaches, several
partial 3D structure “seeds” are generated by statistical prediction or distant homology
relationships, and their relative arrangements are subsequently optimized by strategies
deriving from de novo methods.
The most typical TBM approach, “comparative” or “homology” modelling (HM), uses
experimentally elucidated structures of related protein family members as “templates” to
model the structure of the protein under investigation (the “target”). Target sequence can
either be fully covered by one or more templates, exhibiting good homology over most of
the target sequence, or can require a “patchwork” of different templates, each best covering
a different region of the target.
Computational Methods in Mass Spectrometry-Based Protein 3D Studies 139

A further group of approaches, presently under active development and already exhibiting
good performances in CASP and other benchmark and testing experiments, is formed by the
“integrative” or “hybrid” methods. They combine information from a varied set of
computational and experimental sources, often acting as/based on “metaservers”, i.e.
servers that submit a prediction request to several other servers, then averaging their results
to provide a consensus that in many cases is more reliable than the single predictions from
which it originated. Some metaservers use the consensus as input to their own prediction
algorithms to further elaborate the models.
In order to provide some guidelines for structural prediction/refinement tasks in the
presence of MS-based data, a general procedure will be outlined for protein fold/structure
modelling. The starting step in protein modelling is usually represented by a search for
already structurally-characterized similar sequences. Sensitive methods for sequence
homology detection and alignment have been developed, based on iterative profile searches,
e.g. PSI-Blast (Altschul et al., 1997), Hidden Markov Models, e.g. SAM (K. Karplus et al.
1998), HMMER (Eddy, 1998), or profile-profile alignment such as FFAS03 (Jaroszewski et al.,
2005), profile.scan (Marti-Renom et al., 2004), and HHsearch (Soding, 2005).
When homology with known templates is over 40%, HM programs can be used rather
confidently. In this case, especially when alignments to be used in modelling have already
been obtained, local programs represent a more viable alternative to web-based methods
than in TFM processes. If analysis is limited to most popular programs and web services
capable of implementing user MS-based restraints (strategy S1 in Fig. 1), the number of
possible candidates considerably decreases. Among web servers, on the basis of identified
homologies with templates, Robetta is automatically capable of switching from ab initio to
comparative modelling, while I-TASSER requires user-provided alignment or templates to
activate comparative modelling mode. A very powerful, versatile and popular HM
program, available both as a standalone application, and as a web service, and embedded in
many modelling servers, is MODELLER (http://www.salilab.org/modeller/). It include
routines for template search, sequence and structural alignments, determination of
homology-derived restraints, model building, loop modelling, model refinement and
validation. MS-based distance restraints can be added to those produced from target-
template alignments, as well as to other restraints enforcing secondary structures, symmetry
or part of the structure that must not be allowed to change upon modelling. However, some
scripting ability is required to fully exploit MODELLER versatility.
The overall accuracy of HM models calculated from alignments with sequence identities of
40% or higher is almost always good (typical root mean square deviations (RMSDs) from
corresponding experimental structures less than 2Å). The frequency of models deviating by
more than 2Å RMSD from experimental structures rapidly increases when target–template
sequence identity falls significantly below 30–40%, the so-called “twilight zone” of HM (Blake
& Cohen, 2001; Melo & Sali, 2007). In such cases, the quality of resulting modelled structures
significantly increases by combining additional information, both of statistical origin, such as
SS prediction profiles, and from sparse experimental data (low resolution NMR or chemical
crosslinking, limited proteolysis, chemical/isotopical labelling coupled with MS).
If the search does not produce templates with sufficient homology and/or covering of the
target sequence, TFM or mixed TFM/TBM methods must be used. Many programs based on
ab initio, fold recognition and threading methods are presently offered as web services; this
is because very often they use a metaserver approach for some steps, need extensive
140 Computational Biology and Applied Bioinformatics

searches in large databases, require huge computational resources, or to better protect

underlying programs and algorithms, currently under very active development. Although
this may offer some advantages, especially to users less-experienced in biocomputing or
endowed with limited computing facilities, it may also imply strong limitations in the full
exploitation of the features implemented in the different methods, with particularly serious
implications in MS3D. Only few servers either include a NMR structure determination
module (not always suitable for MS-based data), or explicitly allow the optional usage of
user-provided distance restraints in the main input form. Fortunately, two of the most used
and versatile servers, Robetta (http://robetta.bakerlab.org/) and I-TASSER
(http://zhanglab.ccmb.med.umich.edu/I-TASSER/), good performers at the last CASP
rounds (http://predictioncenter.org/), allow input of distance restraints in the modelling
procedure, via a NMR-dedicated service for Robetta (Rosetta-NMR, suitable for working
with sparse restraint) (Bowers et al., 2000), or directly in the main prediction submission
page (I-TASSER). Other servers can still allow the implementation of MS-based information
in the model generation step if they can save intermediate results, such as sequence
alignments, SS or fold predictions. These latter, after addition of MS-based restraints, can be
then included into suitable modelling programs, to be run either locally or on web servers.
A successful examples of modelling with MS-based information in a low-homology case is
Gadd45β. A model was built, despite the low sequence identity (<20%) with template
identified by fold recognition programs, through the introduction of additional SS restraints,
which were based on SS profiles and experimental data from limited proteolysis and
alkylation reactions combined with MS analysis (Papa et al., 2007). Model robustness was
confirmed by comparison with the homolog Gadd45γ structure solved later (Schrag JD et al.,
2008), where the only divergence in SS profiles was the occurrence of two short 310 helices
(three residues each long) and an additional two-residues β-strand in predicted loop regions
(Fig. 2). Furthermore, this latter β-strand is so distorted that only a few SS assignment
programs could identify it, and the corresponding sequence in Gadd45β, predicted
unstructured and outside the template alignment, was not modelled at all.

Fig. 2. Comparison between the MS3D model of Gadd45β (light green) and the
crystallographic structure of its homolog Gadd45γ (light blue). Sequences with different SS
profiles are painted green in Gadd45β and magenta in Gadd45γ.
Computational Methods in Mass Spectrometry-Based Protein 3D Studies 141

3.2 Docking
Usually, methods for protein docking involve a six-dimensional search of the rotational and
translational space of one protein with respect to the other where the molecules are treated
as rigid or semirigid-bodies. However, during protein-protein association, the interface
residues of both molecules may undergo conformational changes that sometimes involve
not only side-chains, but also large backbone rearrangements. To manage at least in part
these conformational changes, protein docking protocols have introduced some degree of
protein flexibility by either use of "soft" scoring functions allowing some steric clash, or
explicit inclusion of domain movement/side chain flexibility. Biological information from
experimental data on regions or residues involved in complexation can guide the search of
complex configurations or filter out wrong solutions. Among the programs most frequently
used for protein-protein docking, recently reviewed by Moreira and colleagues (Moreira et
al., 2010), some of them can manage biological information and will be discussed in this
context.
In the Attract program (http://www.t38.physik.tu-muenchen.de/08475.htm ), proteins are
represented with a reduced model (up to 3 pseudoatoms per amino acid) to allow the
systematic docking minimization of many thousand starting structures. During the docking,
both partner proteins are treated as rigid-body and the protocol is based on energy
minimization in translational and rotational degrees of freedom of one protein with respect
to the other. Flexibility of critical surface side-chains as well as large loop movements are
introduced in the calculation by using a multiple conformational copy approach (Bastard et
al., 2006). Experimental data can be taken into account at various stages of the docking
procedure.
The 3D-Dock algorithm (http://www.sbg.bio.ic.ac.uk/docking/) performs a global scan of
translational and rotational space of the two interacting proteins, with a scoring function
based on shape complementarity and electrostatic interaction. The protein is described at
atomic level, while the side-chain conformations are modelled by multiple copy
representation using a rotamer library. Biological information can be used as distance
restraints to filter final complexes.
HADDOCK (http://www.nmr.chem.uu.nl/haddock/) makes use of biochemical or
biophysical interaction data, introduced as ambiguous intermolecular distance restraints
between all residues potentially involved in the interaction. Docking protocol consists of
four steps: 1) topology and structure generation; 2) randomization of orientations and rigid
body energy minimization; 3) semi-flexible simulated annealing (SA) in torsion angle space;
4) flexible refinement in Cartesian space with explicit solvent (water or DMSO). The final
structures are clustered using interface backbone RMSD and scored by their average
interaction energy and buried interface area. Recently, also explicit inclusion of water
molecules at the interface was incorporated in the protocol.
Molfit (http://www.weizmann.ac.il/Chemical_Services/molfit/) represents each molecule
involved in docking process by a 3-dimensional grid of complex numbers and estimates the
extent of geometric and chemical surface complementarity by correlating the grids using
Fast Fourier Transforms (FFT). During the search, contacts involving specified surface
regions of either one or both molecules are up- or down-weighted, depending on available
structural and biochemical data or sequence analysis (Ben-Zeev et al., 2003). The solutions
are sorted by their complementarity scores and the top ranking solutions are further refined
by small rigid body rotations around the starting position.
142 Computational Biology and Applied Bioinformatics

PatchDock (http://bioinfo3d.cs.tau.ac.il/PatchDock/) is based on shape complementarity.

First, the surfaces of interacting molecules are divided according to the shape in concave,
convex and flat patches; then, complementarity among patches are identified by shape-
matching techniques. The algorithm is a rigid body docking, but some flexibility is
indirectly considered by allowing some steric clashes. The resulting complexes are ranked
on the basis of the shape complementarity score. PatchDock allows integration of external
information by a list of binding site residues, thus restricting the matching stage to their
corresponding patches.
RosettaDock (http://rosettadock.graylab.jhu.edu/) try to mimics the two stages of a
docking process, recognition and binding, as hypothesized in Camacho & Vajda, 2001.
Recognition is simulated by a low resolution phase in which a coarse-grained representation
of proteins, with side chains replaced by single pseudoatoms, undergoes a rigid body Monte
Carlo (MC) search on translations and rotations. Binding is emulated by a high-resolution
refinement phase where explicit sidechains are added by using a backbone-dependent
rotamer packing algorithm. The sampling problem is handled by supercomputing clusters
to ensure a very large number of decoys that are discriminated by scoring functions at the
end of both stages of docking. The docking search problem can be simplified when
biological information is available on the binding region of one or both interacting proteins.
The reduction of conformational space to be sampled could be pursued by: i) opportunely
pre-orienting the partner proteins, or ii) reducing docking sampling to the high-affinity
domain, in the case of multidomain proteins, or iii) using loose distance constraints.
ZDOCK (http://zdock.bu.edu/) is a rigid body docking program based on FFT algorithm
and an energy function that combines shape complementarity, electrostatics and desolvation
terms. RDOCK (http://zdock.bu.edu/) is a refinement program to minimize and rerank the
solutions found by ZDOCK. The complexes are minimized by CHARMm (Brooks et al.,
1983) to remove clashes and improve energies, then electrostatic and desolvation terms are
recalculated in a more accurate fashion with respect to ZDOCK. Biological information can
be used either to avoid undesirable contacts between certain residues during ZDOCK
calculations or to filter solutions after RDOCK.
As in protein folding, also for docking the use of MS-based information allowed the
modelling of several complexes even in the lack of suitable templates with high homology.
The fold of prohibitin proteins PHB1 and PHB2 was predicted (Back et al., 2002) by SS and
fold recognition algorithms, while crosslinking allowed to model the relative spatial
arrangement of the two proteins in their 1:1 complex. Another example of combined use of
SS information, chemical crosslinking, limited proteolysis and MS analysis results with a
low sequence identity (~ 20%) template is the modelling of porcine aminoacylase 1 dimer; in
this case, standard modelling procedures based on automatic alignment had failed to
produce a dimeric model consistent with experimental data (D'Ambrosio et al., 2003).
In the case of protein-small ligand docking, the conformational space to be explored is
reduced by the small size of the ligand, whose full flexibility can usually be allowed, and by
the limited fraction of protein surface to be sampled, corresponding to the binding site, often
already known. Among the programs for ligand-flexible docking that allow protein side-
chains flexibility, Autodock is one of most popular (http://autodock.scripps.edu/).
AutoDock combines a grid-based method with a Lamarckian Genetic Algorithm to allow a
rapid evaluation of the binding energy. A simulated annealing method and a traditional
genetic algorithm are also available in Autodock4.
Computational Methods in Mass Spectrometry-Based Protein 3D Studies 143

In general, MS-based data can be used to limit the protein region to be sampled (Kessl et al.,
2009) or can be explicitly considered in the docking procedure, as in the case of the mapping
of Sso7d ATPase site (Renzone et al., 2007b). In this case, three independent approaches for
molecular docking/MD studies were followed, considering both FSBA-derivatives and the
ATP-Sso7d non-covalent complex: i) unrestrained MD, starting from a full-extended,
external conformation for Y7-FSBA and K39-FSBA residue sidechains, and from several
random orientations for ATP, with an initial distance of 20 Å from Sso7d surface, in regions
not involved in protein binding; ii) restrained MD, by gradually imposing distance restraints
corresponding to a H-bond between adenine NH2 group and each accessible (i.e., within a
distance lower or equal to the maximum length of the corresponding FSBA-derivative)
donor sidechain; iii) rigid ligand docking, by calculating 2000 ZDOCK models of the non-
covalent complex of Sso7d with an adenosine molecule. The rigid ligand docking
reproduced only in part features from other approaches, as rigid docking correctly predicted
the anchoring point for adenosine ring, but failed to achieve a correct position for the ribose
moiety, due to the required concerted rearrangement of two Sso7d loops involved in the
binding. This latter feature represents one of the main advantages of modelling strategies
involving MD (in particular, in cartesian coordinates) because MD-based simulation
techniques are the best or the only approaches that reproduce medium-to-large scale
concerted rearrangements of non-contiguous regions.

3.3 Model simulation, refinement and validation

Refinement (R stage in Fig.1) and validation of final models (FF stage) represent very
important steps, especially in cases of low homologies with known templates and when fine
details of the models are used to predict or explain functional properties of the investigated
system. In addition, very often the modelled structures are aimed at understanding the
structural effects of point mutations or other local sequence alterations (sequence
deletions/insertions, addition or deletion of disulphide bridges, formation of covalent
constructs between two molecules and post-translational modifications), or of changes in
environmental parameters (temperature, pressure, salt concentration and pH). In these
cases, techniques are required to simulate the static or dynamic behaviour of the
investigated system in its perturbed and unperturbed states.

3.3.1 Computational techniques and programs for model simulation and refinement
Model refinement, when not implemented in the modelling procedure, can be performed by
energy minimization (EM) or, better, by different molecular simulation methods, mostly
based on variants of molecular dynamics (MD) or Monte Carlo (MC) techniques. They are
also commonly used to characterize dynamic properties and structural changes upon local
or environmental perturbations.
Structures deriving from folding or docking procedures need, in general, at least a structural
regularization by EM before final validation steps, to avoid meaningless results from many
methods. Scoring functions of the latter evaluate the probity of parameters, such as dihedral
angle distributions, presence and distribution of steric bumps, voids in the molecular core,
specific nonbonded interactions (H-bonds, hydrophobic clusters). Representing a
mandatory step in most MC/MD protocols, EM programs are included in all the molecular
simulation packages, and they share with MC/MD most input files and part of the setup
parameters. Thus, unless they are be explicitly discussed, all system- and restraint-related
features or issues illustrated for simulation methods also implicitly held for EM.
144 Computational Biology and Applied Bioinformatics

As we are mostly interested in techniques implementing experimentally-derived constraints

or restraints, some of the most popular methods for constraints-based modelling will be
briefly described. These methods have been developed and optimized mainly to identify
and refine 3D structures consistent with spatial constraints from diffraction and resonance
experiments (de Bakker et al., 2006). They have also been extensively applied to both TBM
(Fiser & Sali, 2003) and free modelling prediction and simulation (Bradley et al., 2005;
Schueler-Furman et al., 2005), and are often used to refine/validate models produced in
TFM and TBM approaches described in sections 3.1 and 3.2. There are two main categories
of constraint-based modelling algorithms: i) distance geometry embedding, which uses a
metric matrix of distances from atomic coordinates to their collective centroid, to project
distance space to 3D space (Havel et al. 1983; Aszodi et al. 1995, 1997); ii) minimization,
which incorporates distance constraints in variable energy optimization procedures, such as
molecular dynamics (MD) and Monte Carlo (MC). For both MD and MC, it is possible to
work both in full cartesian coordinates, or in the restricted torsion angle (TA) space, with
covalent structure parameters kept fixed at their reference values, thus originating the
Torsional Angle MD (TAMD) and Torsional Angle MC (TAMC) approaches. They are
currently implemented in several modelling and refinement packages, developed for
structural refinement of X-ray or NMR structures (Rice & Brünger, 1994; Stein et al. 1997;
Güntert et al., 1997), folding prediction (Gray et al., 2003), or more general packages
(Mathiowetz et al., 1994; Vaidehi et al., 1997). Standard MC/MD methods are only useful for
structural refinement, local exploration and to characterize limited global rearrangements.
However, they are also widely used as sampling techniques in folding/docking approaches,
although in those cases enhanced sampling extensions of both methods are employed.
Simulated annealing (SA) (Kirkpatrick et al., 1983) and replica exchange (RE) approaches
(Nymeyer et al., 2004) are the most common examples of these MC/MD enhancements, both
potentially overcoming the large energy barriers required for sampling the wide
conformational and configurational spaces to be explored in folding and docking
applications, respectively.
A non-exhaustive list of the most diffused simulation packages including a more-than-basic
treatment of distance-related restraints and also exhibiting good versatility (i.e.
implementation of different algorithms, approaches, force fields and solvent
representations), may include at least: AMBER (http://ambermd.org/), CHARMM
(http://www.charmm.org/), DESMOND (http://deshawresearch.com/resources.html),
GROMACS (http://www.gromacs.org/) and TINKER (http://dasher.wustl.edu/tinker).
CYANA (http:www.cyana.org) and XPLOR/CNS (http://cns-online.org/v1.3/), although
originally more specialized for structural determination and refinement from NMR and
NMR/X-ray data, respectively, have been recently included in several TFM and TBM
protocols, thanks to their efficient implementations of TAMD and distance or torsional angle
restraints. The choice of a simulation program should ideally keep into account several
criteria, ranging from computational efficiency, to support of sampling or refinement
algorithms, to integration with other tools for TFM or TBM applications.
The main problems associated with simulation methods having relevant potential
implications on MS3D are: i) insufficient sampling; ii) inaccuracy in the potential energy
functionals driving the simulations; iii) influence of the approach used to implement
experimentally-derived information on final structure sets.
Sampling problem can be approached both by increasing the sampling efficiency with
MC/MD variations like SA and RE, and by decreasing the size of the space to be explored.
Computational Methods in Mass Spectrometry-Based Protein 3D Studies 145

This latter result can be reached by reducing the overall number of degrees of freedom to be
explicitly sampled and/or by reducing the number of possible values per variable to a
small, finite number (discretization, like in grid-based methods), and/or by restraining
acceptable variable ranges. Reduction of the total number of degrees of freedom can be
accomplished by switching to coarse-grained representations of the system, where a number
of explicit atoms, ranging from connected triples, to amino acid sidechains, to whole
residues, up to full protein subdomains, are replaced by a single particle. This method is
frequently used in initial stages of ab initio folding modelling, or in the simulation of very
large systems, such as giant structural proteins of huge protein aggregates.
Another possible way to reduce the number of degrees of freedom is the aforementioned TA
approach, requiring for a N atom system only N/3 torsional angles compared with 3N
coordinates in atomic cartesian space (Schwieters & Clore, 2001). Moreover, as the high
frequency motions of bending and stretching are removed, TAMD can use longer time steps
in the numerical integration of equations of motion than that required for a classical
molecular dynamics in cartesian space. Its main limitation may derive from neglecting
covalent geometry variations (in particular, bending centred on protein Cα atoms) that are
known to be associated with conformational variations (Berkholz et al., 2009), for instance
from α-helix to β-strand, and that can be important in concerted transitions or in large
structures with extensive and oriented SS regions. Discretization is mostly employed in the
initial screening of computationally intensive problems, such as ab initio modelling.
Restraining variable value ranges in MS3D is usually associated with either predictive
methods (SS, H-bond pattern, residue exposure), or to homology analysis, or to
experimentally-derived information. Origin, nature and form of these restraints have
already been discussed in previous sections, while some more detail on the implementation
of distance-related information into simulation programs will be given at the end of this
section.
While the implementation of restraints can be very variable in methods where the scoring
function does not intend to mimic or replicate a physical interaction between involved
entities, in methods based on physically-sounding molecular potential functions
(forcefields) have DRs implemented by a more limited number of approaches. At its
simplest, a DR will be represented as a harmonic restraint, for which only the target distance
and the force constant need to be specified in input. This functional form is present in
practically all most common programs, but either requires a precise knowledge of the target
distance, or it will result in a very loose restraint if the force constant is lowered too much to
account for low-precision target values, the usual case in MS-based data. In a more complex
and useful form, implemented with slight variations in several programs (AMBER,
CHARMM, GROMACS, XPLOR/CNS, DESMOND, TINKER), the restraint is a well with a
square bottom with parabolic sides out to a defined distance, and then linear beyond that on
both (AMBER) or just the upper limit side (CHARMM, GROMACS, XPLOR/CNS,
DESMOND). In some programs (CHARMM, AMBER, XPLOR/CNS), it is possible to select
an alternative behaviour when a distance restraint gets very large (Nilges et al,1988b) by
“flattening out” the potential, thus leading to no force for large violations; this allows for
errors in constraint lists, but might tend to ignore constraints that should be included to pull
a bad initial structure towards a more correct one.
Other forms for less-common applications can also be available in the programs or be
implemented by an user. However, the most interesting additional features of versatile DR
146 Computational Biology and Applied Bioinformatics

implementations are the different averages that can be used to describe DRs: i) complex
restraints can involve atom groups rather than single atoms at either or both restraint sides;
ii) time-averaged DRs, where target values are satisfied on average within a given time lapse
rather than instantaneously; iii) ambiguous DRs, averaged on different distance pairs. The
latter two cases are very useful when the overall DRs are not fully consistent each other,
because they are observed in the presence of conformational equilibria and, as such, they are
associated with different microstates of the system. In addition, complex and versatile
protocols can be simply developed in those programs where different parameters can be
smoothly varied during the simulation (AMBER).

3.3.2 Programs for model validation

A validation of the final models, very often included in part in the available automated
modelling protocols, represents a mandatory step, especially for more complex (low-
homology, few experimental data) modelling tasks. A huge number of protein and nucleic
acid structural analysis and validation tools exists, based on many different criteria, and
subjected to continuous development and testing; thus, even a CASP section is dedicated to
structural assessment tools (http://www.predictioncenter.org/), and the “Bioinformatics
Links Directory” site alone currently reports 76 results matching “3-D structural analysis”
(Brazas et al., 2010). Being outside the scope of the present report, information on 3D
structural validation tools can be searched on specialized sites such as
http://bioinformatics.ca/links_directory/ . However, similarly to what stated on prediction
metaservers, a general principle for validation is to possibly use several tools, based on
different criteria, looking for emergent properties and consensus among the results.
Specific parameters associated with MS-based data can be usually analysed with available
tools. Distance restraints and their violations can be analysed both on single structures and
on ensembles (sets of possible solutions of prediction methods, frames from molecular
dynamics trajectories) with several graphic or textual programs, the most specialized
obviously being those tools developed for the analysis of NMR-derived structures.
Surface information can be analysed by programs like:
DSSP (http://swift.cmbi.ru.nl/gv/dssp/),
NACCESS (http://www.bioinf.manchester.ac.uk/naccess/),
GETAREA (http://curie.utmb.edu/getarea.html/),
ASA-VIEW (http://gibk26.bse.kyutech.ac.jp/jouhou/shandar/netasa/asaview/)
that calculate different kinds of molecular surfaces, such as van der Waals, accessible, or
solvent excluded surfaces for overall systems and contact surfaces for complexes are used.
However, differently from distance restraints, available programs usually work on a single
input structure at a time, thus making structure filtering and analysis on the large ensembles
of models potentially produced by conformational prediction, molecular simulation or
docking calculations, a painful or impossible task. In these cases, scripts or programs to
automate the surface calculations and to average or filter the results must be developed.

4. Modelling with sparse experimental restraints

In the previous section many of the computational methods that can concur to produce
structural models in MS3D applications have been outlined, together with different ways to
integrate MS-based experimental information into them. Here we will refocus on the overall
Computational Methods in Mass Spectrometry-Based Protein 3D Studies 147

computational approach in MS3D, to illustrate some of its peculiar features and issues, its
present potentialities and the variety of possible combinations of data and protocols that can
be devised to optimally handle different types of structural problems. Depending on nature
and quantity of available experimental information and on previous knowledge of the
investigated system, different combinations of the methods mentioned in previous sections
can be optimally employed. We will start illustrating examples of methods for de novo
protein folding, a frontier application of modelling with sparse restraints, because it is based
on minimal additional information on the system under investigation.
The MONSSTER program (Skolnick et al. 1997) only makes use of SS profiles and a limited
number of long-distance restraints. By employing system discretization and coarse-graining
to reduce required sampling, a protein is represented by a lattice-based Cα-backbone trace
with single interaction center rotamers for the side-chains. By using N/7 (N is the protein
length) long-range restraints, this method is able to produce folds of moderate resolution,
falling in the range from 4 to 5.5 Å of RMSD for Cα-traces with respect to the native
conformation for all α and α/β proteins, whereas β-proteins require, for the same resolution,
N/4 restraints. A more recent method for de novo protein modelling (Latek et al., 2007)
adopts restrained folding simulations supported by SS predictions, reinforced with sparse
experimental data. Authors focused on NMR chemical-shift-based restraints, but also sparse
restraints from different sources can be employed. A significant improvement of model
quality was already obtained by using a number of DRs equal to N/12.
As already stated by Latek and colleagues, the introduction of DRs in protein folding
protocol represents a critical step that in principle could negatively affect the sampling of
conformational space. In fact, restraint application at too early stages of calculations can trap
the protein into local minima, where restraints are satisfied, but the native conformation is
not reached. In addition to the number, even the specific distribution of long-range
restraints along the sequence can affect the sampling efficiency. To test the influence of data
sets in folding problem, we applied a well-tested protocol of SA, developed for AMBER
program and mainly oriented to NMR structure determination, to the folding simulation of
bovine pancreatic trypsin inhibitor (BPTI), by using different sets of ten long-distance
restraints, randomly selected from available NMR data (Berndt et al., 1992), with optional
inclusion of a SS profile. Fig. 3 shows representative structures for each restraint set.
The four native BPTI disulphide bridges were taken into account by additional distance and
angle restraints. BPTI represents a typical benchmark for this kind of studies, due to its
peculiar topology (an α/β fold with long connection loops, stabilized by disulphide bonds)
still associated with a limited size (58 residues), and to the availability of both X-ray and
NMR accurate structures. SA cycles of 50 structures each were obtained and compared for
four combinations of three sets (S1-3) of ten long distance restraints, totally non-redundant
among different sets and SS profiles: a) S1+SS profile; b) S1 alone; c) S2+SS profile; d) S3+SS
profile. S1 set performed definitely better than the other two, its best model exhibiting a
RMSD value of 2.4 Å on protein backbone of residues 3-53 from the representative NMR
structure.
This set was also able to provide a reasonable low-resolution fold even in the absence of SS
restraints (b). S3 resulted in almost correctly folded models, but with significantly worse
RMSD values than S1 (c). In S3 pseudomirror images (d) of the BPTI fold occurred several
times and only one model out of 50 was correctly folded (not shown).
148 Computational Biology and Applied Bioinformatics

Fig. 3. Ab initio modelling from sparse restraints of BPTI. Representative models from
different restraint sets (S1, S2, S3), with optional SS dihedral restraints are shown in ribbon
representation, coloured in red/yellow (helices), cyan (strands) and grey (loops). Models are
best-fitted on Cα atoms of residues 3-53 to the representative conformation of the NMR
high-resolution structure (PDB code: 1pit) (green), except for the S3+SS set, where
superposition with β-sheet only is shown, to better illustrate pseudo-mirroring effect,
although RMSD values are calculated on the same 3-53 residue range as other models.
Computational Methods in Mass Spectrometry-Based Protein 3D Studies 149

These results suggest a strong dependency of results upon both the exact nature of
experimental data used in structure determination, and the protocol followed for model
building. Thus, the number of restraints estimated in the aforementioned studies as
necessary/sufficient for a reliable structural prediction should be prudently interpreted for
practical purposes. If a proper protocol is adopted, increasing quantity, quality and
distribution homogeneity of data should decrease this dependency, but the problem still
remains severe when using very sparse restraints, such as those associated with many MS3D
applications. A careful validation of the models and, possibly, execution of more modelling
cycles with variations in different protocol parameters, can help to identify and solve these
kinds of problems.
However, in spite of these potential issues, ab initio MS3D can provide a precious insight
into systems that are impossible to study with other structural methods. In addition to
increases in the number of experimental data, also homology-based information and other
statistically-derived constraints can substantially increase the reliability of MS3D
predictions. Thus, suitable combinations of experimental data, predictive methods and
computational approaches have allowed the modelling of many different proteins and
protein complexes spanning a wide range of sizes and complexity. The illustrative examples
shown in Table 1 represents just a sample of systems affordable with current computational
MS3D techniques and a guideline to select possible approaches for different problem
classes. Heterogeneity of reported systems, data and methods, while suggesting the
enormous potentialities of MS3D approaches, practically prevents any really meaningful
critical comparison among methods, whose description in applicative papers is often
incomplete. A standardization of MS3D computational methods is still far from being
achieved, since it requires considerable computational effort to tackle with the considerable
number of strategies and parameters that should be tested in a truly exhaustive analysis.
Furthermore, the extreme sensitivity of modelling with sparse data to constraint
distribution, as seen in the example shown in Fig. 3, either introduces some degree of
arbitrariness in comparative analyses, or make them even more computationally-intensive,
by requiring the use of more subsets for each system setup to be sampled.
Advancements in MS3D experimental approaches continuously change the scenarios for
computational procedures, by substantially increasing the number of data, as well as the
types of crosslinking or labelling agents and proteolytic enzymes. The large number of
crosslinks obtained for apolipoproteins (Silva et al., 2005; Tubb et al., 2008) or CopA copper
ATPase (Lübben et al., 2009) represent good examples of these trends (Table 1).

5. Conclusion
As already stated in the preceding section, the compared analysis of computational
approaches involved in MS3D is still considerably limited, because of the complexity both of
the systems to be investigated, and of the methods themselves, especially when they are
used in combination with restraints as sparse as those usually available in MS3D studies.
The continuous development in all involved experimental and computational techniques
considerably accelerates the obsolescence of the results provided by any accurate
methodological analysis, thus representing a further disincentive to these usually very time
consuming studies. In this view, rather than strict prescriptions, detailed recipes or sharp
critical compared analysis of available approaches, this study was meant to provide an
150 Computational Biology and Applied Bioinformatics

PSF/NA
Gadd45β-MKK7
PSF/MD,EM
Computational Methods in Mass Spectrometry-Based Protein 3D Studies

Table 1. Some examples of MS3D studies from literature. The following abbreviations have been used (in bold, non standard
abbreviations): aXL: crosslinking, CL: chemical labelling, PL: photoaffinity labeling, LP: limited proteolysis; AA: alkylation
analyses, bsee a, PSF: post-sampling filtering with experimental data, IIS: esperimental data integrated in sampling; cXR: x-ray
crystallography; dSSp: secondary structure prediction, esee a,b, HM: Homology modeling, aiM: ab-initio modeling, FP: fold
prediction, MD: molecular dynamics, SA: Simulate Annealing, EM: energy minimization, PPD: protein-protein docking, PLD:
protein-small ligand docking; DR: distance restraints, TM: trans-membrane; NA: not available.
151
152 Computational Biology and Applied Bioinformatics

overall and as wide as possible picture of the state-of-art approaches in MS3D

computational techniques and their potential application fields. However, in spite of these
limitations, some general conclusions can still be drawn.
For predictive methods that stay behind the most ambitious MS3D applications (ab initio
folding, folding prediction, threading), at least when used in the absence of experimental
data, metaservers exhibit on average best performances than the single employed servers, as
also shown by the results of the last CASP rounds on automatic servers
(http://predictioncenter.org/). This suggests two distinct considerations: 1) the accuracy of
sampling and scoring exhibited by each single method, as well the rationale behind them,
are still so limited to prevent reliable predictions on best performing methods in any given
case; 2) nevertheless, most methods tend to locate correct solutions, or, in general, groups of
solutions including the correct one or a close analogue. Therefore, a consensus among the
predictions from different servers generally improves the final solutions, by smoothing
down both extreme results and random fluctuations associated with each single approach.
Well consolidated metaservers, such as Robetta or I-TASSER, can be regarded as reasonable
starting guesses for general folding problems, also considering that they both include
distance-related restraints in their available options. However, special classes of systems
(e.g. transmembrane proteins or several enzyme families) can instead benefit from
employing specifically-devised approaches.
In comparing server-based applications to standalone programs (often available in
alternative for a given approach), potential users should also consider that the former
require less computational skill and resources, but are intrinsically less flexible than the
latter, and that legal and secrecy issues may arise, because several servers consider
submitted prediction requests and the corresponding results as public data, as usually
clearly stated in submission pages. In addition to possible information “leakage” in projects,
the public status of the models would prevent their use in patents.
When considering more specifically MS3D procedures, it has been shown that even a small
number of MS-based restraints can significantly help in restricting the overall space to be
explored and in identifying the correct fold/complexation mode, especially if they are
introduced in early modelling stages of a computational procedure optimized to deal with
both the investigated system and the available data. Thus, experimental restraints can allow
the use of a single model generation procedure, rather than a multiple/metaserver
approach, at least in non-critical cases. In fact, they should filter out all wrong solutions
deriving from the biases of the modelling method, leaving only those close to the “real” one,
if it is included in the sampled set. In particular, since the lowest energy structure should
ideally also be associated with a minimum violation of experimentally-derived restraints,
the coincidence of minimum energy structures with least violated restraints should be
suggestive of correct modelling convergence and evaluation of experimental data. However,
particular care must be adopted not only in the choice of the overall computational
procedure, but especially of the protocol used to introduce experimental information,
because a too abrupt build up of the restraints can easily bring to local minima far from the
correct solution. Comparison of proper scoring functions other than energy between
experimentally-restrained and unrestrained solutions may provide significant help in
identifying potential issues in data or protocols. Estimates of the sensitivity of solutions to
changes in protocols may also enforce the reliability of best converged cases. In particular,
when other restraints are also present, the relative strength and/or introduction order of the
Computational Methods in Mass Spectrometry-Based Protein 3D Studies 153

different sets could play an important role in the final result; thus, their weight should be
carefully evaluated by performing more modelling runs with different setups.
When evaluating the overall modelling procedures, their corresponding caveats and
performance issues, the importance of many details in setup and validation of MS3D
computational procedures fully emerges, thus suggesting that they still requires a
considerable human skill, although many full automated programs and servers allow in
principle the use of MS3D protocols even to inexperienced users. This is also demonstrated
for the pure ab initio modelling stage by the still superior performances obtained by human-
guided predictions in CASP rounds, when compared to fully automated servers.
Future improvements in MS3D are expected as a natural consequence of continuous
development in biochemical/MS techniques for experimental data, and in hardware/
software for molecular simulations and predictive methods. However, some specific, less
expensive and, possibly, quicker evolution in MS3D could be propelled by targeted
development of computational approaches more directly related to the real nature of the
experimental information on which MS3D is based, notably algorithms implementing
surface-dependent contributions and more faithful representations of crosslinkers than
straight distance restraints.

6. References
Aebersold, R. & Mann, M. (2003). Mass spectrometry-based proteomics. Nature, Vol.422, pp.
198–207.
Altschul, S.F.; Madden, T.L.; Schaffer, A.A.; Zhang, J.; Zhang, Z.; Miller, W. & Lipman, D.J.
(1997). Gapped BLAST and PSI-BLAST: A new generation of protein database
search programs. Nucleic Acids Research, Vol.25, pp.3389–3402.
Aszodi, A.; Gradwell, M.J. & Taylor, W.R.(1995). Global fold determination from a small
number of distance restraints. Journal of Molecular Biology Vol.251, pp.308–326.
Aszodi, A.; Munro, R.E. & Taylor, W.R.(1997). Protein modeling by multiple sequence
threading and distance geometry. Proteins, Vol. 29, pp.38–42.
Back, J.W.; de Jong, L.; Muijsers, A.O. & de Koster, C.G. (2003). Chemical crosslinking and
mass spectrometry for protein structural modeling. Journal of Molecular Biology,
Vol.331,pp.303–313.
Back, J.W.; Sanz, M.A.; De Jong, L.; De Koning, L.J.; Nijtmans, L.G.; De Koster, C.G.; Grivell,
L.A.; Van Der Spek, H. & Muijsers, A.O.(2002). A structure for the yeast prohibitin
complex: Structure prediction and evidence from chemical crosslinking and mass
spectrometry. Protein Science, Vol. 11, pp.2471–2478.
Balasu, M.C.; Spiridon, L.N.; Miron, S. ; Craescu, C.T.; Scheidig, A.J., Petrescu, A.J. &
Szedlacsek, S.E. (2009). Interface Analysis of the Complex between ERK2 and PTP-
SL. Plos one, Vol. 4, pp. e5432.
Bastard, K.; Prévost, C. & Zacharias, M. (2006). Accounting for loop flexibility during
protein-protein docking. Proteins, Vol.62, pp. 956-969.
Ben-Zeev, E. & Eisenstein, M. (2003). Weighted geometric docking: incorporating external
information in the rotation-translation scan. Proteins, Vol.52, pp. 24-27.
Berndt, K.D.; Güntert, P.; Orbons, L.P. & Wüthrich, K. (1992). Determination of a high-
quality nuclear magnetic resonance solution structure of the bovine pancreatic
154 Computational Biology and Applied Bioinformatics

trypsin inhibitor and comparison with three crystal structures. Journal of Molecular
Biology, Vol.227, pp.757-775.
Blake, J.D. & Cohen, F.E. (2001). Pairwise sequence alignment below the twilight zone.
Journal of Molecular Biology, Vol. 307, pp. 721-735.
Bowers, P.M.; Strauss, C.E.M. & Baker, D. (2000). De novo protein structure determination
using sparse NMR data. Journal of Biomolecular NMR, Vol.18, pp.311–318.
Brazas, M.D.; Yamada, J.T. & Ouellette, B.F.F. (2010). Providing web servers and training in
Bioinformatics: 2010 update on the Bioinformatics Links Directory . Nucleic Acids
Research, Vol. 38, pp.W3–W6.
Brooks, B.R.; Bruccoleri, R.E.; Olafson, B.D.; States, D.J.; Swaminathan, S. & Karplus
M.(2003). CHARMM: A Program for Macromolecular Energy, Minimization, and
Dynamics Calculations. Journal of Computational Chemistry, Vol.4, pp.187-217.
Camacho, C. J. & Vajda, S. (2001). Protein docking along smooth association pathways.
PNAS USA, Vol.98, pp.10636–10641.
Carlsohn, E.; Ångström, J. ; Emmett, M.R.; Marshall, A.G. & Nilsson, C.L. (2004). Chemical
cross-linking of the urease complex from Helicobacter pylori and analysis by
Fourier transform ion cyclotron resonance mass spectrometry and molecular
modeling International Journal of Mass Spectrometry, Vol.234, pp. 137–144.
Chu, F.; Shan, S.; Moustakas, D.T.; Alber, F.; Egea, P.F.; Stroud, R.M.; Walter, P. &
Burlingame A.L. (2004). Unraveling the interface of signal recognition particle and
its receptor by using chemical cross-linking and tandem mass spectrometry. PNAS,
Vol.101, pp. 16454-16459.
D’Ambrosio, C.; Talamo, F.; Vitale, R.M.; Amodeo, P.; Tell, G.; Ferrara, L. & Scaloni, A.
(2003). Probing the Dimeric Structure of Porcine Aminoacylase 1 by Mass
Spectrometric and Modeling Procedures. Biochemistry, Vol. 42, pp. 4430-4443.
de Bakker, P.I.; Furnham, N.; Blundell, T.L. & DePristo, M.A. (2006). Conformer generation
under restraints. Current Opinion in Structural Biology, Vol. 16, pp.160–165.
Dimova, K; Kalkhof, S.; Pottratz, I.; Ihling, C.; Rodriguez-Castaneda, F.; Liepold, T.;
Griesinger, C.; Brose, N.; Sinz, A. & Jahn, O. (2009). Structural Insights into the
Calmodulin-Munc13 Interaction Obtained by Cross-Linking and Mass
Spectrometry. Biochemistry, Vol.48, pp. 5908-5921.
Eddy, S.R. (1998). Profile hidden Markov models. Bioinformatics, Vol.14, pp.755–763.
Fabris, D. & Yu, E.T. (2010). The collaboratory for MS3D:a new cyberinfrastructure for the
structural elucidation of biological macromolecules and their assemblies using
mass spectrometry-based approaches. Journal Proteome Research, Vol.7, pp. 4848-
4857.
Fiser, A. & Sali, A. (2003). Modeller: generation and refinement of homology base protein
structure models. Methods in Enzymology, Vol. 374, pp.461–491.
Förster, F.; Webb, B.; Krukenberg, K.A.; Tsuruta, H.; Agard, D.A. & Sali A.(2008).
Integration of Small-Angle X-Ray Scattering Data into Structural Modeling of
Proteins and Their Assemblies. Journal of Molecular Biology, Vol.382, pp.1089–
1106.
Friedhoff, P. (2005). Mapping protein–protein interactions by bioinformatics and
crosslinking.. Analitycal & Bioanalitycal Chemistry, Vol.381,pp.78–80.
Computational Methods in Mass Spectrometry-Based Protein 3D Studies 155

Giron-Monzon, L.; Manelyte, L.; Ahrends, R.; Kirsch, D.; Spengler, B. & Friedhoff, P. (2004).
Mapping Protein-Protein Interactions between MutL and MutH by Cross-linking.
The Journal of Biochemical Chemistry, Vol.279, pp. 49338–49345.
Gray, J.J.; Moughon, S.; Wang, C.; Schueler-Furman, O.; Kuhlman, B.; Rohl, C.A. & Baker, D.
(2003). Protein-protein docking with simultaneous optimization of rigid-body
displacement and side-chain conformations. Journal of Molecular Biology, Vol.331,
pp.281-299.
Green, N.S.; Reisler, E. & Houk, K.N. (2001). Quantitative evaluation of the lengths of
homobifunctional protein cross-linking reagents used as molecular rulers. Protein
Science, Vol.10, pp.1293-1304.
Grintsevich, E.E.; Benchaar, S.A.; Warshaviak, D.; Boontheung, P.; Halgand, F.; Whitelegge,
J.P.; Faull, K.F.; Ogorzalek Loo, R.R; Sept, D.; Loo, J.A. & Reisler, E. (2008). Mapping
the Cofilin Binding Site on Yeast G-Actin by Chemical Cross-Linking. Journal of
Molecular Biology, Vol.377, pp. 395-409.
Güntert, P.; Mumenthaler, C. & Wüthrich, K. (1997). Torsion angle dynamics for NMR
structure calculation with the new program Dyana. Journal of Molecular Biology, Vol.
273, pp. 283–298.
Havel, T.F.; Kuntz, I.D. & Crippen, G.M.(1983). The combinatorial distance geometry
method for the calculation of molecular conformation. I. A new approach to an old
problem. Journal of Theoretical Biology, Vol. 310, pp.638–642.
Jaroszewski, L.; Rychlewski, L.; Li, Z.; Li, W. & Godzik, A. (2005). FFAS03: a server for
profile– profile sequence alignments. Nucleic Acids Research, Vol.33, pp.W284–288.
Karplus, K.; Barrett, C. & Hughey R. (1998). Hidden Markov models for detecting remote
protein homologies. Bioinformatics, Vol.14, pp.846–856.
Kessl, J.J.; Eidahl, J.O.; Shkriabai, N.; Zhao, Z.; McKee, C.J.; Hess, S.; Burke, T.R. Jr &
Kvaratskhelia, M. (2009). An allosteric mechanism for inhibiting HIV-1 integrase
with a small molecule.Molecular Pharmacology, Vol. 76, pp.824–832.
Kirkpatrick, S.; Gelatt, C.D. Jr. & Vecchi, M.P. (1983). Optimization by Simulated Annealing.
Science, Vol 220,pp. 671-680.
Latek, D.; Ekonomiuk, D. & Kolinski , A.(2007). Protein structure prediction: combining de
novo modeling with sparse experimental data. Journal of Computational Chemistry,
Vol. 28, pp.1668–1676.
Leitner, A.; Walzthoeni, T.; Kahraman, A.; Herzog, F.; Rinner, O.; Beck, M. & Aebersolda, R.
(2010). Probing Native Protein Structures by Chemical Cross-linking, Mass
Spectrometry, and Bioinformatics. Molecular & Cellular Proteomics, Vol.24, pp. 1634-
1649.
Lin, M.; Lu, H.M.; Rong Chen,R. & Liang, J.(2008). Generating properly weighted ensemble
of conformations of proteins from sparse or indirect distance constraints. The
Journal of Chemical Physics, Vol.129, pp.094101–094114.
Marti-Renom, M.A.; Madhusudhan, M.S. & Sali, A. (2004). Alignment of protein sequences
by their profiles. Protein Science, Vol.13, pp.1071–1087.
Lübben, M.; Portmann, R.; Kock, G.; Stoll, R.; Young, M.M. & Solioz, M. (2009). Structural
model of the CopA copper ATPase of Enterococcus hirae based on chemical cross-
linking. Biometals, Vol.22, pp. 363-375.
156 Computational Biology and Applied Bioinformatics

Mathiowetz, A.M.; Jain, A.; Karasawa, N. & Goddard, W.A. III. (1994). Protein simulation
using techniques suitable for very large systems: The cell multipole method for
nonbond interactions and the Newton–Euler inverse mass operator method for
internal coordinate dynamics. Proteins, Vol. 20, pp. 227–247.
Melo, F. & Sali, A. (2007). Fold assessment for comparative protein structure modeling.
Protein Science, Vol. 16, pp. 2412–2426.
Millevoi, S.; Thion, L.; Joseph, G.; Vossen, C.; Ghisolfi-Nieto, L. & Erard, M. (2001). Atypical
binding of the neuronal POU protein N-Oct3 to noncanonical DNA targets.
Implications for heterodimerization with HNF-3b. European Journal Biochemistry,
Vol.268, pp. 781-791.
Moreira, I.S.; Fernandes, P.A. & Ramos, M.J. (2010). Protein-protein docking dealing with
the unknown. Journal of Computational Chemistry, Vol. 31, pp.317–342.
Mouradov, D.; Craven, A.; Forwood, J.K.; Flanagan, J.U.; García-Castellanos, R.; Gomis-
Rüth, F.X.; Hume, D.A.; Martin, J.L.; Kobe, B. & Huber, T. (2006). Modelling the
structure of latexin–carboxypeptidase. A complex based on chemical cross-
linking and molecular docking. Protein Engineering, Design & Selection, Vol.19, pp.
9-16.
Nikolova, L.; Soman, K. ; Nichols, J.C.; Daniel, D.S., Dickey, B.F. & Hoffenberg, S.
(1998). Conformationally variable Rab protein surface regions mapped by
limited proteolysis and homology modelling. Biochemical Journal, Vol.336, pp.
461–469.
Nilges, M.; Clore, G.M. & Gronenborn, A.M.(1988a). Determination of three dimensional
structures of proteins from interproton distance data by hybrid distance
geometry-dynamical simulated annealing calculations. FEBS Letters, Vol.229,
pp.317–324.
Nilges, M.; Gronenborn, A.M.; Brünger, A.T. & Clore, G.M. (1988b). Determination of
three- dimensional structures of proteins by simulated annealing with
interproton distance restraints: application to crambin, potato carboxypeptidase
inhibitor and barley serine proteinase inhibitor 2. Protein Engineering, Vol.2,
pp.27-38.
Nymeyer, H.; Gnanakaran, S. and García, A.E. (2004). Atomic simulations of protein
folding using the replica exchange algorithm. Methods in Enzymology, Vol.383,
pp.111-149.
Papa, S.; Monti, S.M.; Vitale, R.M.; Bubici, C.; Jayawardena, S.; Alvarez, K.; De Smaele, E.;
Dathan, N.; Pedone, C.; Ruvo M. & Franzoso, G. (2007). Insights into the structural
basis of the GADD45beta-mediated inactivation of the JNK kinase, MKK7/JNKK2..
Journal of Biological Chemistry, Vol. 282, pp. 19029-19041.
Potluri, S.; Khan, A.A.; Kuzminykh, A.; Bujnicki, J.M., Friedman, A.M. & Bailey-Kellogg, C.
(2004). Geometric Analysis of Cross-Linkability for Protein Fold Discrimination.
Pacific Symposium on Biocomputing, Vol.9, pp.447-458.
Renzone, G.; Salzano, A.M.; Arena, S.; D’Ambrosio, C. & Scaloni, A.(2007a). Mass
Spectrometry-Based Approaches for Structural Studies on Protein Complexes at
Low-Resolution. Current Proteomics, Vol. 4, pp. 1-16.
Computational Methods in Mass Spectrometry-Based Protein 3D Studies 157

Renzone, G.; Vitale, R.M.; Scaloni, A.; Rossi, M., Amodeo, P. & Guagliardi A. (2007b).
Structural Characterization of the Functional Regions in the Archaeal Protein
Sso7d. Proteins: Structure, Function, and Bioinformatics, Vol. 67, pp. 189-197.
Rice, L.M & Brünger, A.T. (1994). Torsion angle dynamics: Reduced variable conformational
sampling enhances crystallographic structure refinement. Proteins, Vol. 19, pp. 277–
290.
Russell, R.B.; Alber, F.; Aloy, P.; Davis, F.P.; Korkin, D.;Pichaud, M; Topf, M. & Sali, A.
(2004). A structural perspective on protein-protein interactions. Current Opinion in
Structural Biology, Vol.14, pp. 313-324.
Scaloni, A; Miraglia, N.; Orrù, S.; Amodeo, P.; Motta, A.; Marino, G. & Pucci, P.(1998).
Topology of the calmodulin-melittin complex. Journal of Molecular Biology, Vol. 277,
pp.945–958.
Schrag, J.D.; Jiralerspong, S.; Banville, M; Jaramillo, M.L. & O'Connor-McCourt, M.D. (2007).
The crystal structure and dimerization interface of GADD45gamma. PNAS, Vol.
105, pp. 6566-6571.
Schueler-Furman, O.; Wang, C.; Bradley, P.; Misura, K. & Baker, D. (2005). Progress in
modeling of protein structures and interactions. Science, Vol. 310, pp.638–642.
Schulz,D.M.; Kalkhof, S.; Schmidt, A.; Ihling, C.; Stingl, C.; Mechtler, K.; Zschörnig, O &
Sinz, A. (2007). Annexin A2/P11 interaction: New insights into annexin A2
tetramer structure by chemical crosslinking, high-resolution mass spectrometry,
and computational modeling. Proteins: Structure Function & Bioinformatics, Vol.69,
pp. 254-269.
Schwieters, C.D. & Clore, G.M. (2001). Internal Coordinates for Molecular Dynamics and
Minimization in Structure Determination and Refinement. Journal of Magnetic
Resonance, Vol. 152, pp.288–302.
Silva, R.A.G.D.; Hilliard, G.M.; Fang, J.; Macha, S. & Davidson, W.S. (2005). A Three-
Dimensional Molecular Model of Lipid-Free Apolipoprotein A-I Determined by
Cross-Linking/Mass Spectrometry and Sequence Threading. Biochemistry, Vol.44,
pp. 2759-2769.
Singh, P.; Panchaud, A. & Goodlett, D.R. (2010) Chemical Cross-Linking and Mass
Spectrometry As a Low-Resolution Protein Structure Determination Technique.
Analytical Chemistry, Vol. 82, pp. 2636–2642
Sinz, A. (2006). Chemical cross-linking and mass spectrometry to map three dimensional
protein structures and protein-protein interactions. Mass Spectrometry Reviews,
Vol.25, pp. 663-682.
Skolnick,J.; Kolinski, A. & Ortiz, A.R.(1997). MONSSTER: A Method for Folding Globular
Proteins with a Small Number of Distance Restraints. Journal of Molecular Biology,
Vol. 265 pp.217-241.
Soding, J. (2005). Protein homology detection by HMM-HMM comparison. Bioinformatics,
Vol.21, pp.951–960.
Stein, E.G.; Rice, L.M & Brünger, A.T. (1997). Torsion-angle molecular dynamics as a new
efficient tool for NMR structure calculation. Journal of Magnetic Resonance, Vol. 124,
pp. 154–164.
158 Computational Biology and Applied Bioinformatics

Tubb, M.R.; Silva, R.A.G.D.; Fang, J.; Tso, P. & Davidson, W.S. (2008). A Three-dimensional
Homology Model of Lipid-free Apolipoprotein A-IV Using Cross-linking and Mass
Spectrometry. The Journal of Biochemical Chemistry, Vol.283, pp. 17314--17323.
Vaidehi, N., Jain, A. & Goddard, W.A. III (1996). Constant temperature constrained
molecular dynamics: The Newton–Euler inverse mass operator method. Journal of
Physical Chemistry, Vol. 100, pp.10 508–10517.
Van Dijk, A.D.J.; Boelens, R. & Bonvin, A.M.J.J. (2005). Data-driven docking for the study of
biomolecular complexes. FEBS Journal, Vol.272, pp.293–312.
Young, M.M.; Tang, N.; Hempel, J.C.; Oshiro, C.M.; Taylor, E.W.; Kuntz, I.D.; Gibson, B.W.
& Dollinger G. (2000). High throughput protein fold identification by using
experimental constraints derived from intramolecular cross-links and mass
spectrometry. PNAS, Vol.97, pp. 5802-2806.
Zheng, X.; Wintrode, P.L. & Chance M.R. (2007). Complementary Structural Mass
Spectrometry Techniques Reveal Local Dynamics in Functionally Important
Regions of a Metastable Serpin. Structure, Vol.16, pp. 38-51.
 8

Synthetic Biology & Bioinformatics

Prospects in the Cancer Arena
Lígia R. Rodrigues and Leon D. Kluskens
IBB – Institute for Biotechnology and Bioengineering, Centre of Biological Engineering,
University of Minho, Campus de Gualtar, Braga
Portugal

1. Introduction
Cancer is the second leading cause of mortality worldwide, with an expected 1.5-3.0 million
new cases and 0.5-2.0 million deaths in 2011 for the US and Europe, respectively (Jemal et
al., 2011). Hence, this is an enormously important health risk, and progress leading to
enhanced survival is a global priority. Strategies that have been pursued over the years
include the search for new biomarkers, drugs or treatments (Rodrigues et al., 2007).
Synthetic biology together with bioinformatics represents a powerful tool towards the
discovery of novel biomarkers and the design of new biosensors.
Traditionally, the majority of new drugs has been generated from compounds derived from
natural products (Neumann & Neumann-Staubitz, 2010). However, advances in genome
sequencing together with possible manipulation of biosynthetic pathways, constitute
important resources for screening and designing new drugs (Carothers et al. 2009).
Furthermore, the development of rational approaches through the use of bioinformatics for
data integration will enable the understanding of mechanisms underlying the anti-cancer
effect of such drugs (Leonard et al., 2008; Rocha et al., 2010).
Besides in biomarker development and the production of novel drugs, synthetic biology can
also play a crucial role in the level of specific drug targeting. Cells can be engineered to
recognize specific targets or conditions in our bodies that are not naturally recognized by
the immune system (Forbes, 2010).
Synthetic biology is the use of engineering principles to create, in a rational and systematic
way, functional systems based on the molecular machines and regulatory circuits of living
organisms or to re-design and fabricate existing biological systems (Benner & Sismour,
2005). The focus is often on ways of taking parts of natural biological systems, characterizing
and simplifying them, and using them as a component of a highly unnatural, engineered,
biological system (Endy, 2005). Virtually, through synthetic biology, solutions for the unmet
needs of humankind can be achieved, namely in the field of drug discovery. Indeed,
synthetic biology tools enable the elucidation of disease mechanisms, identification of
potential targets, discovery of new chemotherapeutics or design of novel drugs, as well as
the design of biological elements that recognize and target cancer cells. Furthermore,
through synthetic biology it is possible to develop economically attractive microbial
production processes for complex natural products.
160 Computational Biology and Applied Bioinformatics

Bioinformatics is used in drug target identification and validation, and in the development
of biomarkers and tools to maximize the therapeutic benefit of drugs. Now that data on
cellular signalling pathways are available, integrated computational and experimental
projects are being developed, with the goal of enabling in silico pharmacology by linking the
genome, transcriptome and proteome to cellular pathophysiology. Furthermore,
sophisticated computational tools are being developed that enable the modelling and design
of new biological systems. A key component of any synthetic biology effort is the use of
quantitative models (Arkin, 2001). These models and their corresponding simulations enable
optimization of a system design, as well as guiding their subsequent analysis. Dynamic
models of gene regulatory and reaction networks are essential for the characterization of
artificial and synthetic systems (Rocha et al., 2008). Several software tools and standards
have been developed in order to facilitate model exchange and reuse (Rocha et al., 2010).
In this chapter, synthetic biology approaches for cancer diagnosis and drug development
will be reviewed. Specifically, examples on the design of RNA-based biosensors, bacteria
and virus as anti-cancer agents, and engineered microbial cell factories for the production of
drugs, will be presented.

2. Synthetic biology: tools to design, build and optimize biological processes

Synthetic biology uses biological insights combined with engineering principles to design
and build new biological functions and complex artificial systems that do not occur in
Nature (Andrianantoandro et al., 2006). The building blocks used in synthetic biology are
the components of molecular biology processes: promoter sequences, operator sequences,
ribosome binding sites (RBS), termination sites, reporter proteins, and transcription factors.
Examples of such building blocks are given in Table 1.
Great developments of DNA synthesis technologies have opened new perspectives for the
design of very large and complex circuits (Purnick & Weiss, 2009), making it now affordable
to synthesize a given gene instead of cloning it. It is possible to synthesize de novo a small
virus (Mueller et al., 2009), to replace the genome of one bacterium by another (Lartigue et
al., 2007) and to make large chunks of DNA coding for elaborate genetic circuits. Software
tools to simulate large networks and the entire panel of omics technologies to analyze the
engineered microorganism are available (for details see section 3). Finally, the repositories of
biological parts (e.g. Registry of Standard Biological Parts (http://partsregistry.org/)) will
increase in complexity, number and reliability of circuits available for different species.
Currently, the design and synthesis of biological systems are not decoupled. For example,
the construction of metabolic pathways or any circuit from genetic parts first requires a
collection of well characterized parts, which do not yet fully exist. Nevertheless, this
limitation is being addressed through the development and compilation of standard
biological parts (Kelly et al., 2009). When designing individual biological parts, the base-by-
base content of that part (promoter, RBS, protein coding region, terminator, among others) is
explicitly dictated (McArthur IV & Fong, 2010). Rules and guidelines for designing genetic
parts at this level are being established (Canton et al., 2008). Particularly, an important issue
when designing protein-coding parts is codon optimization, encoding the same amino acid
sequence with an alternative, preferred nucleotide sequence. Although a particular
sequence, when expressed, may be theoretically functional, its expression may be far from
optimal or even completely suppressed due to codon usage bias in the heterologous host.
Synthetic Biology & Bioinformatics Prospects in the Cancer Arena 161

Genetic part Examples Rationale

Transcriptional control
Constitutive promoters lacIq, SV40, T7, sp6 “Always on” transcription
Regulatory regions tetO, lacO, ara, gal4, rhl box Repressor and activator sites
Inducible promoters ara, ethanol, lac, gal, rhl, lux, Control of the promoter by
fdhH, sal, glnK, cyc1 induction or by cell state
Cell fate regulators GATA factors Control cell differentiation
Transcriptional control
RNA interference Logic functions, RNAi Genetic switch, logic evaluation
(RNAi) repressor and gene silencing
Riboregulators Ligand-controlled Switches for detection and
ribozymes actuation
Ribosome binding site Kozak consensus sequence Control the level of translation
mutants
Post-transcriptional control
Phosphorylation Yeast phosphorylation Modulate genetic circuit behavior
cascades pathway
Protein receptor design TNT, ACT and EST Control detection thresholds and
receptors combinatorial protein function
Protein degradation Ssra tags, peptides rich in Protein degradation at varying
Pro, Glu, Ser and Thr rates
Localization signals Nuclear localization, Import or export from nucleus and
nuclear export and mitochondria
mitochondrial localization
signals
Others
Reporter genes GFP, YFP, CFP, LacZ Detection of expression
Antibiotic resistance ampicilin, Selection of cells
chloramphenicol
Table 1. Genetic elements used as components of synthetic regulatory networks (adapted
from McArthur IV & Fong, 2010 and Purnick & Weiss, 2009). Legend: CFP, cyan fluorescent
protein; GFP, green fluorescent protein; YFP, yellow fluorescent protein.
Codon optimization of coding sequences can be achieved using freely available algorithms
such as Gene Designer (see section 3). Besides codon optimization, compliance with
standard assembly requirements and part-specific objectives including activity or specificity
modifications should be considered. For example, the BioBrick methodology requires that
parts exclude four standard restriction enzyme sites, which are reserved for use in assembly
(Shetty et al., 2008). Extensive collections of parts can be generated by using a naturally
occurring part as a template and rationally modifying it to create a library of that particular
genetic part. Significant progress in this area has been recently demonstrated for promoters
and RBS (Ellis et al., 2009; Salis et al., 2009). Ellis and co-workers (2009) constructed two
promoter libraries that can be used to tune network behavior a priori by fitting mathematical
promoter models with measured parameters. By using this model-guided design approach
the authors were able to limit the variability of the system and increase predictability.
162 Computational Biology and Applied Bioinformatics

However, it is well-known that noisy or leaky promoters can complicate the system design.
In these cases a finer control over expression can be established by weakening the binding
strength of the downstream gene (Ham et al., 2006), or by using two promoter inputs to
drive transcription of an output via a modular AND gate (Anderson et al., 2006).
Additionally, modular and scalable RNA-based devices (aptamers, ribozymes, and
transmitter sequences) can be engineered to regulate gene transcription or translation (Win
& Smolke, 2007).
Design at the pathway level is not only concerned with including the necessary parts, but
also with controlling the expressed functionality of those parts. Parts-based synthetic
metabolic pathways will require tunable control, just as their natural counterparts which
often employ feedback and feed-forward motifs to achieve complex regulation (Purnick &
Weiss, 2009). Using a synthetic biology approach, the design of DNA sequences encoding
metabolic pathways (e.g. operons) should be relatively straightforward. Synthetic scaffolds
and well-characterized families of regulatory parts have emerged as powerful tools for
engineering metabolism by providing rational methodologies for coordinating control of
multigene expression, as well as decoupling pathway design from construction (Ellis et al.,
2009). Pathway design should not overlook the fact that exogenous pathways interact with
native cellular components and have their own specific energy requirements. Therefore,
modifying endogenous gene expression may be necessary in addition to balancing cofactor
fluxes and installing membrane transporters (Park et al., 2008).
After designing parts, circuits or pathways, the genomic constructs ought to be
manufactured through DNA synthesis. Nucleotide’s sequence information can be
outsourced to synthesis companies (e.g. DNA2.0, GENEART or Genscript, among others).
The convenience of this approach over traditional cloning allows for the systematic
generation of genetic part variants such as promoter libraries. Also, it provides a way to
eliminate restriction sites or undesirable RNA secondary structures, and to perform codon
optimization. The ability to make large changes to DNA molecules has resulted in
standardized methods for assembling basic genetic parts into larger composite devices,
which facilitate part-sharing and faster system-level construction, as demonstrated by the
BioBrick methodology (Shetty et al., 2008) and the Gateway cloning system (Hartley, 2003).
Other approaches based on type II restriction enzymes, such as Golden Gate Shuffling,
provide ways to assemble many more components together in one step (Engler et al., 2009).
A similar one-step assembly approach, circular polymerase extension cloning (CPEC),
avoids the need for restriction-ligation, or single-stranded homologous recombination
altogether (Quan & Tian, 2009). Not only is this useful for cloning single genes, but also for
assembling parts into larger sequences encoding entire metabolic pathways and for
generating combinatorial part libraries. On a chromosomal level, disruption of genes in
Escherichia coli and other microorganisms has become much faster with the development of
RecBCD and lambda RED-assisted recombination systems (Datsenko & Wanner, 2000),
allowing the insertion, deletion or modification by simply using linear gene fragments.
Additionally, multiplex automated genome engineering (MAGE) has been introduced as
another scalable, combinatorial method for producing large-scale genomic diversity (Wang
et al., 2009). This approach makes chromosomal modification easier by simultaneously
mutating target sites across the chromosome. Plasmid-based expression and chromosomal
integration are the two common vehicles for implementing synthetic metabolic pathways.
Recently, the chemically inducible chromosomal evolution (CIChE) was proposed as a long-
Synthetic Biology & Bioinformatics Prospects in the Cancer Arena 163

term expression alternative method (Tyo et al., 2009). This new method avoids
complications associated with plasmid replication and segregation, and can be used to
integrate multiple copies of genes into the genome. All these techniques will provide
technical platforms for the rapid synthesis of parts and subsequent pathways.
The majority of the synthetic biology advances has been achieved purely in vitro (Isalan et
al., 2008), or in microorganisms involving the design of small gene circuits without a direct
practical application, although scientifically very exciting. These studies have offered
fundamental insight into biological processes, like the role and sources of biological noise;
the existence of biological modules with defined properties; the dynamics of oscillatory
behavior; gene transcription and translation; or cell communication (Alon, 2003; Kobayashi
et al., 2004). An interesting example of a larger system that has been redesigned is the
refactoring of the T7 bacteriophage (Chan et al., 2005). Another successful example has been
the production of terpenoid compounds in E. coli (Martin et al., 2003) and Saccharomyces
cerevisiae (Ro et al., 2006) that can be used for the synthesis of artemisinin. Bacteria and fungi
have long been used in numerous industrial microbiology applications, synthesizing
important metabolites in large amounts. The production of amino acids, citric acid and
enzymes are examples of other products of interest, overproduced by microorganisms.
Genetic engineering of strains can contribute to the improvement of these production levels.
Altogether, the ability to engineer biological systems will enable vast progress in existing
applications and the development of several new possibilities. Furthermore, novel
applications can be developed by coupling gene regulatory networks with biosensor
modules and biological response systems. An extensive RNA-based framework has been
developed for engineering ligand-controlled gene regulatory systems, called ribozyme
switches. These switches exhibit tunable regulation, design modularity, and target
specificity and could be used, for example, to regulate cell growth (Win & Smolke, 2007).
Engineering interactions between programmed bacteria and mammalian cells will lead to
exciting medical applications (Anderson et al., 2006). Synthetic biology will change the
paradigm of the traditional approaches used to treat diseases by developing “smart”
therapies where the therapeutic agent can perform computation and logic operations and
make complex decisions (Andrianantoandro et al., 2006). There are also promising
applications in the field of living vectors for gene therapy and chemical factories (Forbes,
2010; Leonard et al., 2008).

3. Bioinformatics: a rational path towards biological behavior predictability

In order to evolve as an engineering discipline, synthetic biology cannot rely on endless trial
and error methods driven by verbal description of biomolecular interaction networks.
Genome projects identify the components of gene networks in biological organisms, gene
after gene, and DNA microarray experiments discover the network connections (Arkin,
2001). However, these data cannot adequately explain biomolecular phenomena or enable
rational engineering of dynamic gene expression regulation. The challenge is then to reduce
the amount and complexity of biological data into concise theoretical formulations with
predictive ability, ultimately associating synthetic DNA sequences to dynamic phenotypes.

3.1 Models for synthetic biology

The engineering process usually involves multiple cycles of design, optimization and
revision. This is particularly evident in the process of constructing gene circuits (Marguet et
164 Computational Biology and Applied Bioinformatics

al., 2007). Due to the large number of participating species and the complexity of their
interactions, it becomes difficult to intuitively predict a design behavior. Therefore, only
detailed modeling can allow the investigation of dynamic gene expression in a way fit for
analysis and design (Di Ventura et al., 2006). Modeling a cellular process can highlight
which experiments are likely to be the most informative in testing model hypothesis, and for
example allow testing for the effect of drugs (Di Bernardo et al., 2005) or mutant
phenotypes (Segre et al., 2002) on cellular processes, thus paving the way for individualized
medicine.
Data are the precursor to any model, and the need to organize as much experimental data as
possible in a systematic manner has led to several excellent databases as summarized in
Table 2. The term “model” can be used for verbal or graphical descriptions of a mechanism
underlying a cellular process, or refer to a set of equations expressing in a formal and exact
manner the relationships among variables that characterize the state of a biological system
(Di Ventura et al., 2006). The importance of mathematical modeling has been extensively
demonstrated in systems biology (You, 2004), although its utility in synthetic biology seems
even more dominant (Kaznessis, 2009).

Name Website
BIND (Biomolecular Interaction Network Database) http://www.bind.ca/
Brenda (a comprehensive enzyme information http://www.brenda.uni-koeln.de/
system)
CSNDB (Cell Signaling Networks Database) http://geo.nihs.go.jp/csndb/
DIP (Database of Interacting Proteins) http://dip.doe-mbi.ucla.edu/
EcoCyc/Metacyc/BioCyc (Encyclopedia of E. coli http://ecocyc.org/
genes and metabolism)
EMP (Enzymes and Metabolic Pathways Database) http://www.empproject.com/
GeneNet (information on gene networks) http://wwwmgs.bionet.nsc.ru/mgs/s
ystems/genenet/
Kegg (Kyoto Encyclopedia of Genes and Genomes) http://www.genome.ad.jp/kegg/kegg
.html
SPAD (Signaling Pathway Database) http://www.grt.kyushu-u.ac.jp/eny-
doc/
RegulonDB (E. coli K12 transcriptional network) http://regulondb.ccg.unam.mx/
ExPASy-beta (Bioinformatics Resource Portal) http://beta.expasy.org/
Table 2. Databases of molecular properties, interactions and pathways (adapted from Arkin,
2001).
Model-driven rational engineering of synthetic gene networks is possible at the level of
topologies or at the level of molecular components. In the first one, it is considered that
molecules control the concentration of other molecules, e.g. DNA-binding proteins regulate
the expression of specific genes by either activation or repression. By combining simple
regulatory interactions, such as negative and positive feedback and feed-forward loops, one
may create more complex networks that precisely control the production of protein
Synthetic Biology & Bioinformatics Prospects in the Cancer Arena 165

molecules (e.g. bistable switches, oscillators, and filters). Experimentally, these networks can
be created using existing libraries of regulatory proteins and their corresponding operator
sites. Examples of these models are the oscillator described by Gardner et al (2000) and
repressilator by Elowitz and Leibler (2000). In the second level, the kinetics and strengths of
molecular interactions within the system are described. By altering the characteristics of the
components, such as DNA-binding proteins and their corresponding DNA sites, one can
modify the system dynamics without modifying the network topology. Experimentally, the
DNA sequences that yield the desired characteristics of each component can be engineered
to achieve the desired protein-protein, protein-RNA, or protein-DNA binding constants and
enzymatic activities. For example, Alon and co-workers (2003) showed how simple
mutations on the DNA sequence of the lactose operon can result in widely different
phenotypic behavior.
Various mathematical formulations can be used to model gene circuits. At the population
level, gene circuits can be modeled using ordinary differential equations (ODEs). In an ODE
formulation, the dynamics of the interactions within the circuit are deterministic. That is, the
ODE formulation ignores the randomness intrinsic to cellular processes, and is convenient
for circuit designs that are thought to be less affected by noise or when the impact of noise is
irrelevant (Marguet et al., 2007). An ODE model facilitates further sophisticated analyses,
such as sensitivity analysis and bifurcation analysis. Such analyses are useful to determine
how quantitative or qualitative circuit behavior will be impacted by changes in circuit
parameters. For instance, in designing a bistable toggle switch, bifurcation analysis was
used to explore how qualitative features of the circuit may depend on reaction parameters
(Gardner et al., 2000). Results of the analysis were used to guide the choice of genetic
components (genes, promoters and RBS) and growth conditions to favor a successful
implementation of designed circuit function. However, in a single cell, the gene circuit’s
dynamics often involve small numbers of interacting molecules that will result in highly
noisy dynamics even for expression of a single gene. For many gene circuits, the impact of
such cellular noise may be critical and needs to be considered (Di Ventura et al., 2006). This
can be done using stochastic models (Arkin, 2001). Different rounds of simulation using a
stochastic model will lead to different results each time, which presumably reflect aspects of
noisy dynamics inside a cell. For synthetic biology applications, the key of such analysis is
not necessarily to accurately predict the exact noise level at each time point. This is not
possible even for the simplest circuits due to the “extrinsic” noise component for each circuit
(Elowitz et al., 2002). Rather, it is a way to determine to what extent the designed function
can be maintained and, given a certain level of uncertainty or randomness, to what extent
additional layers of control can minimize or exploit such variations. Independently of the
model that is used, these can be evolved in silico to optimize designs towards a given
function. As an example, genetic algorithms were used by Francois and Hakim (2004) to
design gene regulatory networks exhibiting oscillations.
In most attempts to engineer gene circuits, mathematical models are often purposefully
simplified to accommodate available computational power and to capture the qualitative
behavior of the underlying systems. Simplification is beneficial partially due to the limited
quantitative characterization of circuit elements, and partially because simpler models may
better reveal key design constraints. The limitation, however, is that a simplified model may
fail to capture richer dynamics intrinsic to a circuit. Synthetic models combine features of
mathematical models and model organisms. In the engineering of genetic networks,
166 Computational Biology and Applied Bioinformatics

synthetic biologists start from mathematical models, which are used as the blueprints to
engineer a model out of biological components that has the same materiality as model
organism but is much less complex. The specific characteristics of synthetic models allow
one to use them as tools in distinguishing between different mathematical models and
evaluating results gained in performing experiments with model organisms (Loettgers,
2007).

3.2 Computational tools for synthetic biology

Computational tools are essential for synthetic biology to support the design procedure at
different levels. Due to the lack of quantitative characterizations of biological parts, most
design procedures are iterative requiring experimental validation to enable subsequent
refinements (Canton et al., 2008). Furthermore, stochastic noise, uncertainty about the
cellular environment of an engineered system, and little insulation of components
complicate the design process and require corresponding models and analysis methods (Di
Ventura et al., 2006). Many computational standards and tools developed in the field of
systems biology (Wierling et al., 2007) are applicable for synthetic biology as well.
As previously discussed, synthetic gene circuits can be constructed from a handful of basic
parts that can be described independently and assembled into interoperating modules of
different complexity. For this purpose, standardization and modularity of parts at different
levels is required (Canton et al., 2008). The Registry of Standard Biological Parts constitutes
a reference point for current research in synthetic biology and it provides relevant
information on several DNA-based synthetic or natural building blocks. Most
computational tools that specifically support the design of artificial gene circuits use
information from the abovementioned registry. Moreover, many of these tools share
standardized formats for the input/output files. The System Biology Markup Language
(SBML) (http://sbml.org) defines a widely accepted, XML-based format for the exchange of
mathematical models in biology. It provides a concise representation of the chemical
reactions embraced by a biological system. These can be translated into systems of ODEs or
into reaction systems amenable to stochastic simulations (Alon, 2003). Despite its large
applicability to simulations, SBML currently lacks modularity, which is not well aligned
with parts registries in synthetic biology. Alternatively, synthetic gene systems can be
described according to CellML language which is more modular (Cooling et al., 2008).
One important feature to enable the assembly of standard biological parts into gene
circuits is that they share common inputs and outputs. Endy (2005) proposed RNA
polymerases and ribosomes as the molecules that physically exchange information
between parts. Their fluxes, measured in PoPS (Polymerase Per Second) and in RiPS
(Ribosomes Per Second) represent biological currents (Canton et al., 2008). This picture,
however, does not seem sufficient to describe all information exchanges even in simple
engineered gene circuits, since other signal carriers like transcription factors and
environmental “messages” should be explicitly introduced and not indirectly estimated
by means of PoPS and RiPS (Marchisio & Stelling, 2008). Based on the assumption that
parts share common input/output signals, several computational tools have been
proposed for gene circuit design, as presented in Table 3. Comparing these circuit design
tools it is obvious that we are still far from an ideal solution. The software tools differ in
many aspects such as scope of parts and circuit descriptions, the mode of user interaction,
and the integration with databases or other tools.
Synthetic Biology & Bioinformatics Prospects in the Cancer Arena 167

Circuit design and simulation

Biojade http://web.mit.edu/jagoler/www/biojade/
Tinkercell http://www.tinkercell.com/Home
Asmparts http://soft.synth-bio.org/asmparts.html
ProMoT http://www.mpimagdeburg.mpg.de/projects/promot
GenoCAD http://www.genocad.org/genocad/
GEC http://research.microsoft.com/gec
TABASCO http://openwetware.org/wiki/TABASCO
Circuit optimization
Genetdes http://soft.synth-bio.org/genetdes.html
RoVerGeNe http://iasi.bu.edu/~batt/rovergene/rovergene.htm
DNA and RNA design
Gene Designer https://www.dna20.com/index.php?pageID=220
GeneDesign http://www.genedesign.org/
UNAFold http://www.bioinfo.rpi.edu/applications/hybrid/download.php
Vienna RNA package http://www.tbi.univie.ac.at/~ivo/RNA/
Zinc Finger Tools http://www.scripps.edu/mb/barbas/zfdesign/zfdesignhome.php
Protein Design
Rosetta http://www.rosettacommons.org/main.html
RAPTOR http://www.bioinformaticssolutions.com/products/raptor/index.
php
PFP http://dragon.bio.purdue.edu/pfp/
Autodock 4.2 http://autodock.scripps.edu/
HEX 5.1 http://webloria.loria.fr/~ritchied/hex/
Integrated workflows
SynBioSS http://synbioss.sourceforge.net/
Clotho http://biocad-server.eecs.berkeley.edu/wiki/index.php/Tools
Biskit http://biskit.sf.net/
Table 3. Computational design tools for synthetic biology (adapted from Marchisio &
Stelling, 2009; Matsuoka et al., 2009; and Prunick & Weiss, 2009)
Biojade was one of the first tools being reported for circuit design (Goler, 2004). It provides
connections to both parts databases and simulation environments, but it considers only one
kind of signal carrier (RNA polymerases). It can invoke the simulator TABASCO (Kosuri et
al., 2007), thus enabling genome scale simulations at single base-pair resolution.
CellDesigner (Funahashi et al., 2003) has similar capabilities for graphical circuit
composition. However, parts modularity and consequently circuit representation do not
appear detailed enough. Another tool for which parts communicate only by means of PoPS,
but not restricted to a single mathematical framework, is the Tinkercell. On the contrary, in
Asmparts (Rodrigo et al., 2007a) the circuit design is less straightforward and intuitive
because the tool lacks a Graphic User Interface. Nevertheless, each part exists as an
independent SBML module and the model kinetics for transcription and translation permit
to limit the number of parameters necessary for a qualitative system description. Marchisio
and Stelling (2008) developed a new framework for the design of synthetic circuits where
168 Computational Biology and Applied Bioinformatics

each part is modeled independently following the ODE formalism. This results in a set of
composable parts that communicate by fluxes of signal carriers, whose overall amount is
constantly updated inside their corresponding pools. The model also considers transcription
factors, chemicals and small RNAs as signal carriers. Pools are placed among parts and
devices: they store free signal carriers and distribute them to the whole circuit. Hence,
polymerases and ribosomes have a finite amount; this permits to estimate circuit scalability
with respect to the number of parts. Mass action kinetics is fully employed and no
approximations are required to depict the interactions of signal carriers with DNA and
mRNA. The authors implemented the corresponding models into ProMoT (Process
Modeling Tool), software for the object-oriented and modular composition of models for
dynamic processes (Mirschel et al., 2009). GenoCAD (Czar et al., 2009) and GEC (Pedersen &
Phillips, 2009) introduce the notions of a grammar and of a programming language for
genetic circuit design, respectively. These tools use a set of rules to check the correct
composition of standard parts. Relying on libraries of standard parts that are not necessarily
taken from the Registry of Standard Biological Parts, these programs can translate a circuit
design into a complete DNA sequence. The two tools differ in capabilities and possible
connectivity to other tools.
The ultimate goal of designing a genetic circuit is that it works, i.e. that it performs a given
function. For that purpose, optimization cycles to establish an appropriate structure and a
good set of kinetic parameters values are required. These optimization problems are
extremely complex since they involve the selection of adequate parts and appropriate
continuous parameter values. Stochastic optimization methods (e.g. evolutionary
algorithms) attempt to find good solutions by biased random search. They have the
potential for finding globally optimal solutions, but optimization is computationally
expensive. On the other hand, deterministic methods (e.g. gradient descent) are local search
methods, with less computational cost, but at the expense of missing good solutions.
The optimization problem can be tackled by tools such as Genetdes (Rodrigo et al., 2007b)
and OptCircuit (Dasika & Maranas, 2008). They rely on different parts characterizations and
optimization algorithms. Genetdes uses a stochastic method termed “Simulated Annealing”
(Kirkpatrick et al., 1983), which produces a single solution starting from a random circuit
configuration. As a drawback, the algorithm is more likely to get stuck in a local minimum
than an evolutionary algorithm. OptCircuit, on the contrary, treats the circuit design
problem with a deterministic method (Bansal et al., 2003), implementing a procedure
towards a “local’ optimal solution. Each of these optimization algorithms requires a very
simplified model for gene dynamics where, for instance, transcription and translation are
treated as a single step process. Moreover, the current methods can cope only with rather
small circuits. Another tool that has been described by Batt and co-workers (2007),
RoVerGeNe, addresses the problem of parameter estimation more specifically. This tool
permits to tune the performance and to estimate the robustness of a synthetic network with
a known behavior and for which the topology does not require further improvement.
Detailed design of synthetic parts that reproduce the estimated circuit kinetics and
dynamics is a complex task. It requires computational tools in order to achieve error free
solutions in a reasonable amount of time. Other than the placement/removal of restriction
sites and the insertion/deletion of longer motifs, mutations of single nucleotides may be
necessary to tune part characteristics (e.g. promoter strength and affinity toward regulatory
factors). Gene Designer (Villalobos et al., 2006) is a complete tool for building artificial DNA
Synthetic Biology & Bioinformatics Prospects in the Cancer Arena 169

segments and codon usage optimization. GeneDesign (Richardson et al., 2006) is another
tool to design long synthetic DNA sequences. Many other tools are available for specific
analysis of the DNA and RNA circuit components. The package UNAFold (Markham &
Zuker, 2008) predicts the secondary structure of nucleic acid sequences to simulate their
hybridizations and to estimate their melting temperature according to physical
considerations. A more accurate analysis of the secondary structure of ribonucleic acids can
be performed through the Vienna RNA package (Hofacker, 2003). Binding sites along a
DNA chain can be located using Zinc Finger Tools (Mandell & Barbas, 2006). These tools
allows one to search DNA sequences for target sites of particular zinc finger proteins
(Kaiser, 2005), whose structure and composition can also be arranged. Thus, gene control by
a class of proteins with either regulation or nuclease activity can be improved. Furthermore,
tools that enable promoter predictions and primers design are available, such as BDGP and
Primer3. Another relevant task in synthetic biology is the design and engineering of new
proteins. Many tools have been proposed for structure prediction, homology modeling,
function prediction, docking simulations and DNA-protein interactions evaluation.
Examples include the Rosetta package (Simons et al., 1999); RAPTOR (Xu et al., 2003); PFP
(Hawkins et al., 2006); Autodock 4.2 (Morris et al., 2009) and Hex 5.1 (Ritchie, 2008).
Further advance in computational synthetic biology will result from tools that combine and
integrate most of the tasks discussed, starting with the choice and assembly of biological
parts to the compilation and modification of the corresponding DNA sequences. Examples
of such tools comprise SynBioSS (Hill et al., 2008); Clotho and Biskit (Grunberg et al., 2007).
Critical elements are still lacking, such as tools for automatic information integration
(literature and databases), and tools that re-use standardized model entities for optimal
circuit design. Overall, providing an extended and integrated information technology
infrastructure will be crucial for the development of the synthetic biology field.

4. A roadmap from design to production of new drugs

Biological systems are dynamic, that is they mutate, evolve and are subject to noise.
Currently, the full knowledge on how these systems work is still limited. As previously
discussed, synthetic biology approaches involve breaking down organisms into a hierarchy
of composable parts, which is useful for conceptualization purposes. Reprogramming a cell
involves the creation of synthetic biological components by adding, removing, or changing
genes and proteins. Nevertheless, it is important to notice that assembly of parts largely
depends on the cellular context (the so-called chassis), thus restraining the abstraction of
biological components into devices and modules, and their use in design and engineering of
new organisms or functions.
One level of abstraction from the DNA synthesis and manipulation is parts production,
which optimization can be accomplished through either rational design or directed
evolution. Applying rational design to parts alteration or creation is advantageous, in that it
cannot only generate products with a defined function, but it can also produce biological
insights into how the designed function comes about. However, it requires prior structural
knowledge of the part, which is frequently unavailable. Directed evolution is an alternative
method that can effectively address this limitation. Many synthetic biology applications will
require parts for genetic circuits, cell–cell communication systems, and non-natural
metabolic pathways that cannot be found in Nature, simply because Nature is not in need of
them (Dougherty & Arnold, 2009). In essence, directed evolution begins with the generation
170 Computational Biology and Applied Bioinformatics

of a library containing many different DNA molecules, often by error-prone DNA

replication, DNA shuffling or combinatorial synthesis (Crameri et al., 1998). The library is
next subjected to high-throughput screening or selection methods that maintain a link
between genotype and phenotype in order to enrich the molecules that produce the desired
function. Directed evolution can also be applied at other levels of biological hierarchy, for
example to evolve entire gene circuits (Yokobayashi et al., 2002). Rational design and
directed evolution should not be viewed as opposing methods, but as alternate ways to
produce and optimize parts, each with their own unique strengths and weaknesses.
Directed evolution can complement this technique, by using mutagenesis and subsequent
screening for improved synthetic properties (Brustad & Arnold, 2010). In addition, methods
have been developed to incorporate unnatural amino acids in peptides and proteins
(Voloshchuk & Montclare, 2009). This will expand the toolbox of protein parts, and add
beneficial effects, such as increased in vivo stability, when incorporated in proteinaceous
therapeutics. Also, the development of de novo enzymes has seen a significant increase
lately. The principle of computational design uses the design of a model, capable of
stabilizing the transition state of a reaction. From there on, individual amino acids are
positioned around it to create a catalytic site that stabilizes the transition state. The mRNA
display technique resembles phage display and is a technique for the in vitro selection and
evolution of proteins. Translated proteins are associated with their mRNA via a puromycin
linkage. Selection occurs by binding to an immobilized substrate, after which a reverse
transcriptase step will reveal the cDNA and thus the nucleotide sequence (Golynskiy &
Seelig, 2010). If the selection step includes measurement of product formation from the
substrate, novel peptides with catalytic properties can be selected.
For the design, engineering, integration and testing of new synthetic gene networks, tools
and methods derived from experimental molecular biology must be used (for details see
section 2). Nevertheless, progress on these tools and methods is still not enough to
guarantee the complete success of the experiment. As a result, design of synthetic biological
systems has become an iterative process of modeling, construction, and experimental testing
that continues until a system achieves the desired behavior (Purnick & Weiss, 2009). The
process begins with the abstract design of devices, modules, or organisms, and is often
guided by mathematical models (Koide et al., 2009). Afterwards, the newly constructed
systems are tested experimentally. However, such initial attempts rarely yield fully
functional implementations due to incomplete biological information. Rational redesign
based on mathematical models improves system behavior in such situations (Koide et al.,
2009; Prather & Martin, 2008). Directed evolution is a complimentary approach, which can
yield novel and unexpected beneficial changes to the system (Yokobayashi et al., 2002).
These retooled systems are once again tested experimentally and the process is repeated as
needed. Many synthetic biological systems have been engineered successfully in this fashion
because the methodology is highly tolerant to uncertainty (Matsuoka et al., 2009). Figure 1
illustrates the above mentioned iterative approach used in synthetic biology.
Since its inception, metabolic engineering aims to optimize cellular metabolism for a
particular industrial process application through the use of directed genetic modifications
(Tyo et al., 2007). Metabolic engineering is often seen as a cyclic process (Nielsen, 2001),
where the cell factory is analyzed and an appropriate target is identified. This target is then
experimentally implemented and the resulting strain is characterized experimentally and, if
necessary, further analyses are conducted to identify novel targets. The application of
Synthetic Biology & Bioinformatics Prospects in the Cancer Arena 171

synthetic biology to metabolic engineering can potentially create a paradigm shift. Rather
than starting with the full complement of components in a wild-type organism and
piecewise modifying and streamlining its function, metabolic engineering can be attempted
from a bottom-up, parts-based approach to design by carefully and rationally specifying the
inclusion of each necessary component (McArthur IV & Fong, 2010). The importance of
rationally designing improved or new microbial cell factories for the production of drugs
has grown substantially since there is an increasing need for new or existing drugs at prices
that can be affordable for low-income countries. Large-scale re-engineering of a biological
circuit will require systems-level optimization that will come from a deep understanding of
operational relationships among all the constituent parts of a cell. The integrated framework
necessary for conducting such complex bioengineering requires the convergence of systems
and synthetic biology (Koide et al., 2009). In recent years, with advances in systems biology
(Kitano, 2002), there has been an increasing trend toward using mathematical and
computational tools for the in silico design of enhanced microbial strains (Rocha et al., 2010).

MODEL FORMULATION Kinetic model (ODE/SDE)

FROM MASS ACTION (experimental data; existing model
KINETICS parameters)

In silico testing
(parameter space; dynamics)

SIMULATION
DYNAMIC PROPERTIES
EXPLOITATION

re-engineer
Model organism (chassis)
CIRCUIT CONSTRUCTION •Platform (microscopy; flow cytometry; microfluidics)
•Read-out (phenotype; morphology; expression)
EXPERIMENTAL •Analysis (equilibrium; cellular context; genetic
VALIDATION background; environment)
•Circuit construction

CIRCUITS DEFINITION OR
IMPROVEMENT
CHARACTERIZATION

Fig. 1. The iterative synthetic biology approach to design a given biological circuit/system.
Current models in both synthetic and systems biology emphasize the relationship between
environmental influences and the responses of biological networks. Nevertheless, these
models operate at different scales, and to understand the new paradigm of rational systems
re-engineering, synthetic and systems biology fields must join forces (Koide et al., 2009).
Synthetic biology and bottom-up systems biology methods extract discrete, accurate,
quantitative, kinetic and mechanistic details of regulatory sub-circuits. The models
generated from these approaches provide an explicit mathematical foundation that can
ultimately be used in systems redesign and re-engineering. However, these approaches are
confounded by high dimensionality, non-linearity and poor prior knowledge of key
dynamic parameters (Fisher & Henzinger, 2007) when scaled to large systems.
172 Computational Biology and Applied Bioinformatics

Consequently, modular sub-network characterization is performed assuming that the

network is isolated from the rest of the host system. The top-down systems biology
approach is based on data from high-throughput experiments that list the complete set of
components within a system in a qualitative or semi-quantitative manner. Models of overall
systems are similarly qualitative, tending toward algorithmic descriptions of component
interactions. Such models are amenable to the experimental data used to develop them, but
usually sacrifice the finer kinetic and mechanistic details of the molecular components
involved (Price & Shmulevich, 2007). Bridging systems and synthetic biology approaches is
being actively discussed and several solutions have been suggested (Koide et al., 2009).
A typical synthetic biology project is the design and engineering of a new biosynthetic
pathway in a model organism (chassis). Generally, E. coli is the preferred chassis since it is
well-studied, easy to manipulate, and its reproduction in biological cultures is very handy.
Initially, relevant databases like Kegg and BioCyc (Table 2) can be consulted for identifying
all the possible metabolic routes that allow the production of a given drug from metabolites
that exist in native E. coli. Then, for each reaction, the species that are known to possess the
corresponding enzymes/genes must be identified. This step is relevant, since most often the
same enzyme exhibits different kinetic behavior among different species. Information on
sequences and kinetic parameters can be extracted from the above-mentioned sources,
relevant literature and also from Brenda and Expasy databases. Afterwards, the information
collected is used to build a family of dynamic models (Rocha et al., 2008, 2010) that enable
the simulation of possible combinations regarding pathway configuration and the origin of
the enzymes (associated with varying kinetic parameters). Optflux tool can be used for
simulations and metabolic engineering purposes (http://www.optflux.org/). Using the
same input (a fixed amount of precursor) it is possible to select the configuration that allows
obtaining higher drug yields. Furthermore, through the use of genome-scale stoichiometric
models coupled with the dynamic model it is possible to understand the likely limitations
regarding the availability of the possible precursors. In fact, if the precursor for the given
drug biosynthesis is a metabolic intermediate, possible limitations in its availability need to
be addressed in order to devise strategies to cope with it. Based on this information, the next
step involves the construction of the enzymatic reactions that will lead to the production of
the drug from a metabolic precursor in E. coli. The required enzymes are then synthesized
based on the gene sequences previously selected from the databases. The cloning strategy
may include using a single plasmid with two different promoters; using two different
plasmids, with different copy numbers and/or origins of replication; or ultimately
integrating it into the genome, in order to allow fine tuning of the expression of the various
enzymes necessary. Finally, a set of experiments using the engineered bacterium needs to be
performed to evaluate its functionality, side-product formation and/or accumulation,
production of intermediate metabolites and final product (desired drug). In the case of the
previously mentioned artemisinin production, DNA microarray analysis and targeted
metabolic profiling were used to optimize the synthetic pathway, reducing the accumulation
of toxic intermediates (Kizer et al., 2008). These types of methodologies enable the validation
of the drug production model and the design of strategies to further improve its production.

5. Novel strategies for cancer diagnosis and drug development

Cancer is a main issue for the modern society and according to the World Health
Organization it is within the top 10 of leading causes of death in middle- and high-income
Synthetic Biology & Bioinformatics Prospects in the Cancer Arena 173

countries. Several possibilities to further improve existing therapies and diagnostics, or to

develop novel alternatives that still have not been foreseen, can be drawn using synthetic
biology approaches. Promising future applications include the development of RNA-based
biosensors to produce a desired response in vivo or to be integrated in a cancer diagnosis
device; the design and engineering of bacteria that can be programmed to target a tumor
and release a therapeutic agent in situ; the use of virus as a tool for recognizing tumors or for
gene therapy; and the large scale production of complex chemotherapeutic agents, among
others.

5.1 RNA-based biosensors

Synthetic biology seeks for new biological devices and systems that regulate gene
expression and metabolite pathways. Many components of a living cell possess the ability to
carry genetic information, such as DNA, RNA, proteins, among others. RNA has a critical
role in several functions (genetic translation, protein synthesis, signal recognition of
particles) due to its functional versatility from genetic blueprint (e.g. mRNA, RNA virus
genomes. Its catalytic function as enzyme (e.g. ribozymes, rRNA) and regulator of gene
expression (e.g. miRNA, siRNA) makes it stand out among other biopolymers with a more
specialized scope (e.g. DNA, proteins) (Dawid et al., 2009). Therefore, non-coding RNA
molecules enable the formation of complex structures that can interact with DNA, other
RNA molecules, proteins and other small molecules (Isaacs et al., 2006).
Natural biological systems contain transcription factors and regulators, as well as several
RNA-based mechanisms for regulating gene expression (Saito & Inoue, 2009). A number of
studies have been conducted on the use of RNA components in the construction of synthetic
biologic devices (Topp & Gallivan, 2007; Win & Smolke, 2007). The interaction of RNA with
proteins, metabolites and other nucleic acids is affected by the relationship between
sequence, structure and function. This is what makes the RNA molecule so attractive and
malleable to engineering complex and programmable functions.

5.1.1 Riboswitches
One of the most promising elements are the riboswitches, genetic control elements that
allow small molecules to regulate gene expression. They are structured elements typically
found in the 5’-untranslated regions of mRNA that recognize small molecules and respond
by altering their three-dimensional structure. This, in turn, affects transcription elongation,
translation initiation, or other steps of the process that lead to protein production (Beisel &
Smolke, 2009; Winkler & Breaker, 2005). Biological cells can modulate gene expression in
response to physical and chemical variations in the environment allowing them to control
their metabolism and preventing the waste of energy expenditure or inappropriate
physiological responses (Garst & Batey, 2009). There are currently at least twenty classes of
riboswitches that recognize a wide range of ligands, including purine nucleobases (purine
riboswitch), amino acids (lysine riboswitch), vitamin cofactors (cobalamin riboswitch),
amino sugars, metal ions (mgtA riboswitch) and second messenger molecules (cyclic di-
GMP riboswitch) (Beisel & Smolke, 2009). Riboswitches are typically composed of two
distinct domains: a metabolite receptor known as the aptamer domain, and an expression
platform whose secondary structure signals the regulatory response. Embedded within the
aptamer domain is the switching sequence, a sequence shared between the aptamer domain
and the expression platform (Garst & Batey, 2009). The aptamer domain is part of the RNA
174 Computational Biology and Applied Bioinformatics

and forms precise three-dimensional structures. It is considered a structured nucleotide

pocket belonging to the riboswitch, in the 5´-UTR, which when bound regulates
downstream gene expression (Isaacs et al., 2006). Aptamers specifically recognize their
corresponding target molecule, the ligand, within the complex group of other metabolites,
with the appropriate affinity, such as dyes, biomarkers, proteins, peptides, aromatic small
molecules, antibiotics and other biomolecules. Both the nucleotide sequence and the
secondary structure of each aptamer remain highly conserved (Winkler & Breaker, 2005).
Therefore, aptamer domains are the operators of the riboswitches.
A strategy for finding new aptamer sequences is the use of SELEX (Systemic Evolution of
Ligands by Exponential enrichment method). SELEX is a combinatorial chemistry technique
for producing oligonucleotides of either single-stranded DNA or RNA that specifically bind
to one or more target ligands (Stoltenburg et al., 2007). The process begins with the synthesis
of a very large oligonucleotide library consisting of randomly generated sequences of fixed
length flanked by constant 5' and 3' ends that serve as primers. The sequences in the library
are exposed to the target ligand and those that do not bind the target are removed, usually
by affinity chromatography. The bound sequences are eluted and amplified by PCR to
prepare for subsequent rounds of selection in which the stringency of the elution conditions
is increased to identify the tightest-binding sequences (Stoltenburg et al., 2007). SELEX has
been used to evolve aptamers of extremely high binding affinity to a variety of target
ligands. Clinical uses of the technique are suggested by aptamers that bind tumor markers
(Ferreira et al., 2006). The aptamer sequence must then be placed near to the RBS of the
reporter gene, and inserted into E. coli (chassis), using a DNA carrier (i.e. plasmid), in order
to exert its regulatory function.
Synthetic riboswitches represent a powerful tool for the design of biological sensors that
can, for example, detect cancer cells, or the microenvironment of a tumor, and in the
presence of a given molecule perform a desired function, like the expression in situ of a
therapeutic agent. Several cancer biomarkers have been identified in the last decade;
therefore there are many opportunities of taking these compounds as templates to design
adequate riboswitches for their recognition. Alternatively, the engineering goal might be the
detection of some of these biomarkers in biological samples using biosensors with aptamers
as the biological recognition element, hence making it a less invasive approach. The
development of aptamer-based electrochemical biosensors has made the detection of small
and macromolecular analytes easier, faster, and more suited for early detection of protein
biomarkers (Hianik & Wang, 2009). Multi-sensor arrays that provide global information on
complex samples (e.g. biological samples) have deserved much interest recently. Coupling
an aptamer to these devices will increase its specificity and selectivity towards the selected
target(s). The selected target may be any serum biomarker that when detected in high
amounts in biological samples can be suggestive of tumor activity.

5.2 Bacteria as anti-cancer agents

Bacteria possess unique features that make them powerful candidates for treating cancer in
ways that are unattainable by conventional methods. The moderate success of conventional
methods, such as chemotherapy and radiation, is related to its toxicity to normal tissue and
inability to destroy all cancer cells. Many bacteria have been reported to specifically target
tumors, actively penetrate tissue, be easily detected and/or induce a controlled cytotoxicity.
The possibility of engineering interactions between programmed bacteria and mammalian
Synthetic Biology & Bioinformatics Prospects in the Cancer Arena 175

cells opens unforeseen progresses in the medical field. Emerging applications include the
design of bacteria to produce therapeutic agents (in vitro or in vivo) and the use of live
bacteria as targeted delivery systems (Forbes, 2010; Pawelek et al., 2003). An impressive
example of these applications is described by Anderson and co-workers (2006). The authors
have successfully engineered E. coli harboring designed plasmids to invade cancer cells in
an environmentally controlled way, namely in a density-dependent manner under
anaerobic growth conditions and arabinose induction. Plasmids were built containing the
inv gene from Yersinia pseudotuberculosis under control of the Lux promoter, the hypoxia-
responsive fdhF promoter, and the arabinose-inducible araBAD promoter. This is significant
because the tumor environment is often hypoxic and allows for high bacterial cell densities
due to depressed immune function in the tumor. Therefore, this work demonstrated, as a
“proof of concept”, that one can potentially use engineered bacteria to target diseased cells
without significantly impacting healthy cells.
Ideally, an engineered bacterium for cancer therapy would specifically target tumors
enabling the use of more toxic molecules without systemic effects; be self-propelled enabling
its penetration into tumor regions that are inaccessible to passive therapies; be responsive to
external signals enabling the precise control of location and timing of cytotoxicity; be able to
sense the local environment allowing the development of responsive therapies that can
make decisions about where and when drugs are administered; and be externally detectable,
thus providing information about the state of the tumor, the success of localization and the
efficacy of treatment (Forbes, 2010). Indeed some of these features naturally exist in some
bacteria, e.g. many genera of bacteria have been shown to preferentially accumulate in
tumors, including Salmonella, Escherichia, Clostridium and Bifidobacterium. Moreover, bacteria
have motility (flagella) that enable tissue penetration and chemotactic receptors that direct
chemotaxis towards molecular signals in the tumor microenvironment. Selective
cytotoxicity can be engineered by transfection with genes for therapeutic molecules,
including toxins, cytokines, tumor antigens and apoptosis-inducing factors. External control
can be achieved using gene promoter strategies that respond to small molecules, heat or
radiation. Bacteria can be detected using light, magnetic resonance imaging and positron
emission tomography. At last, genetic manipulation of bacteria is easy, thus enabling the
development of treatment strategies, such as expression of anti-tumor proteins and
including vectors to infect cancer cells (Pawelek et al., 2003). To date, many different
bacterial strategies have been implemented in animal models (e.g. Salmonella has been tested
for breast, colon, hepatocellular, melanoma, neuroblastoma, pancreatic and prostate cancer),
and also some human trials (e.g. C. butyricum M-55 has been tested for squamous cell
carcinoma, metastic, malignant neuroma and melanoma) have been carried out (Forbes,
2010).
Ultrasound is one of the techniques often used to treat solid tumors (e.g. breast cancer);
however, this technique is not always successful, as sometimes it just heats the tumor
without destroying it. Therefore, we are currently engineering the heat shock response
machinery from E. coli to trigger the release of a therapeutic agent in situ concurrent with
ultrasound treatment. For that purpose, several modeling and engineering steps are being
implemented. The strategy being pursued is particularly useful for drugs that require in situ
synthesis because of a poor bioavailability, thereby avoiding repetitive oral doses to achieve
sufficient concentration inside the cells. The use of live bacteria for therapeutic purposes
naturally poses some issues (Pawelek et al., 2003), but currently the goal is to achieve the
176 Computational Biology and Applied Bioinformatics

proof-of-concept that an engineered system will enable the production of a cancer-fighting

drug triggered by a temperature increase.

5.3 Alternative nanosized drug carriers

The design of novel tumor targeted multifunctional particles is another extremely
interesting and innovative approach that makes use of the synthetic biology principles. The
modest success of the traditional strategies for cancer treatment has driven research towards
the development of new approaches underpinned by mechanistic understanding of cancer
progression and targeted delivery of rational combination therapy.

5.3.1 Viral drug delivery systems

The use of viruses, in the form of vaccines, has been common practice ever since its first use
to combat smallpox. Recently, genetic engineering has enlarged the applications of viruses,
since it allows the removal of pathogen genes encoding virulence factors that are present in
the virus coat. As a result, it can elicit immunity without causing serious health effects in
humans. In the light of gene therapy, the use of virus-based entities hold a promising future,
since by nature, they are being delivered to human target cells, and can be easily
manipulated genetically. As such, they may be applied to target and lyse specific cancer
cells, delivering therapeutics in situ. Bacteriophages are viruses that specifically and only
infect bacteria. They have gained more attention the last decades, mainly in phage display
technology. In anti-cancer therapy, this technique has contributed enormously to the
identification of new tumor-targeting molecules (Brown, 2010). In vivo phage display
technology identified a peptide exhibiting high affinity to hepatocellular carcinoma cells
(Du et al., 2010). In a different approach, a phage display-selected ligand targeting breast
cancer cells was incorporated in liposomes containing siRNA. The delivered liposomes were
shown to significantly downregulate the PRDM14 gene in the MCF7 target cells (Bedi et al.,
2011). In addition, the option to directly use bacteriophages as drug-delivery platforms has
been explored. A recent study described the use of genetically modified phages able to
target tumor cell receptors via specific antibodies resulting in endocytosis, intracellular
degradation, and drug release (Bar et al., 2008). Using phage display a variety of cancer cell
binding and internalizing ligands have been selected (Gao et al., 2003). Bacteriophages can
also be applied to establish an immune response. Eriksson and co-workers (2009) showed
that a tumor-specific M13 bacteriophage induced regression of melanoma target cells,
involving tumor-associated macrophages and being Toll-like receptor-dependent. Finally,
marker molecules or drugs can be chemically conjugated onto the phage surface, making it a
versatile imaging or therapy vehicle that may reduce costs and improve life quality
(Steinmetz, 2010). An M13 phage containing cancer cell-targeting motifs on the surface was
chemically modified to conjugate with fluorescent molecules, resulting in both binding and
imaging of human KB cancer cells (Li et al., 2010). Besides being genetically part of the virus,
anti-tumor compounds can also be covalently linked to it. We are currently, by phage
display, selecting phages that adhere and penetrate tumor cells. Following this selection, we
will chemically conjugate anti-cancer compounds (e.g. doxorubicin) to bacteriophages,
equipped with the cancer cell-recognizing peptides on the phage surface. We anticipate that
such a multifunctional nanoparticle, targeted to the tumor using a tumor “homing” peptide,
will enable a significant improvement over existing anti-cancer approaches.
Synthetic Biology & Bioinformatics Prospects in the Cancer Arena 177

5.4 Microbial cell factories for the production of drugs

On a different perspective, as exemplified in section 4, synthetic biology approaches can be
used for the large scale production of compounds with pharmaceutical applications. One of
the easily employable approaches to develop synthetic pathways is to combine genes from
different organisms, and design a new set of metabolic pathways to produce various natural
and unnatural products. The host organism provides precursors from its own metabolism,
which are subsequently converted to the desired product through the expression of the
heterologous genes (see section 4). Existing examples of synthetic metabolic networks make
use of transcriptional and translational control elements to regulate the expression of
enzymes that synthesize and breakdown metabolites. In these systems, metabolite
concentration acts as an input for other control elements (Andrianantoandro et al., 2006). An
entire metabolic pathway from S. cerevisiae, the mevalonate isoprenoid pathway for
synthesizing isopentyl pyrophosphate, was successfully transplanted into E. coli. In
combination with an inserted synthetic amorpha-4, 11-diene synthase, this pathway
produced large amounts of a precursor to the anti-malarial drug artemisinin. This new
producing strain is very useful since a significant decrease in the drug production time and
costs could be achieved (Martin et al., 2003). In addition to engineering pathways that
produce synthetic metabolites, artificial circuits can be engineered using metabolic
pathways connected to regulatory proteins and transcriptional control elements
(Andrianantoandro et al., 2006). One study describes such a circuit based on controlling
gene expression through acetate metabolism for cell–cell communication (Bulter et al., 2004).
Metabolic networks may embody more complex motifs, such as an oscillatory network. A
recently constructed metabolic network used glycolytic flux to generate oscillations through
the signaling metabolite acetyl phosphate (Fung et al., 2005). The system integrates
transcriptional regulation with metabolism to produce oscillations that are not correlated
with the cell division cycle. The general concerns of constructing transcriptional and protein
interaction-based modules, such as kinetic matching and optimization of reactions for a new
environment, apply for metabolic networks as well. In addition, the appropriate metabolic
precursors must be present. For this purpose, it may be necessary to include other enzymes
or metabolic pathways that synthesize precursors for the metabolite required in a synthetic
network (Leonard et al., 2008; McArthur IV & Fong, 2010).
Many polyketides and nonribosomal peptides are being used as antibiotic, anti-tumor and
immunosuppressant drugs (Neumann & Neumann-Staubitz, 2010). In order to produce
them in heterologous hosts, assembly of all the necessary genes that make up the synthetic
pathways is essential. The metabolic systems for the synthesis of polyketides are composed
of multiple modules, in which an individual module consists of either a polyketide synthase
or a nonribosomal peptide synthetase. Each module has a specific set of catalytic domains,
which ultimately determine the structure of the metabolic product and thus its function.
Recently, Bumpus et al. (2009) presented a proteomic strategy to identify new gene clusters
for the production of polyketides and nonribosomal peptides, and their biosynthetic
pathways, by adapting mass-spectrometry-based proteomics. This approach allowed
identification of genes that are used in the production of the target product in a species, for
which a complete genome sequence is not available. Such newly identified pathways can
then be copied into a new host strain that is more suitable for producing polyketides and
nonribosomal peptides at an industrial scale. This exemplifies that the sources of new
pathways are not limited to species with fully sequenced genomes.
178 Computational Biology and Applied Bioinformatics

The use of synthetic biology approaches in the field of metabolic engineering opens
enormous possibilities, especially toward the production of new drugs for cancer treatment.
Our goal is to design and model a new biosynthetic pathway for the production of natural
drugs in E. coli. Key to this is the specification of gene sequences encoding enzymes that
catalyze each reaction in the pathway, and whose DNA sequences can be incorporated into
devices that lead to functional expression of the molecules of interest (Prather & Martin,
2008). Partial pathways can be recruited from independent sources and co-localized in a
single host (Kobayashi et al., 2004). Alternatively, pathways can be constructed for the
production of new, non-natural products by engineering existing routes (Martin et al., 2003).

6. Conclusion
Despite all the scientific advances that humankind has seen over the last centuries, there are
still no clear and defined solutions to diagnose and treat cancer. In this sense, the search for
innovative and efficient solutions continues to drive research and investment in this field.
Synthetic biology uses engineering principles to create, in a rational and systematic way,
functional systems based on the molecular machines and regulatory circuits of living
organisms, or to re-design and fabricate existing biological systems. Bioinformatics and
newly developed computational tools play a key role in the improvement of such systems.
Elucidation of disease mechanisms, identification of potential targets and biomarkers,
design of biological elements for recognition and targeting of cancer cells, discovery of new
chemotherapeutics or design of novel drugs and catalysts, are some of the promises of
synthetic biology. Recent achievements are thrilling and promising; yet some of such
innovative solutions are still far from a real application due to technical challenges and
ethical issues. Nevertheless, many scientific efforts are being conducted to overcome these
limitations, and undoubtedly it is expected that synthetic biology together with
sophisticated computational tools, will pave the way to revolutionize the cancer field.

7. References
Alon, U. (2003). Biological networks: the tinkerer as an engineer. Science, Vol.301, No.5641,
(September 2003), pp. 1866-1867, ISSN 0036-8075
Anderson, J.C.; Clarke, E.J.; Arkin, A.P. & Voigt, C.A. (2006). Environmentally controlled
invasion of cancer cells by engineered bacteria. Journal of Molecular Biology, Vol.355,
No.4, (January 2006), pp. 619–627, ISSN 00222836
Andrianantoandro, E.; Basu, S.; Karig, D.K. & Weiss, R. (2006). Synthetic biology: new
engineering rules for an emerging discipline. Molecular Systems Biology Vol.2, No.
2006.0028, (May 2006), pp. 1-14, ISSN 1744-4292
Arkin, A.P. (2001). Synthetic cell biology. Current Opinion in Biotechnology Vol.12, No.6,
(December 2001), pp. 638-644, ISSN 0958-1669
Bansal, V.; Sakizlis, V.; Ross, R.; Perkins, J.D. & Pistikopoulos, E.N. (2003). New algorithms
for mixed-integer dynamic optimization. Computers and Chemical Engineering
Vol.27, No.5, (May 2003), pp. 647-668, ISSN 0098-1354
Bar, H.; Yacoby, I. & Benhar, I. (2008). Killing cancer cells by targeted drug-carrying phage
nanomedicines. BMC Biotechnology Vol.8, No.37, (April 2008), pp. 1-14, ISSN 1472-
6750
Synthetic Biology & Bioinformatics Prospects in the Cancer Arena 179

Batt, G., Yordanov, B., Weiss, R. & Belta, C. (2007). Robustness analysis and tuning of
synthetic gene networks. Bioinformatics Vol.23, No.18, (July 2007), pp. 2415-2422,
ISSN 1367-4803
Bedi, D.; Musacchio, T.; Fagbohun, O.A.; Gillespie, J.W.; Deinnocentes, P.; Bird, R.C.;
Bookbinder, L.; Torchilin, V.P. & Petrenko, V.A. (2011). Delivery of siRNA into
breast cancer cells via phage fusion protein-targeted liposomes. Nanomedicine:
Nanotechnology, Biology, and Medicine, doi:10.1016/j.nano.2010.10.004, ISSN 1549-
9634
Beisel, C.L. & Smolke, C.D. (2009). Design principles for riboswitch function. PLoS
Computational Biology Vol.5, No.4, (April 2009), e1000363, pp. 1-14, ISSN 1553-734X
Benner, S.A. & Sismour, A.M. (2005). Synthetic biology. Nature Reviews Genetics Vol.6, No.7,
(July 2005), pp. 533–543, ISSN 1471-0056
Brown, K.C. (2010). Peptidic tumor targeting agents: the road from phage display peptide
selections to clinical applications. Current Pharmaceutical Design Vol.16, No.9,
(March 2010), pp. 1040-1054, ISSN 1381-6128
Brustad, E.M. & Arnold, F.H. (2010). Optimizing non-natural protein function with directed
evolution. Current Opinion in Chemical Biology Vol.15, No.2, (April 2010), pp. 1-10,
ISSN 1367-5931
Bulter, T.; Lee, S.G.; Wong, W.W.; Fung, E.; Connor, M.R. & Liao, J.C. (2004). Design of
artificial cell–cell communication using gene and metabolic networks. PNAS
Vol.101, No.8 (February 2004), pp. 2299–2304, ISSN 0027-8424
Bumpus, S.; Evans, B.; Thomas, P.; Ntai, I. & Kelleher, N. (2009). A proteomics approach to
discovering natural products and their biosynthetic pathways. Nature Biotechnology
Vol.27, No.10, (September 2009), pp. 951-956, ISSN 1087-0156
Canton, B.; Labno, A. & Endy, D. (2008). Refinement and standardization of synthetic
biological parts and devices. Nature Biotechnology Vol.26, No. 7, (July 2008), pp. 787-
793, ISSN 1087-0156
Carothers, J.M.; Goler, J.A. & Keasling, J.D. (2009). Chemical synthesis using synthetic
biology. Current Opinion in Biotechnology Vol.20, No.4, (August 2009), pp. 498-503,
ISSN 0958-1669
Chan, L.Y.; Kosuri, S. & Endy, D. (2005). Refactoring bacteriophage T7. Molecular Systems
Biology Vol.1, No. 2005.0018, (September 2005), pp. 1-10, ISSN 1744-4292
Cooling, M.T.; Hunter, P. & Crampin, E.J. (2008). Modelling biological modularity with
CellML. IET Systems Biology Vol.2, No.2, (March 2008), pp. 73-79, ISSN 1751-8849
Crameri, A.; Raillard, S.A.; Bermudez, E. & Stemmer, W.P. (1998). DNA shuffling of a family
of genes from diverse species accelerates directed evolution. Nature Vol.391,
No.6664, (January 1998), pp. 288-291, ISSN 0028-0836
Czar, M.J.; Cai, Y. & Peccoud, J. (2009). Writing DNA with GenoCAD. Nucleic Acids Research
Vol.37, No.2, (May 2009), pp. W40-W47, ISSN 0305-1048
Dasika, M.S. & Maranas, C.D. (2008). OptCircuit: an optimization based method for
computational design of genetic circuits. BMC Systems Biology Vol.2, No.24, pp. 1-
19, ISSN 1752-0509
Datsenko, K.A. & Wanner, B.L. (2000). One-step inactivation of chromosomal genes in
Escherichia coli K-12 using PCR products. PNAS Vol.97, No. 12, (June 2000), pp.
6640-6645, ISSN 0027-8424
180 Computational Biology and Applied Bioinformatics

Dawid, A.; Cayrol, B. & Isambert, H. (2009). RNA synthetic biology inspired from bacteria:
construction of transcription attenuators under antisense regulation. Physical
Biology Vol.6, No. 025007, (July 2009), pp. 1-10, ISSN 1478-3975
Di Bernardo, D.; Thompson, M.J.; Gardner, T.S.; Chobot, S.E.; Eastwood, E.L.; Wojtovich,
A.P.; Elliott, S.E.; Schaus, S.E. & Collins, J.J. (2005). Chemogenomic profiling on a
genome-wide scale using reverse-engineered gene networks. Nature Biotechnology
Vol.23, No.3, (March 2005), pp. 377-383, ISSN 1087-0156
Di Ventura, B.; Lemerle, C.; Michalodimitrakis, K. & Serrano, L. (2006). From in vivo to in
silico biology and back. Nature Vol.443, No.7111, (October 2006), pp. 527-533, ISSN
0028-0836
Dougherty, M.J. & Arnold, F.H. (2009). Directed evolution: new parts and optimized
function. Current Opinion in Biotechnology Vol.20, No.4, (August 2009), pp. 486-491,
ISSN 0958-1669
Du B, Han H, Wang Z, Kuang L, Wang L, Yu L, Wu M, Zhou Z, Qian M. (2010). Targeted
drug delivery to hepatocarcinoma in vivo by phage-displayed specific binding
peptide. Molecular Cancer Research Vol.8, No.2, (February 2010), pp.135-144, ISSN
1541-7786
Ellis, T.; Wang, X. & Collins, J.J. (2009). Diversity-based, model guided construction of
synthetic gene networks with predicted functions. Nature Biotechnology Vol.27,
No.5, (May 2009), pp. 465– 471, ISSN 1087-0156
Elowitz, M.B. & Leibler, S. (2000). A synthetic oscillatory network of transcriptional
regulators. Nature Vol.403, No.6767, (January 2000), pp. 335-338, ISSN 0028-0836
Elowitz, M.B.; Levine, A.J.; Siggia, E.D. & Swain, P.S. (2002). Stochastic gene expression in a
single cell. Science Vol.297, No.5584, (August 2002), pp. 1183–1186, ISSN 0036-8075
Endy, D. (2005). Foundations for engineering biology. Nature Vol.438, No. 7067, (November
2005), pp. 449-453, ISSN 0028-0836
Engler, C.; Gruetzner, R.; Kandzia, R. & Marillonnet, S. (2009). Golden gate shuffling: a one
pot DNA shuffling method based on type ils restriction enzymes. PLoS ONE Vol.4,
No.5, (May 2009), e5553, pp. 1-9, ISSN 1932-6203
Eriksson, F.; Tsagozis, P.; Lundberg, K.; Parsa, R.; Mangsbo, S.M.; Persson, M.A.; Harris,
R.A. & Pisa, P.J. (2009). Tumor-specific bacteriophages induce tumor destruction
through activation of tumor-associated macrophages. The Journal of Immunology
Vol.182, No.5, (March 2009), pp. 3105-3111, ISSN 0022-1767
Ferreira, C.S.; Matthews, C.S. & Missailidis, S. (2006). DNA aptamers that bind to MUC1
tumour marker: design and characterization of MUC1-binding single-stranded
DNA aptamers. Tumour Biology Vol.27, No.6, (October 2006), pp. 289-301, ISSN
1010-4283
Fisher, J. & Henzinger, T.A. (2007). Executable cell biology. Nature Biotechnology Vol.25,
No.11, (November 2007), pp. 1239–1249, ISSN 1087-0156
Forbes, N.S. (2010). Engineering the perfect (bacterial) cancer therapy. Nature Reviews Cancer
Vol.10, No.11, (October 2010), pp. 785-794, ISSN 1474-175X
Francois, P. & Hakim, V. (2004). Design of genetic networks with specified functions by
evolution in silico. PNAS Vol.101, No.2, (January 2004), pp. 580–585, ISSN 0027-
8424
Synthetic Biology & Bioinformatics Prospects in the Cancer Arena 181

Funahashi, A.; Morohashi, M. & Kitano, H. (2003). CellDesigner: a process diagram editor
for gene-regulatory and biochemical networks. BIOSILICO Vol.1, No.5, (November
2003),pp. 159–162, ISSN 1478-5382
Fung, E.; Wong, W.W.; Suen, J.K.; Bulter, T.; Lee, S.G. & Liao, J.C. (2005). A synthetic gene
metabolic oscillator. Nature Vol.435, No.7038, (May 2005), pp. 118–122, ISSN 0028-
0836
Gao, C.; Mao, S.; Ronca, F.; Zhuang, S.; Quaranta, V.; Wirsching, P. & Janda, K.D. (2003). De
novo identification of tumor-specific internalizing human antibody-receptor pairs
by phage-display methods. Journal of Immunological Methods. Vol.274, No.1-2,
(March 2003), pp. 185-197, ISSN 0022-1759
Gardner, T.S.; Cantor, C.R. & Collins, J.J. (2000). Construction of a genetic toggle switch in
Escherichia coli. Nature Vol.403, No.6767, (January 2000), pp. 339-342, ISSN 0028-
0836
Garst, A.D. & Batey, R.T. (2009). A switch in time: detailing the life of a riboswitch.
Biochimica Biophysica Acta Vol.1789, No.9-10, (September-October 2009), pp. 584-
591, ISSN 0006-3002
Goler, J.A. (2004). A design and simulation tool for synthetic biological systems. Cambridge,
MA: MIT
Golynskiy, M.V. & Seelig, B. (2010). De novo enzymes: from computational design to mRNA
display. Trends in Biotechnology Vol.28, No.7, (July 2010), pp. 340-345, ISSN 0167-
7799
Grunberg, R.; Nilges, M. & Leckner, J. (2007). Biskit —a software platform for structural
bioinformatics. Bioinformatics Vol.23, No.6, (March 2007), pp. 769-770, ISSN 1367-
4803
Ham, T.S.; Lee, S.K.; Keasling, J.D. & Arkin, A.P. (2006). A tightly regulated inducible
expression system utilizing the fim inversion recombination switch. Biotechnology
and Bioengineering Vol.94, No.1, (May 2006), pp. 1–4, ISSN 0006-3592
Hartley, J.L. (2003). Use of the gateway system for protein expression in multiple hosts.
Current Protocols in Protein Science Chap5: Unit 5.17
Hawkins, T.; Luban, S. & Kihara, D. (2006). Enhanced automated function prediction using
distantly related sequences and contextual association by PFP. Protein Science
Vol.15, No.6, (June 2006), pp. 1550-1556, ISSN 1469-896X
Hianik, T. & Wang, J. (2009). Electrochemical aptasensors – recent achievements and
perspectives. Electroanalysis Vol.21, No.11, (June 2009), pp. 1223-1235, ISSN 1521-
4109
Hill, A.D.; Tomshine, J.R.; Weeding, E.M.; Sotiropoulos, V. & Kaznessis, Y.N. (2008).
SynBioSS: the synthetic biology modeling suite. Bioinformatics Vol.24, No.21,
(November 2008),pp. 2551-2553, ISSN 1367-4803
Hofacker, I.L. (2003). Vienna RNA secondary structure server. Nucleic Acids Research Vol.31,
No.13, (July 2003), pp. 3429-3431, ISSN 0305-1048
Isaacs, F.J.; Dwyer, D.J. & Collins, J.J. (2006). RNA synthetic biology. Nature Biotechnology
Vol.24, No.5, (May 2006), pp. 545-554, ISSN 1087-0156
Isalan, M.; Lemerle, C.; Michalodimitrakis, K.; Horn, C.; Beltrao, P.; Raineri, E.; Garriga-
Canut, M.; Serrano, L. (2008). Evolvability and hierarchy in rewired bacterial gene
networks. Nature Vol.452, No.7189, (April 2008), pp. 840–845, ISSN 0028-0836
182 Computational Biology and Applied Bioinformatics

Jemal, A.; Bray, F.; Center, M.M.; Ferlay, J.; Ward, E. & Forman, D. (2011). Global cancer
statistics. CA Cancer Journal for Clinicians Vol.61, No.2, (March-April 2011),pp. 69-
90, ISSN 0007-9235
Kaiser, J. (2005). Gene therapy. Putting the fingers on gene repair. Science Vol.310, No.5756,
(December 2005), pp.1894-1896, ISSN 0036-8075
Kaznessis, Y.N. (2009). Computational methods in synthetic biology. Biotechnology Journal
Vol.4, No.10, (October 2009), pp.1392-1405, ISSN 1860-7314
Kelly, J.; Rubin, A.J.; Davis, J. II; Ajo-Franklin, C.M.; Cumbers, J.; Czar, M.J.; de Mora, K.;
Glieberman, A.I.; Monie, D.D. & Endy, D. (2009). Measuring the activity of BioBrick
promoters using an in vivo reference standard. Journal of Biological Engineering
Vol.3, No.4, (March 2009), pp. 1-13, ISSN 1754-1611
Kirkpatrick, S.; Gelatt, C.D. Jr. & Vecchi, M.P. (1983). Optimization by Simulated Annealing.
Science Vol.220, No.4598, (May 1983), pp. 671-680, ISSN 0036-8075
Kitano, H. (2002). Systems biology: a brief overview. Science Vol.295, No.5560, (March 2002),
pp. 1662-1664, ISSN 0036-8075
Kizer, L.; Pitera, D.J.; Pfleger, B.F. & Keasling, J.D. (2008). Application of functional
genomics to pathway optimization for increased isoprenoid production. Applied
and Environmental Microbiology Vol.74, No.10, (May 2008), pp. 3229–3241, ISSN
0099-2240
Kobayashi, H.; Kaern, M.; Araki, M.; Chung, K.; Gardner, T.S.; Cantor, C.R. & Collins, J.J.
(2004). Programmable cells: interfacing natural and engineered gene networks.
PNAS Vol.101, No.22, (June 2004), pp. 8414–8419, ISSN 0027-8424
Koide, T.; Pang, W.L. & Baliga, N.S. (2009). The role of predictive modeling in rationally
reengineering biological systems. Nature Reviews Microbiology Vol.7, No.4, (April
2009), pp. 297-305, ISSN 1740-1526
Kosuri, S.; Kelly, J.R. & Endy, D. (2007). TABASCO: a single molecule, base pair resolved
gene expression simulator. BMC Bioinformatics Vol.8, No.480, (December 2007), pp.
1-15, ISSN 1471-2105
Lartigue, C.; Glass, J.I.; Alperovich, N.; Pieper, R.; Parmar, P.P.; Hutchison III, C.A.; Smith,
H.O. & Venter, J.C. (2007). Genome transplantation in bacteria: changing one
species to another. Science Vol.317, No.5838, (August 2007), pp. 632-638, ISSN 0036-
8075
Leonard, E.; Nielsen, D.; Solomon, K. & Prather, K.J. (2008). Engineering microbes with
synthetic biology frameworks. Trends in Biotechnology Vol.26, No.12, (December
2008), pp. 674-681, ISSN 0167-7799
Li K, Chen Y, Li S, Nguyen HG, Niu Z, You S, Mello CM, Lu X, Wang Q. (2010). Chemical
modification of M13 bacteriophage and its application in cancer cell imaging.
Bioconjugate Chemistry Vol.21, No.7, (December 2010), pp. 1369-1377, ISSN 1043-
1802
Loettgers, A. (2007). Model organisms and mathematical and synthetic models to explore
regulation mechanisms. Biological Theory Vol.2, No.2, (December 2007), pp. 134-142,
ISSN 1555-5542
Mandell, J.G. & Barbas, C.F. III (2006). Zinc Finger Tools: custom DNA binding domains for
transcription factors and nucleases. Nucleic Acids Research Vol.34, (July 2006), pp.
W516-523, ISSN 0305-1048
Synthetic Biology & Bioinformatics Prospects in the Cancer Arena 183

Marchisio, M.A. & Stelling, J. (2008). Computational design of synthetic gene circuits with
composable parts. Bioinformatics Vol.24, No.17, (September 2008), pp.1903–1910,
ISSN 1367-4803
Marchisio, M.A. & Stelling, J. (2009). Computational design tools for synthetic biology.
Current Opinion in Biotechnology Vol.20, No.4, (August 2009), pp. 479-485, ISSN
0958-1669
Marguet, P.; Balagadde, F.; Tan, C. & You, L. (2007). Biology by design: reduction and
synthesis of cellular components and behavior. Journal of the Royal Society Interface
Vol.4, No.15, (August 2007), pp. 607-623, ISSN 1742-5689
Markham, N.R. & Zuker, M. (2008). UNAFold: software for nucleic acid folding and
hybridization. Methods Molecular Biology Vol.453, No.I, pp. 3-31, ISSN 1064-3745
Martin, V.J.; Pitera, D.J.; Withers, S.T.; Newman, J.D. & Keasling, J.D. (2003). Engineering a
mevalonate pathway in Escherichia coli for production of terpenoids. Nature
Biotechnology Vol.21, No.7, (July 2003), pp.796–802, ISSN 1087-0156
Matsuoka, Y.; Ghosh, S. & Kitano, H. (2009). Consistent design schematics for
biological systems: standardization of representation in biological engineering.
Journal of the Royal Society Interface Vol.6, No.4, (August 2009), pp. S393-S404, ISSN
1742-5689
McArthur IV, G.H. & Fong, S.S. (2010). Toward engineering synthetic microbial metabolism.
Journal of Biomedicine and Biotechnology doi:10.1155/2010/459760, ISSN 1110-7243
Mirschel, S.; Steinmetz, K.; Rempel, M.; Ginkel, M. & Gilles, E.D. (2009). PROMOT: modular
modeling for systems biology. Bioinformatics Vol.25, No.5, (March 2009), pp. 687-
689, ISSN 1367-4803
Morris, G.M.; Huey, R.; Lindstrom, W.; Sanner, M.F.; Belew, R.K.; Goodsell, D.S. & Olson,
A.J. (2009). AutoDock4 and AutoDockTools4: automated docking with selective
receptor flexibility. Journal of Computational Chemistry Vol.30, No.16, (December
2009), pp. 2785-2791, ISSN 0192-8651
Mueller, S.; Coleman, J.R. & Wimmer, E. (2009). Putting synthesis into biology: a viral view
of genetic engineering through de novo gene and genome synthesis. Chemistry &
Biology Vol.16, No.3, (March 2009), pp. 337-347, ISSN 1074-5521
Neumann, H. & Neumann-Staubitz, P. (2010). Synthetic biology approaches in drug
discovery and pharmaceutical biotechnology. Applied Microbiology and Biotechnology
Vol.87, No.1, (June 2010), pp. 75-86, ISSN 0175-7598
Nielsen, J. (2001). Metabolic engineering. Applied Microbiology and Biotechnology Vol.55, No.3,
(April 2001), pp. 263-283, ISSN 0175-7598
Park, J.H.; Lee, S.Y.; Kim, T.Y. & Kim, H.U. (2008). Application of systems biology for
bioprocess development. Trends in Biotechnology Vol.26, No.8, (August 2008), pp.
404–412, ISSN 0167-7799
Pawelek, J.; Low, K. & Bermudes, D. (2003). Bacteria as tumour-targeting vectors. Lancet
Oncology Vol.4, No.9, (September 2003), pp. 548–556, ISSN 1470-2045
Pedersen, M. & Phillips, A. (2009). Towards programming languages for genetic engineering
of living cells. Journal of the Royal Society Interface Vol.6, No.4, (August 2009), pp.
S437-S450, ISSN 1742-5689
184 Computational Biology and Applied Bioinformatics

Prather, K. & Martin, C.H. (2008). De novo biosynthetic pathways: rational design of
microbial chemical factories. Current Opinion in Biotechnology Vol.19, No.5, (October
2008), pp. 468–474, ISSN 0958-1669
Price, N.D. & Shmulevich, I. (2007). Biochemical and statistical network models for systems
biology. Current Opinion in Biotechnology Vol.18, No.4, (August 2007), pp. 365–370,
ISSN 0958-1669
Purnick, P.E. & Weiss, R. (2009). The second wave of synthetic biology: from modules to
systems. Nature Reviews Molecular Cell Biology Vol.10, No.6, (June 2009), pp. 410-422,
ISSN 1471-0072
Quan, J. & Tian, J. (2009). Circular polymerase extension cloning of complex 1 gene libraries
and pathways. PLoS ONE Vol.4, No.7, (July 2009), e6441, pp.1-6, ISSN 1932-6203
Richardson, S.M.; Wheelan, S.J.; Yarrington, R.M. & Boeke, J.D. (2006). GeneDesign: rapid,
automated design of multikilobase synthetic genes. Genome Research Vol.16, No.4,
(April 2006), pp. 550-556, ISSN 1088-9051
Ritchie, D.W. (2008). Recent progress and future directions in protein–protein
docking.Current Protein & Peptide Science Vol.9, No.1, (February 2008), pp. 1-15,
ISSN 1389-2037
Ro, D.K.; Paradise, E.M.; Ouellet, M.; Fisher, K.J.; Newman, K.L.; Ndungu, J.M.; Ho, K.A.;
Eachus, R.A.; Ham, T.S.; Kirby, J.; Chang, M.C.; Withers, S.T.; Shiba, Y.; Sarpong, R.
& Keasling, J.D. (2006). Production of the antimalarial drug precursor artemisinic
acid in engineered yeast. Nature Vol.440, No.7086, (April 2006), pp. 940–943, ISSN
0028-0836
Rocha, I.; Forster, J. & Nielsen, J. (2008). Design and application of genome-scale
reconstructed metabolic models. Methods in Molecular Biology Vol.416, No.IIIB, pp.
409-431, ISSN 1064-3745
Rocha, I.; Maia, P.; Evangelista, P.; Vilaça, P.; Soares, S.; Pinto, J.P.; Nielsen, J.; Patil, K.R.;
Ferreira, E.C. & Rocha, M. (2010). OptFlux: an open-source software platform for in
silico metabolic engineering. BMC Systems Biology Vol.4, No. 45, pp. 1-12, ISSN
1752-0509
Rodrigo, G.; Carrera, J. & Jaramillo, A. (2007a). Asmparts: assembly of biological model
parts. Systems and Synthetic Biology Vol.1, No.4, (December 2007), pp. 167-170, ISSN
1872-5325
Rodrigo, G.; Carrera, J. & Jaramillo, A. (2007b). Genetdes: automatic design of
transcriptional networks. Bioinformatics Vol.23, No.14, (July 2007), pp.1857-1858,
ISSN 1367-4803
Rodrigues, L.R.; Teixeira, J.A.; Schmitt, F.; Paulsson, M. & Lindmark Måsson, H. (2007). The
role of osteopontin in tumour progression and metastasis in breast cancer. Cancer
Epidemology Biomarkers & Prevention Vol.16, No.6, (June 2007), pp. 1087–1097, ISSN
1055-9965
Saito, H. & Inoue, T. (2009). Synthetic biology with RNA motifs. The International Journal of
Biochemistry & Cell Biology Vol.41, No.2, (February 2009), pp. 398-404, ISSN 1357-
2725
Salis, H.M., Mirsky, E.A. & Voigt, C.A. (2009). Automated design of synthetic ribosome
binding sites to control protein expression. Nature Biotechnology Vol.27, No.10,
(October 2009), pp. 946–950, ISSN 1087-0156
Synthetic Biology & Bioinformatics Prospects in the Cancer Arena 185

Segre, D.; Vitkup, D. & Church, G.M. (2002). Analysis of optimality in natural and perturbed
metabolic networks. PNAS Vol.99, No.23, (November 2001), pp. 15112-15117, ISSN
0027-8424
Shetty, R.P.; Endy, D. & Knight, T.F. Jr. (2008). Engineering BioBrick vectors from BioBrick
parts. Journal of Biological Engineering Vol.2, No.1, (April 2008), pp.5-17 ISSN 1754-
1611
Simons, K.T.; Bonneau, R.; Ruczinski, I. & Baker, D. (1999). Ab initio protein structure
prediction of CASP III targets using ROSETTA. Proteins Vol.3, pp. 171-176, ISSN
0887-3585
Steinmetz, N.F. (2010).Viral nanoparticles as platforms for next-generation therapeutics and
imaging devices. Nanomedicine: Nanotechnology, Biology, and Medicine Vol.6, No.5,
(October 2010), pp. 634-641, ISSN 1549-9634
Stoltenburg, R.; Reinemann, C. & Strehlitz, B. (2007). SELEX—A (r)evolutionary method to
generate high-affinity nucleic acid ligands. Biomolecular Engineering Vol.24, No.4,
(October 2007), pp. 381–403, ISSN 1389-0344
Topp, S. & Gallivan, J.P. (2007). Guiding bacteria with small molecules and RNA. Journal of
the American Chemical Society Vol.129, No.21, (May 2007), pp. 6807-6811, ISSN 0002-
7863
Tyo, K.E.; Alper, H.S. & Stephanopoulos, G. (2007). Expanding the metabolic engineering
toolbox: more options to engineer cells. Trends in Biotechnology Vol.25, No.3, (March
2007), pp. 132-137, ISSN 0167-7799
Tyo, K.E.J.; Ajikumar, P.K. & Stephanopoulos, G. (2009). Stabilized gene duplication enables
long-term selection-free heterologous pathway expression. Nature Biotechnology
Vol.27, No.8, (August 2009), pp. 760–765, ISSN 1087-0156
Villalobos, A.; Ness, J.E.; Gustafsson, C.; Minshull, J. & Govindarajan, S. (2006). Gene
Designer: a synthetic biology tool for constructing artificial DNA segments. BMC
Bioinformatics Vol.7, No.285, (June 2006), pp.1-8, ISSN 1471-2105
Voloshchuk, N. & Montclare, J.K. (2010). Incorporation of unnatural amino acids for
synthetic biology. Molecular Biosystems Vol.6, No.1, (January 2010), pp. 65-80, ISSN
1742-2051
Wang, H.H.; Isaacs, F.J.; Carr, P.A.; Sun, Z.Z.; Xu, G.; Forest, C.R. & Church, G.M. (2009).
Programming cells by multiplex genome engineering and accelerated evolution.
Nature Vol.460, No.7257, (August 2009), pp. 894–898, ISSN 0028-0836
Wierling, C.; Herwig, R. & Lehrach, H. (2007). Resources, standards and tools for systems
biology. Briefings in Functional Genomics and Proteomics Vol.6, No.3, (September
2007), pp. 240-251, ISSN 1473-9550
Win, M.N. & Smolke, C.D. (2007). From the cover: a modular and extensible RNA-based
gene-regulatory platform for engineering cellular function. PNAS Vol.104, No.36,
(September 2007), pp. 14283–14288, ISSN 0027-8424
Winkler, W.C. & Breaker, R.R. (2005). Regulation of bacterial gene expression by
riboswitches. Annual Review of Microbiology Vol.59, pp. 487-517, ISSN 0066-4227
Xu, J.; Li, M.; Kim, D. & Xu, Y. (2003). RAPTOR: optimal protein threading by linear
programming. Journal of Bioinformatics and Computational Biology Vol.1, No.1, (April
2003), pp. 95-117, ISSN 0219-7200
186 Computational Biology and Applied Bioinformatics

Yokobayashi, Y.; Weiss, R. & Arnold, F.H. (2002). Directed evolution of a genetic circuit.
PNAS Vol.99, No.26, (September 2002), pp. 16587–16591, ISSN 0027-8424
You, L. (2004). Toward computational systems biology. Cell Biochemistry and Biophysics
Vol.40, No.2, pp. 167–184, ISSN 1085-9195
 9

An Overview of Hardware-Based Acceleration

of Biological Sequence Alignment
Laiq Hasan and Zaid Al-Ars
TU Delft
The Netherlands

1. Introduction
Efficient biological sequence (proteins or DNA) alignment is an important and challenging
task in bioinformatics. It is similar to string matching in the context of biological data and
is used to infer the evolutionary relationship between a set of protein or DNA sequences.
An accurate alignment can provide valuable information for experimentation on the newly
found sequences. It is indispensable in basic research as well as in practical applications such
as pharmaceutical development, drug discovery, disease prevention and criminal forensics.
Many algorithms and methods, such as, dot plot (Gibbs & McIntyre, 1970), Needleman-Wunsch
(N-W) (Needleman & Wunsch, 1970), Smith-Waterman (S-W) (Smith & Waterman, 1981),
FASTA (Pearson & Lipman, 1985), BLAST (Altschul et al., 1990), HMMER (Eddy, 1998) and
ClustalW (Thompson et al., 1994) have been proposed to perform and accelerate sequence
alignment activities. An overview of these methods is given in (Hasan et al., 2007). Out
of these, S-W algorithm is an optimal sequence alignment method, but its computational
cost makes it inappropriate for practical purposes. To develop efficient and optimal
sequence alignment solutions, the S-W algorithm has recently been implemented on emerging
accelerator platforms such as Field Programmable Gate Arrays (FPGAs), Cell Broadband Engine
(Cell/B.E.) and Graphics Processing Units (GPUs) (Buyukkur & Najjar, 2008; Hasan et al., 2010;
Liu et al., 2009; 2010; Lu et al., 2008). This chapter aims at providing a broad overview of
sequence alignment in general with particular emphasis on the classification and discussion
of available methods and their comparison. Further, it reviews in detail the acceleration
approaches based on implementations on different platforms and provides a comparison
considering different parameters. This chapter is organized as follows:
The remainder of this section gives a classification, discussion and comparison of the available
methods and their hardware acceleration. Section 2 introduces the S-W algorithm which is
the focus of discussion in the succeeding sections. Section 3 reviews CPU-based acceleration.
Section 4 provides a review of FPGA-based acceleration. Section 5 overviews GPU-based
acceleration. Section 6 presents a comparison of accelerations on different platforms, whereas
Section 7 concludes the chapter.

1.1 Classification
Sequence alignment aims at identifying regions of similarity between two DNA or protein
sequences (the query sequence and the subject or database sequence). Traditionally, the
methods of pairwise sequence alignment are classified as either global or local, where pairwise
means considering only two sequences at a time. Global methods attempt to match as many
188
2 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

characters as possible, from end to end, whereas local methods aim at identifying short
stretches of similarity between two sequences. However, in some cases, it might also be
needed to investigate the similarities between a group of sequences, hence multiple sequence
alignment methods are introduced. Multiple sequence alignment is an extension of pairwise
alignment to incorporate more than two sequences at a time. Such methods try to align all of
the sequences in a given query set simultaneously. Figure 1 gives a classification of various
available sequence alignment methods.

Sequence Alignment
Methods

Global Local Multiple

N-W S-W
Dot plot FASTA BLAST HMMER ClustalW
algorithm algorithm

Exact methods Approximate methods

Fig. 1. Various methods for sequence alignment
These methods are categorized into three types, i.e. global, local and multiple, as shown in the
figure. Further, the figure also identifies the exact methods and approximate methods. The
methods shown in Figure 1 are discussed briefly in the following subsection.

1.2 Discussion of available methods

Following is a brief description of the available methods for sequence alignment.

Global methods
Global methods aim at matching as many characters as possible, from end to end between
two sequences i.e. the query sequence (Q) and the database sequence (D). Methods carrying out
global alignment include dot plot and N-W algorithm. Both are categorized as exact methods.
The difference is that dot plot is based on a basic search method, whereas N-W on dynamic
programming (DP) (Giegerich, 2000).

Local methods
In contrast to global methods, local methods attempt to identify short stretches of similarity
between two sequences i.e. Q and D. These include exact method like S-W and heuristics
based approximate methods like FASTA and BLAST.

Multiple alignment methods

It might be of interest in some cases to consider the similarities between a group of sequences.
Multiple sequence alignment methods like HMMER and ClustalW are introduced to handle
such cases.
An Overview
An Overview of Hardware-Based
of Hardware-based Acceleration
Acceleration of Biological of Biological Sequence Alignment
Sequence Alignment 1893

1.3 Comparison
The alignment methods can be compared on the basis of their temporal and spatial
complexities and parameters like alignment type and search procedure. A summary of the
comparison is shown in Table 1. It is interesting to note that all the global and local sequence
alignment methods essentially have the same computational complexity of O( L Q L D ), where
L Q and L D are the lengths of the query and database sequences, respectively. Yet despite this,
each of the algorithms has very different running times, with BLAST being the fastest and
dynamic programming algorithms being the slowest. In case of multiple sequence alignment
methods, ClustalW has the worst time complexity of O( L2Q L2D ), whereas HMMER has a
time complexity of O( L Q L2D ). The space complexities of all the alignment methods are also
essentially identical, around O( L Q L D ) space, except BLAST, the space complexity of which
is O(20w + L Q L D ). In the exact methods, dot plot uses a basic search method, whereas N-W
and S-W use DP. On the other hand, all the approximate methods are heuristic based. It
is also worthy to note that FASTA and BLAST have to make sacrifices on sensitivity to be
able to achieve higher speeds. Thus, a trade off exists between speed and sensitivity and we
must come to a compromise to be able to efficiently align sequences in a biologically relevant
manner in a reasonable amount of time.
Time Space
Method Type Accuracy Search
complexity complexity
Dot plot Global Exact Basic O( L Q L D ) O( L Q L D )
N-W Global Exact DP O( L Q L D ) O( L Q L D )
S-W Local Exact DP O( L Q L D ) O( L Q L D )
FASTA Local Approximate Heuristic O( L Q L D ) O( L Q L D )
BLAST Local Approximate Heuristic O( L Q L D ) O(20w + L Q L D )
HMMER Multiple Approximate Heuristic O( L Q L2D ) O( L Q L D )
ClustalW Multiple Approximate Heuristic O( L2Q L2D ) O( L Q L D )

Table 1. Comparison of various sequence alignment methods

1.4 Hardware platforms

Work has been done on accelerating sequence alignment methods, by implementing them on
various available hardware platforms. Following is a brief discussion about such platforms.

CPUs
CPUs are well known, flexible and scalable architectures. By exploiting the Streaming
SIMD Extension (SSE) instruction set on modern CPUs, the running time of the analyses is
decreased significantly, thereby making analyses of data intensive problems like sequence
alignment feasible. Also emerging CPU technologies like multi-core combines two or more
independent processors into a single package. The Single Instruction Multiple Data-stream
(SIMD) paradigm is heavily utilized in this class of processors, making it appropriate for data
parallel applications like sequence alignment. SIMD describes CPUs with multiple processing
elements that perform the same operation on multiple data simultaneously. Thus, such
machines exploit data level parallelism. The SSE instruction set extension in modern CPUs
contains 70 new SIMD instructions. This extension greatly increases the performance when
exactly the same operations are to be performed on multiple data objects, making sequence
alignment a typical application.
190
4 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

FPGAs
FPGAs are reconfigurable data processing devices on which an algorithm is directly mapped
to basic processing logic elements. To take advantage of using an FPGA, one has to implement
massively parallel algorithms on this reconfigurable device. They are thus well suited for
certain classes of bioinformatics applications, such as sequence alignment. Methods like the
ones based on systolic arrays are used to accelerate such applications.

GPUs
Initially stimulated by the need for real time graphics in video gaming, GPUs have evolved
into powerful and flexible vector processors, ideal for accelerating a variety of data parallel
applications. GPUs have in the last couple of years developed themselves from a fixed
function graphics processing unit into a flexible platform that can be used for high performance
computing (HPC). Applications like bioinformatics sequence alignment can run very efficiently
on these architectures.

2. Smith-Waterman algorithm
In 1981, Smith and Waterman described a method, commonly known as the Smith-Waterman
(S-W) algorithm (Smith & Waterman, 1981), for finding common regions of local similarity.
S-W method has been used as the basis for many subsequent algorithms, and is often quoted
as a benchmark when comparing different alignment techniques. When obtaining the local
S-W alignment, a matrix H is constructed using the following equation.
⎧
⎪
⎪ 0
⎨H
i −1,j −1 + Si,j
Hi,j = max (1)
⎪
⎪ H −d
⎩ i−1,j
Hi,j−1 − d

Where Si,j is the similarity score and d is the penalty for a mismatch. The algorithm can be
implemented using the following pseudo code.

Initialization:

H(0,j) = 0
H(i,0) = 0

Matrix Fill:

for each i,j = 1 to M,N

{
H(i,j) = max(0,
H(i-1,j-1) + S(i,j),
H(i-1,j) - d,
H(i,j-1) - d)
}

Traceback:

H(opt) = max(H(i,j))
traceback(H(opt))
An Overview
An Overview of Hardware-Based
of Hardware-based Acceleration
Acceleration of Biological of Biological Sequence Alignment
Sequence Alignment 1915

The H matrix is constructed with one sequence lined up against the rows of a matrix, and
another against the columns, with the first row and column initialized with a predefined value
(usually zero) i.e. if the sequences are of length M and N respectively, then the matrix for the
alignment algorithm will have ( M + 1) × ( N + 1) dimensions. The matrix fill stage scores
each cell in the matrix. This score is based on whether the two intersecting elements of each
sequence are a match, and also on the score of the cell’s neighbors to the left, above, and
diagonally upper left. Three separate scores are calculated based on all three neighbors, and
the maximum of these three scores (or a zero if a negative value would result) is assigned
to the cell. This is done for each cell in the matrix resulting in O( MN ) complexity for the
matrix fill stage. Even though the computation for each cell usually only consists of additions,
subtractions, and comparisons of integers, the algorithm would nevertheless perform very
poorly if the lengths of the query sequences become large. The traceback step starts at the cell
with the highest score in the matrix and ends at a cell when the similarity score drops below a
certain predefined threshold. For doing this, the algorithm requires to find the maximum cell
which is done by traversing the entire matrix, making the time complexity for the traceback
O( MN ). It is also possible to keep track of the cell with the maximum score, during the
matrix filling segment of the algorithm, although this will not change the overall complexity.
Thus, the total time complexity of the S-W algorithm is O( MN ). The space complexity is also
O( MN ).
In order to reduce the O( MN ) complexity of the matrix fill stage, multiple entries of the
H matrix can be calculated in parallel. This is however complicated by data dependencies,
whereby each Hi,j entry depends on the values of three neighboring entries Hi,j−1, Hi−1,j and
Hi−1,j−1 , with each of those entries in turn depending on the values of three neighboring
entries, which effectively means that this dependency extends to every other entry in the
region Hx,y : x ≤ i, y ≤ j. This implies that it is possible to simultaneously compute all the
elements in each anti-diagonal, since they fall outside each other’s data dependency regions.
Figure 2 shows a sample H matrix for two sequences, with the bounding boxes indicating the
elements that can be computed in parallel. The bottom-right cell is highlighted to show that
its data dependency region is the entire remaining matrix. The dark diagonal arrow indicates
the direction in which the computation progresses. At least 9 cycles are required for this
computation, as there are 9 bounding boxes representing 9 anti-diagonals and a maximum of
5 cells may be computed in parallel.
The degree of parallelism is constrained to the number of elements in the anti-diagonal and
the maximum number of elements that can be computed in parallel are equal to the number
of elements in the longest anti-diagonal (l d ), where,

ld = min( M, N ) (2)
Theoretically, the lower bound to the number of steps required to calculate the entries of
the H matrix in a parallel implementation of the S-W algorithm is equal to the number of
anti-diagonals required to reach the bottom-right element, i.e. M + N − 1 (Liao et al., 2004).
Figure 3 shows the logic circuit to compute an element of the H matrix. The logic contains
three adders, a sequence comparator circuit (SeqCmp) and three max operators (MAX). The
sequence comparator compares the corresponding characters of two input sequences and
outputs a match/mismatch score, depending on whether the two characters are equal or not.
Each max operator finds the maximum of its two inputs. The time to compute an element is
4 cycles, assuming that the time for each cycle is equal to the latency of one add or compare
operation.
192
6 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

G A T T A

0 0 0 0 0 0

G 0 1 0 0 0 0

A 0 0 2 0 0 1

j C 0 0 0 1 0 0

T 0 0 0 1 2 0

C 0 0 0 0 0 1

Fig. 2. Sample H matrix, where the dotted rectangles show the elements that can be
computed in parallel

Hi,j

Cycle 4

MAX

Cycle 3

MAX

Cycle 2

MAX +
Cycle 1
Si,j

Seq
Cmp
+ +

Hi,j-1 d Hi-1,j Hi-1,j-1 Q D 0

Fig. 3. Logic circuit to compute cells in the H matrix, where + is an adder, MAX is a max
operator and SeqCmp is the sequence comparator that generates match/mismatch scores
An Overview
An Overview of Hardware-Based
of Hardware-based Acceleration
Acceleration of Biological of Biological Sequence Alignment
Sequence Alignment 1937

3. CPU-based acceleration
In this section CPU-based acceleration of the S-W algorithm is reviewed. Furthermore, an
estimation of the performance for top-end and future systems is made.

3.1 Recent implementations

The first CPU implementations used a sequential way of calculating all the matrix values.
These implementations were slow and therefore hardly used. In 2007, Farrar introduced
a SSE implementation for S-W (Farrar, 2007). His work used SSE2 instructions for an Intel
processor and was up to six times faster than existing S-W implementations. Two years later,
a Smith-Waterman implementation on Playstation 3 (SWPS3) was introduced (Szalkowski et al.,
2009), which was based on a minor adjustment to Farrar’s implementation. SWPS3 is a
vectorized implementation of the Smith-Waterman local alignment algorithm optimized for
both the IBM Cell/B.E. and Intel x86 architectures. A SWPS3 version optimized for multi
threading has been released recently (Aldinucci et al., 2010). The SSE implementations can
be viewed as being semi parallel, as they constantly calculate sixteen, eight or less values at
the same time, while discarding startup and finish time. Table 2 presents the performance
achieved by these implementations on various CPU platforms.
Peak Benchmark Peak performance
Implementation
performance hardware (per thread)
2.0 GHz, Xeon
3.75 GCUPS
(Farrar, 2007) 2.9 GCUPS
Core 2 Duo
single thread
2.4 GHz
4.08 GCUPS
(Szalkowski et al., 2009) 15.7 GCUPS Core 2 Quad
Q6600, 4 threads
2.5 GHz, 2x Xeon
4.38 GCUPS
(Aldinucci et al., 2010) 35 GCUPS Core Quad
E5420, 8 threads
Table 2. Performance achieved by various S-W CPU implementations (Vermij, 2011)

3.2 Performance estimations for top-end and future CPUs

With the data from Table 2, we make an estimate of the performance on the current top-end
CPUs and take a look into the future. Table 3 gives the estimated peak performances based
on the SIMD register width, the number of cores, clock speed and the known speed per
core. We assumed linear scaling in the number of cores as suggested in Table 2, and the
given performances may therefore not be reliable. Non-ideal inter-core communication,
memory bandwidth limitations and shared caches could lead to a lower peak performance.
Furthermore, no distinction in performance is made between Intel and AMD processors.
Hence, Table 3 must be used as an indication to where the S-W performance could go on
in current and future CPUs (Vermij, 2011).

4. FPGA-based acceleration
FPGAs are programmable logic devices. To map an application on flexible FPGA platforms,
a program is written in a hardware description language like VHDL. The flexibility, difficulty
194
8 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

SIMD Cores Clock Peak performance

System Released
register width (threads) speed (estimated)
Xeon 8 2.26 32
2010 128
Beckton (16) GHz GCUPS
Opteron 12 2.3 48
2010 128
Magny-Cours (12) GHz GCUPS
Opteron 16 2.3 64
2011 128
Interlagos (16) GHz GCUPS
Table 3. Estimated peak performance for current top-end and future CPUs (Vermij, 2011)

of design as well as the performance of FPGA implementations fall typically somewhere

between pure software running on a CPU and an Application Specific Integrated Circuit
(ASIC). FPGAs are widely used to accelerate applications like S-W based sequence alignment.
Implementations rely on the ability to create building blocks called processing elements (PEs)
that can update one matrix cell every clock cycle. Furthermore, multiple PEs can be linked
together in a two dimensional or linear systolic arrays to process huge data in parallel. This
section provides a brief description of traditional systolic arrays followed by a discussion of
existing and future FPGA-based S-W implementations.

4.1 Systolic arrays

Systolic array is an arrangement of processors in an array, where data flows synchronously
across the array between neighbors, usually with data flowing in a specific direction
(Kung & Leiserson, 1979), (Quinton & Robert, 1991). Each processor at each step takes in data
from one or more neighbors (e.g. North and West), processes it and, in the next step, outputs
results to the opposite neighbors (South and East). Systolic arrays can be implemented in
rectangular or 2-dimensional (2D) and linear or 1-dimensional (1D) fashion. Figure 4 gives a
pictorial view of both implementation types.
They best suit compute-intensive applications like biological sequence alignment. The
disadvantage is that being highly specialized processors type, they are difficult to implement
and build.
In (Pfeiffer et al., 2005), a concept to accelerate S-W algorithm on the basis of linear systolic
array is demonstrated. The reason for choosing this architecture is outlined by demonstrating
the efficiency and simplicity in combination with the algorithm. Nevertheless, there are two
key methodologies to speedup this massively parallel system. By turning the processing
from bit-parallel to bit-serial, the actual improvement is enabled. This change is performance
neutral, but in combination with the early maximum detection, a considerable speedup is
possible. Another effect of this improvement is a data dependant execution time of the
processing elements. Here, the second acceleration prevents idle times to exploit the hardware
and speeds up the computation. This can be accomplished by a globally asynchronous timing
representing a self-timed linear systolic array. The authors have provided no performance
estimation due to the initial stage of their work, that is why it cannot be compared with other
related work.
In (Vermij, 2011), the working of a linear systolic array (LSA) is explained. Such an array works
like the SSE unit in a modern CPU. But instead of having a fixed length of lets say 16, the
FPGA based array can have any length.
An Overview
An Overview of Hardware-Based
of Hardware-based Acceleration
Acceleration of Biological of Biological Sequence Alignment
Sequence Alignment 1959

N1 N2 N3 N4

M4M3M2M1
M1 U11 U12 U13 U14

U11

N1
M2 U21 U22 U23 U24

(b) Linear (1D) systolic array

U12

N2
M3 U31 U32 U33 U34

U13

N3
M4 U41 U42 U43 U44

U14

N4
(a) Rectangular (2D) systolic array

Fig. 4. Pictorial view of systolic array architectures

4.2 Existing FPGA implementations

In Section 3, we discussed some existing S-W implementations running on a CPU. A
comparable analysis for FPGAs is rather hard. There are very few real, complete
implementations that give usable results. Most research implementations only discuss
synthetic tests, giving very optimistic numbers for implementations that are hardly used in
practice. Furthermore, there is a great variety in the types of FPGAs used. Since every FPGA
series has a different way of implementing circuitry, it is hard to make a fair comparison.
In addition, the performance of the implementations relies heavily on the data widths used.
Smaller data widths lead to smaller PEs, which lead to faster implementations. These numbers
are not usually published. The first, third and fourth implementations shown in Table 4 make
this clear, where the performance is given in terms of Giga Cell Updates Per Second (GCUPS).
Using the same FPGA device, these three implementations differ significantly in performance.
The most reliable numbers are from Convey and SciEngines, as shown in the last two entries
of Table 4. These implementations work the same in practice for real cases and are build for
maximal performance (Vermij, 2011).
196
10 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

Performance Performance
Reference FPGA Frequency PEs
(per FPGA) (per system)
Virtex2 180 1260
(Puttegowda et al., 2003) 7000 —
XC2V6000 MHz GCUPS
Virtex2 742
(Yu et al., 2003) — 4032 —
XCV1000-6 GCUPS
Virtex2 55 13.9
(Oliver et al., 2005) 252 —
XC2V6000 MHz GCUPS
Virtex2 112 54
(Gok & Yilmaz, 2006) 482 —
XC2V6000 MHz GCUPS
Stratix2 66.7 25.6
(Altera, 2007) 384 —
EP2S180 MHz GCUPS
200 24.1
(Cray, 2010) Virtex4 120 —
MHz GCUPS
Virtex5 150 172.8 691.2
(Convey, 2010) 1152
LX330 MHz GCUPS GCUPS
Spartan6 47 6046
(SciEngines, 2010) — —
LX150 GCUPS GCUPS

Table 4. Performance of various FPGA implementations (Vermij, 2011)

4.3 Future FPGA implementations

The performance of S-W implementations on FPGA can foremost be increased by using
larger and faster FPGAs. Larger FPGAs can contain more PEs and therefore deliver higher
performance in terms of GCUPS. The largest Xilinx Virtex 6 FPGA device has roughly 2.5
times more area than the largest Virtex 5 FPGA, so the peak performance of the former can be
estimated at 2.5 × 172.8 = 432 GCUPS (using the numbers from the Convey implementation)
(Vermij, 2011).

5. GPU-based acceleration
The parallelization capabilities of GPUs can be best exploited for accelerating biological
sequence alignment applications. This section provides some brief background information
about GPUs. Furthermore, it presents the current GPU implementations for S-W based
sequence alignment.

5.1 GPU background

Compute Unified Device Architecture (CUDA) is the hardware and software architecture that
enables NVIDIA GPUs (Fermi™, 2009) to execute programs written in C, C++, Fortran,
OpenCL, DirectCompute and other languages. A CUDA program calls kernels that run on
the GPU. A kernel executes in parallel across a set of threads, where a thread is the basic
unit in the programming model that executes an instance of the kernel, and has access to
registers and per thread local memory. The programmer organizes these threads in grids
of thread blocks, where a thread block is a set of concurrently executing threads and has a
shared memory for communication between the threads. A grid is an array of thread blocks
that execute the same kernel, read inputs from and write outputs to global memory, and
synchronize between interdependent kernel calls. Figure 5 gives a block diagram description
of the GPU architecture.
An Overview
An Overview of Hardware-Based
of Hardware-based Acceleration
Acceleration of Biological of Biological Sequence Alignment
Sequence Alignment 197
11

Device
Device (GPU)
Multiprocessor N
Multiprocessor N

Multiprocessor
Multiprocessor 22

Multiprocessor
Multiprocessor 1
1

Shared memory

Registers Registers Registers

Instruction
Processor 1 Processor 2 … Processor M
unit
Instruction
Unit
Processor 1 Processor 2 Processor M

Global memory

Fig. 5. Block diagram description of a GPU architecture

5.2 Current implementations

The first known implementations of S-W based sequence alignment on a GPU are presented
in (Liu, Schmidt, Voss, Schroder & Muller-Wittig, 2006) and (Liu, Huang, Johnson & Vaidya,
2006). These approaches are similar and use the OpenGL graphics API to search
protein databases. First the database and query sequences are copied to GPU texture
memory. The score matrix is then processed in a systolic array fashion, where the data
flows in anti-diagonals. The results of each anti-diagonal are again stored in texture
memory, which are then used as inputs for the next pass. The implementation in
(Liu, Schmidt, Voss, Schroder & Muller-Wittig, 2006) searched 99.8% of Swiss-Prot (almost
180,000 sequences) and managed to obtain a maximum speed of 650 MCUPS compared
to around 75 for the compared CPU version. The implementation discussed in
(Liu, Huang, Johnson & Vaidya, 2006) offers the ability to run in two modes, i.e. one with
and one without traceback. The version with no traceback managed to perform at 241
MCUPS, compared to 178 with traceback and 120 for the compared CPU implementation.
Both implementations were benchmarked using a Geforce GTX 7800 graphics card.
The first known CUDA implementation, ‘SW-CUDA’, is discussed in (Manavski & Valle,
2008). In this approach, each of the GPU’s processors performs a complete alignment instead
of them being used to stream through a single alignment. The advantage of this is that no
communication between processing elements is required, thereby reducing memory reads and
writes. This implementation managed to perform at 1.9 GCUPS on a single Geforce GTX 8800
graphics card when searching Swiss-Prot, compared to around 0.12 GCUPS for the compared
198
12 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

CPU implementation. Furthermore, it is shown to scale almost linearly with the amount of
GPUs used by simply splitting up the database.
Various improvements have been suggested to the approach presented in (Manavski & Valle,
2008), as shown in (Akoglu & Striemer, 2009; Liu et al., 2009). In (Liu et al., 2009), for
sequences of more than 3,072 amino acids an ‘inter-task parallelization’ method similar to the
systolic array and OpenGL approaches is used as this, while slower, requires less memory.
The ‘CUDASW++’ solution presented in (Liu et al., 2009) manages a maximum speed of
about 9.5 GCUPS searching Swiss-Prot on a Geforce GTX 280 graphics card. An improved
version, ‘CUDASW++ 2.0’ has been published recently (Liu et al., 2010). Being the fastest
Smith-Waterman GPU implementation to date, ‘CUDASW++ 2.0’ managed 17 GCUPS on a
single GTX 280 GPU, outperforming CPU-based BLAST in its benchmarks.
In (Kentie, 2010), an enhanced GPU implementation for protein sequence alignment using
database and memory access optimizations is presented. Each processing element in
this implementation is used to independently generate a complete alignment between a
query sequence and a database sequence. This eliminates the need for inter-processor
communication and results in efficient resource utilization. The GPU used for implementation
(i.e. NVIDIA GTX 275) contains 240 processors, while the latest release of Swiss-Prot contains
more than 500,000 protein sequences. Hence, it is possible to keep all processors well occupied
while aligning query sequences with the sequences in the Swiss-Prot database. The results
demonstrate that the implementation presented in (Kentie, 2010) achieves a performance
of 21.4 GCUPS on an NVIDIA GTX 275 graphics card. Table 5 summarizes these GPU
implementations. Besides NVIDIA, ATI/AMD (AMD, 2011) also produces graphics cards
but to our knowledge no S-W implementations on such cards are available.
Database
Implementation Device Performance
searched
(Liu, Schmidt, Voss, Schroder & Muller-Wittig, 2006) GTX 7800 Swiss-Prot 650 MCUPS
983 protein
(Liu, Huang, Johnson & Vaidya, 2006) GTX 7800 241 MCUPS
sequences
(Manavski & Valle, 2008) GTX 8800 Swiss-Prot 1.9 GCUPS
(Liu et al., 2009) GTX 280 Swiss-Prot 9.5 GCUPS
(Liu et al., 2010) GTX 280 Swiss-Prot 17 GCUPS
(Kentie, 2010) GTX 275 Swiss-Prot 21.4 GCUPS

Table 5. Summary of the existing GPU implementations

6. Comparison of acceleration on different platforms

This section compares the performance of S-W implementations on various platforms like
CPUs, FPGAs and GPUs. The comparison is based on parameters like cost, energy
consumption, flexibility, scalability and future prospects. For the CPU, we consider a four
way, 48 core Opteron machine. For GPUs, a fast PC with 4 high end graphics cards, and for
FPGAs the fastest S-W system known, the one from SciEngines (SciEngines, 2010). The results
are shown in Figure 6. Following is a discussion per metric (Vermij, 2011).

Performance/Euro
FPGAs can deliver the best amount of GCUPS per Euro, followed closely by GPUs. The gap
between GPUs and CPUs can be explained by the extra money needed for a 4 way CPU
system, while plugging 4 GPUs on a commodity motherboard is free. This result explains
An Overview
An Overview of Hardware-Based
of Hardware-based Acceleration
Acceleration of Biological of Biological Sequence Alignment
Sequence Alignment 199
13

Performance / € Performance / €

Future prospect Performance / Watt Future prospect Performance / Watt

Scalability Flexibility Scalability Flexibility

CPU implementations GPU implementations

Performance / €

Future prospect Performance / Watt

Scalability Flexibility

FPGA implementations

Fig. 6. Analysis of various S-W metrics for implementations on different platforms (Vermij,
2011)

why FPGAs are used for high performance computing. S-W might not be the algorithm of
choice to show a major performance per Euro gain from using FPGAs. Nevertheless, it shows
the trend.

Performance/Watt
It is clear that here, in contrast to the previous metric, FPGAs are the absolute winner. Full
systems can deliver thousands of GCUPS for around 1000 Watts. This is another important
reason for using FPGAs for sequence alignment. Note that, while not visible in the graphs,
CPUs score around twice as good as GPUs.

Flexibility
This metric represents the effort needed to change a fast linear gap implementation to use
affine gaps. A skilled engineer would manage to do this in a day for a CPU implementation,
in a couple of days for GPU implementations, and many weeks for their FPGA counterpart.

Scalability
Here we took CPUs as baseline. Given a suitable problem, CPUs are very scalable as they
can be connected together using standard networking solutions. By the same token, GPUs are
also scalable, as they can take advantage of the scalability of CPUs, but will introduce some
extra latency. Therefore they score a bit lower than CPUs. Depending on the used platform,
FPGAs can also be made scalable.
200
14 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

Future prospect
In the past few years, there is a trend for CPUs to replace GPUs in high speed systems. This
trend is expected to continue, thereby reducing the market share of GPUs in favour of CPUs.
CPUs therefore score highly on this metric while GPUs score rather low. In the very specific
HPC areas, however, where memory bandwidth requirements are low and the problem is
very composable, FPGAs will likely continue to be the best choice. S-W partially lies in this
category.

7. Conclusions
This chapter provided a classification, discussion and comparison of the available sequence
alignment methods and their acceleration on various available hardware platforms. A
detailed introduction about the S-W algorithm, its pseudo code, data dependencies in the H
matrix and logic circuit to compute values of the cells in the H matrix are provided. A review
of CPU-based acceleration of S-W algorithm was presented. Recent CPU implementations
were discussed and compared. Further, performance estimations for top end current and
future CPUs were provided. FPGA-based acceleration of S-W algorithm and a discussion
about systolic arrays was given. Existing FPGA implementations were discussed and
compared. Further, an insight into the future FPGA implementations was touched upon.
GPU-based acceleration of S-W algorithm and GPU background were presented. Current
GPU implementations were discussed and compared. Furthermore, this chapter presented a
comparison of S-W accelerations on different hardware platforms. The comparison was based
on the following parameters.
• Performance per euro
• Performance per unit watt
• Flexibility
• Scalability
• Future prospects

8. References
Akoglu, A. & Striemer, G. M. (2009). Scalable and highly parallel implementation of
Smith-Waterman on graphics processing unit using CUDA, Cluster Computing Vol.
12(No. 3): 341–352.
Aldinucci, M., Meneghin, M. & Torquati, M. (2010). Efficient Smith-Waterman on multi-core
with fastflow, Proceedings of the 2010 IEEE International Symposium on Parallel and
Distributed Processing, IEEE, Pisa, Italy, pp. 195–199.
Altera (2007). Implementation of the Smith-Waterman algorithm on a reconfigurable
supercomputing platform, Altera White Paper, Altera, pp. 1–18.
Altschul, S. F., Gish, W., Miller, W., Myers, E. W. & Lipman, D. J. (1990). A basic local alignment
search tool, Journal of Molecular Biology Vol. 215: 403–410.
AMD (2011). ATI/AMD.
URL: http://www.amd.com/us/products/Pages/graphics.aspx
Buyukkur, A. B. & Najjar, W. (2008). Compiler generated systolic arrays for wavefront
algorithm acceleration on FPGAs, Proceedings of International Conference on Field
Programmable Logic and Applications (FPL08), Heidelberg, Germany, pp. 1–4.
An Overview
An Overview of Hardware-Based
of Hardware-based Acceleration
Acceleration of Biological of Biological Sequence Alignment
Sequence Alignment 201
15

Convey (2010). Convey HC1.

URL: http://www.convey.com
Cray (2010). Cray XD1.
URL: http://www.cray.com
Eddy, S. R. (1998). Profile hidden morkov models, Bioinformatics Review Vol. 14: 755-763.
Farrar, M. (2007). Striped Smith-Waterman speeds database searches six times over other
SIMD implementations, Bioinformatics Vol. 23(2): 156–161.
Fermi™(2009). Nvidia’s next generation cuda™ compute architecture, White paper NVIDIA
corporation .
Gibbs, A. J. & McIntyre, G. A. (1970). The diagram, a method for comparing sequences, its
use with amino acid and nucleotide sequences, European Journal of Biochemistry Vol.
16(No. 22): 1–11.
Giegerich, R. (2000). A systematic approach to dynamic programming in bioinformatics,
Bioinformatics Vol. 16: 665–677.
Gok, M. & Yilmaz, C. (2006). Efficient cell designs for systolic Smith-Waterman
implementation, Proceedings of International Conference on Field Programmable Logic and
Applications (FPL06), Madrid, Spain, pp. 1–4.
Hasan, L., Al-Ars, Z. & Taouil, M. (2010). High performance and resource efficient biological
sequence alignment, Proceedings of 32nd Annual International Conference of the IEEE
EMBS, Buenos Aires, Argentina, pp. 1767–1770.
Hasan, L., Al-Ars, Z. & Vassiliadis, S. (2007). Hardware acceleration of sequence alignment
algorithms - an overview, Proceedings of International Conference on Design & Technology
of Integrated Systems in Nanoscale Era (DTIS’07), Rabat, Morocco, pp. 96–101.
Kentie, M. (2010). Biological sequence alignment using graphics processing units, M.Sc. Thesis
CE-MS-2010-35, Computer Engineering Laboratory, TU Delft, The Netherlands, 2010.
Kung, H. T. & Leiserson, C. E. (1979). Algorithms for VLSI processor arrays, in: C. Mead, L.
Conway (eds.): Introduction to VLSI Systems; Addison-Wesley.
Liao, H. Y., Yin, M. L. & Cheng, Y. (2004). A parallel implementation of the Smith-Waterman
algorithm for massive sequences searching, Proceedings of 26th Annual International
Conference of the IEEE EMBS, San Francisco, CA, USA, pp. 2817–2820.
Liu, W., Schmidt, B., Voss, G., Schroder, A. & Muller-Wittig, W. (2006). Bio-sequence database
scanning on a GPU, Parallel and Distributed Processing Symposium, IEEE, Rhodes
Island, pp. 1–8.
Liu, Y., Huang, W., Johnson, J. & Vaidya, S. (2006). GPU accelerated Smith-Waterman,
Proceedings of International Conference on Computational Science, ICCS 2006, Springer,
Reading, UK, pp. 1–8.
Liu, Y., Maskell, D. & Schmidt, B. (2009). CUDASW++: Optimizing Smith-Waterman sequence
database searches for CUDA-enabled graphics processing units, BMC Research Notes
Vol. 2(No. 1:73).
Liu, Y., Schmidt, B. & Maskell, D. (2010). CUDASW++2.0: Enhanced Smith-Waterman protein
database search on CUDA-enabled GPUs based on SIMT and virtualized SIMD
abstractions, BMC Research Notes Vol. 3(No. 1:93).
Lu, J., Perrone, M., Albayraktaroglu, K. & Franklin, M. (2008). HMMER-cell: High
performance protein profile searching on the Cell/B.E. processor, Proceedings of IEEE
International Symposium on Performance Analysis of Systems and Software (ISPASS-2008),
Austin, Texas, USA, pp. 223–232.
202
16 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

Manavski, S. A. & Valle, G. (2008). CUDA compatible GPU cards as efficient hardware
accelerators for Smith-Waterman sequence alignment, BMC Bioinformatics Vol. 9(No.
2:S10).
Needleman, S. & Wunsch, C. (1970). A general method applicable to the search for similarities
in the amino acid sequence of two proteins, Journal of Molecular Biology Vol. 48(No.
3): 443–453.
Oliver, T., Schmidt, B. & Maskell, D. (2005). Hyper customized processors for bio-sequence
database scanning on FPGAs, Proceedings of FPGA’05, ACM, Monterey, California,
USA, pp. 229–237.
Pearson, W. R. & Lipman, D. J. (1985). Rapid and sensitive protein similarity searches, Science
Vol. 227: 1435–1441.
Pfeiffer, G., Kreft, H. & Schimmler, M. (2005). Hardware enhanced biosequence alignment,
Proceedings of International Conference on METMBS, pp. 1–7.
Puttegowda, K., Worek, W., Pappas, N., Dandapani, A. & Athanas, P. (2003). A run-time
reconfigurable system for gene-sequence searching, Proceedings of 16th International
Conference on VLSI Design, IEEE, USA, pp. 561–566.
Quinton, P. & Robert, Y. (1991). Systolic Algorithms and Architectures, Prentice Hall Int.
SciEngines (2010). Sciengines rivyera.
URL: http://www.sciengines.com
Smith, T. F. & Waterman, M. S. (1981). Identification of common molecular subsequences,
Journal of Molecular Biology Vol. 147: 195–197.
Szalkowski, A., Ledergerber, C., Krhenbhl1, P. & Dessimoz, C. (2009). SWPS3 - A fast
multi-threaded vectorized Smith-Waterman for IBM Cell/B.E. and x86/SSE2, BMC
Research Notes Vol. 1(No. 1:107).
Thompson, J. D., Higgins, D. G. & Gibson, T. J. (1994). ClustalW: Improving the
sensitivity of progressive multiple sequence alignment through sequence weighting,
position-specific gap penalties and weight matrix choice, Nucleic Acids Research Vol.
22(No. 22): 4673–4680.
Vermij, E. (2011). Genetic sequence alignment on a supercomputing platform, M.Sc. Thesis,
Computer Engineering Laboratory, TU Delft, The Netherlands, 2011.
Yu, C. W., Kwong, K. H., Lee, K. H. & Leong, P. H. W. (2003). A Smith-Waterman systolic cell,
International Workshop on Field Programmable Logic and Applications (FPL03), Springer,
pp. 375–384.
 Part 2

Case Studies
 10

Retrieving and Categorizing Bioinformatics

Publications through a MultiAgent System
Andrea Addis1, Giuliano Armano1, Eloisa Vargiu1 and Andrea Manconi2
1University of Cagliari - Dept. of Electrical and Electronic Engineering
2Institute for Biomedical Technologies, National Research Council
Italy

1. Introduction
The huge and steady increase of available digital documents, together with the corresponding
volume of daily updated contents, makes the problem of retrieving and categorizing
documents and data a challenging task. To this end, automated content-based document
management systems have gained a main role in the field of intelligent information access
(Armano et al., 2010).
Web retrieval is highly popular and presents a technical challenge due to the heterogeneity
and size of the Web, which is continuously growing (see (Huang, 2000), for a survey). In
particular, it becomes more and more difficult for Web users to select contents that meet their
interests, especially if contents are frequently updated (e.g., news aggregators, newspapers,
scientific digital archives, RSS feeds, and blogs). Supporting users in handling the huge and
widespread amount of Web information is becoming a primary issue.
Among other kinds of information, let us concentrate on publications and scientific literature,
largely available on the Web for any topic. As for bioinformatics, it can be observed
that the steady work of researchers, in conjunction with the advances in technology (e.g.,
high-throughput technologies), has arisen in a growing amount of known sequences. The
information related with these sequences is daily stored in the form of scientific articles.
Digital archives like BMC Bioinformatics1 , PubMed Central2 and other online journals and
resources are more and more searched for by bioinformaticians and biologists, with the goal
of downloading articles relevant to their scientific interests. However, for researchers, it is still
very hard to find out which publications are in fact of interest without an explicit classification
of the relevant topics they describe.
Traditional filtering techniques based on keyword search are often inadequate to express what
the user is really searching for. This principle is valid also in the field of scientific publications
retrieval, where researchers could obtain a great benefit from the adoption of automated tools
able to search for publications related with their interests.
To be effective in the task of selecting and suggesting to a user only relevant publications,
an automated system should at least be able (i) to extract the required information and (ii)
to encode and process it according to a given set of categories. Personalization could also be
provided according to user needs and preferences.

1 http://www.biomedcentral.com/bmcbioinformatics/
2 http://www.pubmedcentral.gov/
206
2 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

In this chapter, we present PUB.MAS, a multiagent system able to retrieve and categorize
bioinformatics publications from selected Web sources. The chapter extends and revises
our previous work (Armano et al., 2007). The main extensions consist of a more detailed
presentation of the information extraction task, a deep explanation of the adopted hierarchical
text categorization technique, and the description of the prototype that has been implemented.
Built upon X.MAS (Addis et al., 2008), a generic multiagent architecture aimed at retrieving,
filtering and reorganizing information according to user interests, PUB.MAS is able to: (i)
extract information from online digital archives; (ii) categorize publications according to a
given taxonomy; and (iii) process user’s feedback. As for information extraction, PUB.MAS
provides specific wrappers able to extract publications from RSS-based Web pages and from
Web Services. As for categorization, PUB.MAS performs Progressive Filtering (PF), the
effective hierarchical text categorization technique described in (Addis et al., 2010). In its
simplest setting, PF decomposes a given rooted taxonomy into pipelines, one for each existing
path between the root and each node of the taxonomy, so that each pipeline can be tuned in
isolation. To this end, a threshold selection algorithm has been devised, aimed at finding a
sub-optimal combination of thresholds for each pipeline. PUB.MAS provides also suitable
strategies to allow users to express what they are really interested in and to personalize
search results accordingly. Moreover, PUB.MAS provides a straightforward approach to user
feedback with the goal of improving the performance of the system depending on user needs
and preferences.
The prototype allows users to set the sources from which publications will be extracted and
the topics s/he is interested in. As for the digital archives, the user can choose between BMC
Bioinformatics and PubMed Central. As for the topics of interest, the user can select one or
more categories from the adopted taxonomy, which is taken from the TAMBIS ontology (Baker
et al., 1999).
The overall task begins with agents able to handle the selected digital archives, which extract
the candidate publications. Then, all agents that embody a classifier trained on the selected
topics are involved to perform text categorization. Finally, the system supplies the user with
the selected publications through suitable interface agents.
The chapter is organized as follows. First, we give a brief survey of relevant related work
on: (i) scientific publication retrieval; (ii) hierarchical text categorization; and (iii) multiagent
systems in information retrieval. Subsequently, we concentrate on the task of retrieving
and categorizing bioinformatics publications. Then, PUB.MAS is illustrated together with
its performances and the implemented prototype. Conclusions end the chapter.

2. Background
Information Retrieval (IR) is the task of representing, storing, organizing, and accessing
information items. IR has changed considerably in recent years with the expansion of the
World Wide Web and the advent of modern and inexpensive graphical user interfaces and
mass storage (Baeza-Yates & Ribeiro-Neto, 1999). The most relevant IR issues that help to
clarify the contextual setting of this chapter are: (i) the work done on scientific publication
retrieval, (ii) the work done on Hierarchical Text Categorization (HTC), and (iii) the work
done on multiagent systems (MAS) for information retrieval.

2.1 Scientific publication retrieval

In the academic area, online search engines are used to find out scientific resources, as journals
and conference proceedings. However, finding and selecting appropriate information on the
Retrieving and Categorizing
Retrieving and Categorizing Bioinformatics
Bioinformatics Publications Publications
through a MultiAgent System through a MultiAgent System 2073

Web is still difficult. To simplify this process, several frameworks and systems have been
developed to retrieve scientific publications from the Web.
Bollacker et al. (2000) developed CiteSeer3 , the well-known automatic generator of digital
libraries of scientific literature. Being aimed at eliminating most of the manual effort of
finding useful publications on the Web, CiteSeer uses sophisticated acquisition, parsing, and
presentation methods. In particular, CiteSeer uses a three-stage process: database creation and
feature extraction; personalized filtering of new publications; and personalized adaptation
and discovery of interesting research and trends. These functions are interdependent:
information filtering affects what is discovered, whereas useful discoveries tune the
information filtering. In (McNee et al., 2002), the authors study how to recommend research
papers using the citation between papers to create the user-item matrix. In particular, they test
the ability of collaborative filtering to recommend citations that could be additional references
for a target research paper. Janssen & Popat (2003) developed UpLib, a personal digital library
system that consists of a full-text indexed repository accessed through an active agent via a
Web interface. UpLib is mainly concerned with the task of collecting personal collections
comprising tens of thousands of documents. In (Mahdavi et al., 2009), the authors start from
the assumption that trend detection in scientific publication retrieval systems helps scholars
to find relevant, new and popular special areas by visualizing the trend of the input topic. To
this end, they developed a semi-automatic system based on a semantic approach.
As for the specific task of retrieving information in the field of bioinformatics, a lot of
work has been done –some of it being recalled hereafter. Tanabe et al. (1999) developed
MedMiner, an Internet-based hypertext program able to filter and organize large amounts of
textual and structured information returned from public search engines –like GeneCards and
PubMed. Craven & Kumlien (1999) applied machine learning techniques to automatically
map information from text sources into structured representations, such as knowledge bases.
Friedman et al. (2001) propose GENIES, a system devised to extract information about cellular
pathways from the biological literature in accordance with a given domain knowledge. Ramu
(2003) developed a Web Server for SIR (Ramu, 2001), a simple indexing and retrieval system
that combines sequence motif search with keyword search. The Web Server, called SIRW, is
a generic tool used by the bioinformatics community for searching and analyzing biological
sequences of interest. Rocco & Critchlow (2003) propose a system aimed at finding classes
of bioinformatics data sources and integrating them behind a unified interface. The main
goal of the system is to eliminate the human effort required to maintain a repository of
information sources. Kiritchenko et al. (2004) propose a system aimed at retrieving Medline
articles that mention genes. After being retrieved, articles are categorized according to the
Gene Ontology (GO) codes. Delfs et al. (2004) developed and implemented GoPubMed, a
system that allows to submit keywords to PubMed, extracts GO terms from the retrieved
abstracts, and supplies the user with the relevant ontology for browsing. Corney et al. (2004)
propose BioRAT (Biological Research Assistant for Text mining), a new information extraction
tool specifically tailored for biomedical tasks. Able to access and analyze both abstracts and
full-length papers, it incorporates a domain specific document search ability.

2.2 Hierarchical text categorization

In recent years several researchers have investigated the use of hierarchies for text
categorization.
Until the mid-1990s researchers mostly ignored the hierarchical structure of categories that
occur in several domains. In 1997, Koller & Sahami (1997) carry out the first proper study

3 http://citeseer.ist.psu.edu/
208
4 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

on HTC on the Reuters-22173 collection. Documents are classified according to the given
hierarchy by filtering them through the single best-matching first-level class and then sending
them to the appropriate second level. This approach shows that hierarchical models perform
well when a small number of features per class is used, as no advantages were found using
the hierarchical model for large numbers of features. McCallum et al. (1998) propose a method
based on naïve Bayes. The authors compare two techniques: (i) exploring all possible paths
in the given hierarchy and (ii) greedily selecting at most two branches according to their
probability, as done in (Koller & Sahami, 1997). Results show that the latter is more error
prone while computationally more efficient. Mladenić & Grobelnik (1998) use the hierarchical
structure to decompose a problem into a set of subproblems, corresponding to categories
(i.e., the nodes of the hierarchy). For each subproblem, a naïve Bayes classifier is generated,
considering examples belonging to the given category, including all examples classified in its
subtrees. The classification applies to all nodes in parallel; a document is passed down to
a category only if the posterior probability for that category is higher than a user-defined
threshold. D’Alessio et al. (2000) propose a system in which, for a given category, the
classification is based on a weighted sum of feature occurrences that should be greater than the
category threshold. Both single and multiple classifications are possible for each document to
be tested. The classification of a document proceeds top-down possibly through multiple
paths. An innovative contribution of this work is the possibility of restructuring a given
hierarchy or building a new one from scratch. Dumais & Chen (2000) use the hierarchical
structure for two purposes: (i) training several SVMs, one for each intermediate node and (ii)
classifying documents by combining scores from SVMs at different levels. The sets of positive
and negative examples are built considering documents that belong to categories at the same
level, and different feature sets are built, one for each category. Several combination rules
have also been assessed. In the work of Ruiz & Srinivasan (2002), a variant of the Hierarchical
Mixture of Experts model is used. A hierarchical classifier combining several neural networks
is also proposed in (Weigend et al., 1999). Gaussier et al. (2002) propose a hierarchical
generative model for textual data, i.e., a model for hierarchical clustering and categorization of
co-occurrence data, focused on documents organization. In (Rousu et al., 2005), a kernel-based
approach for hierarchical text classification in a multi-label context is presented. The work
demonstrates that the use of the dependency structure of microlabels (i.e., unions of partial
paths in the tree) in a Markovian Network framework leads to improved prediction accuracy
on deep hierarchies. Optimization is made feasible by utilizing decomposition of the
original problem and making incremental conditional gradient search in the subproblems.
Ceci & Malerba (2007) present a comprehensive study on hierarchical classification of Web
documents. They extend a previous work (Ceci & Malerba, 2003) considering hierarchical
feature selection mechanisms, a naïve Bayes algorithm aimed at avoiding problems related to
different document lengths, the validation of their framework for a probabilistic SVM-based
classifier, and (iv) an automated threshold selection algorithm. More recently, in (Esuli
et al., 2008), the authors propose a multi-label hierarchical text categorization algorithm
consisting of a hierarchical variant of ADABOOST.MH, a well-known member of the family
of “boosting” learning algorithms. Bennett & Nguyen (2009) study the problem of the error
propagation under the assumption that the higher the node in the hierarchy is the worse is the
mistake, as well as the problem of dealing with increasingly complex decision surfaces. Brank
et al. (2010) deal with the problem of classifying textual documents into a topical hierarchy of
categories. They construct a coding matrix gradually, one column at a time, each new column
being defined in a way that the corresponding binary classifier attempts to correct the most
common mistakes of the current ensemble of binary classifiers. The goal is to achieve good
performance while keeping reasonably low the number of binary classifiers.
Retrieving and Categorizing
Retrieving and Categorizing Bioinformatics
Bioinformatics Publications Publications
through a MultiAgent System through a MultiAgent System 2095

2.3 MultiAgent Systems in information retrieval

Autonomous agents and MAS have been successfully applied to a number of problems and
have been largely used in different application domains (Wooldridge & Jennings, 1995).
As for MAS in IR, in the literature, several centralized agent-based architectures aimed at
performing IR tasks have been proposed. Among others, let us recall NewT (Sheth & Maes,
1993), Letizia (Lieberman, 1995), WebWatcher (Armstrong et al., 1995), and SoftBots (Etzioni
& Weld, 1995). NewT is composed by a society of information-filtering interface agents,
which learn user preferences and act on her/his behalf. These information agents use a
keyword-based filtering algorithm, whereas adaptive techniques are relevance feedback and
genetic algorithms. Letizia is an intelligent user-interface agent able to assist a user while
browsing the Web. The search for information is performed through a cooperative venture
between the user and the software agent: both browse the same search space of linked
Web documents, looking for interesting ones. WebWatcher is an information search agent
that follows Web hyperlinks according to user interests, returning a list of links deemed
interesting. In contrast to systems for assisted browsing or IR, SoftBots accept high-level user
goals and dynamically synthesize the appropriate sequence of Internet commands according
to a suitable ad-hoc language.
Despite the fact that a centralized approach could have some advantages, in IR tasks it may
encompass several problems, in particular how to scale up the architectures to large numbers
of users, how to provide high availability in case of constant demand of the involved services,
and how to provide high trustability in case of sensitive information, such as personal data.
To overcome the above drawbacks, suitable MAS devoted to perform IR tasks have been
proposed. In particular, Sycara et al. (2001) propose Retsina, a MAS infrastructure applied
in many domains. Retsina is an open MAS infrastructure that supports communities of
heterogeneous agents. Three types of agents have been defined: (i) interface agents, able to
display the information to the users; (ii) task agents, able to assist the user in the process of
handling her/his information; and (iii) information agents, able to gather relevant information
from selected sources.
Among other MAS, let us recall IR-agents (Jirapanthong & Sunetnanta, 2000), CEMAS
(Bleyer, 1998) and the cooperative multiagent system for Web IR proposed in (Shaban et al.,
2004). IR-agents implement an XML-based multiagent model for IR. The corresponding
framework is composed of three kinds of agents: (i) managing agents, aimed at extracting
the semantics of information and at performing the actual tasks imposed by coordinator
agents, (ii) interface agents, devised to interact with the users, and (iii) search agents, aimed
at discovering relevant information on the Web. IR-agents do not take into account
personalization, while providing information in a structured form without the adoption of
specific classification mechanisms. In CEMAS, Concept Exchanging MultiAgent System, the
basic idea is to provide specialized agents for exchanging concepts and links, representing the
user, searching for new relevant documents matching existing concepts, and supporting agent
coordination. Although CEMAS provides personalization and classification mechanisms
based on a semantic approach, and it is mainly aimed at supporting scientists while looking
for comprehensive information about their research interests. Finally, in (Shaban et al.,
2004) the underlying idea is to adopt intelligent agents that mimic everyday-life activities of
information seekers. To this end, agents are also able to profile the user in order to anticipate
and achieve her/his preferred goals. Although interesting, the approach is mainly focused on
cooperation among agents rather than on IR issues.
210
6 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

3. The proposed approach

A system for information retrieval must take into account several issues, the most relevant
being:
1. how to deal with different information sources and to integrate new information sources
without re-writing significant parts of it;
2. how to suitably encode data in order to put into evidence the informative content useful
to discriminate among categories;
3. how to control the imbalance between relevant and irrelevant articles (the latter being
usually much more numerous than the former);
4. how to allow the user to specify her/his preferences;
5. how to exploit the user’s feedback to improve the overall performance of the system.
The above issues are typically strongly interdependent in state-of-the-art systems. To better
concentrate on these aspects separately, we adopted a layered multiagent architecture, able to
promote the decoupling among all aspects deemed relevant.
To perform the task of retrieving scientific publications, the actual system –sketched in Figure
1– involves three main activities: extracting the required information from selected online
sources, categorizing it according to a given taxonomy while taking into account also users
preferences, and providing suitable feedback mechanisms.

Fig. 1. PUB.MAS: the multiagent system devised for classifying bioinformatics publications.
Retrieving and Categorizing
Retrieving and Categorizing Bioinformatics
Bioinformatics Publications Publications
through a MultiAgent System through a MultiAgent System 2117

Fig. 2. The structure of a BMC Bioinformatics page.

3.1 Information extraction

This phase is devoted to deal with the huge amount of information provided by information
sources. To this end suitable wrappers have been implemented, able to handle the structure
of a document by saving the information about the corresponding metadata. In general,
given a Web source, a specific wrapper must be implemented, able to map each Web page,
designed according to the constraints imposed by the Web source, to a suitable description,
which contains relevant data in a structured form –such as title, text content, and references.
To make this point clearer, let us consider the structure of the BMC Bioinformatics page of
the paper “A Hybrid Genetic-Neural System for Predicting Protein Secondary Structure”
(Armano et al., 2005) reported in Figure 2. In this case, it is quite easy to implement the
mapping function, since, for each description field, a corresponding tag exists, making it very
simple to process the pages.
A suitable encoding of the text content has also been enforced during this phase: all
non-informative words such as prepositions, conjunctions, pronouns and very common verbs
are deleted using a stop-word list; after that, a standard stemming algorithm (Porter, 1980)
removes the most common morphological and inflexional suffixes. The subsequent step
requires the adoption of suitable domain knowledge. For each category of the underlying
taxonomy, feature selection (based on the information-gain heuristics) has been adopted to
reduce the dimensionality of the feature space.

3.2 Text categorization

Scientific publications are classified according to a high-level taxonomy, which is independent
from the specific user. To this end, classifiers are combined according to the links that hold
within the taxonomy, giving rise to “vertical” and “horizontal” combinations of classifiers.
212
8 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

3.2.1 Vertical combination

The Approach
Vertical combination is currently performed by resorting to Progressively Filtering (PF), a
simple categorization technique framed within the local classifier per node approach, which
admits only binary decisions. In PF, each classifier is entrusted with deciding whether the
input in hand can be forwarded or not to its children. The first proposals in which sequential
boolean decisions are applied in combination with local classifiers per node can be found in
(D’Alessio et al., 2000), (Dumais & Chen, 2000), and (Sun & Lim, 2001). In Wu et al. (2005), the
idea of mirroring the taxonomy structure through binary classifiers is clearly highlighted; the
authors call this technique “binarized structured label learning”.

Fig. 3. An example of PF (highlighted with bold-dashed lines).

In PF, given a taxonomy, where each node represents a classifier entrusted with recognizing all
corresponding positive inputs (i.e., interesting documents), each input traverses the taxonomy
as a “token”, starting from the root. If the current classifier recognizes the token as positive,
it is passed on to all its children (if any), and so on. A typical result consists of activating one
or more branches within the taxonomy, in which the corresponding classifiers have accepted
the token. Figure 3 gives an example of how PF works. A theoretical study of the approach
is beyond the scope of this chapter, the interested reader could refer to (Armano, 2009) for
further details.
A simple way to implement PF consists of unfolding the given taxonomy into pipelines
of classifiers, as depicted in Figure 4. Each node of the pipeline is a binary classifier
able to recognize whether or not an input belongs to the corresponding class (i.e., to the
corresponding node of the taxonomy). Partitioning the taxonomy in pipelines gives rise to
a set of new classifiers, each represented by a pipeline.
Finally, let us note that the implementation of PF described in this chapter performs a sort of
“flattening” though preserving the information about the hierarchical relationships embedded
in a pipeline (Addis et al., 2010). For instance, the pipeline C, C2, C21 actually represents
the classifier C21, although the information about the existing subsumption relationships (i.e.,
C21 ≤ C2 ≤ C) is preserved.
Retrieving and Categorizing
Retrieving and Categorizing Bioinformatics
Bioinformatics Publications Publications
through a MultiAgent System through a MultiAgent System 2139

Fig. 4. The pipelines corresponding to the taxonomy in Figure 3.

The Threshold Selection Algorithm

As we know from classical text categorization, given a set of documents D and a set of labels
C, a function CSVi : D → [0, 1] exists for each ci ∈ C. We assume that the behavior of ci is
controlled by a threshold θi , responsible for relaxing or restricting the acceptance rate of the
corresponding classifier. Given d ∈ D, CSVi (d) ≥ θi permits to categorize d under ci , whereas
CSVi (d) < θi is interpreted as a decision not to categorize d under ci .
In PF, let us still assume that CSVi exists for each ci ∈ C, with the same semantics adopted
in the classical case. Considering a pipeline π, composed of n classifiers, the acceptance
policy strictly depends on the vector θπ = θ1 , θ2 , · · · , θn  that embodies the thresholds of
all classifiers in π. In order to categorize d under π, the following constraint must be satisfied:
∀k = 1 . . . n, CSVi (d) ≥ θk ; otherwise, d is not categorized under ci .
A further simplification of the problem consists of allowing a classifier to have different
behaviors, depending on which pipeline it is embedded in. Each pipeline can be considered
in isolation from the others. For instance, given π1 = C, C2, C21 and π2 = C, C2, C22, the
classifier C is not compelled to have the same threshold in π1 and in π2 (the same holds for
C2).
Given a utility function4 , we are interested in finding an effective and computationally
“light” way to reach a sub-optimum in the task of determining the best vector of thresholds.
Unfortunately, finding the best acceptance thresholds is a difficult task. Exhaustively trying
each possible combination of thresholds (brute-force approach) is unfeasible, the number of
thresholds being virtually infinite. However, the brute-force approach can be approximated
by defining a granularity step that requires to check only a finite number of points in the
range [0, 1], in which the thresholds are permitted to vary with step δ. Although potentially
useful, this “relaxed” brute force algorithm for calibrating thresholds (RBF for short) is still
too heavy from a computational point of view. On the contrary, the threshold selection
algorithm described in this chapter is characterized by low time complexity while maintaining
the capability of finding near-optimum solutions.
Bearing in mind that the lower the threshold the less restrictive is the classifier, we adopt
the greedy bottom-up algorithm for selecting decision threshold that relies on two functions
described in (Addis et al., 2011):

4 Different utility functions (e.g., precision, recall, Fβ , user-defined) can be adopted, depending on the
constraints imposed by the underlying scenario.
214
10 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

• Repair (R), which operates on a classifier C by increasing or decreasing its threshold –i.e.,
R(up, C ) and R(down, C ), respectively– until the selected utility function reaches and
maintains a local maximum.
• Calibrate (C ), which operates going downwards from the given classifier to its offspring. It
is intrinsically recursive and, at each step, calls R to calibrate the current classifier.
Given a pipeline π = C1 , C2 , . . . , CL , TSA is defined as follows (all thresholds are initially
set to 0):

TSA(π ) := f or k = L downto 1 do C(up, Ck ) (1)

which asserts that C is applied to each node of the pipeline, starting from the leaf (k = L).
The Calibrate function is defined as follows:

C(up, Ck ) := R(up, Ck ), k = L
(2)
C(up, Ck ) := R(up, Ck ); C(down, Ck+1 ), k<L
and
C(down, Ck ) := R(down, Ck ), k = L
(3)
C(down, Ck ) := R(down, Ck ); C(up, Ck+1 ), k < L
where the “;” denotes a sequence operator, meaning that in “a;b” action a is performed before
action b. The reason why the direction of threshold optimization changes at each call of
Calibrate (and hence of Repair) lies in the fact that increasing the threshold θk−1 is expected
to forward less false positives to Ck , which allows to decrease θk . Conversely, decreasing
the threshold θk−1 is expected to forward more false positives to Ck , which must react by
increasing θk .
It is worth pointing out that, as also noted in (Lewis, 1995), the sub-optimal combination of
thresholds depends on the adopted dataset, hence it needs to be recalculated for each dataset.

3.2.2 Horizontal combination

To express what the user is really interested in, we implemented suitable horizontal
composition strategies by using extended boolean models (Lee, 1994). In fact, a user is
typically not directly concerned with topics that coincide with classes of the given taxonomy.
Rather, a set of arguments of interest can be obtained by composing such generic topics with
suitable logical operators (i.e., and, or, and not). For instance, a user might be interested in
being kept informed about all articles that involve both “cell” and “nucleus”. This compound
topic can be dealt with by composing the cell and the nucleus classifiers. To address this issue,
we adopted a soft boolean perspective, in which the combination is evaluated using P-norms
(Golub & Loan, 1996).

3.3 Users’ feedback

So far, a simple solution based on the k-NN technology has been implemented and
experimented to deal with the problem of supporting the user’s feedback. When a irrelevant
article is evidenced by the user, it is immediately embedded in the training set of the k-NN
classifier that implements the feedback. A check performed on this training set after inserting
the negative example allows to trigger a procedure entrusted with keeping the number of
negative and positive examples balanced. In particular, when the ratio between negative
and positive examples exceeds a given threshold (by default set to 1.1), some examples are
randomly extracted from the set of “true” positive examples and embedded in the above
training set.
Retrieving and Categorizing
Retrieving and Categorizing Bioinformatics
Bioinformatics Publications Publications
through a MultiAgent System through a MultiAgent System 215
11

4. PUB.MAS
To retrieve and categorize scientific publications, we customized X.MAS (Addis et al., 2008), a
generic multiagent architecture built upon JADE (Bellifemine et al., 2007) devised to facilitate
the implementation of information retrieval and information filtering applications. The
motivation for adopting a MAS lies in the fact that a centralized classification system might
be quickly overwhelmed by a large and dynamic document stream, such as daily-updated
online publications. Furthermore, the Web is intrinsically a pervasive system and offers the
opportunity to take advantage of distributed computing paradigms and spread knowledge
resources.

4.1 The system

PUB.MAS, is organized in the three “virtual” layers depicted in Figure 1, by customizing
X.MAS as follows:
• Information Level. Agents at this level are devoted to deal with the selected information
sources. A wrapper able to deal with the RSS (Really Simple Syndication) format has
been implemented, aimed at extracting publications from BMC Bioinformatics. It is
worth pointing out that the RSS format allows to easily process any given page, since a
corresponding RSS tag exists for each relevant item. Furthermore, the growing amount of
Web Services providing scientific publications requires the implementation of wrappers
explicitly devoted to extract information from them. In order to invoke Web Services
from our multiagent system, required to access the PubMed Central Web Service, we
implemented an ad-hoc wrapper by adopting WSIG (Greenwood & Calisti, 2004).
• Filter Level. Filter agents are devoted to select information deemed relevant to the users,
and to cooperate to prevent information from being overloaded and redundant. A suitable
encoding of the text content has been enforced at this level to facilitate the work of agents
belonging to the task level. As already pointed out, all non-informative words such
as prepositions, conjunctions, pronouns and very common verbs are removed using a
stop-word list. After that, a standard stemming algorithm removes the most common
morphological and inflexional suffixes. Then, for each category, feature selection, based
on the information-gain heuristics, has been adopted to reduce the dimensionality of the
feature space.
• Task Level. Task agents are devoted to identify relevant scientific publications, depending
on user interests. Agents belonging to this architectural level are devoted to perform
two kinds of actions: to classify any given input in accordance with the selected set of
categories, and to decide whether it may be of interest to the user or not. Each task agent
has been trained by resorting to a state-of-the-art technique, i.e. k-NN, in its “weighted”
variant (Cost & Salzberg, 1993). The choice of adopting weighted k-NN stems from the
fact that it does not require specific training and is very robust with respect to the impact
of noisy data. Furthermore, the adoption of weighted k-NN is related with the choice
of P-norms for implementing the “and” operation, as P-norms combination rules require
values in [0,1].
• Interface Level. Interface agents are devoted to perform the feedback that originates from
the users –which can be exploited to improve the overall ability of discriminating relevant
from irrelevant inputs. To this end, PUB.MAS uses the k-NN solution previously described.
216
12 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

4.2 The prototype

Since our primary interest consists of classifying scientific articles for bioinformaticians or
biologists, a high-level is-a taxonomy has been extracted from the TAMBIS ontology (Baker
et al., 1999). A fragment of the taxonomy is depicted in Figure 5.

Fig. 5. A fragment of the adopted taxonomy

Through a suitable user interface (see Figure 6), the user can set the sources from which
publications will be extracted and the topics s/he is interested in. As for the digital archives,
the user can choose between BMC Bioinformatics and PubMed Central. As for the topics of
interest, the user can select one or more categories in accordance with the adopted taxonomy
and compose them in order to build her/his personal document collection. For instance, in the
example reported in Figure 6, the user queries the system on (cell AND nucleus) OR (organelle).
The search for relevant documents is activated by clicking on the Start Search button.
First, information agents devoted to handle Web sources extract the documents. Then, all
agents that embody a classifier trained on the selected topics are involved to perform text
categorization. Finally, the system supplies the user with the selected articles through suitable
interface agents (see Figure 7).

4.3 Experimental results

To assess the system, different kinds of tests have been performed, each aimed at highlighting
(and getting information about) a specific issue. First, we estimated the normalized confusion
matrix for each classifier belonging to the highest level of the taxonomy. Afterwards, we
tested the importance of defining user’s interests by resorting to a relaxation of the logical
operators. Finally, we assessed the solution devised for implementing user’s feedback, based
on the k-NN technique.
Tests have been performed using selected publications extracted from the BMC Bioinformatics
site and from the PubMed Central digital archive. Publications have been classified by an
expert of the domain according to the proposed taxonomy. For each item of the taxonomy, a
Retrieving and Categorizing
Retrieving and Categorizing Bioinformatics
Bioinformatics Publications Publications
through a MultiAgent System through a MultiAgent System 217
13

Fig. 6. The user interface

Fig. 7. A publication retrieved by PUB.MAS

set of about 100-150 articles has been selected to train the corresponding k-NN classifier, and
300-400 articles have been used to test it.
As for the estimation of the normalized confusion matrices (one for each classifier), we
fed classifiers with balanced sets of positive and negative examples. Given a classifier, we
performed several runs to obtain an averaged confusion matrix. Normalization has been
imposed row by row on the averaged confusion matrix. In particular, true negatives and false
positives are divided by the number of negative examples; conversely, the number of false
negatives and true positives are divided by the number of positive examples. In so doing,
218
14 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

we obtain an estimation of the conditional probability P(ĉ( x )|c( x )), where x is the input to be
classified, ĉ( x ) is the output of the classifier, and c( x ) is the category of x
To assess the impact of exploiting a taxonomy over precision and recall, we selected some
relevant samples of three classifiers in pipeline. They have been tested by imposing
randomly-selected relevant and irrelevant inputs, their ratio being set to 1/100, to better
approximate the expected behavior of the pipelines in real-world conditions. Averaging the
results obtained in all experiments in which a pipeline of three classifiers was involved, PF
allowed to reach an accuracy of 95%, a precision of 80%, and a recall of 44%.
Figure 8 and 9 report experimental results focused on average precision and recall,
respectively. Experimental results are compared with those derived theoretically. Let us
note that results show that the filtering effect of a pipeline is not negligible. In particular,
in presence of imbalanced inputs, a pipeline of three classifiers is able to counteract a lack of
equilibrium of about 10 irrelevant articles vs. one relevant article. Since, at least in principle,
the filtering activity goes with the power of the number of classifiers involved in the pipeline,
it is easy to verify that PF could also counteract a ratio between irrelevant and relevant articles
with an order of magnitude of hundreds or thousands, provided that the number of levels of
the underlying taxonomy is deep enough (at least 3-4).
Precision
1
experimental
theoretical

0.8

0.6
precision

0.4

0.2

0
0 1 2 3 4 5 6 7 8 9 10
k

Fig. 8. Average precision using three classifiers in pipeline

To test the validity of the horizontal composition mechanisms, the system has been tested on
20 selected users. The behavior of the system has been monitored over a two-week period
by conducting regular interviews with each user to estimate her/his satisfaction and the
correctness of the process. All users stated their satisfaction with the system after just one
or two days.
As for the user’s feedback, we obtained an improvement of about 0.3% on the precision of
the system by populating a k-NN classifier with examples selected as relevant by the system,
taking care of balancing true positives with false positives.

5. Conclusions
It becomes more and more difficult for Web users to search for, find, and select contents
according to their preferences. The same happens when researchers surf the Web searching
Retrieving and Categorizing
Retrieving and Categorizing Bioinformatics
Bioinformatics Publications Publications
through a MultiAgent System through a MultiAgent System 219
15

Recall
1
experimental
theoretical

0.8

0.6
recall

0.4

0.2

0
0 1 2 3 4 5 6 7 8 9 10
k

Fig. 9. Average recall using three classifiers in pipeline

for scientific publication of interest. Hence, supporting users in the task of dealing with
the information provided by the Web is a primary issue. In this chapter, we focused on
automatically retrieving and categorizing scientific publications and presented PUB.MAS, a
system devoted to provide personalized search results in terms of bioinformatics publications.
The system encompasses three main tasks: extracting scientific publications from online
repositories, classifying them using hierarchical text categorization, and providing suitable
feedback mechanisms. To validate the approach, we performed several experiments. Results
show that the approach is effective and can be adopted in practice.

6. Acknowledgements
This work has been supported by the Italian Ministry of Education under the projects: FIRB
ITALBIONET (RBPR05ZK2Z), Bioinformatics analysis applied to Populations Genetics
(RBIN064YAT 003) and SHIWA.

7. References
Addis, A., Armano, G. & Vargiu, E. (2008). From a generic multiagent architecture to
multiagent information retrieval systems, AT2AI-6, Sixth International Workshop, From
Agent Theory to Agent Implementation, pp. 3–9.
Addis, A., Armano, G. & Vargiu, E. (2010). Assessing progressive filtering to perform
hierarchical text categorization in presence of input imbalance, Proceedings of
International Conference on Knowledge Discovery and Information Retrieval (KDIR 2010),
pp. 14–23.
Addis, A., Armano, G. & Vargiu, E. (2011). A comparative experimental assessment
of a threshold selection algorithm in hierarchical text categorization, Advances in
Information Retrieval. The 33rd European Conference on Information Retrieval (ECIR 2011),
pp. 32–42.
Armano, G. (2009). On the progressive filtering approach to hierarchical text categorization,
Technical report, DIEE - University of Cagliari.
220
16 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

Armano, G., de Gemmis, M., Semeraro, G. & Vargiu, E. (2010). Intelligent Information Access,
Springer-Verlag, Studies in Computational Intelligence series.
Armano, G., Manconi, A. & Vargiu, E. (2007). A multiagent system for retrieving
bioinformatics publications from web sources, IEEE TRANSACTIONS ON
NANOBIOSCIENCE 6(2): 104–109. Special Session on GRID, Web Services, Software
Agents and Ontology Applications for Life Science.
Armano, G., Mancosu, G., Milanesi, L., Orro, A., Saba, M. & Vargiu, E. (2005). A
hybrid genetic-neural system for predicting protein secondary structure, BMC
BIOINFORMATICS 6 (suppl. 4):s3.
Armstrong, R., Freitag, D., Joachims, T. & Mitchell, T. (1995). Webwatcher: A learning
apprentice for the world wide web, AAAI Spring Symposium on Information Gathering,
pp. 6–12.
Baeza-Yates, R. A. & Ribeiro-Neto, B. (1999). Modern Information Retrieval, Addison-Wesley
Longman Publishing Co., Inc., Boston, MA, USA.
Baker, P., Goble, C., Bechhofer, S., Paton, N., Stevens, R. & Brass, A. (1999). An ontology for
bioinformatics applications, Bioinformatics 15(6): 510–520.
Bellifemine, F., Caire, G. & Greenwood, D. (2007). Developing Multi-Agent Systems with JADE
(Wiley Series in Agent Technology), John Wiley and Sons.
Bennett, P. N. & Nguyen, N. (2009). Refined experts: improving classification in large
taxonomies, SIGIR ’09: Proceedings of the 32nd international ACM SIGIR conference on
Research and development in information retrieval, ACM, New York, NY, USA, pp. 11–18.
Bleyer, M. (1998). Multi-Agent Systems for Information Retrieval on the World Wide Web, PhD
thesis, University of Ulm, Germany.
Bollacker, K. D., Lawrence, S. & Giles, C. L. (2000). Discovering relevant scientific literature
on the web, IEEE Intelligent Systems 15(2): 42–47.
Brank, J., Mladenić, D. & Grobelnik, M. (2010). Large-scale hierarchical text classification using
svm and coding matrices, Large-Scale Hierarchical Classification Workshop.
Ceci, M. & Malerba, D. (2003). Hierarchical classification of HTML documents with
WebClassII, in F. Sebastiani (ed.), Proceedings of ECIR-03, 25th European Conference on
Information Retrieval, Berlin Heidelberg NewYork: Springer, pp. 57–72.
Ceci, M. & Malerba, D. (2007). Classifying web documents in a hierarchy of categories: a
comprehensive study, Journal of Intelligent Information Systems 28(1): 37–78.
Corney, D. P. A., Buxton, B. F., Langdon, W. B. & Jones, D. T. (2004). Biorat: Extracting
biological information from full-length papers, Bioinformatics 20(17): 3206–3213.
Cost, W. & Salzberg, S. (1993). A weighted nearest neighbor algorithm for learning with
symbolic features, Machine Learning 10: 57–78.
Craven, M. & Kumlien, J. (1999). Constructing biological knowledge-bases by extracting
information from text sources, Proceedings of the Seventh International Conference on
Intelligent Systems for Molecular Biology, Germany, pp. 77–86.
D’Alessio, S., Murray, K. & Schiaffino, R. (2000). The effect of using hierarchical classifiers
in text categorization, Proceedings of of the 6th International Conference on Recherche
dŠInformation Assistée par Ordinateur (RIAO), pp. 302–313.
Delfs, R., Doms, A., Kozlenkov, A. & Schroeder, M. (2004). Gopubmed: ontology-based
literature search applied to gene ontology and pubmed, Proc. of German Bioinformatics
Conference, pp. 169–178.
Dumais, S. T. & Chen, H. (2000). Hierarchical classification of Web content, in N. J. Belkin,
P. Ingwersen & M.-K. Leong (eds), Proceedings of SIGIR-00, 23rd ACM International
Conference on Research and Development in Information Retrieval, ACM Press, New York,
US, Athens, GR, pp. 256–263.
Retrieving and Categorizing
Retrieving and Categorizing Bioinformatics
Bioinformatics Publications Publications
through a MultiAgent System through a MultiAgent System 221
17

Esuli, A., Fagni, T. & Sebastiani, F. (2008). Boosting multi-label hierarchical text categorization,
Inf. Retr. 11(4): 287–313.
Etzioni, O. & Weld, D. (1995). Intelligent agents on the internet: fact, fiction and forecast, IEEE
Expert 10(4): 44–49.
Friedman, C., Kra, P., Yu, H., Krauthammer, M. & Rzhetsky, A. (2001). Genies: a
natural-language processing system for the extraction of molecular pathways from
journal articles, Bioinformatics 17: S74–S82.
Gaussier, É., Goutte, C., Popat, K. & Chen, F. (2002). A hierarchical model for clustering
and categorising documents, in F. Crestani, M. Girolami & C. J. V. Rijsbergen (eds),
Proceedings of ECIR-02, 24th European Colloquium on Information Retrieval Research,
Springer Verlag, Heidelberg, DE, Glasgow, UK, pp. 229–247.
Golub, G. & Loan, C. V. (1996). Matrix Computations, Baltimore: The Johns Hopkins University
Press.
Greenwood, D. & Calisti, M. (2004). An automatic, bi-directional service integration gateway,
IEEE Systems, Cybernetics and Man Conference.
Huang, L. (2000). A survey on web information retrieval technologies, Technical report, ECSL.
Janssen, W. C. & Popat, K. (2003). Uplib: a universal personal digital library system, DocEng
’03: Proceedings of the 2003 ACM symposium on Document engineering, ACM Press, New
York, NY, USA, pp. 234–242.
Jirapanthong, W. & Sunetnanta, T. (2000). An xml-based multi-agents model for information
retrieval on www, Proceedings of the 4th National Computer Science and Engineering
Conference (NCSEC2000).
Kiritchenko, S., Matwin, S. & Famili, A. F. (2004). Hierarchical text categorization as a tool of
associating genes with gene ontology codes, 2nd European Workshop on Data Mining
and Text Mining for Bioinformatics, pp. 26–30.
Koller, D. & Sahami, M. (1997). Hierarchically classifying documents using very few
words, in D. H. Fisher (ed.), Proceedings of ICML-97, 14th International Conference on
Machine Learning, Morgan Kaufmann Publishers, San Francisco, US, Nashville, US,
pp. 170–178.
Lee, J. (1994). Properties of extended boolean models in information retrieval, Proceedings of
the 17th Annual international ACM SIGIR Conference on Research and Development in
information Retrieval, pp. 182–190.
Lewis, D. D. (1995). Evaluating and optimizing autonomous text classification systems, SIGIR
’95: Proceedings of the 18th annual international ACM SIGIR conference on Research and
development in information retrieval, ACM, New York, NY, USA, pp. 246–254.
Lieberman, H. (1995). Letizia: An agent that assists web browsing, in C. S. Mellish (ed.),
Proceedings of the Fourteenth International Joint Conference on Artificial Intelligence
(IJCAI-95), Morgan Kaufmann publishers Inc.: San Mateo, CA, USA, Montreal,
Quebec, Canada, pp. 924–929.
Mahdavi, F., Ismail, M. A. & Abdullah, N. (2009). Semi-automatic trend detection in
scholarly repository using semantic approach, Proceedings of World Academy of Science,
Engineering and Technology, pp. 224–226.
McCallum, A. K., Rosenfeld, R., Mitchell, T. M. & Ng, A. Y. (1998). Improving text classification
by shrinkage in a hierarchy of classes, in J. W. Shavlik (ed.), Proceedings of ICML-98,
15th International Conference on Machine Learning, Morgan Kaufmann Publishers, San
Francisco, US, Madison, US, pp. 359–367.
McNee, S. M., Albert, I., Cosley, D., Gopalkrishnan, P., Lam, S. K., Rashid, A. M., Konstan, J. A.
& Riedl, J. (2002). On the recommending of citations for research papers, Proceedings
222
18 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

of the 2002 ACM conference on Computer supported cooperative work, CSCW ’02, ACM,
New York, NY, USA, pp. 116–125.
Mladenić, D. & Grobelnik, M. (1998). Feature selection for classification based on
text hierarchy, Text and the Web, Conference on Automated Learning and Discovery
CONALD-98.
Porter, M. (1980). An algorithm for suffix stripping, Program 14(3): 130–137.
Ramu, C. (2001). SIR: a simple indexing and retrieval system for biological flat file databases,
Bioinformatics 17(8): 756–758.
Ramu, C. (2003). SIRW: A web server for the simple indexing and retrieval system
that combines sequence motif searches with keyword searches., Nucleic Acids Res
31(13): 3771–3774.
Rocco, D. & Critchlow, T. (2003). Automatic discovery and classification of bioinformatics web
sources, Bioinformatics 19(15): 1927–1933.
Rousu, J., Saunders, C., Szedmak, S. & Shawe-Taylor, J. (2005). Learning hierarchical
multi-category text classification models, ICML ’05: Proceedings of the 22nd
international conference on Machine learning, ACM, New York, NY, USA, pp. 744–751.
Ruiz, M. E. & Srinivasan, P. (2002). Hierarchical text categorization using neural networks,
Information Retrieval 5(1): 87–118.
Shaban, K., Basir, O. & Kamel, M. (2004). Team consensus in web multi-agents information
retrieval system, Team Consensus in Web Multi-agents Information Retrieval System,
pp. 68–73.
Sheth, B. & Maes, P. (1993). Evolving agents for personalized information filtering, Proceedings
of the 9th Conference on Artificial Intelligence for Applications (CAIA-93), pp. 345–352.
Sun, A. & Lim, E. (2001). Hierarchical text classification and evaluation, ICDM ’01: Proceedings
of the 2001 IEEE International Conference on Data Mining, IEEE Computer Society,
Washington, DC, USA, pp. 521–528.
Sycara, K., Paolucci, M., van Velsen, M. & Giampapa, J. (2001). The RETSINA
MAS infrastructure, Technical Report CMU-RI-TR-01-05, Robotics Institute Technical
Report, Carnegie Mellon.
Tanabe, L., Scherf, U., Smith, L. H., Lee, J. K., Hunter, L. & Weinstein, J. N. (1999). Medminer:
an internet text-mining tool for biomedical information, BioTechniques 27: 1210–1217.
Weigend, A. S., Wiener, E. D. & Pedersen, J. O. (1999). Exploiting hierarchy in text
categorization, Information Retrieval 1(3): 193–216.
Wooldridge, M. J. & Jennings, N. R. (1995). Agent Theories, Architectures, and Languages:
A Survey, in M. J. Wooldridge & N. R. Jennings (eds), Workshop on Agent Theories,
Architectures & Languages (ECAI’94), Vol. 890 of Lecture Notes in Artificial Intelligence,
Springer-Verlag, Amsterdam, The Netherlands, pp. 1–22.
Wu, F., Zhang, J. & Honavar, V. (2005). Learning classifiers using hierarchically structured
class taxonomies, Proc. of the Symp. on Abstraction, Reformulation, and Approximation,
Vol. 3607, Springer Verlag, pp. 313–320.
 0
11

GRID Computing and Computational Immunology

Ferdinando Chiacchio and Francesco Pappalardo
University of Catania
Italy

1. Introduction
Biological function emerges from the interaction of processes acting across a range
of spatio-temporal scales. Therefore understanding disease and developing potential
therapeutic strategies requires studies that bridge across multiple levels. This requires a
systems biology approach and the tools used must be based on effective mathematical
algorithms and built by combining experimental and theoretical approaches, addressing
concrete problems and clearly defined questions.
Technological revolutions in both biotechnology and information technology have produced
enormous amounts of data and are accelerating the extension of our knowledge of biological
systems. These advances are changing the way biomedical research, development and
applications are done. Mathematical and computational models are increasingly used to help
interpret data produced by high-throughput genomics and proteomics projects, and through
advances in instrumentation. Advanced applications of computer models that enable the
simulation of biological processes are used to generate hypotheses and plan experiments.
Computational modeling of immune processes has emerged as a major support area for
immunology and vaccinology research. Computational models have been developed for the
simulation of immune processes at the molecular, cellular, and system levels. Computer
models are used to complement or replace actual testing or experimentation. They are
commonly used in situations where experimentation is expensive, dangerous, or impossible
to perform. Models of the immune system fall into two categories:
1. molecular interactions (such as peptide binding to receptors), and
2. system-level interactions (such as models of immune response, or cellular models of
immune system).
Until recently, models of molecular level immune processes have been successful
in supporting immunology research, such as antigen processing and presentation.
Computational models of the complete immune system have mainly been developed in the
domain of theoretical immunology and were used to offer possible explanations of the overall
function of the immune system, but were usually not applied in practice.
Newer computational tools that focus on immune interactions include numerous methods
used for the mapping of T-cell epitopes (targets of immune responses). Computational
methods for the simulation of antigen processing include prediction of proteasomal
cleavage and peptide binding to transporters associated with antigen processing (TAP).
The basic methods for the prediction of antigen processing and presentation have been
extended to more sophisticated computational models for identification of promiscuous
224
2 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

Human Leukocyte Antigens (HLA)-restricted T-cell epitopes, identification of Major

Histocompatibility Complex (MHC) supermotifs and T-cell epitope hot-spots.
Computational models of cellular or higher level processes or interactions have a longer
history than those focusing on molecular processes, but are also more complex. These
include models of T- cell responses to viruses, analysis of MHC diversity under host-pathogen
co-evolution, B-cell maturation, or even the dynamic model of the immune system that can
simulate both cellular and humoral immune responses.
The main problems that prevented the use of these models in practical applications,
such as design of vaccines and optimization of immunization regimens are: a) large
combinatorial complexity of the human immune system that could not be supported
by existing computational infrastructures, b) lack of understanding of specific molecular
interactions that resulted in an idealization of representation of molecular interactions as
binary strings, and c) lack of experimental model data and correlation of model parameters to
real-life measurements. Recent developments provide remedies to these problems and we are
in the position to address each of these issues.
Grid computing brought powerful computational infrastructure and the capacity that can
match the complexity of the real human immune system. Models of molecular interactions
have reached high accuracy and we are routinely using prediction methods of antigen
processing and presentation to identify the best targets for vaccine constructs. Finally,
experimental models of immune responses to tumors and infectious diseases have been
successfully modeled computationally.
In this work, we present two different experiences in which we show successful stories
in using computational immunology approaches that have been implemented using GRID
infrastructure:
• modeling atherosclerosis, a disease affecting arterial blood vessels, that is one of most
common disease of the developed countries;
• a biological optimization problem on the Grid, i.e. an optimal protocol search algorithm
based on Simulated Annealing (SA) capable to suggest optimal Triplex vaccine dosage
used against mammary carcinoma induced lung metastasis.

2. Modeling atherogenesis using GRID computing

Atherosclerosis is, in large part, due to the deposition of low density lipoproteins (LDLs),
i.e., plasma proteins carrying cholesterol and triglycerides, that determine the formation of
multiple plaques within the arteries (Hanson, 2002; Ross, 1999). The origin of atherosclerosis
is still not fully understood. However there are risk factors which increase the probability
of developing atherosclerosis in humans. Some of these risk factors are beyond a person’s
control (smoking, obesity), others seem to have genetic origin (familial hypercholesterolemia,
diabetes, hypertension) (Romero-Corral et al., 2006). Common denominator in all the form of
atherosclerosis is the elevated level of LDL, which is subject to oxidation becoming oxidized
low density lipoproteins (ox-LDL), that promotes an inflammatory response and immune
activation in the artery walls (Berliner et al., 1996). The formation of atherosclerotic plaques in
the artery reduces both the internal diameter of vessels and the blood flux leading to a number
of serious pathologies (Vinereanu, 2006). Early studies demonstrated that ox-LDL can induce
activation of monocytes/macrophages, endothelial cells and T cells. Ox-LDLs engulfed by
macrophages form the so called foam cells (Steinberg, 1997). These cells represent the nucleus
GRID Computing
GRID Computing and Immunology
and Computational Computational Immunology 2253

of the plaques formation. Ox-LDL promotes also immune activation of B cells inducing the
production of specific anti ox-LDL antibody (OLAB).
Atherosclerosis and their anatomical consequences cause severe problems. Stenosis
(narrowing) and aneurysm of the artery are chronic, slowly progressing and cumulative
effects indicating the progression of atherosclerotic disease. In both case the result is an
insufficient blood supply to the organ fed by the artery. Most commonly, soft plaque suddenly
ruptures, causes the formation of a thrombus that will rapidly slow or stop blood flow, leading
to death of the tissues fed by the artery. This catastrophic event is called infarction and is not
predictable. The most common event is thrombosis of the coronary artery causing infarction
(a heart attack): However, since atherosclerosis is a body wide process, similar events also
occur in the arteries of the brain (stroke attack), intestines, kidneys, etc. Those atherosclerosis
associated events often cause of dead or serious invalidating diseases and require preventive
treatments. Vaccine research for atherosclerosis is a hot pharmaceutical topic.
Recently we proposed a model based on the Agent Based Model (ABM) paradigm
(Pappalardo et al., 2008) which reproduces clinical and laboratory parameters associated to
atherogenesis. The model and its computer implementation (SimAthero simulator) considers
all the relevant variables that play an important role in atherogenesis and its induced immune
response, i.e., LDL, ox-LDL, OLAB, chitotriosidase and the foam cells generated in the artery
wall.
We present three different situations over a time scale of two years. The standard normal
patients where no foam cells are formed; patients having high level of LDL but who delay to
apply appropriate treatments and finally patients who may have many events of high level of
LDL but takes immediately appropriate treatments.

2.1 Description of the model

2.1.1 The biological scenario
Exogenous and endogenous factors induce in humans a very small, first oxidative process
of blood circulating native LDLs (minimally modified LDLs or mm-LDLs). In endothelium
mm-LDLs are extensively oxidized from intracellular oxidative products and then recognized
by the macrophage scavenger receptor. High level and persistent in time LDLs lead to
macrophages engulfment and their transformation in foam cells. Contrary, low level of
LDLs and their oxidized fraction, lead to the internalization of the oxidized low density
lipoproteins and subsequent presentation by major histocompatibility complex class II at the
macrophages surface. Recognition of ox-LDL by macrophages and naive B cells, leads, by
T helper lymphocytes cooperation, to the activation of humoral response and production
of OLAB. When the OLAB/ox-LDL immune complexes are generated in the vascular wall,
the macrophages catch them by the Fc receptor or via phagocytosis and destroy ox-LDL in
the lysosome system. During this process, the activated macrophage releases chitotriosidase
enzyme, that is then used as a marker of macrophage activation.

2.1.2 The model

To describe the above scenario one needs to include all the crucial entities (cells, molecules,
adjuvants, cytokines, interactions) that biologists and medical doctors recognize as relevant in
the game. The model described in (Pappalardo et al., 2008) contains entities and interactions
which both biologist and MD considered relevant to describe the process.
Atherosclerosis is a very complex phenomenon which involves many components some of
them not fully understood. In the present version of the simulator we considered only in
226
4 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

the immune system processes that control the atherogenesis. These processes may occur in
immune system organs like lymph nodes or locally in the artery endothelium. To describe the
Immune processes we considered both cellular and molecular entities.
Cellular entities can take up a state from a certain set of suitable states and their dynamics
is realized by means of state-changes. A state change takes place when a cell interacts with
another cell or with a molecule or both of them. We considered the relevant lymphocytes
that play a role in the atherogenesis-immune system response, B lymphocytes and helper T
lymphocytes. Monocytes are represented as well and we take care of macrophages. Specific
entities involved in atherogenesis are present in the model: low density lipoproteins, oxidized
low density lipoproteins, foam cells, auto antibodies anti oxidized low density lipoproteins
and chitotriosidase enzyme. Cytotoxic T lymphocytes are not taken into consideration
because they are not involved in the immune response (only humoral response is present
during atherogenesis).
Molecular entities The model distinguishes between simple small molecules like interleukins
or signaling molecules in general and more complex molecules like immunoglobulins and
antigens, for which we need to represent the specificity. We only represent interleukin 2
that is necessary for the development of T cell immunologic memory, one of the unique
characteristics of the immune system, which depends upon the expansion of the number
and function of antigen-selected T cell clones. For what is related to the immunoglobulins,
we represent only type immunoglobulins of class G (IgG). This just because at the actual
state we don’t need to represent other classes of Ig and because IgG is the most versatile
immunoglobulin since it is capable of carrying out all of the functions of immunoglobulins
molecules. Moreover IgG is the major immunoglobulin in serum (75% of serum Ig is IgG) and
IgG is the major Ig in extra vascular spaces.
The actual model does not consider multi-compartments processes and mimics all processes
in a virtual region in which all interactions take place. Our physical space is therefore
represented by a 2D domain bounded by two opposite rigid walls and left and right periodic
boundaries. This biological knowledge is represented using an ABM technique. This allows to
describe, in a defined space, the immune system entities with their different biological states
and the interactions between different entities. The system evolution in space and in time
is generated from the interactions and diffusion of the different entities. Compared to the
complexity of the real biological system our model is still very naive and it can be extended
in many aspects. However, the model is sufficiently complete to describe the major aspects of
the atherogenesis-immune system response phenomenon.
The computer implementation of the model (SimAthero hereafter) has two main classes of
parameters: the first one refers to values known from standard immunology literature (Abbas
et al., 2007; Celada et al., 1996; Goldspy et al., 2000; Klimov et al., 1999); the second one collects
all the parameters with unknown values which we arbitrarily set to plausible values after
performing a series of tests (tuning phase).
The simulator takes care of the main interactions that happens during an immune response
against atherogenesis.
Physical proximity is modeled through the concept of lattice-site. All interactions among cells
and molecules take place within a lattice-site in a single time step, so that there is no correlation
between entities residing on different sites at a fixed time. The simulation space is represented
as a L × L hexagonal (or triangular) lattice (six neighbors), with periodic boundary conditions
to the left and right side, while the top and bottom are represented by rigid walls. All
entities are allowed to move with uniform probability between neighboring lattices in the
GRID Computing
GRID Computing and Immunology
and Computational Computational Immunology 2275

grid with equal diffusion coefficient. In the present release of the simulator chemotaxis is not
implemented.
LDLs values can be fixed in order to simulate different patients both in normolipidic condition
and in hypercholesterolemic condition. The same applies to ox-LDLs. However human habits
change with time and personal life style. A normolipidic patient can change its attitude
becoming an hypercholesterolemic one and vice versa. For this reason we allow the simulator
to accept varying life style conditions and preventive actions to decrease risk factors.

2.2 Results
The model described include the possibility of mimicking biological diversity between
patients. The general behavior of a class of virtual patients arise from the results of a suitable
set of patients, i.e., the mean values of many runs of the simulator of different patients under
the same conditions. The class of virtual patients described by the model were tuned against
human data data collected by (Brizzi et al., 2003; 2004) where different conditions, normal and
hypercholesterolemic diabetic patients were analyzed.
In this section we analyze the behavior of the same patients in three broad class of clinical
conditions to show how SimAthero could be used in order to analyze and predict the effects
of various LDL levels in blood. The normal patient simulation is used as control experiment
for the other simulations. The differences among these four clinical conditions depend on
the LDL level and the time interval which occurs between the time in which concentration of
LDL rise above normal level and the time in which the patient take appropriate treatments
(lifestyle o drug) to reduce it to normal level.
Jobs were launched using the SimAthero simulator on the COMETA Grid. The
submission process was done through the web interface of the ImmunoGrid project
(http://www.immunogrid.eu).
A patient with a LDL level of roughly 950-970 ng/μl of blood is considered normal in clinical
practice and he has with very low risk of atheroslerotique plaque. The results of SimAthero
for a virtual normal patient (Figure 1) show that he will not support the formation of foam
cells and, as a consequence, the beginning of atherogenesis process is absent.
We then simulated a scenario in which a patient, due to several reasons (diet, life style,
oxidative agents and so on, so forth) leads its LDL level at 1300 ng/μl, taking it up to 1700
ng/mul. Looking at figure 2 one can observe about 12 foam cells per μl at the end of in silico
follow up. This leads to a small atherogenesis process due to the high level of LDL.
Lastly (figure 3), we analyzed a virtual patient that initially takes its LDL level to small peaks,
causing no damage. After that, he takes its LDL level to a hypercholesterolemic behavior,
generating a small damage, as shown. This shows that small LDL alteration are completely
taken under control by the normal behavior of the organism, but high LDL peaks lead to foam
cells formation and then to the beginning of the atherogenesis process.

2.3 Remarks on atherosclerosis modeling using GRID computing

Atherosclerosis is a pathology where the immune control plays a relevant role. We presented
studies on the increased atherosclerosis risk using an ABM model of atherogenesis and its
induced immune system response in humans. Very few mathematical models (Cobbold et
al., 2002; Ibragimov et al., 2005) and (to our best knowledge) no computational models of
atherogenesis have been developed to date.
It is well known that the major risk in atherosclerosis is persistent high level of LDL
concentration. However it is not known if short period of high LDL concentration can cause
228
6 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

Fig. 1. Simulation results of a virtual patient with level of LDL considered normal. The
follow-up period is two years. The figure shows that foam cells formation is absent in this
patient.

Fig. 2. Simulation results of a virtual patient with level of LDL considered at high risk. The
follow-up period is two years. The figure shows that foam cells formation is present, leading
to an atherogenesis process.
GRID Computing
GRID Computing and Immunology
and Computational Computational Immunology 2297

Fig. 3. Simulation results of a virtual patient with level of LDL considered quasi-normal at
the beginning and then at high risk. The follow-up period is two years. The figure shows that
foam cells formation is negligible in the first time, but becomes important soon after.

irreversible damage and if reduction of the LDL concentration (either by lfe style or drug) can
drastically or partially reduce the already acquired risk.
Using an ABM cellular model describing the initial phase of plaque formation (atherogenesis)
we are able to simulate the effect of life style which increases the risk of atherosclerosis.

3. Vaccine dosage optimization using GRID

As a second application, we show an example of how a physiological model in conjunction with
some optimization techniques can be used to speed-up the research of an opitmal vaccination
schedule for an immunopreventive vaccine, using the powerful of the GRID computing
paradigm.
A vaccination schedule is usually designed empirically using a combination of immunological
knowledge, vaccinological experience from previous endeavors, and practical constraints. In
subsequent trials, the schedule of vaccinations is then renewed on the basis of the protection
elicited in the first batch of subjects and their immunological responses e.g. kinetics of
antibody titers, cell mediated response, etc. The problem of defining optimal schedules is
particularly important in cancer immunopreventive approaches, which requires a sequence
of vaccine administrations to keep a high level of protective immunity against the constant
generation of cancer cells over very long periods, ideally for the entire lifetime of the host.
The Triplex vaccine represents a clear example of such immunopreventive approaches. It has
been designed to raise the immune response against the breast cancer for the prevention of
the mammary carcinoma formation in HER-2/Neu mouse models using a Chronic schedule in
a follow up time between 52 and 57 weeks.
However it is not known if the Chronic schedule schedule is minimal, i.e. if it can guarantee
survival for the mice population avoiding unnecessary/redoundant vaccine administrations.
230
8 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

Shorter heuristic protocols failed, in in vivo experiments, in fulfilling this requirement, but
between the Chronic and the shorter schedules there is still a huge number of possibilities
which remain yet unexplored.
The SimTriplex () is a physiological computational model developed with the aim to answer
to this question. It demonstrated able to reproduce in silico the in vivo Immune System (IS) -
breast cancer competition elicted by the Triplex vaccine.
Optimal search strategy was biologically guided. Considering that Chronic proved to be
effective for tumor control, the optimal search tried to find a protocol with minimum number
of vaccine administrations able to reproduce, in silico, the time evolution of the chronic
schedule. This strategy was used in the GA optimal search. It is well known that GA are
slowly converging algorithms; the GA optimal search required several days using 32 nodes of
an High Performance Computing infrastructure.
To this end we decided to investigate the applicability of Simulated Annealing (SA), a global
optimization algorithm, widely tested and known for its computational speed and ability
to achieve optimal solutions. (Interested readers can found an extended description of SA
algorithm in (Van Laarhoven at al., 1987)). The combination of the Simulated Annealing
algorithm with biologically driven heuristic strategies, leads to a much faster algorithm and
better results for the optimal vaccination schedule problem for Triplex vaccine. In this context
we remark how the COMETA grid infrastructure demostrated an excellent framework for
protocol search and validation. We first executed the SA algorithm on a subset of the virtual
mice population using MPI jobs. We therefore checked the protocol quality calculating (with
simple jobs) the survivals of the entire population.

3.1 The algorithm

The work done by Kirkpatrick (Kirkpatrick et al., 1983) opened the path to a deep analogy
between Statistical Mechanics (the behavior of systems with many degrees of freedom in
thermal equilibrium at a finite temperature) and Combinatorial Optimization (the method
of finding the minimum, if any, of a given function with respect to many parameters). There
is a close similarity, indeed, between the procedure of annealing in solids and the framework
required for optimization of complex systems.

3.1.1 The optimal vaccination schedule search problem

The SimTriplex model (Pappalardo et al., 2005) has been created with the aim to mimic the
behavior of the immune system stimulated by the Triplex vaccine. It simulates all the major
interactions of cells and molecules of the immune system in vaccinated as well as naive
HER-2/neu mice. In silico experiments showed an excellent agreement with the in vivo ones.
As previously said, a protocol is said to be optimal if it can maintain efficacy with a minimum
number of vaccine administrations. As in standard drug administration, the vaccination
protocol must assure survival for a high percentages of patients. Schedule design is usually
achieved using medical consensus, i.e. a public statement on a particular aspect of medical
knowledge available at the time it was written, and that is generally agreed upon as the
evidence-based, state-of-the-art (or state-of-science) knowledge by a representative group of
experts in that area. Our goal is to improve medical consensus, helping biologists in design
vaccine protocols with simulators and optimization techniques.
Let us consider a time interval [0, T ], in which we study the action of the vaccine on a set of
virtual mice S. We then discretize the given time interval in N − 1 equally spaced subintervals
of width Δt (=8 hours), i.e. {t1 = 0, t2 , . . . , ti , . . . , t N = T }.
GRID Computing
GRID Computing and Immunology
and Computational Computational Immunology 2319

Let x = { x1 , x2 , . . . , xi , . . . x N } be a binary vector representing a vaccine schedule, where

xi = 0/1 means respectively administration/no administration of the vaccine at time ti . The
number of vaccine administrations is given by n = ∑iN=1 xi . With T = 400 days, and Δt = 8
hours the search space D has cardinality 2400 , excluding any exhaustive search.
One of the wet biologists requirements imposes no more than two administrations a week
(monday and thursday) because this is already considered a very intensive schedule from an
immunological point of view. This reduces the cardinality of the search space D, from 2400
(∼ 10120 ) to 2114 (∼ 1034 ). We still have no chance for an exhaustive search.
The time of the carcinoma in situ (CIS) formation is computed through SimTriplex simulator.
It is defined by τ (x, λ j ), which is a function of the vaccination schedule x administered to the
mouse j ∈ S and a parameter λ j which represents the biological diversity. The vaccine will be
obviously effective if τ ≥ T.
As pointed out in § 1, any optimal protocol should try to reproduce , in silico, the chronic
time evolution of cancer cells. This leads us to the need to use two thresholds on the allowed
maximum number of cancer cells.
We also note here that, due to biological variability, a schedule found for a single mouse is not
immunologically effective as it usually does not reveal able to protect high percentages of the
treated patients. Having this in mind, we formulate our optimization problem as follows.
Let { j1 , j2 , . . . , jm } ⊂ S, with m = 8, a random chosen subset of in silico mice, the problem is
defined as:
⎧
⎪
⎪ τ (x̄, λ j1 ) = max(τ (x, λ j1 ))
⎪
⎪
⎪
⎪ τ (x̄, λ j2 ) = max(τ (x, λ j2 ))
⎪
⎪
⎪
⎪ .. ..
⎪
⎪ . .
⎨
τ (x̄, λ jm ) = max(τ (x, λ jm )) (1)
⎪
⎪ n(x̄) = min(n(x))
⎪
⎪
⎪
⎪
⎪
⎪ subject to:
⎪
⎪ M1 (x) ≤ γ1 , t ∈ [0, Tin ]
⎪
⎪
⎩
M2 (x) ≤ γ2 , t ∈ [ Tin , T ]
where M1 (x) and M2 (x) are the maximum number of cancer cells in [0, Tin ] (cellular-mediated
controlled phase) and in [ Tin , T ] (humoral-mediated controlled phase) respectively, and Tin ∼
T/3, while γ1 and γ2 represent cancer cells threshold in [0, Tin ] and in [ Tin , T ], respectively.
We deal with a multi-objective discrete and constrained optimization problem.
We modified this last formulation of the problem grouping all the τ (x, λ jh ) (h = 1, . . . , m) by a
proper statistical indicator. We chose the harmonic mean H of the survivals:
m
H (x, λ j1 , . . . , λ jm ) = m (2)
1
∑ τ (x, λ j )
i =1 i

since it is very frequently used when statistic measurements of time are involved.
Therefore, the system (1) translates as:
⎧
⎪
⎪ H (x̄, λ) = max ( H (x, λ))
⎪
⎪
⎨ n(x̄) = min(n(x))
subject to: (3)
⎪
⎪
⎪
⎪ M1 (x) ≤ γ1 , t ∈ [0, Tin ]
⎩
M2 (x) ≤ γ2 , t ∈ [ Tin , T ]
232
10 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

with λ = (λ j1 , . . . , λ jm ).

3.1.2 Simulated annealing

An acclaimed Monte Carlo method, commonly referred as the Metropolis criterion, has been
designed by Metropolis (Metropolis et al., 1953) with the aim to mimic the evolution of
the complex systems towards equilibrium at a fixed temperature. This method randomly
perturbates the position of the particles of a solid modifying its configuration. If the energy
difference, ΔE, between the unperturbed and perturbed configurations is negative, the new
configuration has lower energy and it’s considered as the new one. Otherwise the probability
of acceptance of the new configuration is given by the Boltzmann factor (which expresses the
"probability" of a state with energy E relative to the probability of a state of zero energy).
After a large number of perturbations the probability distribution of the states should
approach the Boltzmann distribution.
The Metropolis algorithm can also be used in combinatorial optimization problems to
generate sequences of configurations of a system using a cost function C and control parameter
c respectively as the energy and temperature in the physical annealing.
The SA algorithm can be represented as a sequence of Metropolis algorithms evaluated at a
sequence of decreasing values of the control parameter c. A generalization of the method is
given as follows: a generation mechanism is defined so that, given a configuration i, another
configuration j can be obtained by choosing at random an element in the neighborhood of i.
If ΔCij = C ( j) − C (i ) ≤ 0, then the probability that the next configuration is j is given by 1; if
ΔCij > 0 the probability is given by e−ΔCij /c (Metropolis criterion).
This process is continued until the probability distribution, P, of the configurations
approaches the Boltzmann distribution, which translates as:

1
P{con f iguration = i } = · e−C(i)/c (4)
Q(c)

where Q(c) is a normalization constant depending on the control parameter c.

The control parameter is then lowered in steps, with the system being allowed to reach
equilibrium by generating a chain of configurations at each step. The algorithm stops for
some fixed value of the control parameter c where virtually no deteriorations can be accepted
anymore.
At the end the final frozen configuration is assumed as a solution of the problem.

3.1.3 Implementation
In our in silico experiment, we select a population of 200 virtual mice and a simple random
sample of k = 8 mice. To use the SA algorithm for the optimal vaccine schedule search
problem we tried to define the SA relevant concepts (the solid configuration, the temperature,
the energy and the semi-equilibrium condition) in terms of vaccine protocol and to describe
the main protocol elements (the number of injections, the mean survival age, the time
distribution of injections) in terms of a cooling process.
As previously said, we describe any candidate protocol as a binary vector x of cardinality
V = 114. The total number of vaccine administrations n and the total number of possible
schedules with n vaccine administrations M are given by:
GRID Computing
GRID Computing and Immunology
and Computational Computational Immunology 233
11

V
n= ∑ xnl ( j)
j =1

M = V!/[n! (V − n)!]

The configuration distribution is defined by xin , nin , and the initial energy Ein as defined later
on.
The temperature is slowly but constantly lowered to reach a state with minimum energy.
We coupled this entity with n, the number of vaccine administrations of a semi-equilibrium
configuration.
At a given temperature, a semi-equilibrium configuration is reached when its Energy is
minimal. Since we want to maximize survival times of a mice sample set, the concept of
Energy can be easily associated with the harmonic mean H of the survival times τi , H (x̄, λ) (i.e.
E ∝ H). As a matter of fact H decreases when the cumulative survival time of the sample
increases, in perfect accord with the energy definition.
The perturbation of a protocol has been initially implemented as random 1 bits reallocation
(Pennisi et al., 2008b). This perturbation has been heuristically improved using biological
knowledge. As we want to optimize mice survival, scheduling many vaccine administrations
j j +1
after the death of a mouse makes nonsense. So, in moving from xi to xi , we improve random
bits reallocation also moving some “1” at a suitable time t < min{τi } , i = 1, k.
The SA algorithm for protocol optimization. i) starts from a randomly chosen initial vaccine
distribution and finds the initial semi-equilibrium configuration nin , xlninin , Enlinin
ii) Decrease the number of injections of 1 unit; iii) find a semi-equilibrium configuration xi ,
Ei according to Metropolis algorithm; iv) cycle on (ii). The algorithm stops when, once the
algorithm control parameter, i.e. the number of vaccine administrations, is decreased from
n to n − 1, the Metropolis algorithm is not able to find a semi-equilibrium configuration, i.e.
an acceptable value of survivals, in λ iterations. The accepted protocol is the last found at
temperature n.

3.2 Computational results and conclusions

In silico optimal protocol search is a two-step process: search and validation. During the search
step, the optimization tecnique tries to find an optimal protocol. As pointed out in section
3.1.1, optimal search stragies have to be executed on a representative subset of the population
in order to guarantee significative survival percentages for the population.
The search technique therefore needs to test simultaneously every candidate protocol on
the mice subset to compute its fitness function. This process usually requires a relatively
small number of nodes with high communication throughputs, representing a typical massive
parrallel application. In our case it has been implemented using the MPI (Message passing
Interface) libraries.
Validation represents a completely different process. Here the protocol found by the search
technique is tested over the entire population to compute mean survival rates, requiring an
high number of CPUs with almost no need of communication. The “search and validation”
process is represented in figure 4.
In this context the Cometa grid revealed itself as an excellent tool for our needs. It
demonstrated to be highly flexible, giving us the opportunity to execute these so-different
processes on the same infrastructure. We only needed to define the first process as an MPI
234
12 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

Fig. 4. The “in silico” search and validation optimal protocol representation.

job and the second as a sequence of simple jobs in a trasparent way, without worrying of the
hidden the architectures behind the job submission interface.
To compare the results with those obtained using GA optimization technique (Lollini et al.,
2006), we executed the SA algorithm on the same 8 random selected virtual mice sample used
by GA. The protocol was then validated on the same population set (200 virtual mice).
The SA in silico tumor free percentages of the mice population show no substantial difference
with GA results (87% for GA vs 86, 5% for SA). Figure 5 shows the mean number of cancer
cells, computed on the 200-mice set, for the GA-protocol (up lhs) and the SA-protocol (down
lhs). Only the SA-protocol is able to totally fulfill the safety threshold conditions (shown in
red).
Moreover the SA algorithm required a computational effort of about 2 hrs on a 8 processor
unit to find a protocol with 37 vaccine administrations, showing speed-up factor of ∼ 1.4 · 102
in respect to previous GA experiments (Pennisi et al., 2008a).
GRID Computing
GRID Computing and Immunology
and Computational Computational Immunology 235
13

Cancer Cells Tumor Associated Antigens

6 6
10 10
Error Error
105 105
number of entities

number of entities
104 104
3
10 103
102 102
101 101
0 0
10 10
0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400
days days

Cancer Cells Tumor Associated Antigens

6 6
10 10
Error Error
5 5
10 10
number of entities

number of entities

4 4
10 10
3 3
10 10
2 2
10 10
1 1
10 10
100 100
0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400
days days

Fig. 5. Cancer cells behaviors and thresholds in GA (top) and SA (bottom). Small vertical
bars on the x-axis represent vaccine administration times. Broken-line graphs on the lhs
represent safety thresholds.

4. Conclusions
Mathematical models and Cellular Automata are mostly used for cellular level simulations,
while a range of statistical modeling applications are suitable for the analysis of sequences at
molecular level of the immune system.
Grid computing technology brings the possibility of simulating the immune system at the
natural scale. In our opinion, a Grid solution is only as good as the interface provided to
the users. We presented two successful stories in which we have shown successful stories
in using computational immunology approaches that have been implemented using GRID
infrastructure.
We like to conclude by stressing the interdisciplinary nature of the experiences described
above and by noting that the contribution of life scientists needs to go beyond the only data
supply, as it is extremely important in defining the biological scenario and ultimately construct
a robust and validated mathematical or computational model. Only through a common effort
of life and computer scientists it is possible to turn software into a valuable tools in life
sciences.
236
14 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

5. Acknowledgments
This work makes use of results produced by the PI2S2 Project managed by the Consorzio
COMETA, a project co-funded by the Italian Ministry of University and Research (MIUR)
within the Piano Operativo Nazionale "Ricerca Scientifica, Sviluppo Tecnologico, Alta
Formazione" (PON 2000-2006). More information is available at: http://www.pi2s2.it and
http://www.consorzio-cometa.it.

6. References
Abbas, A.K.; Lichtman, A.H.; Pillai, S. (2007) Cellular and molecular immunology, Saunders,
6th edition.
Artieda, M.; Cenarro, A.; Gañàn, A.; Jericó, I.; Gonzalvo, C.; Casado, J.M.; Vitoria, I.; Puzo,
J.; Pocoví,M.; Civeira,F. (2003) Serum chitotriosidase activity is increased in subjects
with atherosclerosis disease, Arterioscler. Thromb. Vasc. Biol., 23, 1645-1652.
Artieda, M.; Cenarro, A.; Gańàn, A.; Lukic, A.; Moreno, E.; Puzo, J.; Pocoví, M.; Civeira, F.
(2007) Serum chitotriosidase activity, a marker of activated macrophages, predicts
new cardiovascular events independently of C-Reactive Protein, Cardiology, 108,
297-306.
Berliner, J.A.; Heinecke, J.W. (1996) The role of oxidized lipoproteins in atherogenesis, Free
Radic. Biol. Med., 20(5), 707-727.
Brizzi, P.; Isaja, T.; D’Agata, A.; Malaguarnera, A.; Malaguarnera, M.; Musumeci, S.
(2002) Oxidized LDL antibodies (OLAB) in patients with beta-thalassemia major, J.
Atheroscler. Thromb., 9(3), 139-144.
Brizzi, P.; Tonolo, G.; Carusillo, F.; Malaguarnera, M.; Maioli, M.; Musumeci, S. (2003) Plasma
Lipid Composition and LDL Oxidation, Clin. Chem. Lab. Med., 41(1), 56-60.
Brizzi, P.; Tonolo, G.; Bertrand, G.; Carusillo, F.; Severino, C.; Maioli, M.; Malaguarnera,
L.; Musumeci, S. (2004) Autoantibodies against oxidized low-density lipoprotein
(ox-LDL) and LDL oxidation status, Clin. Chem. Lab. Med., 42(2), 164-170.
Celada, F.; Seiden, P.E. (1996) Affinity maturation and hypermutation in a simulation of the
humoral immune response, Eur. J. Immunol., 26, 1350.
Cobbold, C.A.; Sherratt, J.A.; Maxwell, S.R.J. (2002) Lipoprotein Oxidation and its Significance
for Atherosclerosis: a Mathematical Approach, Bulletin of Mathematical Biology, 64,
65-95.
Goldsby, R.A. et al. (2000) In Austen,K.F., Frank,M.M., Atkinson,J.P. and Cantor,H. (eds.) Kuby
Immunology. W.H. Freeman and Company, New York.
Hanson, G.K. (2002) Inflammation, atherosclerosis, and coronary artery disease, N. Engl. J.
Med., 352(16), 1685-1695.
Ibragimov, A.I.; McNeal, C.J.; Ritter, L.R.; Walton,J.R. (2005) A mathematical model of
atherogenesis as an inflammatory response, Math Med Biol., 22(4), 305-333.
Klimov, A.N.; Nikul’cheva, N.G. (1999) Lipid and Lipoprotein Metabolism and Its
Disturbances, St. Petersburg: Piter Kom.
Lollini,P.-L. (2008) Private communication.
Orem, C.; Orem, A.; Uydu, H.A.; Celik, S.; Erdol, C.; Kural,B.V. (2002) The effects of
lipid-lowering therapy on low-density lipoprotein auto-antibodies: relationship with
low-density lipoprotein oxidation and plasma total antioxidant status, Coron. Artery
Dis., 13(1), 56-71.
GRID Computing
GRID Computing and Immunology
and Computational Computational Immunology 237
15

Pappalardo, F.; Musumeci, S.; Motta, S. (2008) Modeling immune system control of
atherogenesis, Bioinformatics, 24:15, 1715-1721.
Romero-Corral, A.; Somers, V.K.; Korinek, J.; Sierra-Johnson, J.; Thomas, R.J.; Allison,
T.G.; Lopez-Jimenez,F. (2006) Update in prevention of atherosclerotic heart disease:
management of major cardiovascular risk factors, Rev. Invest. Clin., 58(3), 237-244.
Ross,R. (1999) Atherosclerosis–an inflammatory disease, N. Engl. J. Med., 340(2), 115-126.
Tonolo,G. (2008) Private communication.
Shaw, P.X.; Hörkkö, S.; Tsimikas, S.; Chang, M.K.; Palinski, W.; Silverman, G.J.; Chen, P.P.;
Witztum, J.L. (2001) Human-derived anti-oxidized LDL autoantibody blocks uptake
of oxidized LDL by macrophages and localizes to atherosclerotic lesions in vivo,
Arterioscler. Thromb. Vasc. Biol., 21(8), 1333-1339.
Shoji, T.; Nishizawa, Y.; Fukumoto, M.; Shimamura, K.; Kimura, J; Kanda, H.; Emoto, M.;
Kawagishi, T.; Morii,H. (2000) Inverse relationship between circulating oxidized low
density lipoprotein (oxLDL) and anti-oxLDL antibody levels in healthy subjects,
Atherosclerosis, 148(1), 171-177.
Steinberg,D. (1997) Low density lipoprotein oxidation and its pathobiological significance, J.
Biol. Chem., 272(34), 20963-20966.
Tinahones, F.J.; Gomez-Zumaquero, J.M.; Rojo-Martinez, G.; Cardona, F.; Esteva de Antonio,
I.E.; Ruiz de Adana, M.S.; Soriguer, F.K. (2002) Increased levels of anti-oxidized
low-density lipoprotein antibodies are associated with reduced levels of cholesterol
in the general population, Metabolism., 51(4), 429-431.
Tinahones, F.J.; Gomez-Zumaquero, J.M.; Garrido-Sanchez, L.; Garcia-Fuentes, E.;
Rojo-Martinez, G.; Esteva, I.; Ruiz de Adana, M.S.; Cardona, F.; Soriguer,F. (2005)
Influence of age and sex on levels of anti-oxidized LDL antibodies and anti-LDL
immune complexes in the general population, J. Lipid Res., 46(3), 452-457.
Vinereanu,D. (2006) Risk factors for atherosclerotic disease: present and future, Herz, Suppl.
3, 5-24.
Metropolis, N.; Rosenbluth, A.W.; Rosenbluth, M.N.; Teller, A.H.; Teller, E. (1953) "Equation
of State Calculations by Fast Computing Machines", Journal of Chemical Physics, 21,
1087-1092.
Kirkpatrick, S.; Gelatt, C. D.; Vecchi, Jr., M. P. Optimization by simulated Annealing Science,
Vol. 220, No. 4598, 671-680, 1983
Van Laarhoven, P.J.M.; Aarts, E.H.L. (1987) Simulated Annealing: Theory and Applications,
Springer Edition.
Kirschner, D.; Panetta, J.C. (1998) Modeling immunotherapy of the tumor-immune interaction.
J. Math. Biol., (37), 235-252.
Nani,F.,Freedman,S. (2000) A mathematical model of cancer treatment by immunotherapy,
Math. Biosci., 163, 159-199.
Pappalardo, F.; Lollini, P.L.; Castiglione, F; Motta, S. Modeling and simulation of cancer
immunoprevention vaccine. Bioinformatics. 21, 2891-2897
Agur, Z.; Hassin, R.; Levy, S. Optimizing chemotherapy scheduling using local search
heuristics, Operations Research, 54:5, 829-846.
Kumar, N.; Hendriks, B.S.; Janes, K.A.; De Graaf, D.; Lauffenburger D.A. Applying
computational modeling to drug discovery and development Drug Discovery Today
Volume 11, Issues 17-18, September 2006, Pages 806-811
Lollini, P.L.; Motta, S.; Pappalardo, F. Discovery of cancer vaccination protocols with a genetic
algorithm driving an agent based simulator. BMC. Bioinformatics. 7, 352.
238
16 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

Davies, M.N.; Flower, D.R. (2007) Harnessing bioinformatics to discover new vaccines, Drug
Discovery Today, 12(9-10), 389-395.
Castiglione, F.; Piccoli, B. (2007) Cancer immunotherapy, mathematical modeling and optimal
control, J. Theo. Biol., 247, 723-732.
Pennisi, M.; Catanuto, R.; Pappalardo, F.; Motta, S. (2008) Optimal vaccination schedules using
Simulated Annealing, Bioinformatics, 24:15, 1740-1742.
Pennisi, M.; Catanuto, R.; Mastriani, E.; Cincotti, A.; Pappalardo,F.; Motta,S. (2008) Simulated
Annealing And Optimal Protocols, Journal of Circuits Systems, and Computers, 18:8,
1565-1579.
 0
12

A Comparative Study of Machine Learning and

Evolutionary Computation Approaches for
Protein Secondary Structure Classification
César Manuel Vargas Benítez, Chidambaram Chidambaram,
Fernanda Hembecker and Heitor Silvério Lopes
Bioinformatics Laboratory, Federal University of Technology Paraná (UTFPR)
Curitiba - PR - Brazil

1. Introduction
Proteins are essential to life and they have countless biological functions. Proteins are
synthesized in the ribosome of cells following a template given by the messenger RNA
(mRNA). During the synthesis, the protein folds into a unique three-dimensional structure,
known as native conformation. This process is called protein folding. The biological function
of a protein depends on its three-dimensional conformation, which in turn, is a function of its
primary and secondary structures.
It is known that ill-formed proteins can be completely inactive or even harmful to the
organism. Several diseases are believed to result from the accumulation of ill-formed proteins,
such as Alzheimer’s disease, cystic fibrosis, Huntington’s disease and some types of cancer.
Therefore, acquiring knowledge about the secondary structure of proteins is an important
issue, since such knowledge can lead to important medical and biochemical advancements
and even to the development of new drugs with specific functionality.
A possible way to infer the full structure of an unknown protein is to identify potential
secondary structures in it. However, the pattern formation rules of secondary structure of
proteins are still not known precisely.
This paper aims at applying Machine Learning and Evolutionary Computation methods to
define suitable classifiers for predicting the secondary structure of proteins, starting from their
primary structure (that is, their linear sequence of amino acids).
The organization of this paper is as follows: in Section 2 we introduce some basic concepts and
some important aspects of molecular biology, computational methods for classification tasks
and the protein classification problem. Next, in Sections 3 and 4, we present, respectively, a
review of the machine learning and evolutionary computation methods used in this work.
In Section 6, we describe the methodology applied to develop the comparison of different
classification algorithms. Next, Section 7, the computational experiments and results are
detailed. Finally, in the last Section 8, discussion about results, conclusions and future
directions are pointed out.
240
2 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

2. Background
2.1 Molecular biology
Proteins are considered the primary components of living beings and they have countless
biological functions. Finding the proteins that make up an organism and understanding their
function is the foundation of molecular Biology (Hunter, 1993).
All proteins are composed by a chain of amino acids (also called residues) that are linked
together by means of peptide bonds. Each amino acid is characterized by a central carbon
atom (also known as Cα) to which are attached (as shown in Figure 1) a hydrogen atom,
an amino group (NH2 ), a carboxyl group (COOH) and a side-chain that gives each amino
acid a distinctive function (also known as radical R). Two amino acids form a peptide bond
when the carboxyl group of one molecule reacts with the amino group of the other. This
process of amino acids aggregation is known as dehydration by releasing a water molecule
(Griffiths et al., 2000). All amino acids have the same backbone and they differ from each
other by the side-chain, which can range from just a hydrogen atom (in glycine) to a
complex heterocyclic group (in tryptophan). The side-chain defines the physical and chemical
properties of the amino acids of a protein (Cooper, 2000; Nelson & Cox, 2008).

Fig. 1. General structure of an α-amino acid. The side-chain (R element) attached to the Cα
defines the function of the amino acid
There are numerous amino acids in the nature, but only 20 are proteinogenic. They are shown
in Table 1. The first to be discovered was asparagine, in 1806. The last, threonine, was
identified in 1938 (Nelson & Cox, 2008).
To understand the structures and functions of proteins, it is of fundamental importance to
have knowledge about the properties of the amino acids, defined by their side-chain. Thus,
the amino acids can be grouped into four categories: hydrophobic (also called non-polar),
hydrophilic (also called polar), neutral, basic and acid (Cooper, 2000). Kyte & Doolittle
(1982) proposed an hydrophobicity scale for all 20 amino acids. See the Table 1 for detailed
information about the proteinogenic amino acids.
Polar amino acids can form hydrogen bonds with water and tend to be positioned preferably
outwards of the protein, i.e., they are capable to interact with the aqueous medium (which is
polar). On the other hand, hydrophobic amino acids tend to group themselves in the inner
part of the protein, in such a way to get protected from the aqueous medium by the polar
amino acids.
According to this behavior in aqueous solution, one can conclude that the polarity of the side
chain directs the process of protein structures formation (Lodish et al., 2000).
From the chemical point of view, proteins are structurally complex and functionally
sophisticated molecules. The structural organization of proteins is commonly described
into four levels of complexity (Cooper, 2000; Griffiths et al., 2000; Lodish et al., 2000;
Nelson & Cox, 2008), in which the upper cover the properties of lower: primary, secondary,
tertiary and quaternary structures.
The primary structure is the linear sequence of amino acids. This is the simplest level of
organization, it represents only the peptide bonds between amino acids.
A Comparative Study of Machine Learning
and Evolutionary
A Comparative Computation
Study of Machine Approaches
Learning and Evolutionary Computation for Protein
Approaches Secondary
for Protein Structure
Secondary Structure Classification
Classification 2413

Amino acid Type

K&D Normalised Value
name symbol
Isoleucine ILE +4.5 10.00 Hydrophobic
Valine VAL +4.2 9.68 Hydrophobic
Leucine LEU +3.8 9.26 Hydrophobic
Phenylalanine PHE +2.8 8.20 Hydrophobic
Cysteine CYS +2.5 7.88 Hydrophobic
Methionine MET +1.9 7.25 Hydrophobic
Alanine ALA +1.8 7.14 Hydrophobic
Glycine GLY -0.4 4.81 Neutral
Threonine THR -0.7 4.49 Neutral
Serine SER -0.8 4.39 Neutral
Tryptophan TRP -0.9 4.28 Neutral
Tyrosine TYR -1.3 3.86 Neutral
Proline PRO -1.6 3.54 Neutral
Histidine HIS -3.2 1.85 Hydrophilic
Glutamine GLN -3.5 1.53 Hydrophilic
Asparagine ASN -3.5 1.53 Hydrophilic
Glutamic acid GLU -3.5 1.53 Hydrophilic
Aspartic acid ASP -3.5 1.53 Hydrophilic
Lysine LYS -3.9 1.10 Hydrophilic
Arginine ARG -4.0 1.00 Hydrophilic

Table 1. Kyte and Doolittle (K & D) hydrophobicity scale and the normalized scale used in
the computational experiments

The secondary structure of a protein refers to the local conformation of some part of a
three-dimensional structure. There are, basically, three main secondary structures: α-helices
(Pauling et al., 1951a), β-sheets (Pauling et al., 1951b) and turns (Lewis et al., 1973). In the
structure of an α-helix, the backbone is tightly turned around an imaginary helix (spiral)
and the side-chains of the amino acids protrude outwards the backbone (Figure 2(a)). The
β-sheet is formed by two or more polypeptide segments of the same molecule, or different
molecules, arranged laterally and stabilized by hydrogen bonds between the NH and CO
groups (Figure 2(b)). Adjacent polypeptides in a β-sheet can have same direction (parallel
β-sheet) or opposite directions (antiparallel β-sheet). Functionally, the antiparallel β-sheets
are present in various types of proteins, for example, enzymes, transport proteins, antibodies
and cell-surface proteins (Branden & Tooze, 1999). Turns are composed by a small number of
amino acids and they are usually located in the surface of proteins forming folds that redirect
the polypeptide chain into the protein. They allow large proteins to fold in highly compact
structures.
242
4 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

Secondary structures can be associated through side-chain interactions to motifs

(Branden & Tooze, 1999; Griffiths et al., 2000; Nölting, 2006). Motifs are patterns often
found in three-dimensional structures that perform specific functions. For instance, the
helix-turn-helix motif is important in DNA-protein interactions.

(a) (b)

Fig. 2. α-helix (a) and β-sheet (b) structures. Adapted from (Alberts et al., 2002)
The tertiary structure represents the conformation of a polypeptide chain, i.e. the
three-dimensional arrangement of the amino acids. The tertiary structure is the folding of
a polypeptide as a result of interactions between the side chains of amino acids that are in
different regions of the primary structure. Figure 3(a) shows an example of tertiary structure,
where one can observe the presence of two secondary structures: α-helix and β-sheet.
Finally, the arrangement of three-dimensional structures constitutes quaternary structure (as
shown in Figure 3(b)). Figures 3(a) and 3(b) were drawn using RasMol 1 from PDB files (see
Section 2.2).
The Proteins can be classified into two major groups, considering higher levels of structure
(Nelson & Cox, 2008): fibrous and globular proteins. Both groups are structurally different:
fibrous proteins consist of a single type of secondary structure; globular proteins have a
nonrepetitive sequence and often contain several types of secondary structure. Helices are
the most abundant form of secondary structure in globular proteins, followed by sheets, and
in the third place, turns (Nölting, 2006).

2.2 Protein databases and classification

Finding protein functions has been, since long ago, an important topic in the Bioinformatics
community. As mentioned in Section 2, the function of a protein is directly related to its
structure. Due to its great importance for Medicine and Biochemistry, many research has been
done about proteins (including the many genome sequencing projects) and, consequently,
many information is available. There are many resources related to protein structure and
function. Table 2 lists some protein databases. Basically, the protein databases can be classified
into two classes: sequence and structure databases.

1 RasMol is a molecular visualization software. Available at http://www.rasmol.org

A Comparative Study of Machine Learning
and Evolutionary
A Comparative Computation
Study of Machine Approaches
Learning and Evolutionary Computation for Protein
Approaches Secondary
for Protein Structure
Secondary Structure Classification
Classification 2435

(a) (b)

Fig. 3. Tertiary structure of Ribonuclease-A (a) and quaternary structure of Hemoglobin (b).

Database Description Web Address

PDB repository of protein http://www.pdb.org
structures
UniProtKB/TrEMBL repository of protein http://www.uniprot.org/
amino acid sequences,
name/description,
taxonomic data and citation
information
PIR protein sequence databases http://pir.georgetown.edu/
PROSITE documentation entries http://www.expasy.org/prosite/
describing sequence
motif definitions, protein
domains, families and
functional patterns
PRINTS Fingerprints information on http://www.bioinf.man.ac.uk/dbbrowser/
protein sequences
BLOCKS Multiple-alignment blocks http://blocks.fhcrc.org/
eMOTIF protein motif database, http://motif.stanford.edu/emotif/
derived from PRINT and
BLOCKS
PRODOM protein domain databases http://protein.toulouse.inra.fr/prodom.html
InterPro protein families and http://www.ebi.ac.uk/interpro/
domains

Table 2. Some important protein databases

In this work we used the Protein Data Bank (PDB) (Berman et al., 2000; Bernstein et al.,
1977) that is an international repository of three-dimensional structure of biological
244
6 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

macromolecules. The data is, typically, obtained by X-ray crystallography (Drenth, 1999;
Sunde & Blake, 1997) or NMR spectroscopy (Nuclear Magnetic Resonance) (Jaroniec et al.,
2004; Wüthrich, 1986). Almost all protein structures known today are stored in the PDB
(Bernstein et al., 1977).
Despite the growing number of protein sequences already discovered, only a few portion
of them have their three-dimensional and secondary structures unveiled. For instance, the
UniProtKB/TrEMBL repository (The UniProt Consortium, 2010) of protein sequences has
currently around 13,89 million records (as in March/2011), and the PDB registers the structure
of only 71,794 proteins. This fact is due to the cost and difficulty in unveiling the structure of
proteins, from the biochemical and biological point of view. Therefore, computer science has
an important role here, proposing models and methods for studying the Protein Structure
Prediction problem (PSP).
There are two basic approaches that are used to the prediction of protein functions:
prediction of the protein structure and then prediction of function from the structure, or
else, classifying proteins and supposing that similar sequences will have similar functions
(Tsunoda et al., 2011). Several approaches to solve the PSP exist, each addressing the
problem by using a computational method to obtain optimal or quasi-optimal solutions,
such as Molecular Dynamics with ab initio model (Hardin et al., 2002; Lee et al., 2001),
neural nets (Yanikoglu & Erman, 2002) and evolutionary computation methods with lattice
(Benítez & Lopes, 2010; Lopes, 2008; Scapin & Lopes, 2007; Shmygelska & Hoos, 2005) and
off-lattice (Kalegari & Lopes, 2010) models.
However, there is no consensus about protein classification which is done using different
properties of proteins through several approaches. Many computational techniques have been
used to classify proteins into families, such as structural transformations (Ohkawa et al., 1996),
data compression (Chiba et al., 2001), genetic programming (Koza, 1996), Markov chains
(Durbin et al., 1998) and neural networks (Wang & Ma, 2000; Weinert & Lopes, 2004).
Other papers focus on motifs discovery, as a starting step for protein classification. For
instance, (Tsunoda & Lopes, 2006) present a system based on a genetic algorithm that was
conceived to discover motifs that occur very often in proteins of a given family but rarely occur
in proteins of other families. Also, Tsunoda et al. (2011) present an evolutionary approach
for motif discovery and transmembrane protein classification, named GAMBIT. Techniques
of clustering for sequences analysis were presented by (Manning et al., 1997). Chang et al.
(2004) proposed a Particle Swarm Optimization (PSO) approach to motif discovery using two
protein families from the PROSITE.
Wolstencroft et al. (2006) describes the addition of an ontology that captures human
understanding of recognizing members of protein phosphatases family by domain
architecture as an ontology. G.Mirceva & Davcev (2009) present an approach based on Hidden
Markov Models (HMMs) and the Viterbi algorithm to domain classification of proteins.
Davies et al. (2007) present an hierarchical classification of G protein-coupled receptors
(GPCRs) The GPCRs are a common target for therapeutic drugs (Klabunde & Hessler, 2002).

3. Machine learning methods

In this section, we focus on the Machine Learning (ML) algorithms applied to the classification
of a secondary protein data set. Most of the methods discussed in this section can be
considered as important supervised ML techniques. It is frequently cited in the literature that
the efficiency of ML algorithms can be quite different from data set to data set. Therefore,
it is usual to test the data set with many different ML algorithms. For this purpose,
A Comparative Study of Machine Learning
and Evolutionary
A Comparative Computation
Study of Machine Approaches
Learning and Evolutionary Computation for Protein
Approaches Secondary
for Protein Structure
Secondary Structure Classification
Classification 2457

Weka2 (Witten et al., 2011) is an appropriate and a flexible tool which contains a collection
of state-of-the-art ML algorithms and data preprocessing tools. It allows users to test the
algorithms with new data sets quickly.
It is not an easy task to select which algorithm is the most suitable for a specific domain or
problem. Also, as there are many algorithms that use numerical data, an expert is needed to
tune the data. In this context, the comparison of many different algorithms is not an easy task.
Such a comparison can be performed by statistical analysis of the accuracy rate obtained from
trained classifiers on some specific data set (Kotsiantis, 2007). To support this comparison
and to evaluate the quality of the methods, three main measures can be considered (Mitchell,
1997):
• The classification rate of the training data
• The correct classification rate of some test data
• The average performance using cross validation
The most well-known classification algorithms are grouped in the Weka workbench,
including Bayesian classifiers, Neural networks, Meta-learners, Decision trees, Lazy classifiers
and rule-based classifiers (Witten et al., 2011). Bayesian methods include Naïve Bayes,
complement Naïve Bayes (CNB), multinomial Naïve Bayes, Bayesian networks and AODE.
Both decision trees and rule-based classifiers belong to the category of logic-based supervised
classification algorithms (Kotsiantis, 2007). Decision trees algorithms include several variants,
such as NBTree, ID3, J48 (also known as C4.5) and alternative decision trees (ADTree).
PART, decision tables, Rider, JRip, and NNge are included in the group of rule learners.
Lazy Bayesian rules (LBR), Instance-Based learning schemes (IB1 and IBk), Kstar, and
Locally-Weighted Learning (LWL) are part of the lazy learning algorithms.
Besides these basic classification learning algorithms, there are some meta-learning schemes
such as LogitBoost, MultiBoostAB, and ADABoost. These boosting algorithms enable users
to combine instances of one or more of the above-mentioned algorithms. In addition to
these all algorithms, Weka workbench includes neural networks methods such as Multiplayer
Perceptron (MLP), Sequential Minimal Optimization algorithm (SMO), Radial Basis Function
(RBF) network, and logistic and voted perceptron (Witten et al., 2011). In the following
subsections, we discuss specific aspects and present a brief comparison of some classification
methods mentioned previously and used in the protein classification task of this work.

3.1 Bayesian methods

Bayesian methods are the most practical approaches amongst many learning algorithms and
those that provide learning based on statistical approaches. These methods are characterized
by induction of probabilistic networks from the training data (Kotsiantis, 2007). In the Weka
workbench there are several methods, such as: CNB, NaïveBayes, NaïveBayesSimple and
AODE. Bayesian methods emerged as alternative approaches for decision trees and neural
networks and, at the same time, being competitive with them for real-world applications
(John & Langley, 1995). However, it is known that, to apply such methods, prior knowledge
of many probabilities is required (Mitchell, 1997). Naïve Bayes (NB) is one of the learning
algorithms widely applied to classification tasks. The NB classifier is the most effective
algorithm of the Bayesian group of algorithms, and it can easily handle unknown or missing
values (Mitchell, 1997). The main advantage of the NB is that it does not require a long

2 Available at: http://www.cs.waikato.ac.nz/ml/weka/

246
8 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

time to train the classifier with the data set. This classifier is based on the assumption of
attribute independence. Some Bayesian variants have tried to alleviate this assumption, such
as the NBTree (Zheng & Webb, 2000) algorithm, which has a high computational overhead.
To improve the prediction accuracy without increasing the computational cost, Webb et al.
(2005) developed a variant algorithm of NB called AODE (Aggregating One-Dependence
Estimators) that uses a weaker attribute independence assumption than NB. NB classifier
traditionally relies on the assumption that the numerical attributes are generated by a
single Gaussian distribution. However, this is not always the best approximation for the
density estimation on continuous attributes. Following this direction, John & Langley (1995)
proposed a new approach described as flexible Bayes, in which a variety of nonparametric
density estimation methods were used instead of Gaussian distribution. This approach is
implemented through NaïveBayesUpdateable class in the Weka tool. Bayesian networks (BN)
methods use several search algorithms that works under conditions of uncertainty. All BN
learners are considered as slow and not suitable for large data sets (Cheng et al., 2002). Unlike
decision trees and neural networks, the main aspect of this method is that it obtains prior
information of the data set from the structural relationships among the features (Kotsiantis,
2007).

3.2 Decision trees

Basically, the methods of this category classify data by building decision trees, in which each
node represents a feature in an instance and each branch represents the value of a node.
They are heuristic, non-incremental, and do not use any world knowledge. In this category,
there are many learning methods, amongst of which, ID3, J48 and C4.5 are most well-known
(Kotsiantis, 2007). Not only these algorithms are based on decision trees. There are also some
hybrid versions like NBTree, LMTree, RandomForest and ADTree. In general, these versions
have competitive performance with the traditional C4.5 algorithm. The earliest version of
ID3 was developed by Quinlan (1993) and its improved version is C4.5 (Martin, 1995). ID3
allows to work with errors in the training data, as well as missing attribute values. This
method uses a hill-climbing strategy to search in the hypothesis space, so as to find out a
decision tree that correctly classifies the training data. This learning method can handle noisy
training data and is less sensitive to errors of individual training examples (Mitchell, 1997).
From the decision trees, during the learning process, a set of rules in the disjunctive form are
obtained by traversing the different possible paths in the entire tree. A limitation of the ID3
algorithm is that it cannot guarantee the optimal solution. Another improved version of the
algorithm, J48, implements Quinlan’s C4.5 algorithm by generating a pruned or an unpruned
decision tree (Witten et al., 2011). The NBTree algorithm was proposed by Kohavi (Kohavi,
1996) to overcome the accuracy problem encountered with both Naïve-Bayes and decision
trees in small data sets. NBTree is a hybrid algorithm which induces a mix of decision-tree and
Naïve-Bayes classifiers. Eventually, this algorithm outperforms both C4.5 and Naïve-Bayes.
To predict numeric quantities, Landwehr et al. (2005) introduced a new learning method
combining tree induction methods and logistic regression models into decision trees. This
method is denominated logistic model trees (LMTree). LMTree produces a single tree which is
not easily interpretable, but better than multiple trees. This algorithm achieved performance
higher than decision trees with C4.5 and logistic regression.
RandomForest is a classifier consisting of a collection of tree-structured classifiers in which
each tree depends on the values of a random vector, sampled independently and with the
same distribution for all trees in the forest (Breiman, 2001).
A Comparative Study of Machine Learning
and Evolutionary
A Comparative Computation
Study of Machine Approaches
Learning and Evolutionary Computation for Protein
Approaches Secondary
for Protein Structure
Secondary Structure Classification
Classification 2479

Freund & Schapire (1999) presented a new type of classification algorithm, combining
decision trees and boosting, called alternative decision trees (ADTree). An important feature
of the ADTree algorithm is the measure of confidence or classification margin. This learning
algorithm was compared with other improved version of C4.5 with boosting, denominated
as C5.0. According to the experiments, it can be realized that the ADTree is competitive with
C5.0 and generate easily interpretable small rules.

3.3 Neural Networks

Neural Network (NN) learning methods are robust to errors in the training data. They provide
a good approximation to real-valued, discrete-valued, and vector-valued target functions.
High accuracy and high speed rate of classification are some relevant aspects of NN classifiers
(Kotsiantis, 2007). Algorithms like Multiplayer Perceptron (MLP), SMO, RBFNetwork and
Logistic are part of the Weka workbench. A RBF neural networks is a particular NN
constructed from spatially localized kernel functions, and uses a different error estimation
and gradient descent function called the radial basis function (RBF). This method uses a
cross-validation technique to stop the training which is not present in other NN algorithms
(Mitchell, 1997). Platt (1998) proposed an improved algorithm for training support vector
machines (SVMs) called Sequential Minimal Optimization SMO. The results obtained for
real-world test sets showed that the SMO is 1200 times faster than linear SVMs and 15 times
faster than non-linear SVMs.

3.4 Meta-learning methods

Most of the methods of this category such as boosting, bagging and wagging are common
committee learning approaches that reduce the classification error from learned classifiers.
Boosting is the one of the most important recent advancements in classification algorithms,
since it can improve dramatically their performance (Friedman et al., 2000). It is also
known as machine learning meta-algorithm or discrete AdaBoost. AdaBoost, as a boosting
algorithm, can efficiently convert a weak learning algorithm into a short learning algorithm.
Furthermore, it is an adaptive behavior with error rates of the individual weak hypotheses
(Freund & Schapire, 1999). AdaBoost calls a given weak learning algorithm repeatedly in
a series of rounds and trains the classifiers by over-weighting the training samples that
were misclassified in the next iteration. A complete algorithm description can be found
in Friedman et al. (2000). One of the important properties of the ADABoost algorithm is
the identification of outliers which are either mislabeled in the training data or inherently
ambiguous and hard to categorize (Freund & Schapire, 1999). Many other developments
on AdaBoost resulted in variants such as Discrete AdaBoost, Real AdaBoost, LPBoost and
LogitBoost. Both ADABoost and bagging generic techniques can be employed together
with any baseline classification technique. Wagging is variant of bagging, that requires a
base learning algorithm capable of using training cases with differing weights (Webb, 2000).
MultiBoosting is an extension technique of AdaBoost with wagging. It offers a potential
computational advantage over AdaBoost (Webb, 2000).

3.5 Rule-based methods

Rules can be extracted from the data set using many different machine learning algorithms.
The IF-THEN rules is a convenient way to represent the underlying knowledge present
in the data set, which can be easily understood by domain experts. Among a variety of
rule-based methods that were investigated, decision trees and separate-and-conquer strategy
248
10 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

are considered the most important (Frank & Witten, 1998). Based on these approaches,
emerged two dominant rule-based learning methods: C4.5 (Quinlan, 1993) and RIPPER
(Cohen, 1995). In fact, RIPPER is an optimized algorithm which is very efficient in large
samples with noisy data, and produce error rates lower than or equivalent to C4.5 (Cohen,
1995). In the Weka workbench, rule-based classifiers have many learning algorithms,
including PART, DecisionTable, Ridor, JRip and NNge. Decision tables, which are simple
space hypotheses, are represented by DTM (Decision Table Majority). Kohavi (1995) evaluated
the power of decision tables through Inducer DTM (IDTM). IDTM, on some data sets, obtained
comparable performance accuracy as C4.5. JRip implements a propositional rule learner
called Repeated Incremental Pruning to Produce Error Reduction (RIPPER), proposed by
Cohen (1995). Frank & Witten (1998) proposed a new approach, by combining the paradigms
of decision trees and separate-and-conquer, called PART. PART is based on the repeated
generation of partial decision trees in a separate-and-conquer manner. Not only accuracy
of a learning algorithm, but also the size of a rule set resulted from the learning process
is important. The size of a rule strongly influences on the degree of comprehensibility.
Rule sets produced by PART are generally smaller as C4.5 and more accurate than RIPPER
(Frank & Witten, 1998). Another algorithm that is part of Weka workbench is the Non-Nested
Generalized Exemplars (NNge), proposed by Martin (1995). NNge generalizes exemplars
without nesting (exemplars contained within one another) or overlaping. By generalization,
examples in the data set that belongs to the same class are grouped together. When tested
against domains containing both large and small disjuncts, NNge performs better than C4.5.
NNge performs well on data sets that combine small and large disjuncts, but it performs
poorly in domains with a high degree of noise (Martin, 1995).

3.6 Lazy methods

Instance-based methods are considered as lazy learning methods because the classification
or induction process is done only after receiving a new instance or a training example.
Learning process will be started on the stored examples only after when a new query instance
is encountered (Mitchell, 1997). Nearest Neighbor algorithm is the one of the most basic
instance-based methods. The instance space is defined in terms of Euclidean distance.
However, since Euclidean distance is inadequate for many domains, several improvements
were proposed to the instance-based nearest neighbor algorithm which are known as IB1 to
IB5 (Martin, 1995). Zheng & Webb (2000) proposed a novel algorithm called Lazy Bayesian
Rule learning algorithm (LBR) in which lazy learning techniques are applied to Bayesian
tree induction. Error minimization was maintained as an important criteria in this algorithm.
Experiments done with different domains showed that, on average, LBR overcomes mostly all
other algorithms including Naïve Bayes classifier and C4.5. The Locally Weighted Learning
algorithm (LWL) is similar to other lazy learning methods, however it behaves differently
when classifying a new instance. LWL algorithm constructs a new Naïve Bayes model using
a weighted set of training instances (Frank et al., 2003). It empirically outperforms both
standard Naïve Bayes as well as nearest-neighbor methods on most data sets tested by those
authors.

4. Gene expression programming and GEPCLASS

Gene expression programming (GEP) is proposed by Ferreira (2001), and it has features of
both genetic algorithms (GAs) and genetic programming (GP). The basic difference between
the three algorithms is the way the individuals are defined in each algorithm. In the
A Comparative Study of Machine Learning
and Evolutionary
A Comparative Computation
Study of Machine Approaches
Learning and Evolutionary Computation for Protein
Approaches Secondary
for Protein Structure
Secondary Structure Classification
Classification 249
11

traditional GAs individuals are called chromosomes and are represented as linear binary
strings of fixed length. On the other side, in GP, individuals or trees are non-linear entities
of different sizes and shapes. On the other hand, in GEP, the individuals are encoded as linear
strings of fixed length (chromosomes) which are afterwards expressed as non-linear entities
of different sizes and shapes (simple diagram representations or expression trees (ETs))
(Ferreira, 2001). In fact, in GEP, chromosomes are simple, compact, linear and relatively small
entities, that are manipulated by means of special genetic operators (replication, mutation,
recombination and transposition). ETs, in turn, are the phenotypical representation of the
chromosome. Unlike genetic algorithms, selection operates over ETs (over the phenotype, not
the genotype). During the reproduction phase, only chromosomes are generated, modified
and transmitted to the next generations. The interplay of chromosomes and ETs allows to
translate the language of chromosomes into the language of ETs. The use of varied set of
genetic operators introduce genetic diversity in GEP populations always producing valid
ETs. In the same way as other evolutionary algorithms, in GEP, the initial population must
be defined either randomly or using some previous knowledge collected from the problem.
Next, chromosomes are expressed into ETs which will be then evaluated according to the
definition of the problem resulting in a fitness measure. During the iteration process, the best
individual(s) is(are) kept and the rest are submitted to a fitness-based selection procedure.
Selected individuals naturally go through modifications by means of genetic operators leading
to a new generation of individuals. The whole process is repeated until a stopping criterion
is met (Weinert & Lopes, 2006). The structural organization of GEP genes are based on Open
Reading Frames (ORF), a biological terminology. Although the length of genes is constant, the
length of ORFs are not. GEP genes are composed of a head that contain symbols that represent
both function and terminals, and a tail that contains only terminals. GEP chromosomes are
basically formed by more than one gene of different lengths. Unlike GP, in which an individual
of the population is modified by only one operator at a time, in GEP, an individual may be
changed by one or several genetic operators. Besides the genetic operations like replication,
mutation and recombination, GEP includes operations based on transpositions and insertion
of elements.
In this work, we used the GEPCLASS system3 that was specially developed for
data classification with some modifications regarding the original GEP algorithm
(Weinert & Lopes, 2006). Prior implementing data classification using evolutionary
algorithms, it is necessary to definet whether an individual represents a single rule (Michigan
approach) or a complete solution composed by a set of rules (Pittsburg approach) (Freitas,
1998). This system can implement both approaches, either by an explicit decision of a
user, or allowing the algorithm decide by itself which one is the most suitable for a given
classification task, during the evolutionary process. GEPCLASS can manage both continuous
and categorical (nominal) attributes in the data set. If a given attribute is nominal, GEPCLASS
uses only = or = as relational operators. Otherwise, if the attribute is continuous or ordered
categorical, all relational operators can be used.
Figure 4 shows a 2-genes chromosome with different lengths for heads and tails. In the
chromosome part, upward arrows show the points delimiting the coding sequence of each
gene. This chromosome transformed into an ET and, later, to a candidate rule. GEPCLASS
uses variable-length chromosomes that can have one or more genes. Genes within a
given chromosome are of the same size. Using chromosomes with different lengths in the
population can introduce healthy genetic diversity during the search.

3 Freely available at: http://bioinfo.cpgei.ct.utfpr.edu.br/en/softwares.htm

250
12 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

Fig. 4. Example of chromosome structure, expression tree (ET) and rule in GEPCLASS.
Adapted from (Weinert & Lopes, 2006).

5. Hidden Markov models

A hidden Markov model (HMM) has an underlying stochastic process that is not observable (it
is hidden), but can only be observed through another set of stochastic processes that produce
the sequence of observations (Rabiner, 1989). A HMM can be visualized as a finite state
machine composed of a finite set of states, a1 , a2 , . . . , an , including a beginning and an ending
states (G.Mirceva & Davcev, 2009). The HMM can generate a protein sequence by emitting
symbols as it progresses through a series of states. Any sequence can be represented by a
path through the model. Each state has probabilities associated to it: the transition and the
emission probabilities. Although the states are hidden, there are several applications in which
the states of the model can have physical meaning.
HMMs offer the advantages of having strong statistical foundations that are well-suited
to natural language domains and they are computationally efficient (Seymore et al., 1999).
Therefore, HMMs have been used for pattern recognition in many domains and, in special, in
Bioinformatics (Durbin et al., 1998; Tavares et al., 2008).

6. Methodology
6.1 The data set and sequence encoding
A number of records of human globular proteins was selected and downloaded from the
PDB. The original files were scanned and all annotated secondary structures were extracted
by using a parser developed in Java programming language.
These primary sequences extracted from the files had a variable length from 2 to 29 amino
acids, and they were divided into five classes, following the PDB nomenclature: HELIX,
SHEET0, SHEET1 and SHEET2 and TURN with 1280, 284, 530, 498 and 119 sequences,
respectively. HELIX, SHEET0 (SHEET1 and SHEET2) and TURN represent α-helices,
β-sheets and turns, respectively. In order to properly train the classifiers it is also necessary a
“negative” class. That is, a class dissimilar to the others that represent no secondary structure.
A Comparative Study of Machine Learning
and Evolutionary
A Comparative Computation
Study of Machine Approaches
Learning and Evolutionary Computation for Protein
Approaches Secondary
for Protein Structure
Secondary Structure Classification
Classification 251
13

This was accomplished by creating a NULL class of 3000 variable-sized sequences. Therefore,
the database created for this work had 5711 records, with six unbalanced classes.
The natural encoding of the protein sequence is a string of letters from the alphabet of letters
representing the 20 proteinogenic amino acids. However, this encoding is not appropriate for
some classification algorithms. Thus, another encoding was proposed, converting the string
of amino acid symbols into a real-valued vector, by using a physico-chemical property of the
amino acids, as suggested by (Weinert & Lopes, 2004). This was accomplished using the Kyte
and Doolittle hydrophobicity scale. The real-valued vector was normalized in the range 1.00
– 10.00 (as shown in Table 1).

6.2 Computational methods

Three different computational approaches were used in this work. First, we applied
several machine learning algorithms using the Weka workbench, as mentioned before.
The algorithms in Weka were grouped into Bayesian, neural networks, meta-learners,
trees, lazy-learners and rule-based. The input file for this algorithms was formatted to
Attribute-Relation File Format (ARFF). This is an ASCII text file that describes a set of
instances sharing a set of attributes.
An ARFF file has two sections: the header and data information. The header of the ARFF
file contains the name of the relation, a list of attributes and their types. The convention used
for data was: p1 , p2 ,. . . , p29 , corresponding to positions in the amino acid sequence from 1 to
29, followed by a nominal class attribute, that identifies the class of each instance (HELIX,
SHEET0, SHEET1, SHEET2, TURN and NULL). Where pi are real-valued, as explained in
Section 6.1.
A second approach used was HMMs. To test this approach, we have used the HMM package
for Weka4 . Similar to other input files, the input file for this approach was also formatted as
ARFF. The HMM classifiers only work on a sequence of data which in Weka is represented
as a relational attribute. Data instances have a single nominal class attribute and a single
relational sequence attribute. The instances in this relational attribute consist of single nominal
data instances using the natural encoding of the protein sequence, i.e., the string of letters
representing the amino acid sequence (for example, “GLY, TRP, LEU, . . . , LEU”).
Finally, Gene Expression Programming (GEP) was used for generating classification rules by
means of the software GEPCLASS (Weinert & Lopes, 2006). The input file to GEP is similar to
the ARFF file that was used before, but without header information, just with the real-valued
data instances as detailed in the section 6.1.

7. Experiments and results

The main objective of this work is to compare the performance of ML algorithms and
evolutionary computation approaches for protein secondary structure classification. The main
motivation to test several algorithms is to identify the most suitable one for protein secondary
structure classification.
The performance of the methods are measured according to their predictive accuracy and
other statistical parameters drawn from confusion matrix. The processing time and memory
load were not considered for the experimental analysis.
The training/testing methodology includes a 10-fold cross-validation (Kohavi, 1995) in which
the data set is divided into 10 parts. In the first round, one part is used for testing and the

4 Available at http://www.doc.gold.ac.uk/ mas02mg/software/hmmweka/

252
14 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

remaining, nine parts are used for training. This procedure is repeated until each partition
has been used as the test set. The reported result is the weighted average of the 10 runs. The
purpose of cross-validation is to avoid biased results when a small sample of data is used.
The outcome of a classifier can lead to four different results, according to what was expected
and what, in fact, was obtained. Therefore, when classifying unknown instances, four possible
outcomes can be computed:
• tp: true positive: the number of instances (proteins) that are correctly classified, i.e., the
algorithm predicts that the protein belongs to a given class and the protein really belongs
to that class;
• f n: false positive: the number of instances that are wrongly classified, i.e., the algorithm
predicts the protein that belongs to a given class but it does not belong to it;
• tn: true negative: the number of instances that are correctly classified as not belonging to
a given class, i.e., the algorithm predicts that the protein does not belong to a given class,
and indeed it does not belong to it.
• f p: false negative: the number of instances of a given class that are wrongly classified, i.e.,
the algorithm predicts that the protein does not belong to a given class but it does belong
to it.
Combining the above-cited four outcomes obtained from a classifier, we have then calculated,
for each class, metrics commonly used in ML: sensitivity (Se), specificity (Sp), the predictive
accuracy and the Matthews Correlation Coefficient (MCC) (Matthews, 1975), defined in
Equations 1, 2, 3 and 4, respectively. Then, the weighted average of these metrics was
calculated. The results are shown in Table 3. The best results, according to these metrics,
are shown in bold in the table.
tp
Se = (1)
tp + f n
tn
Sp = (2)
tn + f p
tp + tn
Accuracy = (3)
tp + f p + f n + tn
tp × tn − f p × f n
MCC =  (4)
(tp + f p) × (tp + f n ) × (tn + f p) × (tn + f n )
The visual comparison of performance of the classifiers was done using a ROC (Receiver
Operating Characteristics) plot (Fawcett, 2006). The ROC plot is a useful technique for
visualizing and comparing classifiers and is commonly used in decision making in ML, data
mining and Bioinformatics (Sing et al., 2005; Tavares et al., 2008). It is constructed using the
performance rate of the classifiers. The ROC analysis can be used for visualizing the behavior
of diagnostic systems and medical decisions (Fawcett, 2006). In a ROC plot axes x and y are
defined as (1-Sp) and Se, respectively. These axes can be interpreted as the relative trade-offs
between the benefits and costs of a classifier. Thus, each classifier correspond to their (1-Sp,
Se) pairs. The best prediction would be that lying as close as possible to the upper left corner,
representing 100% of sensitivity (no fn) and specificity (no fp). Figure 5 shows the ROC
plot for the classifiers evaluated in this work. It should be noted that some of the classifiers
have achieved almost the same performance. Therefore, some points are superimposed in
A Comparative Study of Machine Learning
and Evolutionary
A Comparative Computation
Study of Machine Approaches
Learning and Evolutionary Computation for Protein
Approaches Secondary
for Protein Structure
Secondary Structure Classification
Classification 253
15

Group Method Se Sp MCC Accuracy rate (%)

NaiveBayes 0.601 0.910 0.47 56.58
Bayes
AODE 0.688 0.292 0.57 68.78
MLP 0.599 0.774 0.39 59.87
Neural Net Logistic 0.559 0.604 0.21 55.90
SMO 0.792 0.792 0.67 79.16
ADABoost 0.774 0.875 0.65 77.43
Meta logitBoost 0.769 0.879 0.65 76.89
Bagging 0.741 0.854 0.60 74.07
J48 0.719 0.87 0.58 71.88
Trees RandomForest 0.774 0.874 0.65 77.41
RepTree 0.717 0.868 0.58 71.72
IB1 0.727 0.869 0.59 72.74
Lazy IBk 0.683 0.86 0.53 68.33
Kstar 0.730 0.874 0.59 72.96
Jrip 0.667 0.739 0.45 66.70
Rules PART 0.726 0.862 0.59 72.63
Ridor 0.670 0.831 0.50 66.99
HMM 0.629 0.919 0.54 62.86
GEP 0.750 0.630 0.320 75.43

Table 3. Classifier performance

the graph. The top classifiers are identified in the ROC space: SMO (Sequential Minimal
Optimization algorithm), ADABoost (ML meta-algorithm) and RandomForest (collection of
tree-structured classifiers).

8. Conclusion and future works

Prediction of secondary structure protein has become an important research area in
Bioinformatics. Since the beginning, similar sequences of proteins are used to predict
the function of new proteins. In this work, several methods from ML and evolutionary
computation for classifying protein secondary structures were compared. Considering
sensitivity and specificity alone, SMO and NaiveBayes achieved the highest values, meaning
that the first was good to detect secondary structures when they are present, and the latter
was good to classify sequences that are not secondary structures. The highest sensitivity value
was below 0.8, suggesting the presence of semantic noise in the data set given by the inherent
variability of amino acid sequences in the secondary structure of proteins.
According to the results shown in Table 3 and considering the accuracy rate, we can conclude
that SMO, Meta classifiers (ADABoost and logitBoost), RandomForest, and GEP methods
achieved better performances, all above 75% of accuracy. These results can be considered
very expressive, taking into account the difficulty of the classification problem, imposed, as
mentioned before, mainly by biological variability of the secondary structure of proteins.
It is also possible to observe that the ROC plot shows the differences between methods more
clearly than Table 3, when considering both, sensitivity and specificity. For instance, it is
possible to observe that the HMM and NaiveBayes achieved the lowest number of false
254
16 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

0.8
SMO Bayes
0.78 Lazy
Adaboost Meta
0.76 RandomForest Neural Nets
Rules
0.74 Trees
GEP
0.72 HMM

0.7

0.68
Se

0.66

0.64

0.62

0.6

0.58

0.56

0.54
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 0.6 0.65 0.7 0.75
1−Sp
Fig. 5. ROC plot

negatives (highest specificity), however, they perform badly considering sensitivity. In the
upper-left corner are the three best-performing methods. However, GEP is slightly distant
from them, making clear that that considering only the accuracy rate may be misleading. The
analysis of the classifiers using the MCC leads to similar results as the ROC curve. Therefore,
based on our results, we can conclude that ROC and MCC are the best way to analyze the
performance of methods in this classification problem.
It is important to mention that no effort was done to fine-tune parameters of any method. In
all cases, the default parameters of the methods were used. However, it is a matter of fact
that adjusting parameters of classifiers (for instance, in GEP and HMM methods), the overall
performance can be significantly improved. On the other hand, such procedure could lead to
a biased comparison of classifiers.
Although SMO and Meta-learners achieved the best classification performance, it should be
highlighted the importance of rule-based methods (including GEP), since a set of classification
rules are more comprehensive and expressive to humans than other type of classification
method expressed by numbers.
Future work will include the use of hybrid techniques incorporating ML and evolutionary
computation methods, as well as hierarchical classification methods.

9. Acknowledgements
This work was partially supported by the Brazilian National Research Council (CNPq) under
grant no. 305669/2010-9 to H.S.Lopes; and a doctoral scholarship do C.M.V. Benítez from
CAPES-DS.
A Comparative Study of Machine Learning
and Evolutionary
A Comparative Computation
Study of Machine Approaches
Learning and Evolutionary Computation for Protein
Approaches Secondary
for Protein Structure
Secondary Structure Classification
Classification 255
17

10. References
Alberts, B., Johnson, A., Lewis, J., Raff, M., Roberts, K. & Walter, P. (2002). Molecular Biology of
The Cell, Garland Science, New York, USA.
Benítez, C. & Lopes, H. (2010). Hierarchical parallel genetic algorithm applied to the
three-dimensional hp side-chain protein folding problem, Proc. of IEEE International
Conference on Systems, Man and Cybernetics, IEEE Computer Society, pp. 2669–2676.
Berman, H., Westbrook, J., Feng, Z., Gilliland, G., Bhat, T., Weissig, H., Shindyalov, I. &
Bourne, P. (2000). The protein databank, Nucleic Acids Research 28(1): 235–242.
Bernstein, F., Koetzle, T., William, G., Meyer, D., Brice, M. & Rodgers, J. (1977). The protein
databank: A computer-based archival file for macromolecular structures, Journal of
Molecular Biology 112: 535–542.
Branden, C. & Tooze, J. (1999). Introduction to Protein Structure, Garland Publishing, New York,
USA.
Breiman, L. (2001). Random forests, Machine Learning 45: 5–32.
Chang, B., Ratnaweera, A. & Watson, S. H. H. (2004). Particle swarm optimisation for protein
motif discovery, Genetic Programming and Evolvable Machines 5: 203–214.
Cheng, J., Greiner, R., Kelly, J., Bell, D. & Liu, W. (2002). Learning bayesian networks from
data: An information-theory based approach, Artificial Intelligence 137(1-2): 43–90.
Chiba, S., Sugawara, K. & Watanabe, T. (2001). Classification and function estimation of
protein by using data compression and genetic algorithms, Proc. of Congress on
Evolutionary Computation, IEEE Press, Piscataway, USA, pp. 839–844.
Cohen, W. W. (1995). Fast effective rule induction, In Proceedings of the Twelfth International
Conference on Machine Learning, Morgan Kaufmann, pp. 115–123.
Cooper, G. (2000). The Cell: A Molecular Approach, Sinauer Associates, Sunderland, UK.
Davies, M. N., Secker, A., Freitas, A. A., Mendao, M., Timmis, J. & Flower, D. R. (2007).
On the hierarchical classification of G protein-coupled receptors, Bioinformatics
23(23): 3113–3118.
Drenth, J. (1999). Principles of Protein X-Ray Crystallography, Springer-Verlag, New York, USA.
Durbin, R., Eddy, S., Krogh, A. & Mitchison, G. (1998). Biological Sequence Analysis: Probabilistic
Models of Proteins and Nucleic Acids, Cambridge University Press, Cambridge, USA.
Fawcett, T. (2006). An introduction to ROC analysis, Pattern Recognition Letters 27: 861–874.
Ferreira, C. (2001). Gene expression programming: a new adaptative algorithm for solving
problems, Complex Systems 13(2): 87–129.
Frank, E., Hall, M. & Pfahringer, B. (2003). Locally weighted naive bayes, Proc. Conference on
Uncertainty in Artificial Intelligence, Morgan Kaufmann.
Frank, E. & Witten, I. H. (1998). Generating accurate rule sets without global
optimization, Proc. 15th International Conference on Machine Learning, Morgan
Kaufmann, pp. 144–151.
Freitas, A. (1998). Data Mining and Knowledge Discovery with Evolutionary Algorithms,
Heidelberg, Germany.
Freund, Y. & Schapire, R. E. (1999). A short introduction to boosting, Journal of Japanese Society
for Artificial Intelligence 14(5): 771–780.
Friedman, J., Hastie, T. & Tibshirani, R. (2000). Additive logistic regression: a statistical view
of boosting, Annals of Statistics 28(2): 337–407.
Mirceva G. & Davcev, D. (2009). HMM-based approach for classifying protein structures,
International Journal of Bio-Science and Bio-Technology 1(1): 37–46.
256
18 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

Griffiths, A., Miller, J., Suzuki, D., Lewontin, R. & Gelbart, W. (2000). An Introduction to Genetic
Analysis, 7th edn, W.H. Freeman, New York, USA.
Hardin, C., Pogorelov, T. & Luthey-Schulten, Z. (2002). Ab initio protein structure prediction,
Current Opinion in Structural Biology 12(2): 176–181.
Hunter, L. (1993). Artificial Intelligence and Molecular Biology, 1st edn, AAAI Press, Boston, USA.
Jaroniec, C., MacPhee, C., Bajaj, V., McMahon, M., Dobson, C. & Griffin, R. (2004). High
resolution molecular structure of a peptide in an amyloid fibril determined by magic
angle spinning NMR spectroscopy, Proceedings of the National Academy of Sciences of
the USA 101(3): 711–716.
John, G. H. & Langley, P. (1995). Estimating continuous distributions in bayesian classifiers,
Proc. of the Eleventh Conference on Uncertainty in Artificial Intelligence, Morgan
Kaufmann, pp. 338–345.
Kalegari, D. & Lopes, H. (2010). A differential evolution approach for protein structure
optimisation using a 2D off-lattice model, International Journal of Bio-Inspired
Computation 2(3/4): 242–250.
Klabunde, T. & Hessler, G. (2002). Drug design strategies for targeting G-protein-coupled
receptors, Chembiochem 3(10): 928–944.
Kohavi, R. (1995). A study of cross-validation and bootstrap for accuracy estimation and
model selection, Proc. of 14th International Joint Conference on Artificial Intelligence,
Morgan Kaufmann, San Francisco, USA, pp. 1137–1143.
Kohavi, R. (1996). Scaling up the accuracy of naive-bayes classifiers: a decision-tree hybrid,
Proc. of 2nd International Conference on Knowledge Discovery and Data Mining Conference,
pp. 202–207.
Kotsiantis, S. B. (2007). Supervised machine learning: A review of classification techniques,
Informatica 31(3): 249–268.
Koza, J. (1996). Classifying protein segments as transmembrane domains using genetic
programming and architecture-altering operations, in P. Angeline & K. Kinnear Jr.
(eds), Advances in Genetic Programming, Vol. II, MIT Press, Cambridge, USA.
Kyte, J. & Doolittle, R. (1982). A simple method for displaying the hydropathic character of
proteins, Journal of Molecular Biology 157: 105–132.
Landwehr, N., Hall, M. & Frank, E. (2005). Logistic model tree, Machine Learning
59(1): 161–205.
Lee, M., Duan, Y. & Kollman, P. (2001). State of the art in studying protein folding and protein
structure prediction using molecular dynamics methods, Journal of Molecular Graphics
and Modelling 19(1): 146–149.
Lewis, P., Momany, F. & Scheraga, H. (1973). Chain reversals in proteins, Biochimica et
Biophysica Acta 303(2): 211–229.
Lodish, H., Berk, A., Matsudaira, P., Kaiser, C., Krieger, M., Scott, M., Zipursky, L. & Darnell,
J. (2000). Molecular Cell Biology, 4th edn, W.H. Freeman, New York, USA.
Lopes, H. (2008). Evolutionary algorithms for the protein folding problem: A review and
current trends, Computational Intelligence in Biomedicine and Bioinformatics, Vol. I,
Springer-Verlag, Heidelberg, pp. 297–315.
Manning, A., Keane, J., Brass, A. & Goble, C. (1997). Clustering techniques in biological
sequence analysis, in J. Komorowski & J. Zytkow (eds), Principles of Data Mining and
Knowledge Discovery, Vol. 1263 of Lecture Notes in Computer Science, Springer-Verlag,
Heidelberg, Germany, pp. 315–322.
A Comparative Study of Machine Learning
and Evolutionary
A Comparative Computation
Study of Machine Approaches
Learning and Evolutionary Computation for Protein
Approaches Secondary
for Protein Structure
Secondary Structure Classification
Classification 257
19

Martin, B. (1995). Instance-based learning: Nearest neighbor with generalisation, Master’s thesis,
Department of Computer Science, University of Waikato, Waikato, New Zealand.
Matthews, B. (1975). Comparison of the predicted and observed secondary structure of T4
phage lysozyme, Biochimica et Biophysica Acta 405(2): 442–451.
Mitchell, T. M. (1997). Machine Learning, McGraw-Hill, New York, USA.
Nelson, D. & Cox, M. (2008). Lehninger Principles of Biochemistry, 5th edn, W.H. Freeman, New
York, USA.
Nölting, B. (2006). Protein Folding Kinetics, 2nd edn, Springer-Verlag, Berlin, Germany.
Ohkawa, T., Namihira, D., Komoda, N., Kidera, A. & Nakamura, H. (1996). Protein structure
classification by structural transformation, Proc. of IEEE International Joint Symposia on
Intelligence and Systems, IEEE Computer Society Press, Piscataway, USA, pp. 23–29.
Pauling, L., Corey, R. & Branson, H. (1951a). Configurations of polypeptide chains with
favored orientations of the polypeptide around single bonds: two pleated sheets,
Proceedings of the National Academy of Sciences of the USA 37(11): 729–740.
Pauling, L., Corey, R. & Branson, H. (1951b). The structure of proteins: two hydrogen-bonded
helicel configurations of the polypeptide chain, Proceedings of the National Academy of
Sciences of the USA 37: 205–211.
Platt, J. C. (1998). Fast training of support vector machines using sequential minimal
optimization, in C. B. B. Schoelkopf & A. Smola (eds), Advances in Kernel Methods,
MIT Press, Cambridge, USA.
Quinlan, J. (1993). C4.5: Programs for Machine Learning, San Francisco, USA.
Rabiner, L. (1989). A tutorial on hidden Markov models and selected applications in speech
recognition, 77(2): 257–286.
Scapin, M. & Lopes, H. (2007). A hybrid genetic algorithm for the protein folding problem
using the 2D-HP lattice model., in A. Yang, Y. Shan & L. Bui (eds), Success in
Evolutionary Computation, Vol. 92 of Studies in Computational Intelligence, Springer,
Heidelberg, Germany, pp. 205–224.
Seymore, K., McCallum, A. & Rosenfeld, R. (1999). Learning hidden Markov model structure
for information extraction, AAAI 99 Workshop on Machine Learning for Information
Extraction, pp. 37–42.
Shmygelska, A. & Hoos, H. (2005). An ant colony optimisation algorithm for the 2D and 3D
hydrophobic polar protein folding problem, BMC Bioinformatics 6: 30.
Sing, T., Sander, O., Beerenwinke, N. & Lengauer, T. (2005). ROCR: visualizing classifier
performance in R, Bioinformatics 21: 3940–3941.
Sunde, M. & Blake, C. (1997). The structure of amyloid fibrils by electron microscopy and
X-ray diffraction, Advances in Protein Chemistry 50: 123–159.
Tavares, L., Lopes, H. & Lima, C. (2008). A comparative study of machine learning methods
for detecting promoters in bacterial DNA sequences, in D.-S. Huang, D. S. L.
Donald C. Wunsch II & K.-H. Jo (eds), Advanced Intelligent Computing Theories
and Applications, Vol. 5227 of Lecture Notes in Computer Science, Springer-Verlag,
Heidelberg, pp. 959–966.
The UniProt Consortium (2010). The universal protein resource (UniProt) in 2010, Nucleic
Acids Research 38: D142–D148.
Tsunoda, D. & Lopes, H. (2006). Automatic motif discovery in an enzyme database using a
genetic algorithm-based approach, Soft Computing 10: 325–330.
258
20 Computational Biology and AppliedWill-be-set-by-IN-TECH
Bioinformatics

Tsunoda, D., Lopes, H. & Freitas, A. (2011). A genetic programming method for
protein motif discovery and protein classification, Soft Computing pp. 1–12. DOI
10.1007/s00500-010-0624-9.
Wang, T. L. J. & Ma, Q. (2000). Application of neural networks to biological data mining:
A case study in protein sequence classification, In Proc. of the Sixth ACM SIGKDD
International Conference on Knowledge Discovery and Data Mining, pp. 305–309.
Webb, G. I. (2000). Multiboosting: a technique for combining boosting and wagging,
40: 159–197.
Webb, G. I., Boughton, J. R. & Wang, Z. (2005). Not so naïve Bayes: aggregating
one-dependence estimators, Machine Learning .
Weinert, W. & Lopes, H. (2004). Neural networks for protein classification, Applied
Bioinformatics 3(1): 41–48.
Weinert, W. & Lopes, H. (2006). GEPCLASS: A classification rule discovery tool using
gene expression programming, in X. Li, O. R. Zaïane & Z.-H. Li (eds), Advanced
Data Mining and Applications, Vol. 4093 of Lecture Notes in Computer Science,
Springer-Verlag, Heidelberg, Germany, pp. 871–880.
Witten, I., Frank, E. & Hall, M. (2011). Data Mining: Practical Machine Learning Tools and
Techniques, 3rd edn, Morgan Kauffmann, San Francisco.
Wolstencroft, K., Lord, P., Tabernero, L., Brass, A. & Stevens, R. (2006). Protein classification
using ontology classification, Bioinformatics 22(14): e530–e538.
Wüthrich, K. (1986). NMR of Proteins and Nucleic Acids, John Wiley & Sons, New York, USA.
Yanikoglu, B. & Erman, B. (2002). Minimum energy configurations of the 2-dimensional
HP-model of proteins by self-organizing networks, Journal of Computational Biology
9(4): 613–620.
Zheng, Z. & Webb, G. I. (2000). Lazy learning of bayesian rules, Machine Learning 41: 53–87.
 13

Functional Analysis of the Cervical

Carcinoma Transcriptome: Networks and
New Genes Associated to Cancer
Mauricio Salcedo et al.*
Laboratorio de Oncología Genómica, Unidad de Investigación en Enfermedades
Oncológicas, Hospital de Oncología, CMN-SXXI, IMSS,
México DF

1. Introduction
Cancer is one of the most important public health problem in Mexico and worldwide,
especially for female population, breast and cervical cancer (CC) types are the most
frequent. Incidence rates of CC are higher in developing countries 40/100,000 women per
year vs. 10/100,000 in developed countries (1). In Mexico there are 12,000 new reports cases
every year (2). The absence of the screening programs or comparatively ineffective screening
programs lead to relatively late diagnosis of the disease and also in differences in the human
papillomavirus (HPV) infection (3). Several types of HPV are associated with CC worldwide
(4, 5), being the HPV16 the most frequent oncogenic type.
Epidemiological and experimental studies suggest that high risk HPV have an important
role in cervical carcinogenesis. Persistent viral infection, genetic background in combination
with constitutive expression of the viral oncogenes as E6 and E7, are decisive steps for
malignant transformation, because these oncoproteins interact with the tumour suppressor
proteins p53 and pRB, respectively for their degradation (6, 7). Finally, these interactions
could induce cellular proliferation and genetic instability for example, which could promote
the accumulation of mutations and aneuploidy (8). In conclusion, viral oncoproteins have a
general impact in global profile of expressed genes, which could be analyzed by high-
throughput methodologies. One of these techniques is DNA oligonucleotide-based
microarray technology, which allows a rapid and high-throughput detection of thousands of
transcripts simultaneously (9-11).
It has been published several studies about gene expression profiles in HPV infected cells.
Mainly these reports are based on gene expression levels altered by E6 and E7 HPV
oncoproteins (12-17). Regarding changes in gene expression profiles in cervical cancer

*Sergio Juarez-Mendez1, Vanessa Villegas-Ruiz1, Hugo Arreola1, Oscar Perez2, Guillermo Gómez3,
Edgar Roman-Bassaure4, Pablo Romero1, Raúl Peralta1
1 Laboratorio de Oncología Genómica, Unidad de Investigación en Enfermedades Oncológicas, Hospital de
Oncología, CMN-SXXI, IMSS, México DF
2 Laboratorio de Oncología Experimental, Instituto Nacional de Pediatría, SS, México
3 Centro Nacional de Clínica de Displasias, Unidad de Oncología, Hospital General de México, SS.
4 Servicio de Oncología, Hospital General de México, SS.
260 Computational Biology and Applied Bioinformatics

samples, there are a few papers comparing normal cervical expressed genes versus tumors
samples (12, 18), the major aim in those studies was to find potential tumor markers with
clinical value. At present the list of the potential markers is short (p16, survivin).
We have published some works about alterations in gene expression in CC (19, 20). In those
reports we observed that WNT pathway, calcium pathway and some cellular proteases
(MMP11, cathepsin F) could be involved in cervical carcinogenesis. Thus, these findings are
contributing to our knowledge about alterations in CC pathogenesis.
In the present work our microarray data obtained from microarray assay on CC samples
(20) were newly managed and analyzed by using new bioinformatics suite programs and
then, to get others genes altered in cervical cancer, as well as to define signaling pathways
probably implicated in this type of cancer.

2. Material and methods

Biological samples. Eight squamous CC tissues stage IIB (according of International
Federation of Obstetrics and Gynecology, FIGO) HPV16 positive were selected and two
healthy “normal” cervical samples were originally studied. The normal samples were
collected after hysterectomy by uterine myomatosis without HPV infection history. All
DNAs from healthy and cervical cancer samples were subjected to PCR by using general
oligonucleotides against to HPV; and to confirm the HPV type, the positive samples were
then sequenced (data not shown). An important point to eliminate negative false, only the
samples harboured at least 70% of tumour cells or normal epithelial cells were analyzed.
Total RNA was extracted from squamous CC and normal tissues using TRIzol reagent
(GIBCO, BRL, USA). RNA was synthesized and labeled with CodeLink Express Assay
Reagent Kit (Applied Microarrays, GE Healthcare, USA).

2.1 Microarray platform

CodeLink™ Human Whole Genome Microarrays offer comprehensive coverage of the
Human genome, this array have ~57,000 probes and consider transcripts and ESTs
(expression tagged sequences). In this system are included: 1,200 genes for oncogenesis
process, 1,400 for cell cycle, 1,000 for cell-signaling, 3,000 for metabolism, 1,400 for
developmental process, 2,700 of transcription and translation, 1,100 for immune and
inflammation response, 800 for protein phosphorylation, 600 for apoptosis, 1,150 for ion
transport, 400 for synaptic transmission, 200 for kinases, among other.
This array harbors 45,674 genes (based on unique UniGene IDs), 360 positive controls, 384
negative controls, 100 housekeeping genes, one specific and functionally validated probe
and the oligonucleotide probe length of 30-mer.
This platform has been applied in different biological models (21-23). CodeLink Bioarray are
recently introduced, single-color oligonucleotide microarrays, which differ from Affymetrix
GeneChips in the following aspects: 1) this Bioarray use a single pre-synthesized, pre-
validated 30-mer probe to detect each target transcript, whereas Gene-Chips use multiple in-
situ synthesized, 25-mer probes; and 2) the surface of CodeLink Bioarrays is made of 3-
dimensional aqueous gel matrix, whereas that of Affymetrix GeneChips is made of 2-
dimensional glass matrix. These characteristics could suggest that CodeLink Bioarrays
behave differently from GeneChips and may require different normalization strategies from
the ones optimized for GeneChips (24)(Figure 1).
Functional Analysis of the Cervical Carcinoma
Transcriptome: Networks and New Genes Associated to Cancer 261

Data set
Partek analysis and PCA
Clustering Analysis
Ingenuity Pathways Analysis (IPA)

Fig. 1. General strategy of microarray analysis on cervical cancer samples. Originally RNA
samples from normal epithelial cells or CC lesions were subjected to hybridization reaction
on arrays. The data were analyzed by using different bioinformatics Tools to obtain
differentially expressed genes.

3. Analysis of the data

Partek® Genomics SuiteTM is a comprehensive suite of advanced statistics and interactive
data visualization specifically designed to reliably extract biological signals from noisy data.
The commercial software is unique in supporting all microarray and next generation
sequencing technologies including gene expression and digital gene expression,
exon/alternative splicing, RNA-Seq, copy number and association, ChIP-chip, ChIP-seq,
and microRNAs in a single software package, allowing for analysis of multiple applications
in one complete solution.
262 Computational Biology and Applied Bioinformatics

This kind of analysis will provide results with minimal noises generated from the internal
controls. To perform this analysis is necessary to apply a software suite which is composed
by three different statistical tests: Probe Statistical algorithm (MAS5), Probe Logarithmic
Intensity Error (Plier) and Robust Multichip Analysis (RMA). The goal of these tests is to
establish differences and similarities between internal controls and to get the most real data.
In the present case, the normalized samples were analyzed RMA statistical tests eliminating
the tags harboring variations.
In the global gene expression is difficult understand what happen with all of genes in
different cellular process including the cancer, in this context a way of visualization data is a
Principal Component Analysis or PCA (25). This method is a mathematical technique to
reduction the effect of the gene expression sample in a small dimensional space, when there
is less changes in the global gene expression the dimensional is smaller. Next for
visualization of changes in gene expression in all samples we made a clustering analysis
(26), in this method we used a K-means algorithm and let to classify to determine similitude
and dissimilitude in all samples, to finish we applied methods of systems biology as
Ingenuity Pathway Analysis) and gene classification to determine new list of candidates for
subsequent lab verification and might help in the search for a cure for cancers.

3.1 Networks by Ingenuity Pathways Analysis (IPA)

This software (www.ingenuity.com) to help life science researchers explore, interpret, and
analyze complex biological Systems, and is used to help researchers analyze 'omics data and
model biological systems. This analysis was to identify Networks of interacting genes and
other functional groups. A cut-off ratio of 2 was used to define genes.

4. Results
The best and most accurate method for identifying disease-causing genes is monitoring gene
expression values in different samples using microarray technology. One of the
shortcomings of microarray data is that they provide a small quantity of samples with
respect to the number of genes. This problem reduces the classification accuracy of the
methods, so gene selection is essential to improve the predictive accuracy and to identify
potential marker genes for a disease. Among numerous existing methods for gene selection,
PARTEK has become one of the leading methods, but its performance can be reduced
because of the small sample size, noisy data and the fact that the methods remove
redundant genes.
The original cervical dataset was already published and available in a previous report (20).
This dataset was obtained by using CodeLink microarray platform and provides the
expression levels of 57,000 probes for 2 normal tissues and 8 cervical cancers HPV16
positive. The data were pre-processed by carrying out a base 10 logarithmic transformation
and normalized (see Figure 2). After first analysis of the data, using significance analysis of
microarrays 3,248 genes well annotated were identified.
After that, the data already normalized were managed and analyzed using Partek genomics
suite version 6.5 (Partek GS) obtaining a list of differentially expressed genes (Tumor versus
Normal). Specific and new procedures are actually performed in analysis of microarray
data. For instance, after internal control analysis, the raw data are normalized using the
common software available on line or some other like Partek GS. This suite provides
Functional Analysis of the Cervical Carcinoma
Transcriptome: Networks and New Genes Associated to Cancer 263

rigorous and easy-to-use statistical tests for differential expression of genes or exons, and a
flexible and powerful statistical test to detect alternative splicing based on a powerful mixed
model analysis of variance.
By one-side ANOVA statistical test a small group of 208 genes were selected with a false
discovery ratio < 10%, from these, 111 overexpressed genes with a fold change >2, and 97
downregulated genes with a fold change of <-2 genes were observed (Tables 1 and 2).

A)

B)
Fig. 2. Normalization data set. The values of gene expression were normalyzed by medians.
A) Left image is showing the normalization of positive internal controls harboured in the
bioarray. The right image is showing the normalization of negative internal controls of the
array. As example, these figures corresponds to a normal samples.
264 Computational Biology and Applied Bioinformatics

Table 1. List of genes and EST’s (without gene name) up-regulated

Functional Analysis of the Cervical Carcinoma
Transcriptome: Networks and New Genes Associated to Cancer 265

Table 1. List of genes up-regulated

266 Computational Biology and Applied Bioinformatics

Id Gene N am e F o ld -C h a n g e
SCGB1D2 s e c r e t o g lo b in , f a m ily 1 D , m e m b e r 2 -1 8 6 .6 8 6
KRT4 k e r a t in 4 -1 7 3 .8 7 6
FO S F B J m u r in e o s t e o s a r c o m a v ir a l o n c o g e n e h o m o lo g -1 3 8 .0 6 2
MAL m a l, T - c e ll d if f e r e n t ia t io n p r o t e in -1 3 3 .1 7 3
W IS P 2 W N T 1 in d u c ib le s ig n a lin g p a t h w a y p r o t e in 2 -9 0 .6 2 4 2
UNQ698 s u p r a b a s in -8 8 .3 6 3 2
FLJ2 2 6 5 5 R E R G / R A S - lik e -8 8 .2 0 8 6
SPARCL1 S P A R C - lik e 1 ( h e v in ) -8 6 .9 8 8 7
M GC45780 s c a v e n g e r r e c e p t o r c la s s A , m e m b e r 5 ( p u t a t iv e ) -7 8 .2 2 0 5
RGS5 r e g u la t o r o f G - p r o t e in s ig n a lin g 5 -7 3 .2 3 8
MAMDC2 M A M d o m a in c o n t a in in g 2 -6 9 .9 8 7 9
IG F B P 6 in s u lin - lik e g r o w t h f a c t o r b in d in g p r o t e in 6 -6 8 .8 7 6 5
C 1 o r f1 0 c o r n u lin -6 7 .0 5 8 4
FH L1 f o u r a n d a h a lf L I M d o m a in s 1 -6 4 .6 3 3 8
TNA C - t y p e le c t in d o m a in f a m ily 3 , m e m b e r B -5 9 .3 4 0 5
W 57655 - -5 7 .5 2 8 8
PLAC9 p la c e n t a - s p e c if ic 9 -5 1 .9 8 9 6
DF c o m p le m e n t f a c t o r D ( a d ip s in ) -5 0 .8 3 1 7
UNQ467 k e r a t in o c y t e d if f e r e n t ia t io n - a s s o c ia t e d p r o t e in -5 0 .2 5 3 3
PTG D S p r o s t a g la n d in D 2 s y n t h a s e 2 1 k D a ( b r a in ) -5 0 .1 3 4 8
SCGB2A2 s e c r e t o g lo b in , f a m ily 2 A , m e m b e r 2 -4 9 .6 3 3
D PT d e r m a t o p o n t in -4 7 .7 0 4 8
C R IS P 3 c y s t e in e - r ic h s e c r e t o r y p r o t e in 3 -4 6 .6 0 4 8
BNC2 b a s o n u c lin 2 -4 6 .1 5 3 8
OGN o s t e o g ly c in -4 4 .5 3 0 8
EDN3 e n d o t h e lin 3 -4 2 .5 1 6 1
D PT d e r m a t o p o n t in -3 8 .5 7 1 2
FXYD 1 F X Y D d o m a in c o n t a in in g io n t r a n s p o r t r e g u la t o r 1 -3 6 .7 6 5 5
PPP1R3C p r o t e in p h o s p h a t a s e 1 , r e g u la t o r y ( in h ib it o r ) s u b u n it 3 C -3 5 .3 0 7 3
OSR2 o d d - s k ip p e d r e la t e d 2 ( D r o s o p h ila ) -3 4 .3 5 1 2
ANKRD25 K N m o t if a n d a n k y r in r e p e a t d o m a in s 2 -3 2 .6 4 0 6
BF941677 - -3 0 .9 5 0 3
M IT F m ic r o p h t h a lm ia - a s s o c ia t e d t r a n s c r ip t io n f a c t o r -3 0 .7 0 0 2
BQ007074 - -3 0 .1 0 1 1
AL137566 - -2 9 .8 3 3
FBLN 5 f ib u lin 5 -2 9 .4 4 9 4
PGM 5 p h o s p h o g lu c o m u t a s e 5 -2 8 .8 1 9 2
CRYAB c r y s t a llin , a lp h a B -2 8 .3 0 9 4
M FAP4 m ic r o f ib r illa r - a s s o c ia t e d p r o t e in 4 -2 7 .6 3 5 7
AA101632 - -2 7 .2 7 8 4
CO LEC 12 c o lle c t in s u b - f a m ily m e m b e r 1 2 -2 6 .3 8 1 3
PCO LCE2 p r o c o lla g e n C - e n d o p e p t id a s e e n h a n c e r 2 -2 6 .1 8 5 1
C O L14A1 c o lla g e n , t y p e X I V , a lp h a 1 -2 5 .1 7 4 5
ARHGAP6 R h o G T P a s e a c t iv a t in g p r o t e in 6 -2 4 .9 4 1 5
AEBP1 A E b in d in g p r o t e in 1 -2 4 .5 7 3 8
TENC1 t e n s in lik e C 1 d o m a in c o n t a in in g p h o s p h a t a s e ( t e n s in 2 ) -2 2 .8 4 1 3
AN G PTL2 a n g io p o ie t in - lik e 2 -2 2 .7 6 9 2
C O L4A1 c o lla g e n , t y p e I V , a lp h a 1 -2 2 .2 2 7 8
LM O 3 L I M d o m a in o n ly 3 ( r h o m b o t in - lik e 2 ) -2 2 .0 8 2 2
NDRG2 N D R G f a m ily m e m b e r 2 -2 1 .5 6 0 7
FLJ1 0 9 7 0 t r a n s m e m b r a n e p r o t e in 1 0 0 -2 1 .5 0 5 4
CLCA4 c h lo r id e c h a n n e l a c c e s s o r y 4 -2 1 .3 4 1 2
COX7A1 c y t o c h r o m e c o x id a s e s u b u n it V I I a p o ly p e p t id e 1 ( m u s c le ) -2 0 .3 2 9 5
M IT F m ic r o p h t h a lm ia - a s s o c ia t e d t r a n s c r ip t io n f a c t o r -2 0 .1 9 7 6
GNG11 g u a n in e n u c le o t id e b in d in g p r o t e in ( G p r o t e in ) , g a m m a 1 1 -2 0 .1 2 1 1
TACC1 t r a n s f o r m in g , a c id ic c o ile d - c o il c o n t a in in g p r o t e in 1 -1 9 .9 8 5 1
ECG2 s e r in e p e p t id a s e in h ib it o r , K a z a l t y p e 7 ( p u t a t iv e ) -1 9 .9 0 9 5
A I7 3 4 2 1 2 - -1 9 .5 2 7
A I7 6 5 6 3 7 - -1 9 .1 4 1 6
EBF e a r ly B - c e ll f a c t o r 1 -1 8 .9 8 3 2
B A I3 b r a in - s p e c if ic a n g io g e n e s is in h ib it o r 3 -1 8 .8 9 8 5
ARHGAP28 R h o G T P a s e a c t iv a t in g p r o t e in 2 8 -1 8 .5 6 5 4
KCNAB1 t a s s iu m v o lt a g e - g a t e d c h a n n e l, s h a k e r - r e la t e d s u b f a m ily , b e t a m e m b e -1 8 .2 3 7 9
AA578982 - -1 8 .1 5 9 6
CECR6 c a t e y e s y n d r o m e c h r o m o s o m e r e g io n , c a n d id a t e 6 -1 8 .0 4 7 6
AF321976 - -1 7 .2 4 9 9
EDN3 e n d o t h e lin 3 -1 6 .2 8 9 6
M YLK m y o s in lig h t c h a in k in a s e -1 6 .0 1 7 5
C7 c o m p le m e n t c o m p o n e n t 7 -1 5 .7 4 7 3

Table 2. List of genes and EST’s (without gene name) down-regulated

Functional Analysis of the Cervical Carcinoma
Transcriptome: Networks and New Genes Associated to Cancer 267

Table 2. List of genes down-regulated

268 Computational Biology and Applied Bioinformatics

4.1 Analyzing microarray data with novel software suite

If we want to display the data in just two dimensions, we want as much of the variation in
the data as possible captured in just two dimensions. Principal component analysis or PCA
has been developed for this purpose. Applying this PCA method in the cervical data we
observed some expected differences (Figure 3).

Fig. 3. Principal component analysis of cervical cancer samples. As expected, after perform
PCA analysis, the normal samples (red balls) were grouped out of CC Group (blue balls) in
a 3-D image.
In order to obtain a graphical representation of the differences between normal and tumor
tissues, hierarchical cluster analysis was performed on all samples with a pseudo-color
visualization matrix of the 208 selected genes grouping with greater intra-group similarity
and differences between groups. The phylogenetic tree resulting from the hierarchical
complete linkage-clustering algorithm is shown in figure 4. The figure shows those genes
that are changing respect to health cervical tissue. In this method of clustering, allows to do
relationships among objects (genes or samples) and are represented by a tree whose branch
lengths reflect the degree of similarity between the objects, as assessed by a pairwise
similarity function. The computed trees can be used to arrange individual samples in the
original data table; this allows the samples or groups of samples with similar expression
patterns to be shown adjacent to each other.
In general, tumor samples showed heterogeneity among them compared with normal
samples, which had a more homogeneous gene expression profile. So, clustering analysis in
CC failed to show significant segregation of patients based on expression profiling possibly
Functional Analysis of the Cervical Carcinoma
Transcriptome: Networks and New Genes Associated to Cancer 269

Fig. 4. Hierarchical clustering of the gene expression data for cervical tissues. Clustering
analysis were performed for all tumours and normal samples. The data were clustered using
the standard hierarchical method with ward linkage and using the Perason correlation to
determine distance tumour. Before clustering, the data was filtered to remove genes that
were scored absent in 75% or more of the samples as they are likely to be measuring noise in
the system. The cluster of normal samples exhibit nearly identical patterns of gene
expression changes, on contrary, as expected the invasive samples grouped in a different
branch showing a heterogeneous gene expression. Blue color indentify downregulation and
red color an overexpression status.
270 Computational Biology and Applied Bioinformatics

due to the heterogeneous nature of the samples as well as the relatively small numbers of
samples in this study. Even when the samples were subjected to rigorous procedures of
analysis, the special selection of the patients, including age, clinical stage, HPV16 positive
and contraceptive oral status avoiding any bias, in this stage of the carcinogenesis process
(stage IIb) the pattern of gene expression is quite different between samples.

Table 3. Disease and Disorders functions in cervical cancer

Functional Analysis of the Cervical Carcinoma
Transcriptome: Networks and New Genes Associated to Cancer 271

We used IPA to investigate the biological relevant of the observed genome-wide expressed
gene changes by categorizing our data set into biological functions and/or diseases
Ingenuity Pathway analysis was applied. The 208 genes annotated list by PARTEK analysis
was submitted to the visualization IPA tool. This bioinformatics tool is employed for
visualizing expression data in the context of KEGG biological pathways; the importance IPA
is that retrieves an impact factor (IF) of genes that entire pathway involved, which can help
to obtain a clearer notion of the alteration level in each biological pathway, and understand
the complexity of these different process of the cancer cell. We imported a list of
significantly up and down regulated genes (with extension .txt) into the program to convert
the expression data into illustrations in an attempt to explore altered mechanisms in CC. To
overcome any possible incorrect IF in altered pathways due to different size of samples, we
submitted a similar quantity of up and down regulated genes. This allowed confirming that
genes involved in several metabolic pathways were altered in CC (see networks).
We were able to associate biological functions and diseases to the experimental results.
Fifteen pathways were obtained with a high score. Table 3 is showing the genes and the top
three disorders/disease of “small networks” based in the analysis of the data. As can be
seen, a clear route in cancer as it is known was not observed but some genes have been
previously associated; however, these data give important information involving “non
canonical” pathways in cancer.
Finally, in the Figure 5 is showed a “hypothetical network in CC” based from the 15 small
networks. In addition to gene expression values, the proposed method uses Gene Ontology,
which is a reliable source of information on genes. The use of Gene Ontology can
compensate, in part, for the limitations of microarrays, such as having a small number of
samples and erroneous measurement results.

5. Discussion
In our results, non classical “cancer genes” were conserved, respect to expected genes as
MYC, FOS, RB, P53, HIF, etc. However, in the “strict sense of the word” when is considered
a cancer gene? By instance, over-expression, down-regulation, point mutation,
amplification, loss of heterozygosity, polymorphisms, epigenetic changes, etc. Thus, any
gene could be considered like cancer gene, if they are following special criteria as recently
was reported (27).
In this context, we decided to explore two non-related genes in cervical cancer PARK2 gene.
Interestingly, PARK2 gene mutations (point mutations and exonic deletions) were first
identified in autosomal recessive juvenile-onset parkinsonism. This gene is mapped to
6q25.2-q27 containing 12 small exons, and encodes parkin protein which functions as an E3
ligase, ubiquitinating proteins for destruction by the proteosome. Several substrates for
parkin have been identified, including a 22kD glycosolated form of synuclein, parkin-
associated endothelin receptor-like receptor (Pael-R), and CDCrel-1. Over-expression of
Pael-R causes it to become ubiquinated, insoluble, and unfolded, and lead to endoplasmic
reticulum stress and cell death (for review see 28). The location of Parkin is in a
chromosomal region that is frequently deleted in multiple tumor types, including
hepatocellular carcinoma (HCC), ovarian cancer, and breast cancer. The Parkin gene is
within FRA6E, the third most active common fragile site (29,30). Interestingly, all three
fragile sites regions were found consistently deleted in HCC (31) as well as in ovarian,
breast, and prostate cancers. Further PARKIN protein overexpression did not lead to
272 Computational Biology and Applied Bioinformatics

B
Functional Analysis of the Cervical Carcinoma
Transcriptome: Networks and New Genes Associated to Cancer 273

D
Fig. 5. Networks in cervical cancer. Top three networks diagrams generated as graphical
representations of the molecular relationship between genes and gene product. The gene
274 Computational Biology and Applied Bioinformatics

products are represented as nodes (shapes) and the biological relationship between two
nodes is represented as an edge (line). A) Network 1. skeletal and muscular system
development and function, embryonic development, tissue development, B) network 2;
dermatological diseases and conditions, cardiovascular disease, organismal injury and
abnormalities, C) network 3; cell cycle, organismal functions, organismal injury and
abnormalities, D) Ingenuity Pathways Analysis network of genes associated with CC. This
network diagram shows the biological association of 8 focus networks associated with
cervical cancer as graphical representation of the molecular relationship between
genes/gene produtcs.
increased sensitivity to all pro-apoptotic induction but may show specificity for a certain
type of cellular stress. At present, Parkin gene could be considered as new tumor suppressor
gene (32). In our case, per se Parkin expression in CC is interesting. It is widely described
that p53 master gene is constitutively expressed but under stress conditions as HPV
infection, DNA damage or point mutations its half-time life of the protein is increased.
Similar situation might be observed for Parkin gene due to increased expression in CC. This
could be supported because the 6q25 cytogenetic region in CC is not altered as happen for
TP53 gene (33). To this respect, we could hypothesize that a new overexpression of Parkin
gene could be involved in invasion cervical carcinogenesis. These findings could
demonstrate that the genetic context in which a mutation occurs can play a significant role
in determining the type of illness produced or associated.
It has been established that although we inherit two copies of all genes (except those that
reside on the sex chromosomes), there is a subset of these genes in which only the paternal
or maternal copy is functional. This phenomenon of monoallelic, parent-of-origin expression
of genes is termed genomic imprinting. Imprinted genes are normally involved in
embryonic growth and behavioral development, but occasionally they also function
inappropriately as oncogenes and tumor suppressor genes (34). Furthermore, it is well know
that a variety of genetic changes influence the development and progression of cancer.
These changes may result from inherited or spontaneous mutations that are not corrected by
repair mechanisms prior to DNA replication. It is increasingly clear that so called epigenetic
effects that do not affect the primary sequence of the genome also play an important role in
tumorigenesis (35).
Other gene overexpressed seen in this analysis was PEG10 gene. This gene is mapped in
chromosme 7q21. PEG10 protein prevents apoptosis in hepatocellular carcinoma cells
through interaction with SIAH1, a mediator of apoptosis. May also have a role in cell
growth promotion and hepatoma formation. Inhibits the TGF-beta signaling by interacting
with the TGF-beta receptor ALK1. This is a paternally expressed imprinted gene that
encodes transcripts containing two overlapping open reading frames (ORFs), RF1 and
RF1/RF2, as well as retroviral-like slippage and pseudoknot elements, which can induce a -1
nucleotide frame-shift. Increased expression of this gene is associated with hepatocellular
carcinomas. These findings link to cancer genetics and epigenetic by showing that a classic
proto-oncogene, MYC, acts directly upstream of a proliferation-positive imprinted gene,
PEG10 (36,37).
The HOX genes are a family of transcription factors that bind to specific sequences of DNA
in target genes regulating their expression. The role of HOX genes in adult cell
differentiation is still obscure, but growing evidence suggests that they may play an
important role in the development of cancer. We have previously reported that some HOX
Functional Analysis of the Cervical Carcinoma
Transcriptome: Networks and New Genes Associated to Cancer 275

genes could be related to CC. Specifically, HOXA9 was observed expressed in cervical
cancer by RT-PCR end point. In the present work, the data are showing that statistically
significative HOXA9 gene is differentially expressed in CC. Together to HOXB13, D9, D10,
and HOXC cluster (HOXC9, C11–C13) genes this family of genes might be an important
factor involved in CC (35).
It is clear that the most altered genes in CC are not commonly associated to cancer process.
This fact could suggest: 1) the “classic genes of cancer” are statistically significant altered
with tiny values, but there are some exceptions and specific tumor types as neuroblastomas
and N-MYC gene, Her2/neu in breast cancer. 2) At least in stage IIb of cervical carcinogenesis
could be involved genes related to “cellular economy” but not belonging to genes of cancer.
This is supported by recent reports showing molecular alterations in genes not previously
related to cancer (27). 3) The extreme values (high or low) in microarray analysis not always
represent strong candidates of markers in the models performed. What about the most
frequent?. (4) Integrative genomics, multicentre protocols in well selected samples, stratified
stages and clinical follow-up, will be the clue to get cancer hallmarks. In addition, the study
of the molecular function of selected genes strengthened the hypothesis that these genes are
involved in the process of cancer growth.
The data information obtained from microarray analysis should be validated because can
appear errors in positive and negative false due to nature of the massive assays. In this
context, in order to confirm the microarray data, additional molecular tool as end point
PCR, real time PCR, northern blot, immunohistochemistry should be performed and to
obtain results. (Mendez S. An Integrative microarray gene expression analysis, approach
identifies candidates’ array multi-experiments in Ovary Tumours, submitted to publication
2011).

6. Acknowledgements
This work was partially supported by CONACYT (México) grants 69719 AND 87244 from
FONDOS SECTORIALES. We appreciate the technical assistance of Laboratorio de
Oncología Genómica, CIS, HO-IMSS. Sergio JUAREZ and Mauricio SALCEDO made
similar efforts in the present work.

7. References
[1] Parkin DM, Bray F, Ferlay J, Pisani P. Global cancer statistics, 2002. CA Cancer J Clin.
2005, 55(2):74-108
[2] Secretaría de Salud (Méx). Registro Histopatológico de Neoplasias en México. México:
Secretaría de Salud, México D.F; 1999.
[3] Munoz N, Xavier Bosch F. Cervical cancer and human papillomavirus: epidemiological
evidence and perspectives for prevention. Salud Publica Mex 1997; 39:274–282.
[4] Boffetta P, Parkin DM. Cancer in developing countries. CA Cancer J Clin 1994 44:81-90
[5] Walboomers JM, Jacobs MV, Manos MM, Bosch FX, Kummer JA, Shah K, et al. Human
papillomavirus is a necessary cause of invasive cervical cancer worldwide. J Pathol
1999, 189: 12–19.
276 Computational Biology and Applied Bioinformatics

[6] Scheffner, M., B. A. Werness, J. M. Huibregtse, A. J. Levine, and P. M. Howley. The E6

oncoprotein encoded by human papillomavirus types 16 and 18 promotes the
degradation of p53. Cell 1990, 63:1129–1136.
[7] Boyer SN, Wazer DE, Band V. E7 protein of human papilloma virus-16 induces
degradation of retinoblastoma protein through the ubiquitin-proteasome pathway.
Cancer Res. 1996, 56(20):4620-4624.
[8] Duensing S, Munger K. The human papillomavirus type 16 E6 and E7 oncoproteins
independently induce numerical and structural chromosome instability. Cancer
Res. 2002, 62(23):7075-7082.
[9] Nees M, van Wijngaarden E, Bakos E, Schneider A, Dürst M. Identifcation of novel
molecular markers which correlate with HPVinduced tumor progression.
Oncogene 1998, 16:2447–2458.
[10] Brown P, Botstein R. Exploring the new world of the genome with DNA microarrays.
Nature Genet 1999, 21 (suppl):33-37.
[11] Lockhart DJ, Winzeler EA. Genomics, gene expression and DNA arrays. Nature, 2000,
405:827-836.
[12] Ruutu M, Peitsaro P, Johansson B, Syrjanen S: Transcriptional profiling of a human
papillomavirus 33-positive squamous epithelial cell line which acquired a selective
growth advantage after viral integration. Int J Cancer 2002, 100: 318-326.
[13] Duffy CL, Phillips SL, Klingelhutz AJ: Microarray analysis identifies differentiation-
associated genes regulated by human papillomavirus type 16 E6. Virol 2003, 314:
196-205.
[14] Thomas JT, Oh ST, Terhune SS, Laimins LA: Cellular changes induced by low-risk
human papillomavirus type 11 in keratinocytes that stably maintain viral episomes.
J Virol 2001, 75: 7564-7571.
[15] Garner-Hamrick PA, Fostel JM, Chien WM, Banerjee NS, Chow LT, Broker TR, Fisher
C: Global effects of human papillomavirus type 18 E6/E7 in an organotypic
keratinocyte culture system. J. Virol 2004, 78: 9041-9050.
[16] Toussaint-Smith E, Donner DB, Roman A: Expression of human papillomavirus type
16 E6 , E7 oncoproteins in primary foreskin keratinocytes is sufficient to alter the
expression of angiogenic factors. Oncogene 2004, 23: 2988-2995.
[17] Nees M, Geoghegan JM, Hyman T, Frank S, Miller L, Woodworth CD: Papillomavirus
type 16 oncogenes downregulate expression of interferon-responsive genes and
upregulate proliferation-associated, NF-kappaB-responsive genes in cervical
keratinocytes. J Virol 2001, 75: 4283-4296.
[18] Shim C, Zhang, Hun C, Lee. Profiling of differentially expressed genes in human
primary cervical cancer by complementary DNA expression array. Clin Cancer Res
1998, 4:3045-3050.
[19] Vazquez-Ortiz G, García JA, Ciudad CJ, Noé V, Peñuelas S, López-Romero R,
Mendoza-Lorenzo P, Piña-Sánchez P, Salcedo MDifferentially expressed genes
between high-risk human papillomavirus types in human cervical cancer cells. Int J
Gynecol Cancer. 2007, 17:484-491
Functional Analysis of the Cervical Carcinoma
Transcriptome: Networks and New Genes Associated to Cancer 277

[20] Pérez-Plasencia C, Vázquez-Ortiz G, López-Romero R, Piña-Sanchez P, Moreno J,

Salcedo M. Genome wide expression analysis in HPV16 cervical cancer:
identification of altered metabolic pathways. Infect Agent Cancer. 2007, 6;2:16
[21] Davidson LA, Nguyen DV, Hokanson RM, Callaway ES, Isett RB, Turner ND,
Dougherty ER, Wang N, Lupton JR, Carroll RJ, Chapkin RS. Chemopreventive n-3
polyunsaturated fatty acids reprogram genetic signatures during colon cancer
initiation and progression in the rat. Cancer Res. 2004, 64:6797-6804.
[22] Ramakrishnan R, Dorris D, Lublinsky A, Nguyen A, Domanus M, Prokhorova A,
Gieser L, Touma E, Lockner R, Tata M, Zhu X, Patterson M, Shippy R, Sendera TJ,
Mazumder A. An assessment of Motorola CodeLink microarray performance for
gene expression profiling applications. Nucleic Acids Res. 2002, 30:e30.
[23] Nie AY, McMillian M, Parker JB, Leone A, Bryant S, Yieh L, Bittner A, Nelson J,
Carmen A, Wan J, Lord PG. Predictive toxicogenomics approaches reveal
underlying molecular mechanisms of nongenotoxic carcinogenicity. Mol Carcinog.
2006, 45:914-933.
[24] Wu W, Dave N, Tseng G, Richards T, Xing EP, Kaminski N. Comparison of
normalization methods for CodeLink Bioarray data. BMC Bioinformatics 2005,
6:309
[25] Reich D, Price Al, Pattterson N. Principal component analysis of genetic data. Nature
Genetics. 2008, 40(5): 491-492
[26] Eisen MB, Spellman PT, Brown PO, Botstein D. Cluster analysis and display of
genome-wide expression patterns. Proc Natl Acad Sci U S A 1998, 95: 14863–14868
[27] Seltzer MJ, Bennett BD, Joshi AD, Gao P, Thomas AG, Ferraris DV, Tsukamoto T, Rojas
CJ, Slusher BS, Rabinowitz JD, Dang CV, Riggins GJ. Inhibition of glutaminase
preferentially slows growth of glioma cells with mutant IDH1. Cancer Res. 2010,
70:8981-8987.
[28] Schapira A. Etiology of Parkinson’s disease Neurol 2006, 4:S10-S23
[29] Kahkonen M. Population cytogenetics of folate-sensitive fragile sites. Common fragile
sites. Hum Genet 1988, 80:344–348.
[30] Denison SR, Wang F, Mueller B, Kock N, Phillips LA, Klein C, Smith DI. Genetic
alteration in the common fragile site gene PARK2 in ovarian and other cancers.
Oncogene 2003, 22:8370– 8378.
[31] Zhao P, Song X, Nin YY, Lu YL, Li XH. Loss of fragile histidine triad protein in human
hepatocellular carcinoma. World J Gastroenterol 2003, 9:1216–1219.
[32] Wang F, Denison S, Lai JP, Philips LA, Montoya D, Kock N, Schule B, Klein C, Shridhar
V, Roberts LR, Smith D. Parkin Gene Alterations in Hepatocellular Carcinoma.
Genes Chromosome & cancer 2004, 40:85–96
[33] Hidalgo A, Baudis M, Petersen I, Arreola H, Piña P, Vazquez G, Hernandez D,
Gonzalez JL, Lazos M, Lopez R, Salcedo M. Microarray comparative genomic
hybridization detection of chromosomal imbalances in uterine cervix carcinoma.
BMC Cancer 2005, 5, 77
[34] Jirtle RL. Genomic imprinting and cancer. Exp Cell Res. 1999, 248:18-24.
[35] Plass C, Soloway PD. DNA methylation, imprinting and cancer. Eur J Hum Genet
2002, 10:6-16
278 Computational Biology and Applied Bioinformatics

[36] Li CM, Margolin AA, Salas M, Memeo L, Mansukhani M, Hibshoosh H, Szabolcs M,

Klinakis A, Tycko B. PEG10 Is a c-MYC Target Gene in Cancer Cells. Cancer Res
2006, 66: 665-672
[37] Tsuji K, et al. PEG10 is a probable target for the amplification at 7q21 detected in
hepatocellular carcinoma. Cancer Genet Cytogenet, 2010, In press.
[38] López R, Garrido E, Vázquez G, Piña P, Pérez C, Alvarado I, Salcedo M. A subgroup of
HOX Abd-B gene is differentially expressed in cervical cancer. Int J Gynecol
Cancer. 2006, 16:1289-96
 14

Number Distribution of Transmembrane

Helices in Prokaryote Genomes
Ryusuke Sawada and Shigeki Mitaku
Nagoya University
Japan

1. Introduction
Number distribution of transmembrane helices represents genetic feature of survival
strategy, because the number of transmembrane helices is closely related to the functional
group of membrane proteins: for example, most of membrane proteins that have six
transmembrane helices belong to transporter functional group. Survival strategies were
obtained by evolutionary mechanism that changes the genome sequences. Comparisons of
number distributions of transmembrane helices among species that have different survival
strategies help us to understand the evolutionary mechanism that has increased the
categories of membrane proteins.
Some studies about how the categories of protein functions have been increased during
evolution were performed using protein database (Chothia et al., 2003; Huynen & van
Nimwegen, 1998; Koonin et al., 2002; Qian et al., 2001; Vogel et al., 2005). However, these
studies were carried out by the analysis almost for soluble proteins. Classification of protein
function groups are often carried out by the empirical methods such as sequence homology
that use sequence information of three-dimensional structure resolved proteins as template
sequences for each functional group. However three-dimensional structure resolved
membrane proteins were much less than that for the soluble proteins because of
experimental difficulty of membrane proteins.
In the previous study, we developed membrane protein prediction system SOSUI and signal
peptide prediction system SOSUIsignal (Gomi et al., 2004; Hirokawa et al., 1998). By
combination of those systems, number of transmembrane helices can be predicted based not
on empirical but on physicochemical parameters. Therefore, it is possible to investigate the
number distribution of transmembrane regions in membrane proteins comprehensively
among various genomes by using SOSUI and SOSUIsignal.

2. Membrane protein prediction systems

SOSUI prediction software (Hirokawa et al., 1998; Mitaku et al., 2002) for transmembrane
helix regions uses physicochemical features of transmembrane helix segments.
Transmembrane helix regions have three common features: (1) a hydrophobic core at the
center of the helix segment; (2) amphiphilicity at both termini of each helix region; and (3)
length of transmembrane helix regions. These features are essential factors for the
280 Computational Biology and Applied Bioinformatics

transmembrane segment to stably present at the cell membrane. The SOSUI system first
enumerates candidates of transmembrane regions by the average hydrophobicity of
segments which are then discriminated by the distributions of the hydrophobicity and the
amphiphilicity around the candidate segments.
SOSUIsignal (Gomi et al., 2004) predicts signal peptides that are removed from proteins that
are secreted to the extracellular space via the secretary process. Signal peptides are present
at the amino terminal segment of their respective proteins; the physicochemical features N-
terminal structure is recognized by molecular modules during the cleavage process. The
SOSUIsignal system is similar to the SOSUI system in that candidates are first enumerated
by the average hydrophobicity at the amino terminal region and then real signal peptides
are discriminated by several parameters.
By focusing on these physicochemical features, accuracy of the prediction systems is very
good: approximately 95%for SOSUI and 90% for SOSUIsignal. By using these softwares, we
can estimate not only function unknown protein sequence but also simulated ones.

3. Typical number distribution of transmembrane regions in membrane

proteins
We investigated the population for number groups of transmembrane helices for 557
prokaryote genomes using SOSUI and SOSUIsignal. Figure 1 shows the results of the
analysis of the membrane protein encoded in the E. coli genome as a typical example; the

1000
Population

100

10

1
0 5 10 15 20
Number of transmembrane helices
Fig. 1. Number distribution of transmembrane helices for E. coli. Estimation of the number of
transmembrane helices for membrane proteins was performed using SOSUI and
SOSUIsignal. Transmembrane helices number zero means soluble proteins.
Number Distribution of Transmembrane Helices in Prokaryote Genomes 281

largest category of membrane proteins comprised proteins with only one transmembrane-
spanning helix, and the second largest category comprised proteins with two
transmembrane-spanning helices. The populations within each category decreased
gradually up to 4 transmembrane helices and then there is a plateau from 4-13 helices. The
population within each category decreased rapidly for categories comprising proteins with
more than 13 transmembrane helices and there were apparently no proteins with more than
16 transmembrane helices. These results indicated that membrane proteins that have a
particular number of transmembrane helices, such as 12, are important for E. coli.

4. Variety of number distribution of transmembrane regions among

organisms
The general trend of the number distribution of transmembrane helices were very similar
among 557 prokaryotic genomes, but the fine structures of the number distribution of
transmembrane helices can change during the evolution. Four graphs in Fig.2 show the
results for the analysis for four prokaryotic genomes: A, Pyrobaculum calidifontis; B,
Thermoplasma volcanium; C, Pseudomonas putida and D, Thermotoga petrophila. We selected
these four kinds of organisms for showing how the number distributions of transmembrane
helices are different among organisms. The number distribution of transmembrane helices
for P. calidifontis did not show significant shoulder at 13 helices as E. Coli. The shape of the
distribution for T. volcanium was very similar to that for E. Coli, although the population was
much smaller. A significant peak was observed at 12 helices for P. putida. A peak at 6 helices
was observable for T. petrophila. Despite of the difference in the number distribution of
transmembrane helices among organisms, the general trend of the distribution suggests the
existence of a target distribution.

5. Number distribution of transmembrane helices in proteins in organisms

grouped by GC contents
If the difference in the number distribution is due to the fluctuation around a target
distribution, the difference would decrease by averaging of the distribution of many
organisms. In contrast if the difference is due to some systematic change among
organisms, the difference would not disappear by a simple procedure of the averaging.
The GC content of genomes differs widely among species, from 0.3 to 0.7, and it is well
known that various characteristics of prokaryotic cells systematically change according to
GC content. Therefore, we investigated whether the distribution in the number of
transmembrane helices per protein changed according to the GC content. Genomes for 557
prokaryotes were classified into nine groups with different GC content. In Fig. 3, the
average number distributions of transmembrane helices in the nine groups indicated that
the distributions were unchanged despite differences in GC content. The membrane-
protein profile of the nine groups shared a common feature in that the general shapes of
the curves were the same; the curves gradually decreased in the population of each
category of membrane protein and there was a shoulder at the categories with 12
transmembrane helices. This result strongly suggests that the difference in the fine
structures of the number distribution is due to the fluctuation around a general curve of
Fig. 3. Then, aquestion arises about the natural selections: Is the general curve formed by
the pressure of natural selection?
282 Computational Biology and Applied Bioinformatics

1000
Population

100

10

1
0 5 10 15 20
Number of transmembrane helices

(a)

1000

100
Population

10

1
0 5 10 15 20
Number of transmembrane helices

(b)
Number Distribution of Transmembrane Helices in Prokaryote Genomes 283

1000
Population

100

10

1
0 5 10 15 20
Number of transmembrane helices
(c)

1000
Population

100

10

1
0 5 10 15 20
Number of transmembrane helices

(d)

Fig. 2. Number distributions of transmembrane helices for four prokaryotes P. calidifontis

(A), T. volcanium (B), P. putida (C) and T. petrophila (D).
284 Computational Biology and Applied Bioinformatics

0
10

-1
10 0.7

0.6
Population

GC content
-2
10
0.5
-3
10
0.4
-4
10
0.3
-5
10
0 5 10 15 20
Number of transmembrane helices
Fig. 3. Number distributions in nine groups of organisms classified by GC content. 557
prokaryotes were divided into nine groups according to GC content (0.3, 0.35, 0.4, 0.45, 0.5,
0.55, 0.6, 0.65 and 0.7).

6. Random sequence simulation

Presumably, functionally important proteins are maintained in biological genomes by
natural selection. Therefore, the general curve of the number distribution of transmembrane
helices must reflect natural selection that occurs during biological evolution. The prediction
systems, SOSUI and SOSUIsignal, have the great advantage that they are applicable to any
amino acid sequences independent of empirical information because they are based mainly
on the physicochemical parameters of amino acids. So, we planned to use the prediction
system for comparing the number distributions of transmembrane helices between the real
genomes and the simulated genomes in which comprehensive mutations are introduced
with any pressure of the natural selection. Therefore, we investigated the effect of random
mutation uncoupled from natural selection on the number distribution of transmembrane
helices using random sequence simulations. The E. coli genome was used for the random
sequence simulation. At each simulation step, one in every 100 amino acids in all protein
sequences was mutated randomly. When the amino acids were mutated, the new amino
acids were determined according to the genomic amino acid composition of the E. coli
genome. Distributions of number of transmembrane helices were estimated by using
membrane protein prediction systems SOSUI and SOSUIsignal after each simulation step.
Simulations were reiterated until 500 simulation steps.
Distributions of transmembrane helices in membrane proteins for simulated genomes are
shown in Fig. 4. As the simulation steps proceed, the number of membrane proteins with
more than six transmembrane helices decreased monotonously and the shoulder in the
distribution around 12 transmembrane helices disappeared. Beyond 300 simulation steps at
which the sequences were completely randomized, the distribution became very similar to a
single exponential decay. A broken line in Fig.4 represents the single exponential decay
Number Distribution of Transmembrane Helices in Prokaryote Genomes 285

curve, y = 2090e-0.87x, which was obtained least square deviation analysis for the averaged
distribution between 300 and 500 steps.

1000
500

Simulation times
400
Population

100
300

200
10
100

1
0 5 10 15 20
Number of transmembrane helices
Fig. 4. Number distribution of transmembrane helices in E. coli genomes subjected to
random mutation of amino acid sequences. Distributions of simulated genomes are
represented as gray scaled rectangles. Open circle indicates the distribution for the original
(simulation step zero) genome of E. coli. Dotted line represents the fitting result of y = 2090e-
0.87x for the averaged distribution between 300 and 500 simulation steps.

After 300 simulation steps, shapes of the distributions of number of transmembrane helices
for simulated genomes were almost unchanged in spite of additional mutations. A single
exponential distribution for simulated genome can be explained by a kind of reaction in the
evolutionary time scale changing the number of membrane proteins due to extensive
mutations.

TMi ↔ TMi + 1

in which TMi represents a membrane protein with i transmembrane helices. If the

equilibrium constant is the same among the distinct equilibrium state, the shape would
become the exponential, as follow:

k − TMn = k + TMn − 1
n
k+ ⎛ k+ ⎞
TMn = −
TMn − 1 = ⎜⎜ − ⎟⎟ TM0
k ⎝k ⎠
where <TMn>, <TMn-1> and <TM0> represent the population of the membrane protein, with
n, n-1 and 0 helices, respectively, and k+/k- means the equilibrium constant. In the
simulation, the equilibrium constants for each transmembrane helices number group are the
same from the algorithm of the prediction systems, and a single exponential decay in the
computer experiment is well interpreted by this model. However, in the real genome, the
shape of distribution is not exponential, showing a significant plateau and shoulder. This
286 Computational Biology and Applied Bioinformatics

indicates that there equilibrium constants for each transmembrane helices number groups
are not same. This may be due to the difference of the functional importance among
membrane protein groups.

7. References
Chothia, C., Gough, J., Vogel, C. & Teichmann, S.A. (2003). Evolution of the protein
repertoire. Science, Vol. 300, No. 5626, 1701-1703
Gomi, M., Sonoyama, M. & Mitaku, S. (2004). High performance system for signal peptide
prediction: SOSUIsignal. Chem-Bio Informatics Journal, Vol. 4, No. 4, 142-147
Hirokawa, T., Boon-Chieng, S. & Mitaku, S. (1998). SOSUI: classification and secondary
structure prediction system for membrane proteins. Bioinformatics, Vol. 14, No. 4,
378-379
Huynen, M.A. & van Nimwegen, E. (1998). The frequency distribution of gene family sizes
in complete genomes. Mol Biol Evol, Vol. 15, No. 5, 583-589
Koonin, E.V., Wolf, Y.I. & Karev, G.P. (2002). The structure of the protein universe and
genome evolution. Nature, Vol. 420, No. 6912, 218-223
Mitaku, S., Hirokawa, T. & Tsuji, T. (2002). Amphiphilicity index of polar amino acids as an
aid in the characterization of amino acid preference at membrane-water interfaces.
Bioinformatics, Vol. 18, No. 4, 608-616
Qian, J., Luscombe, N.M. & Gerstein, M. (2001). Protein family and fold occurrence in
genomes: power-law behaviour and evolutionary model. J Mol Biol, Vol. 313, No. 4,
673-681
Vogel, C., Teichmann, S.A. & Pereira-Leal, J. (2005). The relationship between domain
duplication and recombination. J Mol Biol, Vol. 346, No. 1, 355-365
 15

Classifying TIM Barrel Protein Domain Structure

by an Alignment Approach Using Best Hit
Strategy and PSI-BLAST
Chia-Han Chu1, Chun Yuan Lin2, Cheng-Wen Chang1,
Chihan Lee3 and Chuan Yi Tang4
1NationalTsing Hua University,
2Chang
Gung University,
3Chipboud Technology Corporation,
4Providence University

Taiwan

1. Introduction
High-tech large-scale sequencing projects have identified a massive number of amino acid
sequences for both known and putative proteins, but information on the three-dimensional
(3D) structures of these proteins is limited. Several structure databases, such as the
Structural Classification of Proteins (SCOP (Andreeva et al., 2008), release version 1.73) and
the Class, Architecture, Topology, and Homologous superfamily (CATH (Cuff et al., 2009),
release version 3.2.0), contain fewer than 60,000 entries in the Protein Data Bank (PDB
(Berman et al., 2000), released on 12 May, 2009). This number of entries constitutes only
about 15% of entries in Swiss-Prot (Bairoch et al., 2004), release version 57.2, with more than
400,000 entries). Either X-ray diffraction or NMR can be used to determine the 3D structure
of a protein, but each method has its limitation (Dubchak et al., 1995). As such, extracting
structural information from sequence databases is an important and complementary
alternative to these experimental methods, especially when swiftly determining protein
functions or discovering new compounds for medical or therapeutic purposes.
From ASTRAL SCOP 1.73, it has been estimated that ~10% of known enzymes have
triosephosphate isomerase (TIM) barrel domains. Moreover, TIM barrel proteins have been
identified in five of six enzyme classes, oxidoreductases, transferases, hydrolases, lyases and
isomerases, in the Enzyme nomenclature (ENZYME (Bairoch, 2000), released on 5 May,
2009) database; the ligases class does not contain TIM barrel protein. TIM barrel proteins are
diverse in sequence and functionality and thus represent attractive targets for protein
engineering and evolutionary studies. It is therefore important to examine TIM barrel
protein domain structure classification in SCOP and ENZYME.
In SCOP, there are six levels of hierarchy: class, fold, superfamily, family, protein domain
and species. The classification of protein structures has, more recently, been facilitated by
computer-aided algorithms. Previous research (Chou & Zhang, 1995; Dubchak et al., 1995;
Lin et al., 2005, 2007) has shown that an overall prediction accuracy rate of 70-90% can be
288 Computational Biology and Applied Bioinformatics

easily achieved by using only amino acid sequence information to classify most of proteins
into four major classes in SCOP (all-alpha (α), all-beta (β), alpha/beta (α/β) and alpha+beta
(α+β)) (Murzin, 1995). For the α/β class (constituting TIM barrel proteins), the overall
prediction accuracy rate achieved 97.9% (Lin et al., 2005, 2007). However, less optimal
results were obtained if a more complicated category was used, such as protein folding
patterns. The overall prediction accuracy rate for classifying 27 fold categories in SCOP only
achieved only 50-70% using amino acid sequence information (Ding & Dubchak, 2001;
Huang et al., 2003; Lin et al., 2005, 2007; Shen & Chou, 2006; Vapnik, 1995; Yu et al., 2003).
Although the classification for the SCOP fold category is still a challenge, the overall
prediction accuracy rate for the TIM barrel fold is 93.8% (Yu et al., 2003). Based on the above
results, it is possible to further classify TIM barrel proteins into the SCOP superfamily and
family categories. Four projection methods, PRIDE (Carugo & Pongor, 2002; Gáspári et al.,
2005), SGM (Rogen & Fain, 2003), LFF (Choi et al., 2004) and SSEF (Zotenko et al., 2006,
2007), have been proposed for protein structure comparisons. Zotenko et al. (Zotenko et al.,
2006) compared these four methods for classifying proteins into the SCOP fold, superfamily
and family categories and showed that the SSEF method had the best overall prediction
accuracy rate. The SSEF method utilizes 3D structure information to generate the triplet of
secondary structure elements as the footprints in the comparisons.
Hence, in this chapter, an alignment approach using the pure best hit strategy, denoted
PBH, is proposed to classify the TIM barrel protein domain structures in terms of the
superfamily and family categories in SCOP. This approach requires only amino acid
sequence information to generate alignment information, but secondary and 3D structure
information is also applied in this approach, respectively, to compare the performances with
each other. This work is also used to perform the classification for the class category in
ENZYME. Two testing data sets, TIM40D and TIM95D from ASTRAL SCOP 1.71
(Chandonia et al., 2004), were tested to evaluate this alignment approach. First, for any two
proteins, we adopt the tools CLUSTALW (Thompson et al., 1994), SSEA (Fontana et al.,
2005) and CE (Shindyalov & Bourne, 1998) to align the amino acid sequences, secondary and
3D structures, respectively, to obtain the scores of sequence identity, secondary structure
identity and RMSD. These scores are then used to build an alignment-based protein-protein
identity score network. Finally, a PBH strategy is used to determine the prediction result of
a target protein by selecting the protein having the best score for the target protein
according to this network. This score can be calculated by a single parameter, such as
sequence identity, or mixed parameters by combing two or three single parameters, such as
combining sequence identity and secondary structure identity. In this chapter, we only
consider the single parameter. To verify the stability of the proposed alignment approach,
we also use the novel TIM barrel proteins in TIM40D and TIM95D from ASTRAL SCOP 1.73
that do not existed in ASTRAL SCOP 1.71. For this test, the alignment-based protein-protein
identity score network constructed by the TIM barrel proteins from ASTRAL SCOP 1.71 and
the PBH strategy are used to predict the classification result for each novel TIM barrel
protein. In addition, we further adopt the PSI-BLAST method as a filter for the PBH
strategy, denoted the BHPB strategy, to reduce the number of false positives. The
experimental results demonstrated that the alignment approach with the PBH strategy or
BHPB strategy is a simple and stable method for TIM barrel protein domain structure
classification, even when only the amino acid sequence information is available.
Classifying TIM Barrel Protein Domain Structure by an
Alignment Approach Using Best Hit Strategy and PSI-BLAST 289

2. Materials
2.1 TIM barrel proteins from ASTRAL SCOP 1.71
Two data sets, TIM40D and TIM95D, were used to evaluate the proposed PBH and BHPB
alignment strategies. TIM40D contains 272 TIM barrel protein domain sequences (abbreviated
to TIM sequences) extracted from the 40D set in ASTRAL SCOP 1.71, in which any two
proteins must have ≤40% sequence identity based on PDB SEQRES records. TIM95D contains
439 TIM sequences extracted from the 95D set in ASTRAL SCOP 1.71, in which any two
proteins must have ≤95% sequence identity based on PDB SEQRES records. For TIM40D and
TIM95D, we directly retrieved amino acid sequences and 3D structures from ASTRAL SCOP
1.71 but excluded redundant and possible mutant data. Secondary structure information for
each TIM barrel protein with eight states (H, I, G, E, B, S, T and _) was first derived from the
digital shape sampling and processing (DSSP (Kabsch & Sander, 1983)) program. Then the
eight states for each TIM barrel protein were then reduced to three states (H, E and C)

Superfamily categories Index N40D* N95D*

Triosephosphate isomerise (TIM) 1 3 16
Ribulose-phosphate binding barrel 2 19 30
Thiamin phosphate synthase 3 2 2
FMN-linked oxidoreductases 4 15 22
Inosine monophosphate dehydrogenase (IMPDH) 5 3 5
PLP-binding barrel 6 8 10
NAD(P)-linked oxidoreductase 7 8 21
(Trans)glycosidases 8 82 134
Metallo-dependent hydrolases 9 18 22
Aldolase 10 31 48
Enolase C-terminal domain-like 11 12 24
Phosphoenolpyruvate/pyruvate domain 12 12 22
Malate synthase G 13 1 2
RuBisCo, C-terminal domain 14 4 10
Xylose isomerise-like 15 7 15
Bacterial luciferase-like 16 7 9
Nicotinate/Quinolinate PRTase C-terminal domain-like 17 4 5
PLC-like phosphodiesterases 18 5 5
Cobalamin (vitamin B12)-dependent enzymes 19 5 6
tRNA-guanine transglycosylase 20 2 2
Dihydropteroate synthetase-like 21 4 6
UROD/MetE-like 22 4 4
FAD-linked oxidoreductase 23 3 3
Pyridoxine 5’-phosphate synthase 24 1 1
Monomethylamine methyltransferase MtmB 25 1 1
Homocysteine S-methyltransferase 26 2 3
(2r)-phospho-3-sulfolactate synthase ComA 27 1 2
Radical SAM enzymes 28 3 3
GlpP-like 29 1 1
CutC-like 30 1 1
ThiG-like 31 1 2
TM1631-like 32 2 2
*N40D: the number of TIM sequences in TIM40D
*N95D: the number of TIM sequences in TIM95D

Table 1. Non-redundant data sets, TIM40D and TIM95D, of superfamily categories in SCOP
290 Computational Biology and Applied Bioinformatics

Index N40D* N95D* Index N40D* N95D* Index N40D* N95D*

1.1 3 16 9.6 4 5 16.3 2 3
2.1 2 5 9.7 1 1 16.4 2 2
2.2 2 4 9.8 1 1 17.1 2 3
2.3 4 7 9.9 1 1 17.2 2 2
2.4 10 13 9.11 1 1 18.1 1 1
2.5 1 1 9.12 1 1 18.2 2 2
3.1 2 2 9.13 1 1 18.3 2 2
4.1 15 22 10.1 18 29 19.1 2 2
5.1 3 5 10.2 2 3 19.2 1 1
6.1 7 9 10.3 3 5 19.3 1 2
6.2 1 1 10.4 3 6 19.4 1 1
7.1 8 21 10.5 3 3 20.1 2 2
8.1 25 48 10.6 2 2 21.1 2 4
8.3 26 41 11.1 1 6 21.2 2 2
8.4 4 12 11.2 11 18 22.1 2 2
8.5 13 18 12.1 1 5 22.2 2 2
8.6 3 3 12.2 1 2 23.1 2 2
8.7 2 2 12.3 1 2 23.2 1 1
8.8 3 3 12.5 4 4 24.1 1 1
8.9 1 1 12.7 4 6 25.1 1 1
8.10 1 2 12.8 1 3 26.1 2 3
8.11 1 1 13.1 1 2 27.1 1 2
8.12 1 1 14.1 4 10 28.1 1 1
8.13 1 1 15.1 1 1 28.2 1 1
8.14 1 1 15.2 1 1 28.3 1 1
9.1 1 2 15.3 2 10 29.1 1 1
9.1 2 2 15.4 1 1 30.1 1 1
9.2 1 3 15.5 1 1 31.1 1 2
9.3 2 2 15.6 1 1 32.1 2 2
9.4 1 1 16.1 2 2 - - -
9.5 1 1 16.2 1 2 - - -
*N40D: the number of TIM sequences in TIM40D
*N95D: the number of TIM sequences in TIM95D

Table 2. Non-redundant data sets, TIM40D and TIM95D, of family categories in SCOP

Class categories Index N40D* N95D*

Oxidoreductases 1 27 46
Transferases 2 31 53
Hydrolases 3 68 106
Lyases 4 58 97
Isomerases 5 23 49
undefined - 67 91
*N40D: the number of TIM sequences in TIM40D
*N95D: the number of TIM sequences in TIM95D
The sum of N40D and N95D are 274 and 442, respectively. TIM sequences:
“d1pii_2” and “d1pii_1” in TIM40D and TIM95D have multiple EC numbers for class categories;
“d1b9ba_” and “d1jvna1” in TIM95D have multiple EC numbers for class categories

Table 3. Non-redundant data sets, TIM40D and TIM95D, of class categories in ENZYME
Classifying TIM Barrel Protein Domain Structure by an
Alignment Approach Using Best Hit Strategy and PSI-BLAST 291

according to the scheme outlined by Jones (Jones, 1999). The TIM sequence “d1cwn_” (SCOP
id) in TIM95D was excluded because of lack of secondary structure information (only 438 TIM
sequences in TIM95D were tested). The TIM barrel proteins (from ASTRAL SCOP 1.71 and the
Universal Protein Resource (UniProt (Bairoch et al., 2005))) for each of TIM40D and TIM95D
were classified into 32 superfamily categories, 91 family categories and 5 class categories
(Tables 1, 2 and 3; supplemental Table S1(Chu, 2011)).

2.2 Novel TIM barrel proteins from ASTRAL SCOP 1.73

Novel TIM barrel proteins from ASTRAL SCOP 1.73 that do not exist in ASTRAL SCOP 1.71
were also tested. The intersection among the TIM barrel proteins from ASTRAL SCOP 1.71
and 1.73 for TIM40D (Figure 1(A)) and TIM95D (Figure 1(B)) are shown. The number of TIM
sequences are represented in green (ASTRAL SCOP 1.71), light blue (ASTRAL SCOP 1.73)
and orange (ASTRAL SCOP 1.73 that are not presented in 1.71). In TIM40D (Figure 1(A)),
we identified 258 TIM sequences (ASTRAL SCOP 1.71 and 1.73), 14 TIM sequences
(exclusively ASTRAL SCOP 1.71) and 64 novel TIM sequences (exclusively ASTRAL SCOP
1.73: 12 of 64 were categorized as new). In TIM95D (Figure 1(B)), we identified 439 TIM
sequences (ASTRAL SCOP 1.71 and 1.73) and 79 novel TIM sequences (exclusively ASTRAL
SCOP 1.73: 12 of 79 were categorized as new). These 12 novel TIM sequences within the new
categories were identical and thus were excluded in the alignment approach. Hence, 52
(TIM40D) and 67 (TIM95D) novel TIM sequences from ASTRAL SCOP 1.73 were used to
evaluate the stability of the proposed PBH alignment strategy, respectively. (see
supplemental Table S2 (Chu, 2011)).

Fig. 1. Intersection among TIM sequences for TIM40D and TIM95D between ASTRAL SCOP
1.71 and 1.73. (A) In TIM40D, there are 272 (ASTRAL SCOP 1.71) and 322 (ASTRAL SCOP
1.73) TIM sequences. (B) In TIM95D, there are 439 (ASTRAL SCOP 1.71) and 518 (ASTRAL
SCOP 1.73) TIM sequences.

3. Results and discussion

3.1 Performance analysis
The standard percentage prediction accuracy rate Qi (Rost & Sander, 1993) was used to
evaluate the proposed alignment approach and Qi is defined as
292 Computational Biology and Applied Bioinformatics

pi
Qi = ¯100, (1)
ni

where ni is the number of test proteins in the ith superfamily/family/class category and pi is
the number of test proteins being correctly predicted in the ith superfamily/family/class.
The overall prediction accuracy rate Q is given by

k
Q = ∑ qiQi , (2)
i =1
where qi = ni/K, where K is the total number of test proteins. Qi is equivalent to Recall
(Gardy et al., 2003), which is defined as

TPi
Recalli = , (3)
(TPi + FN i )

where TPi (true positives) is the number of correctly predicted proteins in the ith
superfamily/family/class category, and FNi (false negatives) is the number of missed
proteins in the ith superfamily/family/class category. Precision (Gardy et al., 2003) was also
used to evaluate the proposed alignment approach. Precision is defined as

TPi
Precisioni = , (4)
(TPi + FPi )

where FPi (false positives) is the number of pseudo proteins predicted in the ith
superfamily/family/class category. In addition, the Matthews Correlation Coefficient (MCC
for short) (Matthews, 1975) was used to measure the prediction quality of classifications by
utilizing the proposed PBH and BHPB alignment strategies. MCC accounts for TPi, FPi, TNi
and FNi as a balanced measure, which can be used for categories with varying sizes. MCC
returns a value +1 for the perfect prediction quality, 0 for the average random prediction
quality, or -1 for an inverse prediction quality. The formula of MCC is defined as

TPi × TN i − FPi × FN i
MCC = (5)
(TPi + FPi ) × (TPi + FN i ) × (TN i + FPi ) × (TN i + FN i )

3.2 Alignment approach with the PBH strategy

Zotenko et al. (Zotenko et al., 2006) compared four projection methods, PRIDE, SGM, LFF
and SSEF, to classify the 40D data set (ASTRAL SCOP 1.69) into superfamily and family
categories in SCOP. There are 246 TIM barrel proteins classified into 24 superfamily
categories and 210 TIM barrel proteins classified into 42 family categories. Based on the
overall Q values, SSEF outperformed LFF, SGM and PRIDE for TIM barrel protein structure
classification (Table 4).
For the proposed PBH alignment strategy, the overall Q values for TIM40D and TIM95D
from ASTRAL SCOP 1.71 are shown in Table 5. In Table 5, the threshold (see Methods) is
determined without decreasing the Q value, which is achieved without a threshold. The Q
Classifying TIM Barrel Protein Domain Structure by an
Alignment Approach Using Best Hit Strategy and PSI-BLAST 293

Method SSEF LFF SGM PRIDE

Q (%) Q (%) Q (%) Q (%)
Superfamily* 78.0 73.6 63.0 41.5
Family* 74.3 72.9 57.1 45.2
*: the performances of SSEF, LFF, SGM and PRIDE are extracted from the additional file (Zotenko et al.,
2006)
Table 4. Overall Q values for SSEF, LFF, SGM and PRIDE in TIM40D (ASTRAL SCOP 1.69)
value will decrease when a score, which is higher or lower than the threshold given in Table
5, is assigned as the threshold. For TIM40D, the best Q value (84.2%) for the superfamily
classification is derived according to secondary structure identity, 76.1% for the family
classification is derived according to sequence identity and 48.2% for the class classification
is derived according to sequence identity (Table 5). The Q value of 48.2% for the class
classification is not valid. In TIM40D, 67 of 274 TIM sequences with undefined class
categories (derived from UniProt) were initially assumed to be false negatives before the test
(see Discussion). Using amino acid sequence or secondary structure information, the PBH
alignment strategy yields results as good as SSEF (footprint information). This alignment
approach will be useful for TIM barrel proteins lacking 3D structure information. Moreover,
for the class classification, the Q value of 48.2% is better than the Q value of 35% (under
Rank 1 condition) by a non-alignment method proposed by Dobson and Doig (Dobson &
Doig, 2005). For TIM95D, the best Q value (93.2%) for the superfamily classification is
derived according to secondary structure identity, 90.0% for the family classification is
derived according to sequence identity and 65.2% for the class classification is derived
according to secondary structure identity. Similarly, for the class classification, 91 of 442
TIM sequences with undefined class categories in TIM95D were initially assumed to be false
negatives before the test.

Secondary structure
Method Sequence identity RMSD
identity
Q (%) Threshold Q (%) Threshold Q (%) Threshold
Superfamily 83.1 <14 84.2 <67 40.4 >1.9
Family 76.1 <14 75.0 <67 37.1 >1.9
TIM40D
Class
48.2 <13 47.4 <67 21.2 >1.8
(ENZYME)
Superfamily 92.5 <14 93.2 <68 68.0 >2.0
Family 90.0 <14 89.3 <68 66.4 >2.0
TIM95D
Class
64.0 <16 65.2 <72 48.0 >1.8
(ENZYME)
Table 5. Overall Q values for the PBH alignment strategy in TIM40D and TIM95D (ASTRAL
SCOP 1.71)
Overall, the Q values of the PBH alignment strategy using secondary structure information
are similar to those using amino acid sequence information in TIM40D and TIM95D. For
practical purposes, however, it may be best to use only amino acid sequence information. In
addition, the Q values of the PBH alignment strategy using the RMSD do not yield valid
results. The RMSD (global alignment result in this chapter) may not be a valid feature for
294 Computational Biology and Applied Bioinformatics

the alignment approach to perform TIM barrel protein domain structure classification. In
Table 5, the threshold is too low for sequence identity, suggesting that the sequence identity
of the target and its selected proteins (within the same category) is low. When the threshold
is set higher than the above sequence identity, the target protein becomes a false negative
and then the Q value under the “no threshold” condition will decrease. The threshold is
high for secondary structure identity, suggesting that the secondary structure identity of the
target and its selected proteins (within the same category) is high. These results imply that
although TIM barrel proteins have diverse sequences they have very similar secondary
structures. This inference matches the recent observation for the TIM barrel proteins.
Tables 6 and 7 show overall Q and Precision values for various categories in TIM40D and
TIM95D from ASTRAL SCOP 1.71, respectively. Only the categories with more than ten TIM
sequences are listed; tests for RMSD were omitted because the results were invalid. In
Tables 6 and 7, the threshold is determined with the best Precision value without decreasing
the Q value, which is achieved without a threshold. Precision values with the threshold
outperform or are equals to those without the threshold. However, it is very difficult to
determine the appropriate threshold to obtain the best Precision value for routine alignment
practices. This problem may be omitted by the BHPB strategy to reduce the number of false
positives.

Method Sequence identity Secondary structure identity

Precision1 Precision2 Precision1 Precision2
Index Q (%) Q (%)
(%) (%) (%) (%)
2 94.7 64.3 78.3(17) 84.2 64.0 84.2(78)
4 73.3 78.6 100.0(18-22) 73.3 100.0 100.0(<77)
8 87.8 92.3 93.5(13) 86.6 95.9 95.9(<67)
Superfamily 9 83.3 78.9 78.9(<15) 72.2 86.7 92.9(68-70)
10 83.9 78.8 81.3(16) 96.8 78.9 81.0(74)
11 83.3 76.9 100.0(17-18) 91.7 91.7 100.0(73-75)
12 75.0 90.0 100.0(15) 83.3 100.0 100.0(<77)
2.4 100.0 66.7 100.0(19-29) 100.0 71.4 100.0(80-86)
4.1 73.3 78.6 100.0(18-22) 73.3 100.0 100.0(<77)
8.1 88.0 84.6 88.0(13) 92.0 95.8 95.8(<67)
Family 8.3 84.6 95.7 95.7(<17) 84.6 88.0 88.0(<73)
8.5 100.0 86.7 100.0(17) 92.3 100.0 100.0(<75)
10.1 83.3 78.9 100.0(18-19) 88.9 72.7 84.2(76)
11.2 90.9 76.9 100.0(17-18) 90.9 83.3 100.0(76-79)
1 66.7 64.3 85.7(21-22) 74.1 71.4 71.4(<72)
2 45.2 58.3 58.3(<13) 45.2 66.7 70.0(64-66)
Class
3 60.3 60.3 62.1(14) 54.4 61.7 63.8(68-70)
(ENZYME)
4 77.6 65.2 71.4(16) 75.9 62.0 69.8(74)
5 61.0 53.8 61.0(15-16) 65.2 45.5 48.4(70-71)
Table 6. Overall Q and Precision values for the PBH alignment strategy in TIM40D
(ASTRAL SCOP 1.71)
Classifying TIM Barrel Protein Domain Structure by an
Alignment Approach Using Best Hit Strategy and PSI-BLAST 295

Method Sequence identity Secondary structure identity

Precision1 Precision2 Precision1 Precision2
Index Q (%) Q (%)
(%) (%) (%) (%)
1 100.0 94.1 100.0(17-44) 100.0 88.9 94.1(80-85)
2 96.7 78.4 82.9(17) 90.0 84.4 96.4(79)
4 90.9 87.0 90.9(15-16) 86.4 95.0 100.0(70-81)
6 90.0 100.0 100.0(<22) 100.0 100.0 100.0(<84)
7 100.0 95.2 100.0(16-23) 100.0 100.0 100.0(<82)
8 91.8 98.4 98.4(<14) 95.5 98.5 98.5(<68)
Superfamily
9 81.8 94.7 94.7(<16) 77.3 89.5 94.4(68-74)
10 95.8 86.8 88.5(16) 97.9 87.0 94.0(76-77)
11 100.0 96.0 100.0(17-21) 100.0 96.0 100.0(74-80)
12 100.0 100.0 100.0(<26) 100.0 100.0 100.0(<79)
14 100.0 100.0 100.0(<31) 100.0 83.3 100.0(74-85)
15 93.3 82.4 82.4(<16) 93.3 93.3 93.3(<76)
1.1 100.0 94.1 100.0(17-44) 100.0 88.9 94.1(80-85)
2.4 100.0 76.5 100.0(20-29) 100.0 86.7 100.0(77-86)
4.1 90.9 87.0 90.9(15-16) 86.4 95.0 100.0(70-81)
7.1 100.0 95.2 100.0(16-23) 100.0 100.0 100.0(<82)
8.1 95.8 97.9 97.9(<14) 97.9 97.9 97.9(<68)
8.3 92.7 100.0 100.0(<17) 92.7 92.7 95.0(73-74)
Family
8.4 100.0 92.3 100.0(16-35) 100.0 100.0 100.0(<85)
8.5 100.0 90.0 100.0(17) 94.4 94.4 100.0(72-74)
10.1 96.6 87.5 100.0(18-19) 96.6 84.8 93.3(76-77)
11.2 100.0 94.7 100.0(17-21) 100.0 94.7 100.0(74-80)
14.1 100.0 100.0 100.0(<31) 100.0 83.3 100.0(74-85)
15.3 100.0 66.7 100.0(18-67) 100.0 90.9 100.0(76-95)
1 89.1 80.4 91.1(21-23) 89.1 80.4 82.0(70-71)
2 77.4 75.9 75.9(<16) 79.2 87.5 87.5(<73)
Class
3 67.0 71.7 72.4(14-15) 69.8 71.2 73.3(72-73)
(ENZYME)
4 91.8 80.2 84.0(17) 90.7 79.3 80.7(73)
5 83.7 78.8 82.0(15) 87.8 79.6 79.6(<72)
Table 7. Overall Q and Precision values for the PBH alignment strategy in TIM95D
(ASTRAL SCOP 1.71)
The Q values for Ribulose-phosphate binding barrel, (Trans)glycosidases, Aldolase and
Enolase C-terminal domain-like in the superfamily categories were above 84.2%, whereas
FMN-linked oxidoreductases, Metallo-dependent hydrolases, and
Phosphoenolpyruvate/pyruvate domain had lower Q values (Table 6). All of family
categories except for FMN-linked oxidoreductases had Q values above 76.1%. Only one of
the class categories, Transferases, had a Q value below 48.2%. For the superfamily
categories, Ribulose-phosphate binding barrel, (Trans)glycosidases and Metallo-dependent
hydrolases, the Q values derived according to sequence identity were better than those
derived according to secondary structure identity. In contrast, the Q values of Aldolase,
Enolase C-terminal domain-like and Phosphoenolpyruvate/pyruvate domain derived
296 Computational Biology and Applied Bioinformatics

according to secondary structure identity were better than those derived according to
sequence identity. FMN-linked oxidoreductases yielded the same Q values based on
sequence identity and secondary structure identity. For the family category, Type II
chitinase had a better Q value derived according to sequence identity than derived
according to secondary structure identity. Amylase, catalytic domain and Class I aldolase
produced better Q values derived according to secondary structure identity than derived
according to sequence identity. Tryptophan biosynthesis enzymes, FMN-linked
oxidoreductases, beta-glycanases and D-glucarate dehydratase-like had the same Q values
derived according to sequence identity and secondary structure identity. For the class
categories, the Q values of Hydrolases and Lyases derived according to sequence identity
were better than those derived according to secondary structure. Oxidoreductases and
Isomerases yielded better Q values derived according to secondary structure identity than
derived according to sequence identity. The remaining category, Transferases, had the same
Q values derived according to sequence identity and secondary structure identity. These
results demonstrated that the proposed PBH alignment strategy is more useful for certain
TIM barrel proteins than others.
In TIM95D superfamily categories, FMN-linked oxidoreductases and Metallo-dependent
hydrolases yielded Q values less than 93.2% (Table 7); others all yielded Q values above
93.2%. All of family categories obtained Q values above 90.0% and no class category
obtained Q values below 65.2%. For the superfamily categories, Ribulose-phosphate binding
barrel, FMN-linked oxidoreductases and Metallo-dependent hydrolases, and for the family
categories, FMN-linked oxidoreductases and Type II chitinase, the Q values derived
according to sequence identity were better than those derived according to secondary
structure identity. For other superfamily and family categories, the same Q values were
obtained using sequence identity and secondary structure identity. For the class categories,
Transferases, Hydrolases and Isomerases had better Q values derived according to
secondary structure identity than those derived according to sequence identity; Lyases had
a better Q value derived according to sequence identity than that derived according to
secondary structure identity. The last category, Oxidoreductases, produced the same Q
values derived according to sequence identity and secondary structure identity.

3.3 Estimating stability using the PBH alignment strategy

Novel TIM sequences in TIM40D (n=52) and TIM95D (n=67) from ASTRAL SCOP 1.73
were used to estimate the stability of the proposed PBH alignment strategy. Table 8
presents the overall Q values for novel TIM sequences. The definition and observation of
the threshold in Table 8 is the same as that in Table 5. In Table 8, the best Q values for the
superfamily, family and class classifications in TIM40D and TIM95D from ASTRAL SCOP
1.73 were derived according to sequence identity. For TIM40D, the best Q value was
94.2% for superfamily, 90.4% for family and 40.4% for class; for TIM95D, the best Q value
was 91.0% for superfamily, 88.1% for family and 47.8% for class. Similarly, for the class
classification, 20 of 52 (TIM40D) and 25 of 67 (TIM95D) novel TIM sequences with
undefined class categories were initially assumed to be false negatives before the test (see
supplemental Table S3 (Chu, 2011)). These results suggest that the proposed PBH
alignment strategy is stable and suitable for TIM barrel protein domain structure
classification.
Classifying TIM Barrel Protein Domain Structure by an
Alignment Approach Using Best Hit Strategy and PSI-BLAST 297

Secondary structure
Method Sequence identity RMSD
identity
Q Q
Threshold Q (%) Threshold Threshold
(%) (%)
Superfamily 94.2 <16 90.4 <72 57.7 >1.7
Family 90.4 <16 84.6 <74 55.8 >1.7
TIM40D
Class
40.4 <17 40.4 <72 25.0 >1.7
(ENZYME)
Superfamily 91.0 <16 86.6 <72 59.7 >1.8
Family 88.1 <16 80.6 <78 58.2 >1.8
TIM95D
Class
47.8 <14 44.8 <72 29.9 >1.7
(ENZYME)
Table 8. Overall Q values for novel TIM sequences in TIM40D and TIM95D (ASTRAL SCOP
1.73)

3.4 Alignment strategy with the BHPB strategy

The high Q value derived according to sequence identity using the PBH alignment strategy
can decrease the false positives via the homologous finding method. PSI-BLAST is an
established method that detects subtle relationships between proteins that are structurally
distant or functionally homologous owing to a position-specific scoring matrix generated
from multiple alignments of the top-scoring BLAST responses to a given query sequence.
The PSI-BLAST package was integrated into the NCBI standalone BLAST package (Altschul
et al., 1997). All of our tests were implemented using Perl combined with the CPAN bioperl
package (http://www.cpan.org/).
Table 9 presents the overall Q values for TIM sequences in TIM40D and TIM95D from
ASTRAL SCOP 1.71 using PSI-BLAST as a filter. The definition and observation of the
threshold in Table 9 is the same as that in Table 5. For TIM40D (Table 9), the best Q values
acquired according to sequence identity were 76.1% for superfamily, 73.9% for family, and
41.6% for class. For TIM95D, the secondary structure identity was used obtain the best Q
value of 62.2% for class, whereas sequence identity was used to obtain the best Q values for
superfamily (88.8%) and family (88.4%). Based on Tables 5 and 9, the Q values obtained
using the BHPB alignment strategy were slightly lower than those obtained using the PBH
alignment strategy. The lower Q values may be a consequence of proteins for which no
homolog was found using PSI-BLAST method; such proteins were thus false negatives.
Although the overall Q values using the PBH alignment strategy were higher than those
using the BHPB alignment strategy, Precision values obtained using the BHPB alignment
strategy were higher than those using the PBH alignment strategy. Tables 10 (TIM40D from
ASTRAL SCOP 1.71) and 11 (TIM95D) show the overall Q and Precision values for TIM
sequences within various categories. The definitions of the threshold in Tables 10 and 11 are
the same as that in Tables 6 and 7, respectively.

3.4.1 Q analysis
In Table 10, for the superfamily categories, the same categories as those observed in Table 6
obtained Q values above 76.1%; however, all of the categories obtained better or equal Q
values derived according to sequence identity than derived according to secondary
298 Computational Biology and Applied Bioinformatics

Secondary structure
Method Sequence identity RMSD
identity
Q Q
Threshold Q (%) Threshold Threshold
(%) (%)
Superfamily 76.1 <14 73.5 <67 39.0 >1.9
Family 73.9 <14 70.6 <67 37.1 >1.9
TIM40D
Class
41.6 <17 40.9 <73 20.1 >1.8
(ENZYME)
Superfamily 88.8 <14 87.2 <68 67.4 >2.0
Family 88.4 <14 86.3 <68 66.4 >2.0
TIM95D
Class
61.3 <18 62.2 <73 47.7 >1.8
(ENZYME)
Table 9. Overall Q values for the BHPB alignment strategy in TIM40D and TIM95D
(ASTRAL SCOP 1.71)

Method Sequence identity Secondary structure identity

Precision1 Precision2 Precision1 Precision2
Index Q (%) Q (%)
(%) (%) (%) (%)
2 89.5 100.0 100.0(<20) 84.2 80.0 88.9(79)
4 73.3 100.0 100.0(<23) 73.3 100.0 100.0(<77)
8 78.0 100.0 100.0(<14) 74.4 98.4 98.4(<67)
Superfamily 9 44.4 100.0 100.0(<26) 44.4 88.9 100.0(78-79)
10 83.9 96.3 96.3(<17) 83.9 92.9 92.9(<75)
11 83.3 90.9 100.0(17-18) 83.3 100.0 100.0(<80)
12 75.0 90.0 100.0(15) 75.0 100.0 100.0(<77)
2.4 100.0 100.0 100.0(<30) 100.0 90.9 100.0(77-86)
4.1 73.3 100.0 100.0(<23) 73.3 100.0 100.0(<77)
8.1 88.0 100.0 100.0(<14) 92.0 100.0 100.0(<67)
Family 8.3 84.6 100.0 100.0(<17) 76.9 95.2 95.2(<75)
8.5 92.3 100.0 100.0(<18) 84.6 100.0 100.0(< 81)
10.1 83.3 93.8 100.0(18-19) 83.3 88.2 93.8(75-76)
11.2 90.9 90.9 100.0(17-18) 90.9 100.0 100.0(<80)
1 66.7 78.3 85.7(21-22) 70.4 73.1 76.0(75-76)
2 38.7 60.0 63.2(17) 32.3 71.4 71.4(<73)
Class
3 42.6 61.7 69.0(18) 44.1 60.0 62.5(72-73)
(ENZYME)
4 70.7 78.8 85.4(18-19) 69.0 74.1 76.9(75-76)
5 60.9 63.6 63.6(<17) 56.5 56.5 59.1(78-79)
Table 10. Overall Q and Precision values for the BHPB alignment strategy in TIM40D
(ASTRAL SCOP 1.71)
structure identity. For the family categories, only FMN-linked oxidoreductases had a Q
value less than 73.9%. Amylase, catalytic domain was the only category that had a lower Q
value when using sequence identity instead of secondary structure identity. For the class
categories, only Transferases had a Q value less than 41.6%. For Transferases, Lyases and
Isomerases, the Q values derived according to sequence identity were higher than those
Classifying TIM Barrel Protein Domain Structure by an
Alignment Approach Using Best Hit Strategy and PSI-BLAST 299

derived according to secondary structure identity. In Table 11, for the superfamily
categories, (Trans)glycosidases, Metallo-dependent hydrolases and Xylose isomerase-like,
had Q values less than 88.8%. For the family categories, only beta-glycanases had a Q value
less than 88.4%. For all superfamily and family categories, the Q values derived according to
sequence identity were higher than or equal to those derived according to secondary
structure identity. All of the class categories had Q values higher than 62.2%. For
Hydrolases and Isomerases, the Q values derived according to secondary structure identity
were higher than those derived according to sequence identity.

Method Sequence identity Secondary structure identity

Precision1 Precision2 Precision1 Precision2
Index Q (%) Q (%)
(%) (%) (%) (%)
1 100.0 100.0 100.0(<45) 100.0 94.1 94.1(<86)
2 96.7 100.0 100.0(<18) 90.0 93.1 96.4(77-79)
4 90.9 100.0 100.0(<17) 86.4 100.0 100.0(<82)
6 90.0 100.0 100.0(<22) 90.0 100.0 100.0(<88)
7 100.0 100.0 100.0(<24) 100.0 100.0 100.0(<82)
8 86.6 100.0 100.0(<14) 85.1 99.1 99.1(<72)
Superfamily
9 63.6 100.0 100.0(<26) 63.6 93.3 100.0(78-79)
10 93.8 97.8 97.8(<17) 93.8 93.8 95.7(75-77)
11 100.0 100.0 100.0(<22) 100.0 100.0 100.0(<81)
12 100.0 100.0 100.0(<26) 100.0 100.0 100.0(<79)
14 100.0 100.0 100.0(<31) 100.0 100.0 100.0(<86)
15 80.0 100.0 100.0(<18) 80.0 100.0 100.0(<80)
1.1 100.0 100.0 100.0(<45) 100.0 94.1 94.1(<86)
2.4 100.0 100.0 100.0(<30) 100.0 92.9 100.0(77-86)
4.1 90.0 100.0 100.0(<17) 86.4 100.0 100.0(<82)
7.1 100.0 100.0 100.0(<24) 100.0 100.0 100.0(<82)
8.1 95.8 100.0 100.0(<14) 95.8 100.0 100.0(<68)
8.3 87.8 100.0 100.0(<17) 85.4 97.2 97.2(<75)
Family
8.4 100.0 100.0 100.0(<36) 100.0 100.0 100.0(<85)
8.5 94.4 100.0 100.0(<18) 88.9 100.0 100.0(<83)
10.1 93.1 96.4 100.0(18-25) 93.1 93.1 96.4(75-86)
11.2 100.0 100.0 100.0(<22) 100.0 100.0 100.0(<81)
14.1 100.0 100.0 100.0(<31) 100.0 100.0 100.0(<86)
15.3 100.0 100.0 100.0(<68) 100.0 100.0 100.0(<96)
1 89.1 87.2 91.1(21-23) 87.0 83.3 85.1(78-79)
2 75.4 81.6 83.3(17) 75.4 88.9 88.9(<73)
Class
3 59.4 72.4 74.1(18) 64.2 73.9 74.7(72-73)
(ENZYME)
4 89.7 87.9 88.8(17) 88.7 86.9 88.7(77)
5 81.6 93.0 93.0(<19) 83.7 85.4 85.4(<80)
Table 11. Overall Q and Precision values for the BHPB alignment strategy in TIM95D
(ASTRAL SCOP 1.71)
300 Computational Biology and Applied Bioinformatics

3.4.2 Precision analysis

In Tables 10 and 11, Precision values with the threshold were higher or equal to those
without the threshold, thus making it difficult to determine the feasible threshold to obtain
the best Precision value for routine alignment practices. However, the differences between
Precision values with and without the threshold were greatly reduced by using the BHPB
alignment strategy with the exception of Hydrolases of TIM40D. Using the BHPB alignment
strategy in TIM40D, the average Precision values without the threshold were 96.7% for
superfamily, 97.8% for family, and 68.5% for class (Table 10). Using the BHPB alignment
strategy in TIM95D, the average Precision values without the threshold were 99.8% for
superfamily, 99.7% for family, and 84.4% for class (Table 11). The best average Precision
values were derived according to sequence identity. The PSI-BLAST method in the BHPB
alignment strategy can filter out some of the false positives. Figures 2 and 3 indicate the

Fig. 2. The increase in Precision values for TIM40D (ASTRAL SCOP 1.71) using the BHPB
alignment strategy. (A) Increase in Precision values for TIM40D derived according to
sequence identity. (B) Increase in Precision values for TIM40D derived according to
secondary structure identity. Superscript ‘a’ or ‘b’ indicates the superfamily categories or the
class categories, respecively. Categories without a superscript indicated the family
categories. The blue bar indicates Precision values using the PBH alignment strategy and the
red bar indicates the increase in Precision values using the BHPB alignment strategy.
Classifying TIM Barrel Protein Domain Structure by an
Alignment Approach Using Best Hit Strategy and PSI-BLAST 301

Fig. 3. The increase in Precision values for TIM95D (ASTRAL SCOP 1.71) using the BHPB
alignment strategy. (A) Increase in Precision values for TIM95D derived according to
sequence identity. (B) Increase in Precision values for TIM95D derived according to
secondary structure identity. Superscript ‘a’ or ‘b’ indicates the superfamily categories or the
class categories, respecively. Categories without a superscript indicated the family
categories. The blue bar indicates Precision values using the PBH alignment strategy and the
red bar indicates the increase in Precision values using the BHPB alignment strategy.
increase in Precision values for TIM40D and TIM95D from ASTRAL SCOP 1.71 using the
BHPB alignment strategy as compared to the PBH alignment strategy, respectively. The
302 Computational Biology and Applied Bioinformatics

increase in Precision values was computed by comparing the results shown in Tables 6, 7, 10
and 11 (see supplemental Table S4 (Chu, 2011)). Based on Figures 2 and 3, Precision values for
almost all categories improved when using the BHPB alignment strategy. The average
increases in Precision values for TIM40D using sequence identity were 16.8% for superfamily,
16.7% for family and 8.1% for class. The average increases in Precision values using secondary
structure identity were 7.1% for superfamily, 9.5% for family and 7.0% for class. The average
increases in Precision values derived according to sequence identity were higher than those
derived according to secondary structure identity for TIM40D and TIM95D. Thus, the BHPB
alignment strategy yields higher Precision values than the PBH alignment strategy.

3.4.3 MCC analysis

Figure 4 presents the MCC measures of (1) the PBH alignment strategy derived according to
sequence identity (PBH(1D) for short), (2) the PBH alignment strategy derived according to
secondary structure identity (PBH(2D) for short), (3) the BHPB alignment strategy derived
according to sequence identity (BHPB(1D) for short) and (4) the BHPB alignment strategy
derived according to secondary structure identity (BHPB(2D) for short) for TIM40D and
TIM95D from ASTRAL SCOP 1.71, respectively. (see supplemental Table S5 (Chu, 2011))

Fig. 4. MCC scores of PBH(1D), PBH(2D), BHPB(1D) and BHPB(2D) for TIM40D and
TIM95D (ASTRAL SCOP 1.71). (A) MCC scores for TIM40D. (B) MCC scores for TIM95D.
Superscript ‘a’ or ‘b’ indicates the superfamily categories or the class categories. Categories
without a superscript indicate the family categories.
Classifying TIM Barrel Protein Domain Structure by an
Alignment Approach Using Best Hit Strategy and PSI-BLAST 303

Using the PBH and BHPB alignment strategies, all of the superfamily categories had MCC
scores greater than 0.7 except Metallo-dependent hydrolases when using the BHPB
alignment strategy (Figure 4(A)); Ribulose-phosphate binding barrel, Enolase C-terminal
domain-like, and Phosphoenolpyruvate/pyruvate domain had MCC scores greater than 0.9.
All of the family categories had MCC scores greater than 0.7; Tryptophan biosynthesis
enzymes, Amylase, catalytic domain, beta-glycanases, Type II chitinase, and D-glucarate
dehydratase-like had MCC scores greater than 0.9. All of the class categories had MCC
scores between 0.3~0.7, which is not an optimal score. From Figure 4(B), all of the
superfamily and family categories had MCC scores greater than 0.7; 13 categories had the
optimal MCC score (+1), indicating perfect prediction quality. All of the class categories had
MCC scores between 0.5~0.9. The above results demonstrate that the proposed PBH or
BHPB alignment strategy yielded high prediction quality for TIM barrel protein domain
structure classification.

3.5 Discussion
Here we further investigate why the alignment approach with the PBH or BHPB strategy is
not sufficient to classify the class category. For the above experiments, all of the EC
annotations for TIM sequences in TIM40D and TIM95D from ASTRAL SCOP 1.71 and 1.73
were derived from UniProt. There are 24.5% TIM40D (67 of 274) and 20.6% TIM95D (91 of
442) TIM sequences listed as undefined in class from ASTRAL SCOP 1.71; there are 38.5%
TIM40D (20 of 52) and 37.3% TIM95D (25 of 67) novel TIM sequences listed as undefined in
class from ASTRAL SCOP 1.73. These TIM sequences with undefined class categories were
initially assumed to be false negatives before the test. Therefore, the Q values for class
obtained by the PBH or the BHPB alignment strategy derived according to sequence identity
or secondary structure identity is poor. However, the ENZYME functions of some of these
TIM sequences with undefined class categories derived from UniProt have been described
in PDB. Thus, the EC annotations derived from PDB were integrated into TIM40D and
TIM95D from ASTRAL SCOP 1.71 and 1.73 (see supplemental Table S3 (Chu, 2011)), and the
above experiments for the class classification were repeated. After the PDB integrations,
13.6% TIM40D (38 of 279) and 11.1% TIM95D (50 of 450) TIM sequences remained
undefined from ASTRAL SCOP 1.71; further, 11.5% TIM40D (6 of 52) and 9.0% TIM95D (6 of
67) novel TIM sequences remained undefined from ASTRAL SCOP 1.73. These six novel
TIM sequences were identical in TIM40D and TIM95D from ASTRAL SCOP 1.73.

3.5.1 Improvement in Q
Figure 5 compares the Q values for TIM40D and TIM95D from ASTRAL SCOP 1.71 with
UniProt and PDB EC annotations using the PBH and BHPB alignment strategies. (see
supplemental Table S6 (Chu, 2011)) The Q values for the class classification using the PBH
and BHPB alignment strategies improved after integrating the PDB EC annotations. By
integrating the PDB EC annotations, some of the false negatives from UniProt were
eliminated. The alignment approach using either the PBH or BHPB strategy was useful for
the class classification. For TIM40D, the best Q value of 62.0% (an increase from 48.2%) for
class was derived according to sequence identity or secondary structure identity using the
PBH alignment strategy; the best Q value of 53.4% (an increase from 41.6%) for class was
derived according to sequence identity using the BHPB alignment strategy. For TIM95D, the
best Q value of 78.2% (an increase from 65.2%) for class was derived according to secondary
304 Computational Biology and Applied Bioinformatics

structure identity using the PBH alignment strategy; the best Q value of 72.9% (an increase
from 62.2%) for class was derived according to sequence identity or secondary structure
identity using the BHPB alignment strategy. For the novel TIM sequences in TIM40D and
TIM95D from ASTRAL SCOP 1.73, the best Q values were 73.1% (TIM40D) and 79.1%
(TIM95D) using the PBH alignment strategy (see supplemental Table S6 (Chu, 2011)).

Fig. 5. Comparisons for TIM40D and TIM95D (ASTRAL SCOP 1.71) with UniProt and PDB
EC annotations. (A) The Q values for TIM40D. (B) The Q values for TIM95D. Superscript ‘a’
or ‘b’ indicates TIM sequences available using only UniProt EC annotations or TIM
sequences available using UniProt and PDB EC annotations.

3.5.2 Improvement in MCC

Table 12 presents MCC scores for TIM sequences in TIM40D and TIM95D from ASTRAL
SCOP 1.71 with UniProt and PDB EC annotations using the PBH and BHPB alignment
strategies. All of the class categories had MCC scores between 0.4~0.8 in TIM40D; greater
than 0.7 in TIM95D; Oxidoreductases and Lyases also had MCC scores greater than 0.9 using
the BHPB alignment strategy. Hence, the proposed PBH or BHPB alignment strategy also
yielded high prediction quality for class.
Sequence identity Secondary structure identity
Category Index MCC MCC
PBH BHPB PBH BHPB
1 0.72 0.80 0.75 0.76
2 0.49 0.47 0.50 0.42
Class
TIM40D 3 0.68 0.61 0.67 0.57
(ENZYME)
4 0.67 0.69 0.67 0.65
5 0.50 0.53 0.52 0.49
1 0.89 0.93 0.87 0.89
2 0.76 0.78 0.81 0.80
Class
TIM95D 3 0.79 0.74 0.81 0.73
(ENZYME)
4 0.87 0.91 0.86 0.90
5 0.74 0.78 0.81 0.79
Table 12. MCC scores for TIM40D and TIM95D (ASTRAL SCOP 1.71) with UniProt and PDB
EC annotations
Classifying TIM Barrel Protein Domain Structure by an
Alignment Approach Using Best Hit Strategy and PSI-BLAST 305

3.5.3 Inferring ENZYME function for TIM barrel proteins with undefined class
categories
After integrating the PDB EC annotations into the above tests, there remained 38 (TIM40D)
and 50 (TIM95D) TIM sequences with undefined class categories from ASTRAL SCOP 1.71;
6 novel TIM sequences had undefined class categories from ASTRAL SCOP 1.73. Therefore,
we used the proposed alignment approach to infer the ENZYME functions for TIM barrel
proteins with undefined class.
We first assessed the classification results of the class categories by the PBH alignment
strategy for TIM sequences in TIM40D and TIM95D from ASTRAL SCOP 1.71 with UniProt
and PDB EC annotations. We found that the target protein and its selected protein belong to
the same superfamily category for most of the true positives identified in the alignment.
Table 13 presents statistics for true positives and false negatives for class using the PBH
alignment strategy. For true positives, 94% (162 of 173) and 99% (342 of 344) of the target
and its selected proteins belonged to the same superfamily category derived according to
PBH(1D) in TIM40D and TIM95D, respectively. For false negatives, however, 38% (40 of
106) and 31% (33 of 106) of the target and its selected proteins belonged to the same
superfamily category derived according to PBH(1D) in TIM40D and TIM95D, respectively.

TIM40D TIM95D
Statistic PBH(1D) PBH(2D) PBH(1D) PBH(2D)
TPi FPi TPi FPi TPi FPi TPi FPi

s, f 154.0 32.0 146.0 37.0 335.0 31.0 332.0 31.0

s, f 8.0 8.0 13.0 9.0 7.0 2.0 11.0 2.0

s 11.0 28.0 14.0 22.0 2.0 23.0 9.0 15.0
sum 173.0 106.0 173.0 106.0 344.0 106.0 352.0 98.0
s : Target and its selected proteins belong to the same superfamily category
s : Target and its selected proteins belong to the different superfamily categories
f : Target and its selected proteins belong to the same family category
f : Target and its selected proteins belong to the different family categories
Table 13. Statistical results for true positives and false negatives for class using the PBH
alignment strategy
Overall, 58% (of 279) and 76% (of 450) of the target and its selected proteins belonged to the
same superfamily and class categories derived according to PBH(1D) in TIM40D and
TIM95D, respectively. Similar observations were made based on PBH(2D) in TIM40D and
TIM95D. We observed 19 (PBH(1D)) and 23 (PBH(2D)) TIM sequences with undefined class
categories in TIM40D with the same superfamily category, respectively. We observed 19
(PBH(1D)) and 26 (PBH(2D)) TIM sequences with undefined class categories in TIM95D
with the same superfamily category. Therefore, it may be possible to infer the ENZYME
functions for TIM barrel proteins with undefined class categories, especially for TIM95D,
according to the classification results predicted by the proposed alignment approach. Table
13 also shows that 14% of 279 and 7% of 450 target and selected proteins belong to the same
306 Computational Biology and Applied Bioinformatics

superfamily category, but they belong to different class categories derived according to
PBH(1D) in TIM40D and TIM95D, respectively. Hence, all of TIM sequences of undefined
class may not be correctly inferred by the proposed alignment approach with the PBH or the
BHPB strategy. In the future, information regarding the active sites will be used in the
proposed alignment approach to remedy discrepancies in undefined class. In the following
test cases, all of the alignment results were displayed by DS Visualizer (Accelrys). The split
structure superposition was displayed utilizing PyMol Molecular Viewer (DeLano, 2002).

4. Methods
4.1 The alignment approach with the PBH strategy
An alignment approach with the PBH strategy was proposed to perform TIM barrel protein
domain structure classification (Figure 6). TIM40D and TIM95D can be used as the input for
this alignment approach. In the alignment methods block, three alignment tools,
CLUSTALW, SSEA and CE, were adopted to align any two of proteins by the amino acid
sequences, secondary structures and 3D structures, respectively, to obtain the scores of
sequence identity, secondary structure identity and RMSD. CLUSTALW is an established
multiple sequence alignment tool (global alignment) for DNA/RNA or protein sequences
based on a progressive pair-wise alignment method by considering sequence weighting,
variations in amino acid substitution matrices and residue-specific gap penalty scores. It is
widely used by biologists to investigate evolutional relationships among multiple protein
sequences. CLUSTALW may not be the best choice for the sequence alignment because of
recent advancements in programming, but it is still suitable for this alignment approach for
two reasons. First, we simply want to obtain the score of sequence identity for any two
proteins rather than the actual alignment information. Hence, the sequence identity score
obtained by CLUSTALW is not significantly different from that obtained by other tools.
Second, the design of most of other tools is focused on revising the multiple sequence
alignment results, not improving the pair-wise alignment results, even using the pair-wise
alignment results by CLUSTALW. SSEA is a multiple protein secondary structure alignment
tool (either global or local alignment) that aligns entire elements (rather than residue-based
elements [20]) of multiple proteins based on the H, C, and E states of SSEs. CE is a popular
and accurate pair-wise protein 3D structural alignment tool that aligns residues in
sequential order in space. If a protein domain sequence is not continuous, however, each
continuous fragment in the domain will be aligned against the other protein using the CE
alignment tool. Two criteria were adopted to resolve this problem. First, the sequence length
of the continuous fragment must be at least 30 residues, and second the minimal RMSD of
any two aligned fragments must be chosen. The default parameters of CLUSTALW
(accurate, but slow mode in setting your pairwise alignment options) and SSEA (global
alignment version) were used to align any two proteins in TIM40D and TIM95D to obtain
scores for sequence and secondary structure identities with normalized values ranging from
0-100. The default parameters of CE were used to align any two proteins in TIM40D and
TIM95D to obtain RMSD scores. After using CLUSTALW, SSEA and CE, these scores were
used to build an alignment-based protein-protein identity score network.
In the best hit strategy block, each protein in the network was first considered as a target
protein. Each target protein was then used to map the remaining proteins in the network.
Finally, the prediction result of each target protein was determined by selecting the
remaining proteins in the network according to certain parameters, which are critical for
Classifying TIM Barrel Protein Domain Structure by an
Alignment Approach Using Best Hit Strategy and PSI-BLAST 307

classification of the target protein. In our method, a PBH strategy is used to determine the
prediction result of a target protein by selecting the protein that has the best score for the
target protein according to this network. This score is calculated by a single parameter
(sequence identity, secondary structure identity, or RMSD). For the sequence or secondary
structure identity, the remaining protein with the highest score for the target protein is
selected; for the RMSD, the remaining protein with the lowest score for the target protein is
selected. For n proteins in the network, the time complexity is O(n2) for n target proteins to
find all selected proteins in this network using the PBH strategy owing to the bidirectional
aspect of the network. We used Perl to implement the PBH finding program because it
supports powerful data structures.
The single parameter threshold was applied in this classification model. When a threshold is
given for this approach, a target protein is assigned to a null situation as a false negative if
the highest score of sequence identity or secondary structure identity (or lowest score of
RMSD) for the target protein among all remaining proteins is less than (or larger than, for
RMSD) this threshold. Although the overall prediction accuracy cannot be improved by the
threshold concept, it may be decreased when an unfavorable threshold is given; however,
the number of false positives may be reduced when an appropriate threshold is used. In
other words, Precision values may be improved by the threshold concept. Nevertheless, an
appropriate threshold is very difficult to attain for the classification problem in a practical
setting. Therefore, in the experimental tests, an appropriate threshold was chosen after
processing was complete. Using the threshold concept, we observed the best possible
Precision values by this alignment approach and the properties of TIM barrel proteins.

Fig. 6. Flow chart of the alignment approach with the PBH strategy
To experimentally test the novel TIM sequences from ASTRAL SCOP 1.73, the flow chart of
the alignment approach with the PBH strategy is slightly different than that shown in Figure
6. For this test, the input is the novel TIM sequences from ASTRAL SCOP 1.73; however, the
alignment-based protein-protein identity score network is built by TIM sequences from
308 Computational Biology and Applied Bioinformatics

ASTRAL SCOP 1.71. Therefore, the target protein is a novel TIM sequence from ASTRAL
SCOP 1.73, and the remaining proteins are obtained from the TIM sequences (ASTRAL
SCOP 1.71). All tools and materials used for this research are accessible from (Chu, 2011).

4.2 The alignment approach with the BHPB strategy

PSI-BLAST is a position-specific iterative BLAST that results from refinement of the
position-specific scoring matrix (PSSM) and the next iterative PSSM. The position-specific
scoring matrix is automatically constructed from a multiple alignment with the highest
scoring hits in the BLAST search. The next iterative PSSM is generated by calculating
position-specific scores for each position in the previous iteration. PSI-BLAST is typically
used instead of BLAST to detect subtle relationships between proteins that are structurally
distant or functionally homologous. Therefore, it is possible to utilize PSI-BLAST as a filter
prior to the PBH strategy, denoted the BHPB strategy. The BHPB strategy can filter out
potential false positives, which may improve Precision values. The flow chart of the
alignment approach with the BHPB strategy is also slightly different than that shown in
Figure 6. In the best hit strategy block, each target protein in the network is used to map a
subset, but not all, of the remaining proteins in the network. This subset of remaining
proteins is grouped from the network using PSI-BLAST method for the target protein.
Hence, the selected protein with the best score for any target protein by the BHPB strategy
may not be the same as that by the PBH strategy.

5. Conclusion
At the amino acid sequence level, TIM barrel proteins are very diverse; however, these
proteins contain very similar secondary structures. Our results demonstrate that the
alignment approach with the PBH strategy or BHPB strategy is a simple and stable method
for TIM barrel protein domain structure classification, even when only amino acid sequence
information is available.

6. Acknowledgment
Part of this work was supported by National Science Council (NSC) under contract NSC95-
2627-B-007-002. The authors would like to thank Shu Hao Chang to help us to collect the
TIM barrel proteins from the SCOP 1.71 version.

7. References
Altschul, S.F.; Madden, T.L.; Schäffer, A.A.; Zhang, J.; Zhang, Z.; Miller, W. & Lipman, D.J.
(1997). Gapped BLAST and PSI-BLAST: a new generation of protein database
search programs. Nucleic Acids Research, 12.06.2008, Available from
ftp://ncbi.nlm.nih.gov/blast/.
Andreeva, A.; Howorth, D.; Chandonia, J.-M.; Brenner, S.E.; Hubbard, T.J.P.; Chothia, C. &
Murzin, A.G. (2008). Data growth and its impact on the SCOP database: new
developments. Nucleic Acids Research, Vol.36, pp. D419-D425.
Bairoch, A. (2000). The ENZYME database in 2000. Nucleic Acids Research, Vol.28, pp. 304-
305.
Classifying TIM Barrel Protein Domain Structure by an
Alignment Approach Using Best Hit Strategy and PSI-BLAST 309

Bairoch, A.I.; Apweiler, R.; Wu, C.H.; Barker, W.C.; Boeckmann, B.; Ferro, S.; Gasteiger, E.;
Huang, H.; Lopez, R.; Magrane, M.; Martin, M.J.; Natale, D.A.; O’Donovan, C.;
Redaschi, N. & Yeh, L.S. (2005). The Universal Protein Resource (UniProt). Nucleic
Acids Research, Vol.33, pp. D154-D159.
Bairoch, A.; Boeckmann, B.; Ferro, S. & Gasteiger, E. (2004). Swiss-Prot: Juggling between
evolution and stability. Briefings in Bioinformatics, Vol.5, pp. 39-55.
Berman, H.M.; Westbrook, J.; Feng, Z.; Gilliland, G.; Bhat, T.N.; Weissig, H.; Shindyalov, I.N.
& Bourne, P.E. (2000). The Protein Data Bank. Nucleic Acids Research, Vol.28, pp.
235-242.
Carugo, O. & Pongor, S. (2002). Protein fold similarity estimated by a probabilistic approach
based on Cα-Cα distance comparison. Journal of Molecular Biology, Vol.315, pp. 887-
898.
Chandonia, J.M.; Hon, G.; Walker, N.S.; Lo, C.L.; Koehl, P.; Levitt, M. & Brenner, S.E. (2004).
The ASTRAL compendium in 2004. Nucleic Acids Research, Vol.32, pp. D189-D192.
Choi, I.; Kwon, J. & Kim, S. (2004). Local feature frequency profile: a method to measure
structural similarity in proteins. Proceeding of the National Academy of Sciences of the
United States of America, Vol.101, pp. 3797-3802.
Chou, K.C. & Zhang, C.T. (1995). Prediction of protein structural classes. Critical Reviews in
Biochemistry and Molecular Biology, Vol.30, pp. 275-349.
Chu, C.-H. (2011). TIM barrel supplemental data for the InTech bookchapter. National Tsing
Hua Uinversity, Computational Systems Biology & Bio-Medicine Laboratory, 03.01.2011,
Available from http://oz.nthu.edu.tw/~d938301/InTech/bookchapter/
Cuff, A.L.; Sillitoe, I.; Lewis, T.; Redfern, O.C.; Garratt, R.; Thornton, J. & Orengo, C.A. (2009).
The CATH classification revisited—architectures reviewed and new ways to
characterize structural divergence in superfamilies. Nucleic Acids Research, Vol.37,
pp. D310-D314.
DeLano, W.L. (2002).The PyMOL Molecular Graphics System, DeLano Scientific, Palo Alto,
CA, USA. http://www.pymol.org.
Ding, C.H.Q. & Dubchak, I. (2001). Multi-class protein fold recognition using support vector
machines and neural networks. Bioinformatics, Vol.17, pp. 349-358.
Dobson, P.D. & Doig, A.J. (2005). Predicting enzyme class from protein structure without
alignments. Journal of Molecular Biology, Vol.345, pp. 187-199.
Dubchak, I.; Muchnik, I.; Holbrook, S.R. & Kim, S.H.(1995). Prediction of protein folding
class using global description of amino acid sequence. Proceeding of the National
Academy of Sciences of the United States of America, Vol.92, pp. 8700-8704.
Fontana, P.; Bindewald, E.; Toppo, S.; Velasco, R.; Valle, G. & Tosatto, S.C.E. (2005). The
SSEA server for protein secondary structure alignment. Bioinformatics, Vol.21, pp.
393-395.
Gardy, J.L.; Spencer, C.; Wang, K.; Ester, M.; Tusnday, G.E.; Simon, I.; Hua, S.; deFays, K.;
Lambert, C.; Nakai, K. & Brinkman, F.S.L. (2003). PSORT-B: improving protein
subcellular localization prediction for gram-negative bacteria. Nucleic Acids
Research, Vol.31, pp. 3613-3617.
Gáspári, Z.; Vlahovicek, K. & Pongor, S. (2005). Efficient recognition of folds in protein 3D
structures by the improved PRIDE algorithm. Bioinformatics, Vol.21, pp. 3322-3323.
Gloster, T.M.; Roberts, S.; Ducros, V.M.-A.; Perugino, G.; Rossi, M.; Hoos, R.; Moracci, M.;
Vasella, A. & Davies, G.J. (2004). Structural studies of the β-Glycosidase from
310 Computational Biology and Applied Bioinformatics

Sulfolobus solfataricus in complex with covalently and noncovalently bound

inhibitors. Biochemistry, Vol.43, pp. 6101-6109.
Huang, C.D.; Lin, C.T. & Pal, N.R. (2003). Hierarchical learning architecture with automatic
feature selection for multi-class protein fold classification. IEEE Transactions on
NanoBioscience, Vol.2, pp. 503–517.
Jones, D.T. (1999). Protein secondary structure prediction based on position-specific scoring
matrices. Journal of Molecular Biology, Vol.292, pp. 195-202.
Kabsch, W. & Sander, C. (1983). Dictionary of protein secondary structure: pattern
recognition of hydrogen-bonded and geometrical features. Biopolymers, Vol.22, pp.
2577-2637.
Lin, K.L.; Lin, C.Y.; Huang, C.D.; Chang, H.M.; Yang, C.Y.; Lin, C.T.; Tang, C.Y. & Hsu, D.F.
(2005). Methods of improving protein structure prediction based on HLA neural
network and combinatorial fusion analysis. WSEAS Transations on Information
Science and Applications, Vol.2, pp. 2146-2153.
Lin, K.L.; Lin, C.Y.; Huang, C.D.; Chang, H.M.; Yang, C.Y.; Lin, C.T.; Tang, C.Y. & Hsu, D.F.
(2007). Feature combination criteria for improving accuracy in protein structure
prediction. IEEE Transactions on NanoBioscience, Vol.6, pp. 186-196.
Matthews, B.W. (1975). Comparison of the predicted and observed secondary structure of T4
phage lysozyme. Biochimica et Biophysica Acta, Vol.405, pp. 442-451.
Murzin, A.G.; Brenner, S.E.; Hubbard, T. & Chothia, C. (1995). SCOP: A structural
classification of proteins database for the investigation of sequence and structures.
Journal of Molecular Biology, Vol.247, pp. 536-540.
Rogen, P. & Fain, B. (2003). Automatic classification of protein structure by using gauss
integrals. Proceeding of the National Academy of Sciences of the United States of America,
Vol.100, pp. 119-124.
Rost, B. & Sander, C. (1993). Prediction of protein secondary structure at better than 70%
accuracy. Journal of Molecular Biology, Vol.232, pp. 584-599.
Shen, H.B. & Chou, K.C. (2006). Ensemble classifier for protein fold pattern recognition.
Bioinformatics, Vol.22, pp. 1717-1722.
Shindyalov, I.N. & Bourne, P.E. (1998). Protein structure alignment by incremental
combinatorial extension (CE) of the optimal path. Protein Engineering, Vol.11, pp.
739-747.
Thompson, J.D.; Higgins, D.G. & Gibson, T.J. (1994). CLUSTAL W: improving the sensitivity
of progressive multiple sequence alignment through sequence weighting,
positions-specific gap penalties and weight matrix choice. Nucleic Acids Research,
Vol.22, pp. 4673-4680.
Vapnik, V.N. (1995). The Nature of Statistical Learning Theory, New York: Springer-Verlag.
Yu, C.S.; Wang, J.Y.; Yang, J.M.; Lyu, P.C.; Lin, C.J. & Hwang, J.K. (2003). Fine-grained protein
fold assignment by support vector machines using generalized nPeptide coding
schemes and jury voting from multiple-parameter sets. Proteins, Vol.50, pp. 531-536.
Zotenko, E.; Dogan, R.I.; Wilbur, W.J.; O’Leary, D.P. & Przytycka, T.M. (2007). Structural
footprinting in protein structure comparison: The impact of structural fragments.
BMC Structural Biology, Vol.7, pp. 53.
Zotenko, E.; O’Leary, D.P. & Przytycka, T.M. (2006). Secondary structure spatial
conformation footprint: a novel method for fast protein structure comparison and
classification. BMC Structural Biology, Vol.6, pp. 12.
 16

Identification of Functional Diversity in the

Enolase Superfamily Proteins
Kaiser Jamil1 and M. Sabeena2
1Head, Genetics Department, Bhagwan Mahavir Medical Research Centre, Mahavir Marg,
Hyderabad, & Dean, School of Life Sciences (SLS),
Jawaharlal Nehru Institute of Advanced Studies, (JNIAS)
2Research Associate, Centre for Biotechnology and Bioinformatics- (CBB), Jawaharlal

Nehru Institute of Advanced Studies, (JNIAS), Secunderabad

India

1. Introduction
The Escherichia coli K12 genome is a widely studied model system. The members of the
Enolase superfamily encoded by E.coli catalyze mechanistically diverse reactions that are
initiated by base-assisted abstraction of the α-proton of a carboxylate anion substrate to
form an enodiolate intermediate (Patricia C ,1996). Six of the eight members of the Enolase
superfamily encoded by the Escherichia coli K-12 genome have known functions (John F,
2008). The members share a conserved tertiary structure with a two-domain architecture, in
which three carboxylate ligands for the Mg2+ ion as well as the acid/base catalysts are
located at the C-terminal ends of the β-strands in a (β/α)7β-barrel [modified (β/α)8- or TIM-
barrel] domain and the specificity-determining residues are located in an N-terminal α+β
capping domain.
The rapid accumulation of data has led to an extraordinary problem of redundancy, which
must be confronted in almost any type of statistical analysis. An important goal of
bioinformatics is to use the vast and heterogeneous biological data to extract patterns and
make discoveries that bring to light the ‘‘unifying’’ principles in biology. (Kaiser Jamil,
2008)Because these patterns can be obscured by bias in the data, we approach the problem
of redundancy by appealing to a well known unifying principle in biology, evolution.
Bioinformatics has developed as a data-driven science with a primary focus on storing and
accessing the vast and exponentially growing amount of sequence and structure data (Gerlt
JA, 2005)
Protein sequences and their three-dimensional structures are successful descendants of
evolutionary process. Proteins might have considerable structural similarities even when no
evolutionary relationship of their sequences can be detected (Anurag Sethi, 2005). This
property is often referred to as the proteins sharing only a ‘‘fold”. Of course, there are also
sequences of common origin in each fold, called a ‘‘superfamily”, and in them groups of
sequences with clear similarities, are designated as ‘‘family”.
The concept of protein superfamily was introduced by Margaret Dayholff in the 1970 and
was used to partition the protein sequence databases based on evolutionary consideration
312 Computational Biology and Applied Bioinformatics

(Lindahl E, 2000). The objective of this study was to analyse the functional diversity of the
enolase gene superfamily. The gene superfamily consisting of twelve genes possess
enzymatic functions such as L-Ala-D/L-Glu epimerase, Glucarate dehydratase, D-
galactarate dehydratase, 2-hydroxy-3-oxopropionate reductase,].o-succinylbenzoate
synthase, D-galactonate dehydratase,[12]. 5-keto-4-deoxy-D-glucarate aldolase, L-
rhamnonate dehydratase, 2-keto-3-deoxy-L-rhamnonate aldolase, Probable galactarate
transporter, and Probable glucarate transporter (Steve EB ,1998)
This study was carried out to determine the Probable glucarate transporter (D-glucarate
permease) features relating enolase superfamily sequences to structural hinges, which is
important for identifying domain boundaries, and designing flexibility into proteins
functions also helps in understanding structure-function relationships.

2. Methodology
Enolase Superfamily Study/Analysis

Enolase Sequence Retrieval from Biological Databases

Sequence Analysis and Alignment (Using BLAST Program)

Multiple Sequence Alignment (Clustal W algorithm)

Sequence Alignment retrieval and improving of alignment using Jalview Program

SCI –PHY server for superfamily and subfamily prediction

ConSurf Server for residue Conservation analysis

Pattern Recognition Using ScanProsite

Visualization of the key residues represents superfamily in visualization program Rasmol

Flowchart represents the materials and methods

2.1 UniProt KB for genomic sequence analysis

Enolase sequence from E.coli formed the basis for this study. The protein sequences were
derived from UniProt KB, we found twelve sequences (Table 1). Most of the sequences in
UniProt KB were derived from the conceptual translation of nucleotide sequences. The
advantage of using UniProt KB was that it provides a stable, comprehensive, freely
Identification of Functional Diversity in the Enolase Superfamily Proteins 313

accessible central resource on protein sequences and functional annotation. UniProt

comprises of four major components, each optimized for different uses: the UniProt
Archive, the UniProt Knowledgebase, the UniProt Reference Clusters and the UniProt
Metagenomic and Environmental Sequence Database. We used this knowledge based
computational analysis which helps for the functional annotation for the gene sequences
shown below:

Accession
S.No Sequence Name Sequence
Id
1. P0A6P9 ENO_ECOLI MSKIVKIIGREIIDSRGNPTVEAEVHLEGGFVGMA
Enolase AAPSGASTGSREALELRDGDKSRFLGKGVTKAVA
OS=Escherichia AVNGPIAQALIGKDAKDQAGIDKIMIDLDGTENK
coli (strain K12) SKFGANAILAVSLANAKAAAAAKGMPLYEHIAE
GN=eno PE=1 LNGTPGKYSMPVPMMNIINGGEHADNNVDIQEF
SV=2 MIQPVGAKTVKEAIRMGSEVFHHLAKVLKAKGM
NTAVGDEGGYAPNLGSNAEALAVIAEAVKAAG
YELGKDITLAMDCAASEFYKDGKYVLAGEGNKA
FTSEEFTHFLEELTKQYPIVSIEDGLDESDWDGFAY
QTKVLGDKIQLVGDDLFVTNTKILKEGIEKGIANS
ILIKFNQIGSLTETLAAIKMAKDAGYTAVISHRSGE
TEDATIADLAVGTAAGQIKTGSMSRSDRVAKYN
QLIRIEEALGEKAPYNGRKEIKGQA
2. P51981 AEEP_ECOLI L- MRTVKVFEEAWPLHTPFVIARGSRSEARVVVVEL
Ala-D/L-Glu EEEGIKGTGECTPYPRYGESDASVMAQIMSVVPQL
epimerase EKGLTREELQKILPAGAARNALDCALWDLAARR
OS=Escherichia QQQSLADLIGITLPETVITAQTVVIGTPDQMANSA
coli (strain K12) STLWQAGAKLLKVKLDNHLISERMVAIRTAVPD
GN=ycjG PE=1 ATLIVDANESWRAEGLAARCQLLADLGVAMLEQ
SV=2 PLPAQDDAALENFIHPLPICADESCHTRSNLKAL
KGRYEMVNIKLDKTGGLTEALALATEARAQGFSL
MLGCMLCTSRAISAALPLVPQVSFADLDGPTWLA
VDVEPALQFTTGELHL
3. P0AES2 GUDH_ECOLI MSSQFTTPVVTEMQVIPVAGHDSMLMNLSGAHA
Glucarate PFFTRNIVIIKDNSGHTGVGEIPGGEKIRKTLEDAIP
dehydratase LVVGKTLGEYKNVLTLVRNTFADRDAGGRGLQT
OS=Escherichia FDLRTTIHVVTGIEAMLDLLGQHLGVNVASLLGD
coli (strainK12) GQQRSEVEMLGYLFFVGNRKATPLPYQSQPDDSC
GN=gudD PE=1 DWYRRHEEAMTPDAVVRLAEAAYEKYGFNDFK
SV=2 LKGGVLAGEEEAESIVALAQRFPQARITLDPNGA
WSLNEAIKIGKYLKGSLAYAEDPCGAEQGFSGRE
VMAEFRRATGLPTATNMIATDWRQMGHTLSLQS
VDIPLADPHFWTMQGSVRVAQMCHEFGLTWGS
HSNNHFDISLAMFTHVAAAAPGKITAIDTHWIW
QEGNQRLTKEPFEIKGGLVQVPEKPGLGVEIDMD
QVMKAHELYQKHGLGARDDAMGMQYLIPGWT
FDNKRPCMVR
314 Computational Biology and Applied Bioinformatics

Accession
S.No Sequence Name Sequence
Id
4. P39829 GARD_ECOLI MANIEIRQETPTAFYIKVHDTDNVAIIVNDNGLK
D-galactarate AGTRFPDGLELIEHIPQGHKVALLDIPANGEIIRYG
dehydratase EVIGYAVRAIPRGSWIDESMVVLPEAPPLHTLPLA
OS=Escherichia TKVPEPLPPLEGYTFEGYRNADGSVGTKNLLGITT
coli (strain K12) SVHCVAGVVDYVVKIIERDLLPKYPNVDGVVGLN
GN=garD PE=1 HLYGCVAINAPAAVVPIRTIHNISLNPNFGGEVM
SV=2 VIGLGCEKLQPERLLTGTDDVQAIPVESASIVSLQD
EKHVGFQSMVEDILQIAERHLQKLNQRQRETCPA
SELVVGMQCGGSDAFSGVTANPAVGYASDLLVR
CGATVMFSEVTEVRDAIHLLTPRAVNEEVGKRLL
EEMEWYDNYLNMGKTDRSANPSPGNKKGGLAN
VVEKALGSIAKSGKSAIVEVLSPGQRPTKRGLIYA
ATPASDFVCGTQQVASGITVQVFTTGRGTPYGLM
AVPVIKMATRTELANRWFDLMDINAGTIATGEET
IEEVGWKLFHFILDVASGKKKTFSDQWGLHNQL
AVFNPAPVT
5. P29208 MENC_ECOLI MRSAQVYRWQIPMDAGVVLRDRRLKTRDGLYV
o- CLREGEREGWGEISPLPGFSQETWEEAQSVLLAW
succinylbenzoat VNNWLAGDCELPQMPSVAFGVSCALAELTDTLP
e synthase QAANYRAAPLCNGDPDDLILKLADMPGEKVAK
OS=Escherichia VKVGLYEAVRDGMVVNLLLEAIPDLHLRLDANR
coli (strain K12) AWTPLKGQQFAKYVNPDYRDRIAFLEEPCKTRD
GN=menC PE=1 DSRAFARETGIAIAWDESLREPDFAFVAEEGVRAV
SV=2 VIKPTLTGSLEKVREQVQAAHALGLTAVISSSIESS
LGLTQLARIAAWLTPDTIPGLDTLDLMQAQQVRR
WPGSTLPVVEVDALERLL
6. Q6BF17 DGOD_ECOLI MKITKITTYRLPPRWMFLKIETDEGVVGWGEPVIE
D-galactonate GRARTVEAAVHELGDYLIGQDPSRINDLWQVMY
dehydratase RAGFYRGGPILMSAIAGIDQALWDIKGKVLNAPV
OS=Escherichia WQLMGGLVRDKIKAYSWVGGDRPADVIDGIKTL
coli (strain K12) REIGFDTFKLNGCEELGLIDNSRAVDAAVNTVAQ
GN=dgoD PE=1 IREAFGNQIEFGLDFHGRVSAPMAKVLIKELEPYR
SV=1 PLFIEEPVLAEQAEYYPKLAAQTHIPLAAGERMFS
RFDFKRVLEAGGISILQPDLSHAGGITECYKIAGM
AEAYDVTLAPHCPLGPIALAACLHIDFVSYNAVL
QEQSMGIHYNKGAELLDFVKNKEDFSMVGGFFK
PLTKPGLGVEIDEAKVIEFSKNAPDWRNPLWRHE
DNSVAEW
7. P23522 GARL_ECOLI 5- MNNDVFPNKFKAALAAKQVQIGCWSALSNPIST
keto-4-deoxy-D- EVLGLAGFDWLVLDGEHAPNDISTFIPQLMALKG
glucarate SASAPVVRVPTNEPVIIKRLLDIGFYNFLIPFVETKE
aldolase EAELAVASTRYPPEGIRGVSVSHRANMFGTVADY
OS=Escherichia FAQSNKNITILVQIESQQGVDNVDAIAATEGVDGI
coli (strain K12) FVGPSDLAAALGHLGNASHPDVQKAIQHIFNRA
GN=garL PE=1 SAHGKPSGILAPVEADARRYLEWGATFVAVGSDL
SV=2 GVFRSATQKLADTFKK
Identification of Functional Diversity in the Enolase Superfamily Proteins 315

Accession
S.No Sequence Name Sequence
Id
8. P77215 RHAMD_ECOLI MTLPKIKQVRAWFTGGATAEKGAGGGDYHDQG
L-rhamnonate ANHWIDDHIATPMSKYRDYEQSRQSFGINVLGTL
dehydratase VVEVEAENGQTGFAVSTAGEMGCFIVEKHLNRFI
OS=Escherichia EGKCVSDIKLIHDQMLSATLYYSGSGGLVMNTISC
coli (strain K12) VDLALWDLFGKVVGLPVYKLLGGAVRDEIQFYA
GN=yfaW PE=1 TGARPDLAKEMGFIGGKMPTHWGPHDGDAGIR
SV=2 KDAAMVADMREKCGEDFWLMLDCWMSQDVN
YATKLAHACAPYNLKWIEECLPPQQYESYRELKR
NAPVGMMVTSGEHHGTLQSFRTLSETGIDIMQPD
VGWCGGLTTLVEIAAIAKSRGQLVVPHGSSVYSH
HAVITFTNTPFSEFLMTSPDCSTMRPQFDPILLNEP
VPVNGRIHKSVLDKPGFGVELNRDCNLKRPYSH
9. P76469 KDRA_ECOLI MNALLSNPFKERLRKGEVQIGLWLSSTTAYMAEI
2-keto-3-deoxy- AATSGYDWLLIDGEHAPNTIQDLYHQLQAVAPY
L-rhamnonate ASQPVIRPVEGSKPLIKQVLDIGAQTLLIPMVDTAE
aldolase QARQVVSATRYPPYGERGVGASVARAARWGRIE
OS=Escherichia NYMAQVNDSLCLLVQVESKTALDNLDEILDVEGI
coli (strain K12) DGVFIGPADLSASLGYPDNAGHPEVQRIIETSIRRI
GN=yfaU PE=1 RAAGKAAGFLAVAPDMAQQCLAWGANFVAVG
SV=1 VDTMLYSDALDQRLAMFKSGKNGPRIKGSY

10. P0AA80 GARP_ECOLI MILDTVDEKKKGVHTRYLILLIIFIVTAVNYADRA

Probable TLSIAGTEVAKELQLSAVSMGYIFSAFGWAYLLM
galactarate QIPGGWLLDKFGSKKVYTYSLFFWSLFTFLQGFVD
transporter MFPLAWAGISMFFMRFMLGFSEAPSFPANARIVA
OS=Escherichia AWFPTKERGTASAIFNSAQYFSLALFSPLLGWLTF
coli (strain K12) AWGWEHVFTVMGVIGFVLTALWIKLIHNPTDHP
GN=garP PE=1 RMSAEELKFISENGAVVDMDHKKPGSAAASGPK
SV=1 LHYIKQLLSNRMMLGVFFGQYFINTITWFFLTWFP
IYLVQEKGMSILKVGLVASIPALCGFAGGVLGGVF
SDYLIKRGLSLTLARKLPIVLGMLLASTIILCNYTN
NTTLVVMLMALAFFGKGFGALGWPVISDTAPKEI
VGLCGGVFNVFGNVASIVTPLVIGYLVSELHSFNA
ALVFVGCSALMAMVCYLFVVGDIKRMELQK
11. P0ABQ2 GARR_ECOLI 2- MKVGFIGLGIMGKPMSKNLLKAGYSLVVADRNP
hydroxy-3- EAIADVIAAGAETASTAKAIAEQCDVIITMLPNSP
oxopropionate HVKEVALGENGIIEGAKPGTVLIDMSSIAPLASREI
reductase SEALKAKGIDMLDAPVSGGEPKAIDGTLSVMVGG
OS=Escherichia DKAIFDKYYDLMKAMAGSVVHTGEIGAGNVTKL
coli (strain K12) ANQVIVALNIAAMSEALTLATKAGVNPDLVYQA
GN=garR PE=1 IRGGLAGSTVLDAKAPMVMDRNFKPGFRIDLHIK
SV=1 DLANALDTSHGVGAQLPLTAAVMEMMQALRA
DGLGTADHSALACYYEKLAKVEVTR
316 Computational Biology and Applied Bioinformatics

Accession
S.No Sequence Name Sequence
Id
12. Q46916 GUDP_ECOLI MSSLSQAASSVEKRTNARYWIVVMLFIVTSFNYG
Probable DRATLSIAGSEMAKDIGLDPVGMGYVFSAFSWAY
glucarate VIGQIPGGWLLDRFGSKRVYFWSIFIWSMFTLLQG
transporter FVDIFSGFGIIVALFTLRFLVGLAEAPSFPGNSRIVA
OS=Escherichia AWFPAQERGTAVSIFNSAQYFATVIFAPIMGWLT
coli (strain K12) HEVWSHVFFFMGGLGIVISFIWLKVIHEPNQHPG
GN=gudP PE=1 VNKKELEYIAAGGALINMDQQNTKVKVPFSVKW
SV=1 GQIKQLLGSRMMIGVYIGQYCINALTYFFITWFPV
YLVQARGMSILKAGFVASVPAVCGFIGGVLGGIIS
DWLMRRTGSLNIARKTPIVMGMLLSMVMVFCNY
VNVEWMIIGFMALAFFGKGIGALGWAVMADTA
PKEISGLSGGLFNMFGNISGIVTPIAIGYIVGTTGSF
NGALIYVGVHALIAVLSYLVLVGDIKRIELKPVAG
Q
Table 1. Enolase Sequences from E.coli –K12 Strain (from UNIPROT-KB in Fasta format)

2.2 BLAST program for sequence analysis and alignment

Basic Local Alignment Search Tool (BLAST) is one of the most heavily used sequence
analysis tools we have used to perform Sequence Analysis and Alignment. BLAST is a
heuristic that finds short matches between two sequences and attempts to start alignments.
In addition to performing alignments, BLAST provides statistical information to help
decipher the biological significance of the alignment as ‘expect’ value. (Scott McGinnis,
2004). Using this BLAST program the twelve gene sequences were aligned against archaea
and bacteria. The sequences were sorted out according to the existing gene names with
similarity and the fused genes were removed.

2.3 Clustal W program for multiple sequence alignment

Multiple sequence alignments are widely acknowledged to be powerful tools in the analysis
of sequence data.( Sabitha Kotra et al 2008) Crucial residues for activity and for maintaining
protein secondary and tertiary structures are often conserved in sequence alignments.
Hence, multiple sequence alignment was done for all the enolase gene sequences based on
the ClustalW algorithm using the tool BioEdit software program. We determined the
alignments which is the starting points for evolutionary studies. Similarity is a percentage
sequence match between nucleotide or protein sequences. The basic hypothesis involved
here was that similarity relates to functionality, if two sequences are similar, they will have
related functionalities.
Realigned the obtained Multiple Sequence Alignments (MSA) using ClustalW (Muhummad
Khan and Kaiser Jamil, 2010). Using MSA we could obtain high score for the conserved
regions, compared to the reported query sequences. So we viewed the multiple alignment
result using a program ‘Jalview’ which improved the multiple alignment. With this
program we could extract and get the complete alignment of all sequences for realigning to
the query sequence to get better results (Fig. 1). Jalview is a multiple alignment editor
written in Java. It is used widely in a variety of web pages which is available as a general
purpose alignment editor. The image below shows the result when Jalview has taken the
Identification of Functional Diversity in the Enolase Superfamily Proteins 317

full length sequences and realigned them (using Clustalw) to the query sequence. The
alignment has far fewer gaps and more similarities to the entire portion of the query
sequences.

Fig. 1. Multiple Sequence Alignment as shown in Jalview

2.4 SCI –PHY server for superfamily and subfamily prediction

Using SCI-PHY server we found subfamilies/subclasses present in the aligned sequences,
which merged into five groups. The corresponding pattern for each group of subfamily
sequences was found by using ScanProsite and PRATT. A low-level simple pattern-
matching application can prove to be a useful tool in many research settings (Doron Betel,
2000). Many of these applications are geared toward heuristic searches where the program
finds sequences that may be closely related to the query nucleotide/protein sequences.

2.5 ConSurf server for conservation analysis

For each subfamily sequences the corresponding PDB ID using ConSurf Server was
determined. ConSurf-DB is a repository of ConSurf Server which used for evolutionary
318 Computational Biology and Applied Bioinformatics

conservation analysis of the proteins of known structures in the PDB. Sequence homologues
of each of the PDB entries were collected and aligned using standard methods. The
algorithm behind the server takes into account the phylogenetic relations between the
aligned proteins and the stochastic nature of the evolutionary process explicitly. The server
assigned the conservation level for each position in the multiple sequence alignment (Ofir
Goldenberg, 2002). Identified specific pattern for each of the FASTA format sequence from
PDB files using ScanProsite and some of the key residues that comprise the functionally
important regions of the protein (Ofir Goldenberg, 2002). We determined the residues
present in each of PDB files denoting subfamilies using Swiss PDB Viewer. Mapped out all
the residues in color with the help of Rasmol by finding the specific pattern.

3. Results and discussion

This study is an attempt to determine the functional diversity in enolase superfamily
protein. The approach we used is a all pairwise alignment of the sequences followed by a
clustering of statistically significant pairs into groups or subfamilies by making sure that
there is a common motif holding all the members together. Multiple sequence alignment
and pattern recognition methods were included in this. The study analyzed the possible
subfamilies in Enolase protein superfamily which shares in organisms such as archaea,
bacteria with respect to E.coli and finally predicted five superfamilies which may play a role
in functional diversity in Enolase superfamily protein.
Generally a protein’s function is encoded within putatively functional signatures or motifs
that represent residues involved in both functional conservation and functional divergence
within a set of homologous proteins at various levels of hierarchy that is, super-families,
families and sub-families. Protein function divergence is according to local structural
variation around the active sites (Changwon K, 2006). Even when proteins have similar
overall structure, the function could be different from each other. Accurate prediction of
residue depth would provide valuable information for fold recognition, prediction of
functional sites, and protein design. Proteins might have considerable structural similarities
even when no evolutionary relationship of their sequences can be detected. This property is
often referred to as the proteins sharing ie; a ‘‘fold”. Of course, there are also sequences of
common origin in each fold, called a ‘‘superfamily”, and in them there are groups of
sequences with clear similarities designated as ‘‘family”. These sequence-level superfamilies
can be categorized with many Bioinformatics approaches (LevelErik L , 2002)

3.1 Functional/ structural validation

The functions of the five identified protein family include:

3.1.1 Group 1
Mandelate racemase / muconate lactonizing enzyme family signature-1: which is an
independent inducible enzyme cofactor. Mandelate racemase (MR) and muconate
lactonizing enzyme (MLE) catalyses separate and mechanistically distinct reactions
necessary for the catabolism of aromatic acids Immobilization of this enzyme leads to an
enhanced activity and facilitates its recovery
MR_MLE_1 Mandelate racemase / muconate lactonizing enzyme family signature 1:
(Fig.2)
Identification of Functional Diversity in the Enolase Superfamily Proteins 319

Polymer: 1
Type: polypeptide(L)
Length: 405 Chains:A, B, C, D, E, F, G, H
Functional Protein: PDB ID: 3D46 chain A in E-val 0.0.
Possible amino acid pattern found in chain A
I-x(1,3)-Q-P-D-[ALV]-[ST]-H-[AV]-G-G-I-[ST]-E-x(2)-K-[IV]-A-[AGST]-[LM]-A-E-[AS]-[FY]-
D-V-[AGT]-[FLV]-[AV]-[LP]-H-C-P-L-G-P-[IV]-A-[FL]-A-[AS]-[CS]-L-x-[ILV]-[DG]
Key Residues
THR 136, SER 138, CYS 139,VAL 140, Asp 141, ALA 143, LEU 144, ASP 146, LEU 147, GLY
149, LYS 150, PRO 155, VAL 156, LEU 159, LEU 160, GLY 161

Fig. 2. Functional Protein Information (PDB Id: 3D46) The residues in yellow colour
represents the identified functional residues in Group 1

3.1.2 Group 2
TonB-dependent receptor proteins signature-1 : TonB-dependent receptors is a family of
beta-barrel proteins from the outer membrane of Gram-negative bacteria. The TonB complex
senses signals from outside the bacterial cell and transmits them via two membranes into
the cytoplasm, leading to transcriptional activation of target genes
TONB_DEPENDENT_REC_1 TonB-dependent receptor proteins signature 1 : (Fig.3)
Polymer:1
Type:polypeptide(L)
Length:99
320 Computational Biology and Applied Bioinformatics

Chains:A, B
Functional Protein: PDB ID: 3LAZ
Possible amino acid pattern found in 3LAZ
T-K-R-G-L-I-Y-A-A-T-P-A-S-D-F-V-C-G-T-Q-Q-V-A-S-G-I-T-V-Q-V-F-T-T-G-R-G-T-P-Y-G-
L-M-A-V-P-V-I-K-M-A
Key Residues
GLU 88, SER89, VAL91, VAL92, PRO94, GLU95

Fig. 3. Functional Protein Information (PDB Id: 3LAZ). The residues in yellow colour
represents the identified functional residues in Group 2

3.1.3 Group 3
3-hydroxyisobutyrate dehydrogenase signature : This enzyme is also called beta-
hydroxyisobutyrate dehydrogenase. This enzyme participates in valine, leucine and
isoleucine degradation.
3_HYDROXYISOBUT_DH 3-hydroxyisobutyrate dehydrogenase signature : (Fig.4. a and
Fig.4. b)
Polymer:1
Type:polypeptide(L)
Length:295
Chains:A, B
Functional Protein: PDB ID: 1YB4
Possible amino acid pattern found in 1YB4
Identification of Functional Diversity in the Enolase Superfamily Proteins 321

G-[IMV]-[EK]-F-L-D-A-P-V-T-G-G-[DQ]-K-[AG]-A-x-E-G-[AT]-L-T-[IV]-M-V-G-G-x(2)-
[ADEN]-[ILV]-F-x(2)-[LV]-x-P-[IV]-F-x-A-[FM]-G-[KR]-x-[IV]-[IV]-[HY]-x-G
Key Residues
PHE5, ILE6, GLY7, LEU8, GLY 9, GLY 12, ALA 16, ASN 18
Polymer:1
Type:polypeptide(L)
Length:299
Chains:A
Alternate: 3_HYDROXYISOBUT_DH 3-hydroxyisobutyrate dehydrogenase signature :
Functional Protein: PDB ID: 1VPD
Possible amino acid pattern found in1VPD
G-[ADET]-x-G-[AS]-G-x(1,2)-T-x(0,1)-K-L-[AT]-N-Q-[IV]-[IMV]-V-[AN]-x-[NT]-I-A-A-[MV]-
[GS]-E-A-[FLM]-x-L-A-[AT]-[KR]-[AS]-[GV]-x-[ADNS]-[IP]
OR
K-L-A-N-Q-x(0,1)-I-x(0,1)-V-[AN]-x-N-I-[AQ]-A-[MV]-S-E-[AS]-[FL]-x-L-A-x-K-A-G-[AIV]-
[DENS]-[PV]-[DE]-x-[MV]-[FY]-x-A-I-[KR]-G-G-L-A-G-S-[AT]-V-[LM]-[DN]-A-K
Key Residues
PHE7, ILE8, GLY9, LEU10, GLY11, GLY14, SER18, ASN20

(a)
322 Computational Biology and Applied Bioinformatics

(b)
Fig. 4. a. Functional Protein Information (PDB Id: 1YB4) The residues in yellow colour
represents the identified functional residues in Group 3. Also. b Functional Protein
Information (PDB Id: 1VPD). The residues in yellow colour represents the identified
functional residues in Group 3

3.1.4 Group 4
Enolase signature : Enolase, also known as phosphopyruvate dehydratase, is a
metalloenzyme responsible for the catalysis of the conversion of 2-phosphoglycerate (2-PG)
to phosphoenolpyruvate (PEP), the ninenth and penultimate step of glycolysis. Enolase can
also catalyze the reverse reaction, depending on environmental concentrations of substrates.
Polymer:1
Type:polypeptide(L)
Length:431
Chains:A, B, C, D
Functional Protein: PDB Id: 1E9I
ENOLASE Enolase signature: (Fig.5. a and Fig.5. b)
Possible amino acid pattern found in 1E9I
G-x(0,1)-D-D-[IL]-F-V-T-[NQ]-[PTV]-[DEKR]-x-[IL]-x(2)-G-[IL]-x(4)-[AGV]-N-[ACS]-[ILV]-
L-[IL]-K-x-N-Q-[IV]-G-[ST]-[LV]-x-[DE]-[AST]-[FILM]-[ADES]-A-[AIV]-x(2)-[AS]-x(3)-[GN]
Key Residues
Identification of Functional Diversity in the Enolase Superfamily Proteins 323

ILE 338, LEU339, ILE340, LYS341, ASN343, GLN344, ILE 345, GLY346, SER347, LEU348,
THR349, GLU350, THR351
Alternate : ENOLASE Enolase signature
Polymer:1
Type:polypeptide(L)
Length:427
Chains:A, B
Functional Protein: PDB ID: 2PA6
Possible amino acid pattern found in 2PA6
S-x(1,2)-S-G-[DE]-[ST]-E-[DG]-[APST]-x-I-A-D-[IL]-[AS]-V-[AG]-x-[AGNS]-[ACS]-G-x-I-K-
T-G-[AS]-x-[AS]-R-[GS]-[DES]-R-[NTV]-A-K-Y-N-[QR]-L-[ILM]-[ER]-I-E-[EQ]-[ADE]-L-
[AEGQ]
Key Residues
LEU 336, LEU337, LEU338, LYS339, ASN341, GLN342, ILE343, GLY344,THR345, LEU 346,
SER347, GLU348, ALA 349

(a)
324 Computational Biology and Applied Bioinformatics

(b)
Fig. 5. a Functional Protein Information (PDB Id: 1E91) The residues in yellow colour
represents the identified functional residues in Group 4. Also. b Functional Protein
Information (PDB Id: 2PA6). The residues in yellow colour represents the identified
functional residues in Group 4

3.1.5 Group 5
Glycerol-3-phosphate transporter (glpT) family of transporters signature :(Fig.6)
The major facilitator superfamily represents the largest group of secondary membrane
transporters in the cell.
Molecule:Glycerol-3-phosphate transporter
Polymer:1
Type:polypeptide(L)
Length:451
Chains:A
Functional Protein: PDB ID: 1PW4
Possible amino acid pattern found in 1PW4
P-x(2,3)-R-x(0,1)-G-x-A-x-[AGS]-[FILV]-x(3)-[AGS]-x(3)-[AGS]-x(2)-[AILV]-x-[APST]-[IPV]-
x(2)-[AG]-x-[ILV]-[ASTV]-x(3)-G-x(3)-[ILMV]-[FY]-x(3)-[AGV]-[AGILPV]-x-[GS]-[FILMV]
Key Residues
GLU153, ARG154, GLY155, SER159, VAL160, TRP161, ASN162, ALA164, ASN166, VAL167,
GLY168, GLY169
Identification of Functional Diversity in the Enolase Superfamily Proteins 325

Fig. 6. Functional Protein Information (PDB Id: 1PW4). The residues in yellow colour
represents the identified functional residues in Group 5

4. Conclusion
Identification of the specificity-determining residues in the various protein family studies
has an important role in bioinformatics because it provides insight into the mechanisms by
which nature achieves its astonishing functional diversity, but also because it enables the
assignment of specific functions to uncharacterized proteins and family prediction.
Genomics has posed the challenge of determination of protein function from sequence or 3-
326 Computational Biology and Applied Bioinformatics

D structure. Functional assignment from sequence relationships can be misleading, and

structural similarity does not necessarily imply functional similarity. Our studies on the
analysis of the superfamily revealed, for the first time, that in these species (archaea and
bacteria) using E. coli. as a genomic model, we can contribute important insights for
understanding their structural as well as functional relationships. The computational
prediction of these functional sites for protein structures provides valuable clues for
functional classification.

5. Acknowledgement
The authors gratefully acknowledge the support from JNIAS for the successful completion
of this project.

6. References
Muhummad Khan and Kaiser Jamil (2008) Genomic distribution, expression and pathways
of cancer metasignature genes through knowledge based data mining. International
Journal of Cancer Research 1 (1), PP1-9, ISSN 1811-9727
Muhummad Khan and Kaiser Jamil (2008), Study on the conserved and polymorphic
sites of MTHFR using bioinformatic approaches. Trends in Bioinformatics 1 (1) 7-
17.
Sabitha Kotra, Kishore Kumar Madala and Kaiser Jamil (2008), Homology Models of the
Mutated EGFR and their Response towards Quinazolin Analogues; J. Molecular
Graphics and modeling , Vol-27, pp244-254.
Muhummadh Khan and Kaiser Jamil (2010) Phylogeny reconstruction of ubiquitin
conjugating (E2) enzymes. Biology and Medicine Vol 2 (2), 10-19.
Patricia C, Babbitt, Miriam S. Hasson, Joseph E. Wedekind, David R. J. Palmer, William C.
Barrett, George H. Reed, Ivan Rayment, Dagmar Ringe, George L. Kenyon, and
John A. Gerlt (1996); The Enolase Superfamily: A General Strategy for Enzyme-
Catalyzed Abstraction of the α-Protons of Carboxylic Acid Biochemistry 35 (51), pp
16489–16501
Babbitt PC Hasson MS, Wedekind JE, Palmer DR, Barrett WC, Reed GH, Rayment I, Ringe
D, Kenyon GL, Gerlt JA (1996) The enolase superfamily: a general strategy for
enzyme-catalyzed abstraction of the alpha-protons of carboxylic acids,
Biochemistry. 35(51):16489-50
John F. Rakus, Alexander A. Fedorov, Elena V. Fedorov, Margaret E. Glasner, Brian K.
Hubbard, Joseph D. Delli, Patricia C. Babbitt, Steven C. Almo and John A. Gerlt,
(2008) Evolution of Enzymatic Activities in the Enolase Superfamily: l-Rhamnonate
Dehydratase, Biochemistry 47 (38), pp 9944–9954
Gerlt JA, Babbitt PC, Rayment I.(2005). Divergent evolution in the enolase superfamily: the
interplay of mechanism and specificity. Arch Biochem Biophys. 1;433(1):59-7
Anurag Sethi, Patrick O'Donoghue, and Zaida Luthey-Schulten (2005) Evolutionary profiles
from the QR factorization of multiple sequence alignments, PNAS vol. 102 no. 11
4045-4050
Identification of Functional Diversity in the Enolase Superfamily Proteins 327

Lindahl E, Elofsson A, (2000) Identification of related proteins on family, superfamily and

fold level. Journal of Molecular Biology 295: 3, 613-625
Steven E. Brenner, Cyrus Chothia, and Tim J. P. Hubbard (1998) Assessing sequence
comparison methods with reliable structurally identified distant evolutionary
relationships PNAS May 26: 95 6073-6078
Dayhoff, M.O. (1974) Computer analysis of protein sequences, Fed. Proc. 33, 2314-2316,.
Scott McGinnis (2004) BLAST: at the core of a powerful and diverse set of sequence analysis
tools, Nucleic Acids Research, Vol. 32
Hubbard BK, Koch M, Palmer DR, Babbitt PC, Gerlt JA. (1998) Evolution of enzymatic
activities in the enolase superfamily: characterization of the (D)-
glucarate/galactarate catabolic pathway in Escherichia coli. Biochemistry.
13;37(41):14369-75.
David R. J. Palmer, James B. Garrett,V. Sharma, R. Meganathan, Patricia C. Babbitt, and John
A. Gerlt, (1999) Unexpected Divergence of Enzyme Function and Sequence: ‘‘N-
Acylamino Acid Racemase” Is o-Succinylbenzoate Synthase, Biochemistry, 38 (14),
pp 4252–4258
Satu Kuorelahti Paula Jouhten, Hannu Maaheimo, Merja Penttila and Peter Richard (2006) l-
galactonate dehydratase is part of the fungal path for d-galacturonic acid
catabolism Molecular Microbiology 61:4 1060 – 1068
Brian K. Hubbard, Marjan Koch, David R. J. Palmer, Patricia C. Babbitt, and John A. Gerlt
(1998) Evolution of Enzymatic Activities in the Enolase Superfamily:
Characterization of the (D)-Glucarate/Galactarate Catabolic Pathway in
Escherichia coli Biochemistry, 37 (41) 14369–14375
John F. Rakus, Alexander A. Fedorov, Elena V. Fedorov, Margaret E. Glasner, Brian
K. Hubbard, Joseph D. Delli, Patricia C. Babbitt, Steven C. Almo and John A. Gerlt,
(2008) Evolution of Enzymatic Activities in the Enolase Superfamily: l-Rhamnonate
Dehydratase, Biochemistry, 47 (38), 9944–9954
Robert Belshaw and Aris Katzourakis (2005) Blast to Align: a program that uses blast to
align problematic nucleotide sequences, Bioinformatics 21(1):122-123
Dmitry Lupyan, Alejandra Leo-Macias and Angel R. Ortiz (2005) A new progressive-
iterative algorithm for multiple structure alignment Bioinformatics Volume 21:15
3255-3263
Doron Betel and Christopher WV Hogue, Kangaroo (2002) A pattern-matching program for
biological sequences, BMC Bioinformatics, 1186/1471-2105-3-20
Ofir Goldenberg, Elana Erez, Guy Nimrod, and Nir Ben-Tal (2009) The ConSurf-DB: pre-
calculated evolutionary conservation profiles of protein structures Nucleic Acids
Res. D323–D327.
Changwon Keum and Dongsup Kim (2006) Protein function prediction via ligand
interface residue match, World Congress on Medical Physics and Biomedical
Engineering 2006, August 27 – September 1, COEX Seoul, Korea ‘‘Imaging the
Future Medicine”
LevelErik Lindahl and Arne Elofsson, (2000) Identification of Related Proteins on Family,
Superfamily and Fold Journal of MolecularBiology 295: 3, 613-625
328 Computational Biology and Applied Bioinformatics

Neidhart DJ, Kenyon GL, Gerlt JA, Petsko GA (1990) Mandelate racemase and muconate
lactonizing enzyme are mechanistically distinct and structurally homologous.
Nature. 347(6294):692-4.
 17

Contributions of Structure Comparison Methods

to the Protein Structure Prediction Field
David Piedra1, Marco d'Abramo2 and Xavier de la Cruz1,3
1IBMB-CSIC
2CNIO
3ICREA

Spain

1. Introduction
Since their development, structure comparison methods have contributed to advance our
understanding of protein structure and evolution (Greene et al, 2007; Hasegawa & Holm,
2009), to help the development of structural genomics projects (Pearl et al, 2005), to improve
protein function annotations (D. A. Lee et al), etc, thus becoming an essential tool in
structural bioinformatics. In recent years, their application range has grown to include the
protein structure prediction field, were they are used to evaluate overall prediction quality
(Jauch et al, 2007; Venclovas et al, 2001; Vincent et al, 2005; G. Wang et al, 2005), to identify a
protein’s fold from low-resolution models (Bonneau et al, 2002; de la Cruz et al, 2002), etc. In
this chapter, after briefly reviewing some of these applications, we show how structure
comparison methods can also be used for local quality assessment of low-resolution models
and how this information can help refine/improve them.
Quality assessment is becoming an important research topic in structural bioinformatics
because model quality determines the applicability of structure predictions (Cozzetto et al,
2007). Also, because prediction technology is now easily available and potential end-users of
prediction methods, from template-based (comparative modeling and threading) to de novo
methods, are no longer specialized structural bioinformaticians. Quality assessment
methods have been routinely used for many years in structural biology in the evaluation of
experimental models. These methods focus on several features of the protein structure (see
(Laskowski et al, 1998) and (Kleywegt, 2000) and references therein). Because a number of
quality issues are common to both experimental and predicted models, the use of these
methods has been naturally extended to the evaluation of structure predictions. For
example, in the case of homology modeling, a widely used structure prediction technique,
evaluation of models with PROCHECK (Laskowski et al, 1993), WHAT-CHECK (Hooft et al,
1997), PROSA (Sippl, 1993), and others (see (Marti-Renom et al, 2000) and references therein)
is part of the standard prediction protocol; WHATIF (Vriend, 1990) and PROSA (Sippl, 1993)
have also been used in the CASP experiment to assess comparative models (Venclovas, 2001;
Williams et al, 2001); etc.
Some quality assessment problems are unique to the structure prediction field, given the
specific characteristics of computational models, and have led to the development of
330 Computational Biology and Applied Bioinformatics

methods aimed at: the recognition of near-native predictions from a set of decoys (Jones &
Thornton, 1996; Lazaridis & Karplus, 2000; Sippl, 1995); identification of a target’s protein
family (Bonneau et al, 2002; de la Cruz et al, 2002); overall quality assessment of predictions
(Archie et al, 2009; Benkert et al, 2009; Cheng et al, 2009; Larsson et al, 2009; Lundstrom et al,
2001; McGuffin, 2009; Mereghetti et al, 2008; Wallner & Elofsson, 2003; 2005; Z. Wang et al,
2009; Zhou & Skolnick, 2008); and, more recently, residue-level quality assessment (Benkert
et al, 2009; Cheng et al, 2009; Larsson et al, 2009; McGuffin, 2009; Wallner & Elofsson, 2006;
2007; Z. Wang et al, 2009). However, in spite of these promising efforts, quality assessment
of protein structure predictions remains an open issue(Cozzetto et al, 2009).
Here we focus on the problem of local quality assessment, which consists on the
identification of correctly modeled regions in predicted structures (Wallner & Elofsson,
2006; 2007), or, as stated by Wallner and Elofsson(Wallner & Elofsson, 2007): “The real value
of local quality prediction is when the method is able to distinguish between high and low
quality regions.”. In many cases, global and local quality estimates are produced
simultaneously (Benkert et al, 2009; Cheng et al, 2009; Larsson et al, 2009; McGuffin, 2009).
However, in this chapter we separate these two issues by assuming that, irrespective of its
quality, a structure prediction with the native fold of the corresponding protein is available.
From a structural point of view this is a natural requirement, as a correct local feature
(particularly if it is one which, like a β-strand (Chou et al, 1983), is stabilized by long-range
interactions) in an otherwise wrong structure can hardly be understood. From a practical
point of view, successful identification of correct parts within incorrect models may lead to
costly errors. For example, identifying a correctly modeled binding site within a structurally
incorrect context should not be used for drug design: it would surely have incorrect
dynamics; the long-range terms of the interaction potential, like electrostatics, would be
meaningless; false neighboring residues could create unwanted steric clashes with the
substrate, thus hampering its docking; or, on the contrary, absence of the true neighbors
could lead to unrealistic docking solutions; etc. In the remaining of the chapter we describe
how structure comparison methods can be applied to obtain local quality estimates for low-
resolution models and how these estimates can be used to improve the model quality.

2. A simple protocol for local quality assessment with structure comparison

methods
As mentioned before, an important goal in local quality assessment(Wallner & Elofsson,
2006; 2007) is to partition the residues from a structure prediction in two quality classes:
high and low. This can be done combining several predictions; however, in the last two
rounds of the CASP experiment -a large, blind prediction experiment performed every two
years(Kryshtafovych et al, 2009)- evaluators of the Quality Assessment category stressed
that methods aimed to assess single predictions are needed(Cozzetto et al, 2007; Cozzetto et
al, 2009). These methods are particularly important for users that generate their protein
models with de novo prediction tools, which are still computationally costly(Jauch et al,
2007), particularly for large proteins.
Here we describe a single-molecule approach, based on the use of structure comparison
methods, that allows to partition model residues in two sets, of high and low quality
respectively. In this approach (Fig. 1), the user’s model of the target is first structurally
aligned with a target’s homolog. This alignment, which constitutes the core of the
procedure, is then used to separate the target’s residues in two groups: aligned and
Contributions of Structure Comparison Methods to the Protein Structure Prediction Field 331

unaligned. The main assumption of this approach is that aligned residues are of higher
quality than the average. The validity of this assumption is tested in the next section. In
section 3 we discuss the conditions that determine/limit the applicability and usefulness of
the method.

Fig. 1. Use of structure comparison methods for local quality assessment of structure
predictions

2.1 Performance of structure comparison methods in local quality assessment

To show that structurally aligned residues are usually of higher quality we used a set of de
novo predictions with medium/low to very low resolution, obtained with Rosetta(Simons et
al, 1997). Although several de novo prediction programs have shown promising results in the
CASP experiment(Jauch et al, 2007; Vincent et al, 2005), we used Rosetta predictions
because: (i) Rosetta is a well known de novo prediction program that has ranked first in
successive CASP rounds (Jauch et al, 2007; Vincent et al, 2005); (ii) many Rosetta predictions
are available at the server of Baker’s group, thus allowing a consistent test of our approach,
with predictions from the same source; and (iii) the program is available for interested users
(http://www.rosettacommons.org/software/).
We downloaded the protein structure predictions from the server of Baker’s laboratory
(http://depts.washington.edu/bakerpg/drupal/). This set was constituted by 999 de novo
models generated with Rosetta(Simons et al, 1997) for 85 proteins, i.e. a total of 84915
models. Application of the protocol studied here (Fig. 1) requires that the structure of a
homolog of the target is available, and that the predictions used have the fold of the target.
332 Computational Biology and Applied Bioinformatics

The former was enforced by keeping only those proteins with a CATH relative at the T-
level(Pearl et al, 2005) (homologs with the same fold of the target protein, regardless of their
sequence similarity). The second condition was required to focus only on the local quality
assessment problem, and was implemented by excluding those models not having the fold
of their target protein. Technically, this meant that we only kept those models structurally
similar to any member of the target’s structural family: that is, those models giving a score
higher than 5.25 for the model-homolog structure comparison done with MAMMOTH
(Ortiz et al, 2002), for at least one homolog of the target protein. This step was
computationally costly, as it involved 7,493,499 structure comparisons, and could only be
carried using MAMMOTH(Ortiz et al, 2002); the 5.25 score threshold was taken from
MAMMOTH’s article (Ortiz et al, 2002). The final dataset was constituted by 68 target
proteins and a total of 17180 models.
The properties of the selected target’s residues (STR; to avoid meaningless results we only
considered STR sets larger than 20 residues) were characterized with four parameters: two
structure-based, and two sequence-based. The former were used to check if STR really were
of better quality, comparing their parameters’ values with those obtained for the set of all
the target residues (ATR), i.e. the whole model structure. It has to be noted that: (i) STR and
ATR sets are constituted by residues from the target protein, more precisely STR is a subset
of ATR; and (ii) three possible STR sets were produced, because we checked our procedure
using three structure comparison methods (MAMMOTH (Ortiz et al, 2002), SSAP(Orengo &
Taylor, 1990) and LGA(Zemla, 2003)). The sequence-based properties were utilized to
describe how STR spread along the sequence of the target, which helps to assess the
usefulness of the protocol. Below we provide a brief description of each parameter, together
with the results obtained from their use.

2.1.1 Structural quality: rmsd

Rmsd(Kabsch, 1976) is a quality measure widely employed to assess structure models: it
corresponds to the average distance between model atoms and their equivalent in the native
structure. Small rmsd values correspond to higher quality predictions than larger values.
In Fig. 2 we see the STR and ATR rmsd distributions. Regardless of the structure
comparison method used (MAMMOTH (Ortiz et al, 2002), SSAP(Orengo & Taylor, 1990)
and LGA(Zemla, 2003) in blue, yellow and red, respectively), STR distributions are shifted
towards lower rmsd values relative to ATR distributions (in grey). This confirms the starting
assumption: it shows that model residues structurally aligned to the protein’s homolog
usually have a higher structural quality. A consensus alignment (in black), which combined
the results of the three structure comparison methods, gave better results at the price of
including fewer residues; for this reason we excluded the consensus approach from
subsequent analyses.
An interesting feature of STR rmsd distributions was that their maxima were between 3.5 Å
and 6.5 Å, and that a majority of individual values were between 3 Å and 8 Å, and below 10
Å. To further explore this issue, we plotted the values of rmsd for STR against ATR (Fig. 3,
grey boxes). In accordance with the histogram results, STR rmsd tended to be smaller than
ATR rmsd. We distinguished two regions in the graph: in the first region (ATR rmsd
between 0 Å and 6-8 Å) there was a roughly linear relationship between ATR and STR
rmsds; however, for ATR rmsd values beyond 8 Å, STR rmsd reached a plateau. This
plateau is at the origin of the thresholds observed in the histograms (Fig. 2), and confirms
Contributions of Structure Comparison Methods to the Protein Structure Prediction Field 333

that structure alignments can be used to identify subsets of model residues with better rmsd
than the rest.

Fig. 2. Quality of structurally aligned regions vs. whole model, rmsd frequency histogram.
As a performance reference we used the PROSA program (Sippl, 1993) (white boxes) which
provides a residue-by-residue, energy-based quality assessment, and is a single model
method, therefore comparable to the approach presented here. PROSA was executed with
default parameters, and we took as high quality residues those having energies below zero.
In Fig. 3 we see that for good models, i.e. those with low ATR values, PROSA results (in
white) were as good as those obtained with structure comparison methods (in grey).
However, as models became poorer, PROSA results became worse, particularly after
structure comparison methods reached their plateau. This indicates that when dealing with
poor predictions use of structure alignments can improve/complement other quality
assessment methods.
334 Computational Biology and Applied Bioinformatics

Fig. 3. Quality of structurally aligned (obtained with MAMMOTH (Ortiz et al, 2002)) regions
vs. whole model, rmsd of selected residues vs. all-residues. Grey: structure comparison-
based protocol (Fig. 1); white: PROSA(Sippl, 1993) results.

2.1.2 Structural quality: GDT_TS

GDT_TS is a quality measure routinely utilised by evaluator teams in the CASP community
experiment (Jauch et al, 2007; Vincent et al, 2005): it is equal to the average of the
percentages of model residues at less than 1 Å, 2 Å, 4 Å and 8 Å from their location in the
correct structure. It was computed following the procedure described by Zemla(Zemla,
2003), using Cα atoms to compute residue-residue distances. GDT_TS varies between 0 and
100, with values approaching 100 as models become better.
We found that STR GDT_TS was in general better than ATR GDT_TS (Fig. 4); this was
particularly true when the latter was below 40-50. Overall, this shows that STR is enriched
in good quality sub-structures relative to ATR, particularly for poor models.
Consistency with rmsd analysis was observed when comparing the performance of
structure comparison-based quality assessment (in grey) with that of PROSA (in white): for
good models (GDT_TS values above 60-70) both approaches had a similar behavior;
however, as model quality decreased, use of structure alignments showed increasingly
better performance than PROSA at pinpointing correct substructures.
Contributions of Structure Comparison Methods to the Protein Structure Prediction Field 335

Fig. 4. Quality of structurally aligned (obtained with MAMMOTH (Ortiz et al, 2002)) regions
vs. whole model, GDT_TS of the selected residues vs. all-residues GDT_TS. Grey: structure
comparison-based protocol (Fig. 1); white: PROSA(Sippl, 1993) results.

2.1.3 Distribution of high quality residues along the protein sequence

Usually, STR do not form a continuous block, they tend to scatter along the sequence. The
nature of this distribution is of interest for some applications of quality assessment methods
(like model refinement) for which STR sets may be of little value if the involved residues are
either too close in sequence, or contain too many orphan residues.
To characterize the distribution of STR along the sequence we used two measures:
maximum distance (MD) between STR runs and normalized size distribution of STR runs
(SAS). Both are based on the fact that, for a given model, STR sets are constituted by residue
runs of varying size. MD corresponds to the largest sequence distance between STR runs
(i.e. the number of residues between the rightmost and leftmost STR runs), divided by
whole sequence length. MD values near 1 indicate that STR runs are spread over the whole
protein, while smaller values point to a tighter residue clustering. SAS corresponds to the
normalized (again by whole sequence length) size distribution for all runs constituting STR
sets. SAS gives a view of how the sequence coverage is done: either by large sequence
chunks, by small residue clusters, or by a mixture of both. When the alignment is
constituted by small, evenly distributed residue clusters the SAS distribution will approach
zero.
336 Computational Biology and Applied Bioinformatics

Our results showed that MD values are more frequent above 0.5, and more than 50% of
them were higher than 0.8 (Fig. 5). The three structure comparison methods showed similar
distributions, although LGA was slightly nearer to 1. This indicates that STR spread over a
substantial part of the predicted protein.

Fig. 5. Frequency distribution of the selected residues along the target sequence: normalized
maximum distance between STR runs (unitless parameter).
Results for SAS (Fig. 6) showed that while ~50 % of STR formed clusters of size lower than
10 % of the whole sequence (i.e. SAS values below 0.1), the remaining residues were
grouped in medium to large stretches. This means that for a 100 residue protein, clusters of
more than 10 residues (which is roughly the size of an average α-helix) are frequent. In
addition, for 95 % of the cases, the largest run of adjacent residues was above 30 % of the
target length.
The picture arising from MD and SAS distributions is that STR usually extend over the
protein length. Although STR sets are constituted by somewhat heterogeneous runs they do
not contain too many orphan residues, as they include one large run (the size of a
supersecondary structure motif, or larger) and several, smaller runs (the size of secondary
structure elements).
Contributions of Structure Comparison Methods to the Protein Structure Prediction Field 337

Fig. 6. Frequency distribution of the selected residues along the target sequence: normalized
maximum distance between STR runs (unitless parameter).

3. Applicability range of structure comparison methods in local quality

assessment
The approach described here is easy to use and has few computational requirements;
however, it cannot be arbitrarily applied to any model or in any prediction scenario. In this
section we describe which are its limits regarding prediction methods, target proteins and
protein model nature.

3.1 Prediction methods

As far as the target protein has a homolog of known structure, model-homolog structure
alignments can be computed and the quality assessment protocol (Fig. 1) can be applied,
regardless of the prediction method originating the model. However, the approach
presented here reaches its maximum utility when models are obtained with de novo
structure prediction methods (methods originally devised to work using only the target’s
sequence and a physico-chemical/statistical potential, irrespective of the availability of
homologs). This may seem somewhat contradictory, as one can think that the existence of
target’s homologs favors the use of comparative modeling methods instead of de novo
methods. However, this is not the case: while de novo methods were initially developed with
the de novo scenario in mind (only sequence information is available for the target protein),
this situation is changing rapidly. Actually, when prediction problems become difficult, or a
338 Computational Biology and Applied Bioinformatics

given method gives an unclear answer, using more than one technique is considered a good
approach within the prediction community, as the evaluators of the de novo section in the
CASP6 experiment explain (Vincent et al, 2005): “Many predicting groups now use both de
novo and homology modeling/fold recognition techniques to predict structures in all
categories”. In addition, it has been shown that de novo methods can compete with
template-based methods in the prediction of difficult targets (Jauch et al, 2007; Raman et al,
2009; Vincent et al, 2005). In this situation, which implies the existence of target's homologs,
our method can be used to score the local quality of de novo predictions.
In addition, a completely new field of application for de novo methods has been unveiled by
the growing interest in knowing the structure of alternative splicing isoforms (C. Lee &
Wang, 2005). Due to the very localized nature of sequence changes (Talavera et al, 2007),
structure prediction of alternative splicing variants seems a trivial exercise in comparative
modeling. However, template-based methods fail to reproduce the structure changes
introduced by alternative splicing (Davletov & Jimenez, 2004). De novo approaches with
their ability to combine first principles with deep conformational searches are ideal
candidates to tackle this problem; in this case, availability of the structure of only one
isoform would allow the application of our method.

3.2 Target proteins

Proteins to which our approach can be applied must have a homolog of known structure.
The number of these proteins is increasing due to: (i) the progress of structural genomics
projects (Todd et al, 2005) (this will increase the number of both easy/medium and hard
targets); (ii) the growing number of alternative splicing variants of unknown structure (C.
Lee & Wang, 2005).

3.3 Protein models

The approach proposed (Fig. 1) is a local, not a global, quality assessment method and
should only be applied to models that have the native fold of the target (see above). Present
de novo methods still cannot consistently produce models with a native-like fold (Moult et al,
2009). Therefore, researchers must ascertain that the model’s fold is correct (irrespective of
its resolution). This can be done using global quality assessment methods like PROSA
(Sippl, 1993), the Elofsson’s suite of programs (Wallner & Elofsson, 2007), etc.

4. Applications
Once available, local quality information can be used with different purposes. For example, it
may help to identify those parts of a theoretical model that are more reliable for mutant
design, or to interpret the results of mutagenesis experiments; it may be used for in sillico
docking experiments involving de novo models, to decide which parts of the models must be
employed preferentially; etc. One the most promising applications of quality assessment
methods is the refinement of low-resolution models (Wallner & Elofsson, 2007). In this section
we illustrate how the results of the procedure here described can be used for this purpose.
Among the possible options available for model refinement, we propose to use the
alignment resulting from the structural superimposition between a de novo model and the
target’s homolog (Fig. 1) as input to a comparative modeling program. We applied this
Contributions of Structure Comparison Methods to the Protein Structure Prediction Field 339

strategy to 15 proteins (five from each of the three main CATH structural classes: alpha, beta
and alpha/beta) from our initial dataset. These 15 proteins contributed a total of 2693 de
novo models that resulted in 8033 model-homolog alignments (obtained with MAMMOTH
(Ortiz et al, 2002)). These alignments were subsequently used as input to the standard
homology modeling program MODELLER (Marti-Renom et al, 2000), which was run with
default parameters. For the aligned regions we found (Fig. 7) that most of the refined
models had lower model-native rmsd than the starting de novo models, i.e. they were closer
to the native structure. A similar, although milder, trend was also observed when
considering the whole set of protein residues (i.e. aligned as well as unaligned residues)
(Fig. 8). These results show that this simple, computationally cheap model refinement
protocol, based on the use of structure comparison local quality analysis, clearly helps to
refine/improve low-resolution de novo models to an accuracy determined by the closest
homolog of the target.

Fig. 7. Model refinement using structure comparison-based local quality assessment: rmsd
of refined models vs. rmsd of original de novo models, subset of aligned residues. Points
below the dotted line correspond to refinement-improved models.
340 Computational Biology and Applied Bioinformatics

Fig. 8. Model refinement using structure comparison-based local quality assessment: rmsd
of refined models vs. rmsd of original de novo models, all protein residues. Points below the
dotted line correspond to refinement-improved models.

5. Conclusions
In this chapter we have described and tested a protocol for local quality assessment of low-
resolution predictions based on the use of structure comparison methods. The testing was
carried with de novo predictions, and the results showed that structure comparison methods
allow the partitioning of the model’s residues in two sets of high and low quality,
respectively. This result holds even when only remote homologs of the target protein are
available. The simplicity of the approach leaves room for future improvements and fruitful
combination with other quality assessment methods. Two conditions determine the
application range of this approach: the target protein must have at least one homolog of
known structure, and models reproducing the fold of the target are required. However,
results indicating that we may be near a full coverage of the proteins’ fold space, together
with advances in overall quality scoring indicate that these two problems are likely to
become minor issues in the near future. Finally, our procedure suggests a simple refinement
Contributions of Structure Comparison Methods to the Protein Structure Prediction Field 341

strategy based on the use of comparative modeling programs that may be used to improve
low-resolution de novo models.

6. Acknowledgment
This work is dedicated to the memory of Angel Ramírez Ortíz, leading bioinformatician and
designer of the MAMMOTH program for structure comparison. The authors wish to thank
the CATH team for their support. Xavier de la Cruz acknowledges funding from the
Spanish government (Grants BIO2006-15557 and BFU2009-11527). David Piedra
acknowledges economical support from the Government of Catalonia and the Spanish
Ministerio de Educación y Ciencia.

7. References
Archie, J.G.; Paluszewski M. & Karplus K. (2009) Applying Undertaker to quality
assessment. Proteins Vol.77 Suppl 9191-195, ISSN 0887-3585.
Benkert, P.; Tosatto S.C. & Schwede T. (2009) Global and local model quality estimation at
CASP8 using the scoring functions QMEAN and QMEANclust. Proteins Vol.77
Suppl 9173-180, ISSN 0887-3585.
Bonneau, R.; Strauss C.E.; Rohl C.A.; Chivian D.; Bradley P.; Malmstrom L.; Robertson T. &
Baker D. (2002) De novo prediction of three-dimensional structures for major
protein families. J Mol Biol Vol.322, No.1, pp. 65-78, ISSN 0022-2836.
Cheng, J.; Wang Z.; Tegge A.N. & Eickholt J. (2009) Prediction of global and local quality of
CASP8 models by MULTICOM series. Proteins Vol.77 Suppl 9181-184, ISSN 0887-
3585.
Chou, K.C.; Nemethy G. & Scheraga H.A. (1983) Role of interchain interactions in the
stabilization of the right-handed twist of beta-sheets. J Mol Biol Vol.168, No.2, pp.
389-407, ISSN 0022-2836.
Cozzetto, D.; Kryshtafovych A.; Ceriani M. & Tramontano A. (2007) Assessment of
predictions in the model quality assessment category. Proteins Vol.69 Suppl 8175-
183, ISSN 0887-3585.
Cozzetto, D.; Kryshtafovych A. & Tramontano A. (2009) Evaluation of CASP8 model quality
predictions. Proteins Vol.77 Suppl 9157-166, ISSN 0887-3585.
Davletov, B. & Jimenez J.L. (2004) Sculpting a domain by splicing. Nat Struct Mol Biol
Vol.11, No.1, pp. 4-5,
de la Cruz, X.; Sillitoe I. & Orengo C. (2002) Use of structure comparison methods for the
refinement of protein structure predictions. I. Identifying the structural family of a
protein from low-resolution models. Proteins Vol.46, No.1, pp. 72-84, ISSN 0887-
3585.
Greene, L.H.; Lewis T.E.; Addou S.; Cuff A.; Dallman T.; Dibley M.; Redfern O.; Pearl F.;
Nambudiry R.; Reid A.; Sillitoe I.; Yeats C.; Thornton J.M. & Orengo C.A. (2007)
The CATH domain structure database: new protocols and classification levels give
a more comprehensive resource for exploring evolution. Nucleic Acids Res Vol.35,
No.Database issue, pp. D291-297, ISSN 0305-1048.
Hasegawa, H. & Holm L. (2009) Advances and pitfalls of protein structural alignment. Curr
Opin Struct Biol Vol.19, No.3, pp. 341-348, ISSN 0959-440X.
342 Computational Biology and Applied Bioinformatics

Hooft, R.W.; Sander C. & Vriend G. (1997) Objectively judging the quality of a protein
structure from a Ramachandran plot. Comput Appl Biosci Vol.13, No.4, pp. 425-
430, ISSN 0266-7061.
Jauch, R.; Yeo H.C.; Kolatkar P.R. & Clarke N.D. (2007) Assessment of CASP7 structure
predictions for template free targets. Proteins Vol.69 Suppl 857-67, ISSN 0887-3585.
Jones, D.T. & Thornton J.M. (1996) Potential energy functions for threading. Curr Opin
Struct Biol Vol.6, No.2, pp. 210-216, ISSN 0959-440X.
Kabsch, W.A. (1976) A solution for the best rotation to relate two sets of vectors. Acta
Crystallogr A Vol.A32922-923,
Kleywegt, G.J. (2000) Validation of protein crystal structures. Acta Crystallogr D Biol
Crystallogr Vol.56, No.Pt 3, pp. 249-265, ISSN 1399-0047.
Kryshtafovych, A.; Fidelis K. & Moult J. (2009) CASP8 results in context of previous
experiments. Proteins Vol.77 Suppl 9217-228, ISSN 0887-3585.
Larsson, P.; Skwark M.J.; Wallner B. & Elofsson A. (2009) Assessment of global and local
model quality in CASP8 using Pcons and ProQ. Proteins Vol.77 Suppl 9167-172,
ISSN 0887-3585.
Laskowski, R.A.; MacArthur M.W.; Moss D.S. & Thornton J.M. (1993) PROCHECK: a
program to check the stereochemical quality of protein structures. J Appl
Crystallogr Vol.26283-291, ISSN 1600-5767.
Laskowski, R.A.; MacArthur M.W. & Thornton J.M. (1998) Validation of protein models
derived from experiment. Curr Opin Struct Biol Vol.8, No.5, pp. 631-639, ISSN
0959-440X.
Lazaridis, T. & Karplus M. (2000) Effective energy functions for protein structure prediction.
Curr Opin Struct Biol Vol.10, No.2, pp. 139-145, ISSN 0959-440X.
Lee, C. & Wang Q. (2005) Bioinformatics analysis of alternative splicing. Brief Bioinform
Vol.6, No.1, pp. 23-33,
Lee, D.A.; Rentzsch R. & Orengo C. (2010) GeMMA: functional subfamily classification
within superfamilies of predicted protein structural domains. Nucleic Acids Res
Vol.38, No.3, pp. 720-737, ISSN 0305-1048.
Lundstrom, J.; Rychlewski L.; Bujnicki J. & Elofsson A. (2001) Pcons: a neural-network-based
consensus predictor that improves fold recognition. Protein Sci Vol.10, No.11, pp.
2354-2362,
Marti-Renom, M.A.; Stuart A.C.; Fiser A.; Sanchez R.; Melo F. & Sali A. (2000) Comparative
protein structure modeling of genes and genomes. Annu Rev Biophys Biomol
Struct Vol.29291-325, ISSN 1056-8700.
McGuffin, L.J. (2009) Prediction of global and local model quality in CASP8 using the
ModFOLD server. Proteins Vol.77 Suppl 9185-190, ISSN 0887-3585.
Mereghetti, P.; Ganadu M.L.; Papaleo E.; Fantucci P. & De Gioia L. (2008) Validation of
protein models by a neural network approach. BMC Bioinformatics Vol.966, ISSN
1471-2105.
Moult, J.; Fidelis K.; Kryshtafovych A.; Rost B. & Tramontano A. (2009) Critical assessment
of methods of protein structure prediction - Round VIII. Proteins Vol.77 Suppl 91-4,
ISSN 0887-3585.
Orengo, C.A. & Taylor W.R. (1990) A rapid method of protein structure alignment. J Theor
Biol Vol.147, No.4, pp. 517-551,
Contributions of Structure Comparison Methods to the Protein Structure Prediction Field 343

Ortiz, A.R.; Strauss C.E. & Olmea O. (2002) MAMMOTH (matching molecular models
obtained from theory): an automated method for model comparison. Protein Sci
Vol.11, No.11, pp. 2606-2621,
Pearl, F.; Todd A.; Sillitoe I.; Dibley M.; Redfern O.; Lewis T.; Bennett C.; Marsden R.; Grant
A.; Lee D.; Akpor A.; Maibaum M.; Harrison A.; Dallman T.; Reeves G.; Diboun I.;
Addou S.; Lise S.; Johnston C.; Sillero A.; Thornton J. & Orengo C. (2005) The
CATH Domain Structure Database and related resources Gene3D and DHS provide
comprehensive domain family information for genome analysis. Nucleic Acids Res
Vol.33, No.Database issue, pp. D247-251,
Raman, S.; Vernon R.; Thompson J.; Tyka M.; Sadreyev R.; Pei J.; Kim D.; Kellogg E.; DiMaio
F.; Lange O.; Kinch L.; Sheffler W.; Kim B.H.; Das R.; Grishin N.V. & Baker D.
(2009) Structure prediction for CASP8 with all-atom refinement using Rosetta.
Proteins Vol.77 Suppl 989-99, ISSN 0887-3585.
Simons, K.T.; Kooperberg C.; Huang E. & Baker D. (1997) Assembly of protein tertiary
structures from fragments with similar local sequences using simulated annealing
and Bayesian scoring functions. J Mol Biol Vol.268, No.1, pp. 209-225,
Sippl, M.J. (1993) Recognition of errors in three-dimensional structures of proteins. Proteins
Vol.17, No.4, pp. 355-362, ISSN 0887-3585.
Sippl, M.J. (1995) Knowledge-based potentials for proteins. Curr Opin Struct Biol Vol.5,
No.2, pp. 229-235, ISSN 0959-440X.
Talavera, D.; Vogel C.; Orozco M.; Teichmann S.A. & de la Cruz X. (2007) The
(in)dependence of alternative splicing and gene duplication. PLoS Comput Biol
Vol.3, No.3, pp. e33,
Todd, A.E.; Marsden R.L.; Thornton J.M. & Orengo C.A. (2005) Progress of structural
genomics initiatives: an analysis of solved target structures. J Mol Biol Vol.348,
No.5, pp. 1235-1260, ISSN 0022-2836.
Venclovas, C. (2001) Comparative modeling of CASP4 target proteins: combining results of
sequence search with three-dimensional structure assessment. Proteins Vol.Suppl
547-54, ISSN 0887-3585.
Venclovas, C.; Zemla A.; Fidelis K. & Moult J. (2001) Comparison of performance in
successive CASP experiments. Proteins Vol.Suppl 5163-170, ISSN 0887-3585.
Vincent, J.J.; Tai C.H.; Sathyanarayana B.K. & Lee B. (2005) Assessment of CASP6
predictions for new and nearly new fold targets. Proteins Vol.61 Suppl 767-83, ISSN
0887-3585.
Vriend, G. (1990) WHAT IF: a molecular modeling and drug design program. J Mol Graph
Vol.8, No.1, pp. 52-56, 29, ISSN 0263-7855.
Wallner, B. & Elofsson A. (2003) Can correct protein models be identified? Protein Sci
Vol.12, No.5, pp. 1073-1086, ISSN 0961-8368.
Wallner, B. & Elofsson A. (2005) Pcons5: combining consensus, structural evaluation and
fold recognition scores. Bioinformatics Vol.21, No.23, pp. 4248-4254, ISSN 1367-
4803.
Wallner, B. & Elofsson A. (2006) Identification of correct regions in protein models using
structural, alignment, and consensus information. Protein Sci Vol.15, No.4, pp. 900-
913, ISSN 0961-8368.
Wallner, B. & Elofsson A. (2007) Prediction of global and local model quality in CASP7
using Pcons and ProQ. Proteins Vol.69 Suppl 8184-193, ISSN 0887-3585.
344 Computational Biology and Applied Bioinformatics

Wang, G.; Jin Y. & Dunbrack R.L., Jr. (2005) Assessment of fold recognition predictions in
CASP6. Proteins Vol.61 Suppl 746-66, ISSN 0887-3585.
Wang, Z.; Tegge A.N. & Cheng J. (2009) Evaluating the absolute quality of a single protein
model using structural features and support vector machines. Proteins Vol.75,
No.3, pp. 638-647, ISSN 0887-3585.
Williams, M.G.; Shirai H.; Shi J.; Nagendra H.G.; Mueller J.; Mizuguchi K.; Miguel R.N.;
Lovell S.C.; Innis C.A.; Deane C.M.; Chen L.; Campillo N.; Burke D.F.; Blundell T.L.
& de Bakker P.I. (2001) Sequence-structure homology recognition by iterative
alignment refinement and comparative modeling. Proteins Vol.Suppl 592-97, ISSN
0887-3585.
Zemla, A. (2003) LGA: A method for finding 3D similarities in protein structures. Nucleic
Acids Res Vol.31, No.13, pp. 3370-3374,
Zhou, H. & Skolnick J. (2008) Protein model quality assessment prediction by combining
fragment comparisons and a consensus C(alpha) contact potential. Proteins Vol.71,
No.3, pp. 1211-1218, ISSN 0887-3585.
 18

Functional Analysis of Intergenic

Regions for Gene Discovery
Li M. Fu
Pacific Tuberculosis and Cancer Research Organization, Anaheim, CA
USA

1. Introduction
Gene finding can be defined as a problem of identifying a stretch of the genomic DNA
sequence that is biologically functional. Such a genomic DNA sequence is known as a gene.
A gene performs a function like protein coding or regulation at the molecular level and
plays a biological role, such as growth, metabolism, and intelligence. Traditionally, gene
finding relies on numerous biological experiments and statistical analysis to pinpoint the
location of a new gene in a genetic map. With the advent of bioinformatics, gene finding has
largely become a computational problem. Genes are predictable based on the genomic
sequence alone. However, the determination of the specific function and biological role of a
gene would still demand in vivo experimentation, which is hoped to be reduced or even
replaced by new bioinformatics algorithms in the future.
A newly sequenced genome is annotated thoroughly so that the information it carries can be
utilized. In essence, genome annotation is to identify the locations of genes and all of the
coding regions in a genome, and determine their protein products as well as functions.
Hundreds of bacterial genome sequences are publicly available and the number will soon
reach a new milestone. Gene annotation by hand is almost impossible to handle the deluge
of new genome sequences appearing at this pace. The need for automated, large-scale, high-
throughput genome annotation is imminent (Overbeek, Begley et al. 2005; Van Domselaar,
Stothard et al. 2005; Stothard and Wishart 2006). The basic level of genome annotation is the
use of BLAST (Altschul, Gish et al. 1990) for finding similarities between related genomic
sequences. Integration with other sources of information and experimental data is a trend in
genome annotation.
A recent study indicates that many genomes could be either over-annotated (too many
genes) or under-annotated (too few genes), and a large percentage of genes may have been
assigned a wrong start codon (Nielsen and Krogh 2005). The fact that the original genome
annotation is accurate and complete upon submission does not guarantee that it will not be
changed, as new experimental evidence and knowledge would continue to arrive and
constant updates would be inevitable. However, re-annotation of the whole genome is not
very fruitful, as most of the genes have been identified in the first annotation. For example,
the re-annotation of the H37Rv genome resulted in about 2% of new protein-coding
sequences (CDS) added to the genome. The result reflects the limitation with current
genome annotation technology. To address the issue, we developed a new method for gene
finding in an annotated genome. We select the genome of Mycobacterium tuberculosis, the
346 Computational Biology and Applied Bioinformatics

causative pathogen of tuberculosis, as the experimental genome for this study. The
availability of the complete genome sequence of M. tuberculosis H37Rv (Cole, Brosch et al.
1998) has led to a better understanding of the biology and pathogenicity of the organism,
and new molecular targets for diagnostics and therapeutics can be invented at a fast pace by
focusing on genes with important functions.
In our previous studies, we found that some intergenic sequences in M. tuberculosis genome
exhibited expression signals, as detected by the Affymetrix GeneChip (Fu 2006; Fu and Fu-
Liu 2007; Fu and Shinnick 2007). The same observation has been made for other bacteria,
such as Bacillus subtilis (Lee, Zhang et al. 2001), and also holds true in the eukaryotic system
(Zheng, Zhang et al. 2005). At present, it is not clear whether or how intergenic expression
represents gene activity. Here, we presented our research work concerning gene discovery
in the intergenic sequences based on transcription activity. In this work, new protein-coding
genes were identified by the bioinformatics criteria based on the gene structure, protein
coding potential, and ortholog evidence, in conjunction with microarray-based
transcriptional evidence.

2. Research methods and design

The developed method of gene finding in the intergenic sequences proceeds as follows:
1. Transcription analysis to identify intergenic regions exhibiting significant gene
expression activity.
2. Coding potential and gene structure analysis on active intergenic elements identified
based on transcription evidence.
3. Protein domain search to identify functional domains in each active intergenic element
with significant transcription activity and coding potential.
4. Homology search based on BLAST to seek homologue evidence.
The flowchart of the method is displayed in Figure 1.
The method was applied to the originally annotated M. tuberculosis H37Rv genome (Cole,
Brosch et al. 1998). The genes discovered in the intergenic sequences were validated against
recent findings in the literature. The research protocols (Fu and Shinnick 2007) were
described below.

2.1 RNA isolation

M. tuberculosis strain H37Rv was obtained from the culture collection of the Mycobacteriology
Laboratory Branch, Centers for Disease Control and Prevention at Atlanta. Bacterial lysis and
RNA isolation were performed following the procedure at the CDC lab, Atlanta (Fisher,
Plikaytis et al. 2002). Briefly, cultures were mixed with an equal volume of RNALaterTM
(Ambion, Austin, TX) and the bacteria harvested by centrifugation (1 min, 25000g, 8°C) and
transferred to Fast Prep tubes (Bio 101, Vista, CA) containing Trizol (Life Technologies,
Gaithersburg, MD). Mycobacteria were mechanically disrupted in a Fast Prep apparatus. The
aqueous phase was recovered, treated with Cleanascite (CPG, Lincoln Park, NJ), and extracted
with chloroform-isoamyl alcohol (24:1 v/v). Nucleic acids were ethanol precipitated. DNAase
I (Ambion) treatment to digest contaminating DNA was performed in the presence of Prime
RNase inhibitor (5’-3’, Boulder, CO). The RNA sample was precipitated and washed in
ethanol, and redissolved to make a final concentration of 1 mg/ml. The purity of RNA was
estimated by the ratio of the readings at 260 nm and 280 nm (A260/A280) in the UV. 20 ul
Functional Analysis of Intergenic Regions for Gene Discovery 347

RNA samples were sent to the UCI DNA core and further checked through a quality and
quantity test based on electrophoresis before microarray hybridization.

Gene Finding in Intergenic Regions

−Bioinformatics Analysis Flow Chart−

Transcription analysis based on

microarrays (Affymetrix Chips)
in each intergenic region

No
Significant gene
expression?
Exit

Yes

Coding potential and gene

structure analysis using
GeneMark programs

Meet a gene No Exit

criteria?

Yes

Homology search based on

BLAST

Orthologs No Hypothetical
present? protein

Yes

Protein domain search using

GENE the Pfam program

Fig. 1. The bioinformatics method with a flowchart developed for finding genes in intergenic
regions.
348 Computational Biology and Applied Bioinformatics

2.2 Microarray hybridization

In this study, we used the anti-sense Affymetrix M. tuberculosis genome array (GeneChip).
The probe selection was based on the genome sequence of M. tuberculosis H37Rv (Cole,
Brosch et al. 1998). Each annotated ORF (Open Reading Frame) or IG (Intergenic Region)
was interrogated with oligonucleotide probe pairs. An IG referred to the region between
two consecutive ORFs. The gene chip represented all 3924 ORFs and 740 intergenic regions
of H37Rv. The selection of these IGs in the original design was based on the sequence
length. Twenty 25-mer probes were selected within each ORF or IG. These probes were
called PM (Perfect-Match) probes. The sequence of each PM probe was perturbed with a
single substitution at the middle base. They were called MM (Mismatch) probes. A PM
probe and its respective MM probe constituted a probe pair. The MM probe served as a
negative control for the PM probe in hybridization.
Microarray hybridization followed the Affymetrix protocol. In brief, the assay utilized
reverse transcriptase and random hexamer primers to produce DNA complementary to the
RNA. The cDNA products were then fragmented by DNAase and labeled with terminal
transferase and biotinylated GeneChip DNA Labeling Reagent at the 3' terminal.
Each RNA sample underwent hybridization with one gene array to produce the expression
data of all genes on the array. We performed eleven independent bacterial cultures and
RNA extractions at different times, and collected eleven sets of microarray data for this
study. A global normalization scheme was applied so that each array's median value was
adjusted to a predefine value (500). The scale factor for achieving this transformed median
value for an array was uniformly applied to all the probe set values on a specific array to
result in the determined signal value for all the probe sets on the array. In this manner,
corresponding probe sets can be directly compared across arrays.

2.3 Gene expression analysis

The gene expression data were analyzed by the program GCOS (GeneChip Operating
Software) version 1.4. In the program, the Detection algorithm determined whether a
measured transcript was detected (P Call) or not detected (A Call) on a single array
according to the Detection p-value that was computed by applying the one-sided Wilcoxon’s
signed rank test to test the Discrimination scores (R) against a predefined adjustable
threshold τ. The parameter τ controlled the sensitivity and specificity of the analysis, and
was set to a typical value of 0.015, and the Detection p-value cutoffs, α1 and α2, set to their
typical values, 0.04 and 0.06, respectively, according to the Affymetrix system.

2.4 Gene prediction

Protein-coding region identification and gene prediction were performed by the programs,
GeneMark and GeneMark.hmm (Lukashin and Borodovsky 1998; Besemer and Borodovsky
2005) (http://exon.gatech.edu/GeneMark/), respectively. The prokaryotic version and the
M. tuberculosis H37Rv genome were selected. Both programs use inhomogeneous Markov
chain models for coding DNA and homogeneous Markov chain models for non-coding
DNA. GeneMark adopts Bayesian formalism, while GeneMark.hmm uses a Hidden Markov
Model (HMM).

2.5 Protein domain search

The Pfam program version 20.0 (Finn, Mistry et al. 2006) (http://pfam.wustl.edu/) was
employed to conduct protein domain search after the input DNA sequence was translated
Functional Analysis of Intergenic Regions for Gene Discovery 349

into a protein sequence in six possible frames. The search mode was set to “global and local
alignments merged”, and the cut-off E-value set to 0.001, which was more stringent than the
default value of 1.0. Pfam maintained a comprehensive collection of multiple sequence
alignments and hidden Markov models for 8296 common protein families based on the
Swissprot 48.9 and SP-TrEMBL 31.9 protein sequence databases.

2.6 Homology search

The BLASTx program (Altschul, Gish et al. 1990) (http://www.ncbi.nlm.nih.gov/BLAST/)
was used to identify high-scoring homologous sequences. The program first translated the
input DNA sequence into a protein sequence in six possible frames, and then matched it
against the non-redundant protein sequence database (nr) in the GenBank and calculated
the statistical significance of the matches. The default cut-off E-value was 10.0 but we set it
to 1.0 × 10 −10 . Orthologs refer to homologs in different strains of the same species. Orthologs
provide critical evidence for gene finding and characterization in a new genome sequence. A
typical prokaryotic gene has the following structure: the promoter, transcription initiation,
the 5' untranslated region, translation initiation, the coding region, translation stop, the 3'
untranslated region, and transcription stop.

3. Results
In our previous research, we conducted a genome-wide expression analysis on intergenic
regions using the Affymetrix GeneChip (Fu and Shinnick 2007). The transcriptional activity
of intergenic regions was measured based on a set of eleven independent RNA samples
extracted from M. tuberculosis culture. Each RNA sample contained the information of
genome-wide expression of genes and intergenic elements. The Affymetrix GeneChip was
uniquely suited to this study since it had the advantage of encoding both genes and
intergenic sequences whereas other types of microarray like the cDNA array could not
profile intergenic sequences. As an additional strength, the Affymetrix array was designed
to minimize cross-hybridization by using unique oligonucleotide probes and the pair of PM
(Perfect-Match) and MM (Mismatch) probes. The cross-hybridization of related or
overlapping gene sequences often contributed to false positive signals, especially in the case
when long cDNA sequences were used as probes. A study demonstrated that the Affymetrix
GeneChip produced more reliable results in detecting changes in gene expression than
cDNA microarrays (Li, Pankratz et al. 2002).
In this work, genes in the intergenic sequences were recognized based on transcriptional
activity, structural patterns, and coding potential, and subsequently validated through
sequence comparison with orthologs from other M. tuberculosis strains.

3.1 Transcriptional analysis

An intergenic region was assumed to transcribe if there existed transcripts in the RNA
sample that were bound to the probes encoding that intergenic sequence. The presence or
absence of a given transcript was determined in accordance with the Detection algorithm of
the Affymetrix system. In this study, a gene or intergenic region was determined to express
(transcriptionally active) only if the derived mRNA was present (P-call) in more than 90% of
the collected RNA samples with a Detection p-value < 0.001. The status of active
350 Computational Biology and Applied Bioinformatics

transcription assigned to an intergenic sequence signified the possible presence of a gene

within that sequence. We focused on finding protein-coding genes and neglected regulatory
genes that transcribed into a regulatory RNA instead of mRNA. Furthermore, it was not
clear how much cross-hybridization would occur between genes and intergenic sequences.
As a result, the functional criterion based on expression activity was strengthened by
structural analysis for minimizing false positives in gene identification.

3.2 Sequence-based gene prediction

In the sequence-based approach to gene prediction, gene structure and coding potential are
the two mutually supportive elements. The GeneMark algorithm was applied to an
intergenic sequence for checking whether it contained a probable coding region, and the
GeneMark.hmm algorithm was used for predicting a gene within the sequence. The criteria
based on the predefined transcriptional evidence, coding potential, and gene structure
yielded 65 candidate genes in the intergenic regions of M. tuberculosis H37Rv.

3.3 Protein domain search

The biological function of a gene is determined using in vivo experimentation in a traditional
approach. Recently, the wealth of bioinformatics knowledge in the functional domains of
proteins has enabled the function of a molecular sequence to be characterized directly,
subject to in vivo validation. Thus, the “candidate” genes within the intergenic sequences
that satisfied the criteria based on transcription activity, gene structure, and coding potential
were further examined for embedded functional domains. To this end, Pfam was applied to
search on the protein sequences of the candidate genes. Twelve of them were found to have
a known domain (Tables 1); a found domain was generally located within the predicted
gene sequence, but there were a few exceptions (i.e., IG398 and IG1140) in which a domain
was found within the intergenic sequence but outside the predicted gene sequence. The
biological function and role of a gene is deducible from its associated functional domains;
yet sufficient evidence from homology or biochemistry would serve to corroborate it.

3.4 Homologue evidence

In evolutionary biology, a reliable means for predicting the function of an unknown gene
sequence is based on homologs or orthologs. BLAST is a bioinformatics program for
database search, allowing functional and evolutionary inference between sequences. In this
study, BLAST was employed to retrieve from sequence databases all proteins that produced
statistically significant alignment with a given intergenic sequence under consideration. The
sequences retrieved by BLAST were homologous to the query sequence. It turned out that
the highest-scoring homologous sequences with ≥98% identity were consistently those
belonging to the same strain (H37Rv) or different strains of Mycobacterium tuberculosis (e.g.,
CDC1551, F11, and C). These sequences are coding sequences described in the currently
annotated genome of M. tuberculosis.
A homologous sequence found in different strains of the same species often represents an
ortholog that shares similar function, whereas a homologous sequence found in the same
organism is a paralog (which is produced via gene duplication within a genome) that tends
to have a different function. No evidence suggested paralogs in our analysis, as argued
based on the following observation. We noted that, given an intergenic sequence, when
BLAST returned a homologous sequence pertaining to the H37Rv strain, it was apparently
Functional Analysis of Intergenic Regions for Gene Discovery 351

the same protein-coding sequence contained in the intergenic sequence. This is because the
intergenic sequence used as a query and its homologous sequence returned by BLAST
occupied the same physical location within the genome, as inferred from information given
by the Affymetrix Genechip. This coincidence was further explained by the fact that the
intergenic sequence was named according to the early version of the H37Rv genome
annotation while the homologous sequence was retrieved from the GenBank which
contained all up-to-date genes. The results are significant. First, we demonstrated that our
method was able to identify protein-coding genes in intergenic regions previously
considered as non-coding sequences. Secondly, our method based on bioinformatics and
transcriptional evidence correctly predicted these changes on a high-throughput, genomic
scale. The changes refer to
• IG1061 → (containing) Rv1322A
• IG499 → Rv0634B
• IG617 → Rv0787A
• IG1741 → Rv2219A
• IG2500 → Rv3198A
• IG2053 → Rv2631
• IG1179 → Rv1489A
• IG2522 → Rv3224B
• IG1291 → Rv1638A
• IG398 → Rv0500A
• IG2870 → Rv3678A
• IG188 → Rv0236A
• IG2498 → Rv3196A,
• IG2591 → Rv3312A
• IG595 → Rv0755A
• IG1814 → Rv2309A
• IG1030 → Rv1290A
• IG2141 → Rv2737A
In the above findings, each intergenic region contained an independent gene/CDS with the
only exception that part of IG2053 was incorporated in its left-flanking CDS. The presence of
a gene structure in an IG and its lack of functional correlation with its adjacent genes
suggested that it was not a run-away segment from adjacent genes.
In our analysis, predicted genes located within intergenic sequences that met the criteria
defined based on protein-coding potential, structural patterns, and transcription evidence,
were called “candidate” genes. If a candidate gene of unknown function was homologous to
another gene of know function, the candidate gene was assigned the function associated
with its homologous gene. Nonetheless, the strategy of inferring the function of an
uncharacterized sequence from its orthologs had limited value in analyzing intergenic data
mainly because most of the orthologs found in this study were hypothetical proteins with
unknown function. We did not assign a specific function to a candidate gene until it had an
ortholog of known function, whether or not the candidate gene carried a known functional
domain. In the absence of a specific function assigned, a CDS was termed a hypothetical
protein rather than a gene in our system.
In this work, six intergenic sequences were identified that met the criteria we defined,
including protein coding, structural patterns, transcription, and ortholog evidence: IG499,
352 Computational Biology and Applied Bioinformatics

IG617, IG1741, IG2500, IG1567, and IG2229, among which four genes had been reported in
the M. tuberculosis H37Rv genome (Table 1). A hypothetical protein was found in 52
intergenic sequences and 14 among them had been reported in the H37Rv genome. Overall,
this research discovered two genes with a specific function and 38 hypothetical proteins that
had not been reported in the H37Rv genome (Fu and Shinnick 2007). The two new genes
discovered were a DNA-binding protein in the CopG family and a nickel binding GTPase,
located in IG1567 and IG2229, respectively (Figure 2). It was worth noting that 4.3% of
intergenic regions exhibiting transcriptional activity contained a gene described in the re-
annotated H37Rv genome, compared with 1.0% of intergenic regions in the absence
transcriptional activity. The four-fold increase in likelihood suggested that microarray-based
transcriptional analysis would facilitate genome-wide gene finding.

IG Lt Flank Rt Flank Domain ID Re-annotated H37Rv Gene

IG1061 Rv1322 Rv1323 Glyoxalase Rv1322A*
IG499 Rv0634c Rv0635 Ribosomal_L33 Rv0634B

IG617 Rv0787 Rv0788 PurS Rv0787A

IG398 Rv0500 Rv0501 DUF1713 Rv0500A*

IG1741 Rv2219 Rv2220 RDD Rv2219A

IG2500 Rv3198c Rv3199c Glutaredoxin Rv3198A
IG2053 Rv2631 Rv2632c UPF0027 Rv2631*
IG1179 Rv1489c Rv1490 MM_CoA_mutase Rv1489A*
IG1140 Rv1438 Rv1439c TetR_N None

IG2522 Rv3224 Rv3225c YbaK Rv3224B*

IG1567 Rv1991c Rv1992c RHH_1 None
IG2229 Rv2856 Rv2857c cobW None
*: Hypothetical protein
Table 1. The functional domains of the predicted genes located within the intergenic
sequences of M. tuberculosis H37Rv genome. Each intergenic sequence shown is
characterized by its flanking genes or ORFs and the functional domain identified in the
translated protein sequence. Most of IGs with a functional domain contain a gene in the re-
annotated H37Rv genome (Fu and Shinnick 2007).

4. Discussion
Computational algorithms for gene prediction are divided in two classes: One is based on
sequence similarity and the other based on gene structure and signal. The latter is known as
ab initio prediction. The first class of algorithm, represented by BLAST (Altschul, Gish et al.
1990), finds sequences (DNA, protein, or ESTs) in the database that match the given
sequence, whereas the second class of algorithm, such as Hidden Markov Model (Burge and
Karlin 1997; Lukashin and Borodovsky 1998; Besemer and Borodovsky 2005), builds a model
Functional Analysis of Intergenic Regions for Gene Discovery 353

of gene structure from empirical data. They both have limitations. For instance, the
sequence-based approach is not applicable if no homology is found, whereas the model-
based approach is not workable if no adequate training data is available for model
parameter estimation. To explore an alternative in a different perspective, the method
developed in our research combined sequence alignment, transcriptional evidence, and
homology. In particular, the transcriptional activity of a piece of DNA is direct evidence that
it is functioning. This is important because a gene means a piece of genomic DNA that is
functional. In the absence of functional evidence, any gene computed by whatever
algorithms will remain hypothetical.

Fig. 2. Examples of new genes with a predicted function found in the genome of M.
tuberculosis H37Rv (Fu and Shinnick 2007).
The whole M. tuberculosis H37Rv genome has been sequenced and annotated
comprehensively (Cole, Brosch et al. 1998). Transcriptional analysis of intergenic regions is a
means of exploring unknown genes. Our idea capitalized on transcription analysis in gene
finding, which was useful especially when applied to an annotated genome. Current
genome annotation technology allowed all genes to be identified by a computational
algorithm, and it was unlikely to add new genes through re-annotation at the same time
unless using a different algorithm. Thus, it was within expectation that the number of new
protein-coding sequences due to re-annotation was merely 2% of that in the original
submission of M. tuberculosis genome (Camus, Pryor et al. 2002). Through homology and
pattern-based search, most protein-coding sequences with a predicted function have been
reported. Yet, transcriptional evidence could quickly hint at potential protein-coding genes
in the intergenic regions. It is encouraging that we are still able to find new genes in this
study given the fact that the current knowledge concerning M. tuberculosis genes is derived
from intensive research in the field involving in vivo biological experiments, such as gene
deletion and complementation. Thus the integration of the sequence- and function-based
analyses would be a useful approach to not just gene characterization but also gene
prediction. As the experiment was based on the standard in vitro growth condition of M.
354 Computational Biology and Applied Bioinformatics

tuberculosis, silent genes under this condition were not under examination in this study, but
the same idea should be applicable to other genomes under other conditions, and contribute
to the improvements of current gene databases.
The methods presented here did not address the issue of genes that did not code proteins.
There are a number of regulatory, non-coding RNAs assuming a distinct role from mRNA,
rRNA and tRNA. Many such RNAs have been identified and characterized both in
prokaryotes and eukaryotes and their main functions are posttranscriptional regulation of
gene expression and RNA-directed DNA methylation (Erdmann, Barciszewska et al. 2001;
Pickford and Cogoni 2003). A non-coding RNA has neither a long open reading frame nor a
gene structure. The DNA sequence that encodes a non-coding RNA is called a gene if its
regulatory function can be defined. Thus it is possible that an isolated expression element
lacking a gene structure is a non-coding, regulatory RNA. However, it was confirmed that
the potential protein-coding genes found in this study did not match any RNA family
published in the RNA-families database (www.sanger.ac.uk/ Software/Rfam/).

5. Conclusion
High-throughput gene finding on a newly sequenced genome is enabled through advanced
computational genome annotation software. However, genome annotation does not
guarantee all genes to be identified since knowledge and concepts about what constitutes a
gene are evolving and yet to be perfected. Genome re-annotation using the same kind of
computational heuristics offers limited help, unless supported with new in vivo
experimental evidence, but such evidence often slowly arrives. We developed a method that
integrated sequence-based and transcriptional information for gene finding in the intergenic
regions of an annotated genome. In the experiment with the M. tuberculosis H37Rv genome,
the method discovered genes with a specific function, such as a DNA-binding protein in the
CopG family and a nickel binding GTPase, as well as hypothetical proteins that have not
been reported in the M. tuberculosis H37Rv genome. This work has demonstrated that
microarray-based transcriptional analysis could play an important role in gene finding on
the genomic scale.

6. Acknowledgments
We would like to thank Dr. Thomas Shinnick at CDC for collaboration and the use of the
facilities, and thank UCI for providing service for microarray hybridization. We thank
Thomas R. Gingeras at Affymetrix, Inc. for designing Mycobacterium tuberculosis GeneChip.
Bacterial culture and RNA isolation were performed by Pramod Aryal.

7. References
Altschul, S. F., W. Gish, W. Miller, E. W. Myers and D. J. Lipman (1990). "Basic local
alignment search tool." J Mol Biol 215(3): 403-10.
Besemer, J. and M. Borodovsky (2005). "GeneMark: web software for gene finding in
prokaryotes, eukaryotes and viruses." Nucleic Acids Res 33(Web Server issue):
W451-4.
Functional Analysis of Intergenic Regions for Gene Discovery 355

Burge, C. and S. Karlin (1997). "Prediction of complete gene structures in human genomic
DNA." J Mol Biol 268(1): 78-94.
Camus, J. C., M. J. Pryor, C. Medigue and S. T. Cole (2002). "Re-annotation of the genome
sequence of Mycobacterium tuberculosis H37Rv." Microbiology 148(Pt 10): 2967-73.
Cole, S. T., R. Brosch, J. Parkhill, T. Garnier, C. Churcher, D. Harris, et al. (1998).
"Deciphering the biology of Mycobacterium tuberculosis from the complete
genome sequence." Nature 393(6685): 537-44.
Erdmann, V. A., M. Z. Barciszewska, A. Hochberg, N. de Groot and J. Barciszewski (2001).
"Regulatory RNAs." Cell Mol Life Sci 58(7): 960-77.
Finn, R. D., J. Mistry, B. Schuster-Bockler, S. Griffiths-Jones, V. Hollich, T. Lassmann, et al.
(2006). "Pfam: clans, web tools and services." Nucleic Acids Res 34(Database issue):
D247-51.
Fisher, M. A., B. B. Plikaytis and T. M. Shinnick (2002). "Microarray analysis of the
Mycobacterium tuberculosis transcriptional response to the acidic conditions found
in phagosomes." J Bacteriol 184(14): 4025-32.
Fu, L. M. (2006). "Exploring drug action on Mycobacterium tuberculosis using affymetrix
oligonucleotide genechips." Tuberculosis (Edinb) 86(2): 134-43.
Fu, L. M. and C. S. Fu-Liu (2007). "The gene expression data of Mycobacterium tuberculosis
based on Affymetrix gene chips provide insight into regulatory and hypothetical
genes." BMC Microbiol 7: 37.
Fu, L. M. and T. M. Shinnick (2007). "Genome-Wide Analysis of Intergenic Regions of
Mycobacterium tuberculosis H37Rv Using Affymetrix GeneChips." EURASIP J
Bioinform Syst Biol: 23054.
Fu, L. M. and T. M. Shinnick (2007). "Understanding the action of INH on a highly INH-
resistant Mycobacterium tuberculosis strain using Genechips." Tuberculosis
(Edinb) 87(1): 63-70.
Lee, J. M., S. Zhang, S. Saha, S. Santa Anna, C. Jiang and J. Perkins (2001). "RNA expression
analysis using an antisense Bacillus subtilis genome array." J Bacteriol 183(24):
7371-80.
Li, J., M. Pankratz and J. A. Johnson (2002). "Differential gene expression patterns revealed
by oligonucleotide versus long cDNA arrays." Toxicol Sci 69(2): 383-90.
Lukashin, A. V. and M. Borodovsky (1998). "GeneMark.hmm: new solutions for gene
finding." Nucleic Acids Res 26(4): 1107-15.
Nielsen, P. and A. Krogh (2005). "Large-scale prokaryotic gene prediction and comparison to
genome annotation." Bioinformatics 21(24): 4322-9.
Overbeek, R., T. Begley, R. M. Butler, J. V. Choudhuri, H. Y. Chuang, M. Cohoon, et al.
(2005). "The subsystems approach to genome annotation and its use in the project
to annotate 1000 genomes." Nucleic Acids Res 33(17): 5691-702.
Pickford, A. S. and C. Cogoni (2003). "RNA-mediated gene silencing." Cell Mol Life Sci 60(5):
871-82.
Stothard, P. and D. S. Wishart (2006). "Automated bacterial genome analysis and
annotation." Curr Opin Microbiol 9(5): 505-10.
356 Computational Biology and Applied Bioinformatics

Van Domselaar, G. H., P. Stothard, S. Shrivastava, J. A. Cruz, A. Guo, X. Dong, et al. (2005).
"BASys: a web server for automated bacterial genome annotation." Nucleic Acids
Res 33(Web Server issue): W455-9.
Zheng, D., Z. Zhang, P. M. Harrison, J. Karro, N. Carriero and M. Gerstein (2005).
"Integrated pseudogene annotation for human chromosome 22: evidence for
transcription." J Mol Biol 349(1): 27-45.
 19

Prediction of Transcriptional Regulatory

Networks for Retinal Development
Ying Li1, Haiyan Huang2 and Li Cai1
1Department of Biomedical Engineering, Rutgers University, NJ
2Department of Statistics, University of California, Berkeley, CA
USA

1. Introduction
The formation of the neatly layered retina during embryonic development is dictated by a
series of complicated transcription factor interactions. Retina-specific expression of these
transcription factors is an essential step in establishing retinal progenitor cells (RPCs) from
embryonic stem cells. The transcriptional control of gene expression is largely mediated by
the combinatorial interactions between cis-regulatory DNA elements and trans-acting
transcription factors, which cooperate/interact with each other to form a transcription
regulatory network during this developmental process. Such regulatory networks are
essential in regulating tissue/cell-specific gene expression, e.g., in cell fate determination
and differentiation during embryonic retinal development (Hu et al., 2010; Kumar, 2009;
Swaroop et al., 2010). Many genes, which involved in transcriptional networks for the
specification of a certain retinal cell lineage, have already been identified and characterized
(Corbo et al., 2010; Kim et al., 2008b; Tang et al., 2010). The transcriptional regulatory
networks for specific retinal cell lineages, e.g., the photoreceptors (Corbo et al., 2010; Hsiau
et al., 2007) and bipolar neurons (Kim et al., 2008b), were established in recent years.
However, the transcriptional regulatory network that governs the entire neural retinal
development is still elusive.
Identifying tissue/cell-specific cis-regulatory elements, trans-acting factor binding sites
(TFBSs), and their binding transcription factors (TFs) represent key steps towards
understanding tissue/cell-specific gene expression and further successful reconstruction of
transcriptional regulatory networks. These steps also present major challenges in both fields
of experimental biology and computational biology.
Currently, the prevailing method of studying TFBSs and transcriptional regulatory
networks is to determine the function of tissue-specific trans-acting factors based on data
from genome-wide gene expression profiling and chromatin immunoprecipitation (ChIP).
ChIP is often used to investigate protein-DNA interactions in a cell. Coupled with massive
parallel sequencing, ChIP-seq is capable of mapping the genome-wide protein-DNA
interaction at a finer resolution (Valouev et al., 2008) to identify candidate enhancer
sequences (Visel et al., 2009). Thus, a regulatory cascade can be recognized via consequential
analysis of the factors involved.
Here, we present a new method for the computational analysis of TFBSs and transcriptional
regulatory networks utilizing genome-wide sequencing, expression, and enhancer data. In
358 Computational Biology and Applied Bioinformatics

contrast to the traditional method, which mainly focuses on factor expression analysis, we
emphasize the sequence elements with tissue-specific enhancer activity. Our hypothesis is
that enhancers, non-coding sequences that direct gene expression in a cell-/tissue-specific
manner, contain common TFBSs that allow the key protein factors (important for the
development of that cell/tissue type) to bind. Experimentally verified tissue-specific
enhancer elements selected from enhancer databases were carefully screened for common
trans-acting factor binding sites to predict potential sequence-factor interacting networks.
DNA-binding protein factors that associate with multiple enhancers can be analyzed using
experimental methods.
As proof-of-principle, simple transcriptional regulatory networks of embryonic retinal
development were assembled based on common/key factors and their interacting genes, as
determined by literature search. These resulting networks provide a general view of
embryonic retinal development and a new hypothesis for further experimentation.

2. Methods and results

In this study, we aimed to develop a method for the identification of regulatory networks
for cell/tissue-specific gene expression. To test our hypothesis of the existence of common
TFBSs on the enhancers of cell/tissue specific genes, we employ the mouse developing
retina as a model system. Enhancers that direct retina-specific gene expression were selected
and their sequences were thoroughly screened for common TFBSs and trans-acting factors
(protein factors) using open-access software such as TESS, JASPA, MatInspector, and
MEME, etc. For common TFBS selection, we developed a Matlab program. These common
protein factors were further studied for their expression profiles. Thereafter, we were able to
construct transcriptional regulatory networks based on the known function of the enhancer-
binding trans-acting factors using a network construction tool, BioTapestry (Longabaugh et
al., 2009).

2.1 Retina-specific enhancers

To determine the transcriptional regulatory networks that govern retinal development, we
identified and selected enhancer elements that direct gene expression in the retina by
searching the VISTA Enhancer Browser. The VISTA Enhancer Browser is a central resource
for experimentally validated human and mouse non-coding DNA fragments with enhancer
activity as assessed in transgenic mice. Most of these non-coding elements were selected for
testing based on their extreme conservation in other vertebrates or epigenomic evidence
(ChIP-Seq) of putative enhancer marks. The results of this in vivo enhancer screen are
provided through this publicly available website (Frazer et al., 2004). Of the 1503 non-
coding DNA elements from human and mouse genomes that were tested for their ability to
direct gene expression on embryonic day 11.5 (E11.5) using a reporter assay in transgenic
mouse, a total of 47 elements were shown to possess enhancer activity in the retina (as of
February 10, 2011). Of the 47 elements, 25 show enhancer activities in both the retina and
other tissues, e.g., heart, limb, brain and spinal cord, etc. These 25 elements were separated
for further analysis as we focused on retina-specific sequence elements. In addition, among
the 22 remaining elements, 17 elements were active enhancers in the retina in at least half of
the tested transgenic embryos (Supporting data 1). To further increase the possibility of
identifying the key enhancer elements that are critical for retinal development, we applied
more stringent selection criteria based on the following two properties: 1) the reporter
Prediction of Transcriptional Regulatory Networks for Retinal Development 359

expression is restricted in the retina and not in any other regions of the CNS; 2) there is
expression of at least one of the flanking genes in the retina at E11.5 (or Theiler Stage 18-21).
The second category was applied because a majority of known enhancers were found in the
immediate up- or downstream region of their target genes, and thus the regulatory activities
of enhancers were considered to be most likely associated with their flanking genes. Any
enhancer elements that did not fit into at least one of the two categories was thus eliminated
(details in Supporting data 5). Based on the above criteria, 8 enhancer elements were
identified (Table 1).

Reporter expression Flanking genes of enhancers and

Enhancer Length pattern derived Annotation of their endogenous expression in
Group
ID (bp) from enhancer reporter expression mouse embryos
activity Upstream Downstream
Irx5: E11-19 Irx6: E11-19
hs27 1113 eye(4/6), limb(4/6) Retina + non-CNS
retina retina
Fxr1: E12-14, E19
hs258 1487 eye(3/5), limb, Retina + non-CNS Ccdc39: E14.5
retina
1
Nr2f1: E11-19 in
hs546 1753 eye(7/7), limb, nose Retina + non-CNS Arrdc3: E14-19
retina
Nr2f1: E11-19 in
hs1170 1288 eye(8/8) Retina Arrdc3: E14-19
retina
eye(6/9), limb, nose, AA408296:
hs932 775 Retina + non-CNS Irf6: not in retina
branchial arch unknown
2
Slc2a1: not in
mm165 926 eye(4/5), heart Retina + non-CNS Lao1: unknown
retina
Ascl1: E11-19 in
hs1122 1218 eye(6/7) Retina+ Spinal cord Pah: not in retina
retina (MGI)
3
eye(5/5), heart, Zfand5 (intragenic): E11-13/E13-19
mm269 760 Retina+ Spinal cord
other in retina

Table 1. A list of eight retina-specific enhancers. The 8 retina-specific enhancers selected

from the VISTA Enhancer Browser are listed above. They are grouped into 3 sub-groups
according to the reporter expression in mouse embryos derived from these enhancers and
the expression pattern of their flanking genes. The enhancer IDs, their expression pattern
and the flanking gene names were retrieved directly from VISTA Enhancer Browser, with an
expression pattern described as ‘tissue type (positive sample number/total sample
number)’. Flanking genes with expression in the retina are shown in bold.

2.2 Trans-acting factor binding sites on retina-specific enhancers

The binding of trans-acting factors (e.g., transcription factors) to non-coding regulatory
DNA (e.g., promoters, enhancers, etc.) is an essential process in the control of gene
expression. This protein-DNA interaction helps recruit the DNA polymerase complex and
co-activators to form the transcription machinery. The binding of these protein factors can
also act as repressors to prevent transcription. Identification of a TFBS in the enhancer and
promoter for a gene may indicate the possibility that the corresponding factors play a role in
the regulation of that gene. Importantly, the ability of an enhancer to direct cell/tissue-
specific gene expression is achieved via the binding of tissue-specific trans-acting protein
360 Computational Biology and Applied Bioinformatics

factors. To determine what protein factors regulate retina-specific gene expression, it is

important to determine what TFBSs are located on the retina-specific enhancers. We thus
searched DNA sequences of the 8 retina-specific enhancer elements for their known TFBSs
using TESS (Schug, 2002), JASPA (Portales-Casamar et al., 2010; Sandelin et al., 2004) and
MatInspector (Cartharius et al., 2005). For example, TESS (Transcription Element Search
System - http://www.cbil.upenn.edu/cgi-bin/tess) is a web tool for predicting TFBSs in
DNA sequences. It can identify TFBSs using site or consensus strings and positional weight
matrices mainly from the TRANSFAC (Knuppel et al., 1994). TRANSFAC contains data on
transcription factors, their experimentally-proven binding sites, and regulated genes. Its
broad compilation of binding sites allows the derivation of positional weight matrices
(Knuppel et al., 1994). The following search parameters were set when searching TESS: a
minimum string length of 6, a maximum allowable string mismatch of 10%, a minimum log-
likelihood ratio score of 12, and organism selection of Mus musculus (the house mouse). Our
search results show that there are approximately 150 TFBSs for each of the 8 enhancer
sequences (Supporting data 2). Similar results were reported by JASPA and MatInspector.
The corresponding protein factors of these TFBSs were considered to be capable of binding
with the 8 retina-specific enhancers, and thus they are important in activating/suppressing
gene expression in the retina.

2.3 A motif containing Pou3f2 binding sites

Since all 8 enhancers possess the ability to direct retina-specific gene expression, there may
be key TFBSs shared amongst these retina-specific sequence elements. To test this
hypothesis, we sorted and screened the TFBSs of each of the 8 enhancers to identify common
ones using a Matlab program that we developed for this study (Supporting data 6). This
Matlab program for common TFBS selection was designed to compare the TFBSs on each of
the retina-specific enhancer elements predicted by TESS. TFBSs for two or three different
enhancers can be sequentially compared. A “model” character was used as the comparison
category instead of the binding site name in both TESS and our Matlab program. As defined
in TESS, a model is “the site string or weight matrix used to pick this site” (Schug, 2002), and
thus describes the nature of a binding site. One factor may have multiple models, and one
model may be shared by multiple factors. The model character is the only necessary
parameter to characterize the transcription factors depending on their binding site property.
With this sorting/searching program, we identified a TFBS for Pou3f2 (also known as Brn2)
that was present in all 8 retina-specific enhancers (Fig. 1A). Previous studies have
demonstrated that the Pou3f2 transcription factor plays an important role in the
development of neural progenitor cells (Catena et al., 2004; Kim et al., 2008b; McEvilly et al.,
2002; Sugitani et al., 2002). Furthermore, the literature reports that this motif was first
discovered as a cis-element in the Chx10 enhancer, which can drive reporter expression in
intermediate and late RPCs (Rowan and Cepko, 2005). In this study, Pou3f2 was also shown
to affect bipolar interneuron fate determination through interactions with Chx10 and Nestin.
We thus speculate that this Pou3f2 binding site may exist in regulatory sequences among
genes important for the development of neural retinal progenitor cells (RPCs). Therefore,
the cis-elements of Chx10, Cyclin D1, Pax6, Rax, and Six3 were examined because these
genes are known for their role in regulating RPC development and retinal cell
differentiation (Conte et al., 2010; Martinez-de Luna et al., 2010; Oliver et al., 1995; Rowan
and Cepko, 2005; Sicinski et al., 1995). Confirming our prediction, Pou3f2 binding sites were
also present in the cis-elements of Chx10, Cylin D1, and Pax6 genes (sequences can be found
Prediction of Transcriptional Regulatory Networks for Retinal Development 361

in Supporting data 4). Next, sequence alignment analysis was performed to identify a
common motif for Pou3f2 binding sites using MEME (Multiple Em for Motif Elicitation,
http://meme.sdsc.edu/meme4_4_0/cgi-bin/meme.cgi) (Bailey and Elkan, 1994). It is
commonly believed that there are dependencies among different positions in a motif.
MEME may ignore such kind of dependency. Sequence alignment reveals a 22 bp motif
containing two Pou3f2 binding sites among all 8 enhancer elements (with a stringent E-
value of 7.2e-20 and p-value < 3.02e-6) and also in the cis-elements of RPC-specific genes,
e.g., Chx10, CyclinD1, and Pax6, from multiple vertebrate species (Fig. 1B). In addition, a
line of evidence indicates that Pou3f2 binds to the Rax enhancer to regulate the expression of
Rax in RPCs (Martinez-de Luna et al., 2010). Such prevalent existence of repeated Pou3f2
binding sites among retina-specific enhancer elements and cis-elements of RPC-specific
genes suggests that Pou3f2 is a key regulatory factor in the embryonic retinal development.

Fig. 1. A 22 bp motif is present in all 8 retina-specific enhancers and cis-elements of retinal

progenitor gene Chx10, CyclinD1 and Pax6. Sequence alignment was performed using
MEME and the output shows a motif exists among all above 15 sequences (p-value < 3e-6).
A. a SeqLog presentation of the 22bp motif. Dual binding sites of Pou3f2 were located next
to each other forms Pou3f2 binding site repeats. B. sequence alignment reveals the 22bp
motif among the 8 enhancers and cis-elements of RPC-specific genes (e.g., Chx10, CyclinD1
and Pax6) (Rowan and Cepko, 2005). Abbreviation: r, rat; m, mouse; h, human; q, quail.
RPC, retinal progenitor cell.
362 Computational Biology and Applied Bioinformatics

2.4 Key trans-acting factors involved in transcriptional regulatory networks of retinal

development
It is unlikely that only one factor (i.e., Pou3f2) is involved in regulating retina-specific gene
expression. Trans-acting factors often do not function alone but rather in a cooperative
manner. To identify other key protein factors for retina-specific gene expression, we applied
the following assumptions to the enhancer elements and their binding trans-acting factors:
1. Key TFBSs should be common to all or a subset of retina-specific enhancer elements;
2. The flanking genes of these enhancer elements should be expressed in the retina during
early retinal development, because an enhancer often regulates the expression of its
flanking gene(s);
3. The binding factors should have a known function in retinal development, or
4. If the binding factors have an unknown function in retinal development, they should be
at least expressed in the retina during retinal development. In this case, the expression
of the factor provides novel hypothesis for their function in retinal development, which
needs to be tested by functional studies.
Based on the above assumptions, the information on the expression of the flanking genes of
enhancer elements in the developing retina is necessary for the TFBS analysis. Five
databases of gene expression (see Table 2) were searched. The information on the expression
of these flanking genes were retrieved from these databases (Tables 1, 4 and supporting data
3). The factors that do not express in the retina during embryonic retinal development, e.g.,
around E11.5 (when enhancer elements were active), were set aside for further analysis of
retinal transcriptional regulatory networks. We then searched common TFBSs among
subsets of the 8 retina-specific enhancers. The TFBSs common to individual different sub-
groups were combined. In addition to Pou3f2, five other factors (i.e., Crx, Hes1, Meis1, Pbx2,
and Tcf3) were identified (Table 3). Four of the 6 factors have known functions in retinal
development, which is consistent with our hypothesis. The last two factors do not have
known functions in the retina. However, the prediction of their binding with groups of
enhancers suggests they play a role during retinogenesis. As the binding sites of these 6
factors were shared among a subset of retina-specific enhancer elements, these 6 binding
trans-acting factors were predicted as the key factors that participate in regulating retina-
specific gene expression during embryonic development.
Database Source Reference
Gene expression database (emage)
http://www.emouseatlas.org/e (Richardson et al.,
of Edinburg Mouse Atlas Project
mage/ 2010)
(EMAP, v5.0_3.3)
Gene Expression Database in
Mouse Genome Informatics (MGI, http://www.informatics.jax.org (Finger et al., 2011)
version 4.4)
(Diez-Roux et al.,
Eurexpress http://www.eurexpress.org
2011)
http://genome.ucsc.edu/cgi-
VisiGene Image Browser (Kent et al., 2002)
bin/hgVisiGene
Genome-scale mouse brain
transcription factor expression Supplementary data S4 and S6 (Gray et al., 2004)
analysis
Table 2. A list of gene expression databases used in this study.
Prediction of Transcriptional Regulatory Networks for Retinal Development 363

Presence in
Factor Binding site Known function Expression pattern enhancer
element
E10.5-14.5 in retina;
diencephalon, future
Pou3f2 ATTTGCAT Induce Bipolar cells All 8 elements
midbrain, future SC,
rhombencephalon
tgaggGGATCA hs27, hs258,
Crx Induce Photoreceptors E11-adult in retina
Acagact hs546, hs1170
Repress Amacrine, E11-13 in retina,
Horizontal, and thalamus, hypothalamus, hs27, hs258,
Hes1 CTTGTG
Ganglion cells; striatum, olfactory hs1170
Induce Photoreceptors epithelial
E10.5-14.5 in retina, lens hs27, hs258,
CTGTCActaaga retinal cell fate
Meis1 vesicle, diencephalon, hs546, hs1170,
tgaca determination
future sc, hindbrain hs932, mm165
cacctgagagTGA E10.5-14.5 in retina, hs27, hs258,
No function known in
Pbx-2 CAGaaggaaggc thalamus, midbrain, hs546, hs1170,
retina
agggag hindbrain, sc, ear hs932, mm165
ccaccagCACCT No function known in hs27, hs258,
Tcf3 E13.5 in retina (MGI)
Gtc retina hs546, hs1170
Table 3. A list of binding factors that show their temporal and spatial co-localization of
expression with each group of enhancers. For each factor, elements shown in the last column
indicate the enhancer elements which share a potential common binding with it. The
‘Expression pattern’ column shows available evidence of the co-localization with enhancers.
Corresponding databases are noted because different databases recorded different
expression pattern for the corresponding factor. The general function of each factor is also
included for future reference. Abbreviation: RPC, retinal progenitor cell; B, bipolar cell; A,
amacrine cell; H, horizontal cell; G, ganglion cell; PR, photoreceptor cell; sc, spinal cord.

Gene Function (related to retina development) and reference

With Mash3, regulate the neuron/glia fate determination (Hatakeyama et al.,
Ascl1
2001); with Mahs3 and Chx10, specify Biopolar cell identity (Satow et al., 2001).
Off circuit subsets of bipolar interneuron (Cheng et al., 2005; Cohen et al., 2000;
Irx5
Kerschensteiner et al., 2008).
No known clear function in retina. But It expresses in the in the area lining the
lumen of the otic vesicle including the region giving rise to ganglion complex of
Irx6
CN VII/VIII at E11.5 through E16.5 and overlaps with Mash1 (Cohen et al.,
2000; Mummenhoff et al., 2001)
Fxr1 Retina pigmentation(de Diego Otero et al., 2000); other function not known.
Amacrine development, may involve in cone differentiation; express in a
Nr2f1
unique gradient in retina along D/V axis (Inoue et al., 2010).
Zfand5 No known function in retina.
Table 4. A list of flanking genes with their function and references
364 Computational Biology and Applied Bioinformatics

Interestingly, three common TFBSs (i.e., Pou3f2, Crx, and Meis1) were present among
enhancer elements hs27, hs258, and hs1170 (see Tables 1, 3). Sequence alignment of the three
enhancer elements and cis-elements of RPC-specific genes (e.g., Chx10, Cyclin D1, and Pax6)
revealed another 22 bp motif (Fig. 2). Crx binding sites were present on enhancer elements
hs546 and hs1170, while Hes1 binding sites were present on enhancer element hs546. Since
these two enhancer elements (hs1170 and hs546) were both located in the non-coding region
between Nr2f1 and Arrdc3, and since Arrdc3 was not active at E11.5 in the retina, the
binding of Crx and Hes1 may participate in regulating the expression of Nr2f1. This result is
supported by the finding that both Crx (Peng et al., 2005) and Nr2f1 (Inoue et al., 2010; Satoh
et al., 2009) play a role in inducing photoreceptor cell fate, though at different stages. In
addition, Crx has been shown to be expressed in bipolar cells, paired with Otx2 and Pou3f2,
in binding with a 164bp Chx10 enhancer (Kim et al., 2008a). Hes1 has been shown to be
active during early eye formation. By suppressing Math5, Hes1 was shown to be involved in
the development of cone photoreceptors, amacrine, horizontal and ganglion cells from the
RPCs (Le et al., 2006; Lee et al., 2005).

Fig. 2. A 22 bp motif is present in a subset of enhancer elements and cis-elements of RPC-

specific genes, (e.g., Chx10, Cyclin D1, Pax6). Sequence alignment was performed using
MEME among enhance elements (hs258, hs546, and hs1170) and cis-elements of genes
Chx10, Cyclin D1, Pax6. A. a SeqLog presentation of the 22bp motif . This 22 bp motif
contains binding sites for 3 factors: Pou3f2, Meis1 and Crx. B. Sequence alignment between
elements mentioned above. Abbreviation: r, rat; m, mouse; h, human; q, quail.
Prediction of Transcriptional Regulatory Networks for Retinal Development 365

Meis1 together with Meis2, as members of the TALE-homeodomain protein Homothorax

(Hth) related protein family, were known to be expressed in the RPCs of mouse and chick
(Heine et al., 2008). Meis1 was expressed in RPCs throughout the entire neurogenesis
period, and Meis2 was expressed more specifically in RPCs before the initiation of retina
differentiation. Together, they function to maintain the RPCs in a rapid proliferating state
and control the expression of other ocular genes, e.g., Pax6, CyclinD1, Six3 and Chx10 (Bessa
et al., 2008; Heine et al., 2008). Since Meis1 binding sites are present in a subset of retina-
specific enhancers, Meis1 may function as an RPC-specific factor. Since the onset of mouse
retina neurogenesis is approximately at E10.5 when the ganglion cells first appear (Leo M.
Chalupa, 2008). By E11.5, RPCs of all six cell types are highly active. Therefore, binding of
Meis1 with enhancers might influence the cell fate of these RPCs.
The presence of common Pbx2 binding sites may indicate a novel functional role of Pbx2 in
RPCs, since the function of Pbx2 in retinal development has not been documented. Previous
studies have shown that Pbx2 is expressed in the zebrafish retina and tectum (French et al.,
2007) together with Pbx1 and Meis1, and down-regulation in their expression caused by the
deficiency of Prep, the prolyl endopeptidase will lead to eye anomalies (Deflorian et al.,
2004; Ferretti et al., 2006). Pbx and Meis proteins are major DNA-binding partners that form
abundant complexes (Chang et al., 1997). Thus, there is a possibility that Pbx2 may function
in the development of RPCs via the interaction with Meis1 and also regulate other RPC-
specific genes (e.g., Irx5, Nr2f1, etc) through enhancer binding (Table 1).
Tcf3 is not yet known to have a function in embryonic retinal development. However, since
Tcf3 binding sites are present among the retina-specific enhancer elements, and Tcf3 is
expressed in the retina during embryogenesis, its specific function in retinal development
needs to be confirmed.

2.5 Generation of transcriptional regulatory network for early retinal development

Based on the available expression data from VISTA Enhancer Browser and gene expression
databases (Table 2), it is known that these 8 enhancer elements and their common binding
trans-acting factors are active during embryonic development in the retina. Among the 8
retina-specific enhancer elements, we have identified 6 common trans-acting factors. These 6
predicted factors are experimentally verified key protein factors known to be involved in
regulating gene expression and cell differentiation of progenitor cells during embryonic
retinal development (Table 3).
Retina-specific gene expression is most likely determined by two kinds of interactions: (1)
the enhancers with their binding protein factors, and (2) the protein factors with their
interacting partners. The information about these interactions was used to generate the
transcriptional regulatory networks important for retinal development. Therefore,
transcriptional regulatory networks of embryonic retina were predicted based on these 6
common/key trans-acting factors (Table 3) and their known interacting partners (Table 5).
To construct retinal transcriptional regulatory networks, a java-based software program
named BioTapestry (Longabaugh et al., 2009) (http://www.biotapestry.org/, version 5.0.2)
was used to organize the factors and their known interacting partners. BioTapestry is a
network facilitating software program designed for dealing with systems that exhibit
increasingly complex over time, such as genetic regulatory networks. Its unique annotation
system allows the illustration of enhancer-regulated gene expression and connection
between factors. Experimental evidence can also be added to network elements after the
network was built, as a proof of particular interactions. We only used the presenting
366 Computational Biology and Applied Bioinformatics

function of BioTapestry here to show the networks of retinal development during early
neurogenesis. For better illustration, the network was mapped according to the 3-layer
structure of the mouse retina.
Based on the published information on the interacting factors of these 6 trans-acting factors
and their known functions in retinal cell development, we were able to build transcriptional
regulatory networks for all six major retinal cell types (Fig. 3). In RPCs, Meis1 is regulated
by Prep1 since it has been shown that insufficient Prep1 expression leads to a decrease in
Meis1 expression (Deflorian et al., 2004). Meis1 itself can regulate or interact with 4 other
factors (e.g., Pax6, CyclinD1, Six3 and Chx10) important to retinal cell differentiation (Bessa
et al., 2008; Heine et al., 2008). Crx, interacting with other factors (e.g, Otx2, Nrl and Nr2e3)
plays an important role in both cone and rod photoreceptor cell fate determination (Peng et
al., 2005). Crx can also influence Chx10 and Irx5 expression to affect the bipolar cell
formation (Cheng et al., 2005; Kerschensteiner et al., 2008; Kim et al., 2008a), which is similar
to the function of Pou3f2 (Kim et al., 2008a; Rowan and Cepko, 2005) and Otx2 (Kim et al.,
2008a). Another important factor is Hes1. Hes1 functions in maintaining RPC proliferation
(Wall et al., 2009) and regulating Math5, a critical factor for the generation of amacrine,
horizontal, and ganglion cells (Lee et al., 2005). In addition, Hes1 is regulated by Hes6 (Bae
et al., 2000). The Nr2f1 gene expresses in the retina in a gradient along the dorsal-ventral
axis and has been found to influence the development of amacrine cells. At the same time it
can affect cone and rod photoreceptors differentiation (Inoue et al., 2010). The references of
all genes/factors in the networks are listed in Table 5.

Fig. 3. Examples of transcriptional regulatory networks of embryonic retinal development.

Genes and factors are connected with arrows or bars to indicate the promoting and
suppressing relationship, respectively.
Prediction of Transcriptional Regulatory Networks for Retinal Development 367

Factor name Reference cited

Cabp5 (Kim et al., 2008a)
Chx10 (Hatakeyama et al., 2001)
CyclinD1 (Bessa et al., 2008; Heine et al., 2008)
Foxn4 (Shengguo Li, 2004)
Grk1 (Young and Young, 2007)
Grmb (Kim et al., 2008a)
Hes6 (Bae et al., 2000)
Mash3 (Hatakeyama et al., 2001; Satow et al., 2001)
Math5 (Lee et al., 2005)
Nestin (Rowan and Cepko, 2005)
NeuroD1 (Conte et al., 2010)
Nr2f1 (Inoue et al., 2010; Satoh et al., 2009)
Otx2 (Kim et al., 2008a; Young and Young, 2007)
Pax6 (Ferretti et al., 2006; Lee et al., 2005; Oliver et al., 1995)
Prep1 (Deflorian et al., 2004; Ferretti et al., 2006)
Rax (Heine et al., 2008; Martinez-de Luna et al., 2010)
Six3 (Oliver et al., 1995)
Six6 (Conte et al., 2010)

Table 5. A list of protein factors that interact with the 6 key trans-acting factors in a network
model
In summary, the computational method we developed in this study can be described as
following. First, experimentally verified enhancer elements can be selected from enhancer
databases, e.g., the Vista Enhancer Browser, based on tissue/cell-specific expression
patterns derived from the enhancer element and its flanking gene (Fig. 4A-B). These tissue-
specific enhancer elements can be located in the non-coding regions in inter- or intra-
genetic sequences. Then, the trans-acting factor binding sites (TFBS) of all tissue-specific
enhancer elements can be predicted using TFBS search tools such as TESS and MatInspector.
Common binding sites shared by all tissue-specific enhancer elements or subsets of elements
are later determined as shown in (Fig. 4C). Those common factors can be further analyzed
according to their expression pattern. Only the ones with spatio-temporal expression
patterns can be selected as key factors for constructing a tissue-specific transcriptional
regulatory network. In addition, factors that interact with these key factors can also be
found from the literature. Finally, these enhancers, genes, factors and interactions can be
pieced together to build the network (Fig. 4D).
368 Computational Biology and Applied Bioinformatics

Fig. 4. Computational analysis of TFBSs and transcriptional regulatory networks for

tissue/cell-specific gene expression. A. Selection of experimentally verified enhancer
elements from enhancer databases. B. Depiction of enhancer elements and their flanking
Prediction of Transcriptional Regulatory Networks for Retinal Development 369

genes. Orange boxes represent tissue-specific enhancer elements. Red bars represent
transcript start sites of downstream flanking genes. Black bars represent the end of the last
exon of upstream flanking genes. Enhancer elements in the intronic region of a gene were
not illustrated. C. Comparison of the trans-acting factor binding sites (TFBSs). Two examples
of common TFBS-TF pairs are illustrated by pairs of circle/azure arrows and
pentagon/cyan arrows. Other polygons on the enhancer elements (orange bars) represent
binding sites that are not common to all members. D. A simple example of a tissue-specific
transcriptional regulatory network generated with two key factors (TF1 and TF2) and their
interacting genes (G). The thick horizontal lines on top of factor and gene names (TF1, TF2
and G1) represent the cis-regulatory region of the corresponding genes. Arrows and bars
terminating on this line illustrate the input from other factors, promoters or repressors,
respectively. The arrow derived from the thick line represents the regulatory outputs.
Abbreviation: E, enhancer element; G, gene; Gd, downstream gene; Gu, upstream gene; TF,
trans-acting (e.g., transcription) factor; TFBS, transcription factor binding site.

3. Conclusion
In this study we have explored a new way of using existing TFBS-finding methods to
predict trans-acting factors and networks that regulate tissue-specific gene expression (Fig.
4). As a proof-of-principle study, using this method, we have identified experimentally
verified transcription factors (Pou3f2, Crx, Meis1, and Hes1) and their transcriptional
regulatory networks that are known to be important for mouse embryonic retinal
development. This not only provides a general idea about the development of the RPCs and
retinal cell differentiation, but also validates the effectiveness of our newly developed
method. Furthermore, we predicted that two other factors (Pbx2 and Tcf3, expressed in
retina) may have a novel functional role in embryonic retinal development, which generates
new hypotheses for future experimentation, i.e., to test the involvement of the two factors,
Pbx2 and Tcf3, in the development of RPCs. This method is based on the assumption that
non-coding DNA sequences contain signals regulating tissue-specific gene expression. The
feasibility of this method relies on the existing data of experimentally verified tissue-specific
enhancers (Visel et al., 2007) and available information of tissue-specific gene expression
(Finger et al., 2011; Richardson et al., 2010). The identification of retina-specific trans-acting
factors and networks indicates that our method is useful for the analysis of TFBSs and
transcriptional regulatory networks. Since our method utilizes the existing/known
information about tissue-specific enhancer elements and gene expression, we will not be
able to identify novel factors that are involved in the transcriptional regulatory networks. In
order to reconstruct a more sophisticated network in future studies, we will need to include
all of the retina-specific enhancer elements. Finally, this method can be applied to the
analysis of TFBSs and transcriptional regulatory networks in other tissue types.

4. Acknowledgement
This work is supported in part by grants (EY018738 and EY019094) from the National
Institute of Health, the New Jersey Commission on Spinal Cord Research (10A-003-SCR1
and 08-3074-SCR-E-0), and Busch Biomedical Research Awards (6-49121). The authors thank
the Cai lab members for helpful discussions and for proof-reading the manuscript.
370 Computational Biology and Applied Bioinformatics

5. References
Bae, S., Bessho, Y., Hojo, M., and Kageyama, R. (2000). The bHLH gene Hes6, an inhibitor of
Hes1, promotes neuronal differentiation. Development 127, 2933-2943.
Bailey, T.L., and Elkan, C. (1994). Fitting a mixture model by expectation maximization to
discover motifs in biopolymers. Proc Int Conf Intell Syst Mol Biol 2, 28-36.
Bessa, J., Tavares, M.J., Santos, J., Kikuta, H., Laplante, M., Becker, T.S., Gómez-Skarmeta,
J.L., and Casares, F. (2008). meis1 regulates cyclin D1 and c-myc expression, and
controls the proliferation of the multipotent cells in the early developing zebrafish
eye. Development 135, 799-803.
Cartharius, K., Frech, K., Grote, K., Klocke, B., Haltmeier, M., Klingenhoff, A., Frisch, M.,
Bayerlein, M., and Werner, T. (2005). MatInspector and beyond: promoter analysis
based on transcription factor binding sites. Bioinformatics 21, 2933-2942.
Catena, R., Tiveron, C., Ronchi, A., Porta, S., Ferri, A., Tatangelo, L., Cavallaro, M., Favaro,
R., Ottolenghi, S., Reinbold, R., et al. (2004). Conserved POU binding DNA sites in
the Sox2 upstream enhancer regulate gene expression in embryonic and neural
stem cells. J Biol Chem 279, 41846-41857.
Chang, C., Jacobs, Y., Nakamura, T., Jenkins, N., Copeland, N., and Cleary, M. (1997). Meis
proteins are major in vivo DNA binding partners for wild-type but not chimeric
Pbx proteins. Mol Cell Biol 17, 5679-5687.
Cheng, C.W., Chow, R.L., Lebel, M., Sakuma, R., Cheung, H.O.-L., Thanabalasingham, V.,
Zhang, X., Bruneau, B.G., Birch, D.G., Hui, C.-c., et al. (2005). The Iroquois
homeobox gene, Irx5, is required for retinal cone bipolar cell development.
Developmental Biology 287, 48-60.
Cohen, D.R., Cheng, C.W., Cheng, S.H., and Hui, C.-c. (2000). Expression of two novel
mouse Iroquois homeobox genes during neurogenesis. Mechanisms of Development
91, 317-321.
Conte, I., Marco-Ferreres, R., Beccari, L., Cisneros, E., Ruiz, J.M., Tabanera, N., and
Bovolenta, P. (2010). Proper differentiation of photoreceptors and amacrine cells
depends on a regulatory loop between NeuroD and Six6. Development 137, 2307-
2317.
Corbo, J.C., Lawrence, K.A., Karlstetter, M., Myers, C.A., Abdelaziz, M., Dirkes, W., Weigelt,
K., Seifert, M., Benes, V., Fritsche, L.G., et al. (2010). CRX ChIP-seq reveals the cis-
regulatory architecture of mouse photoreceptors. Genome Res 20, 1512-1525.
de Diego Otero, Y., Bakker, C., Raghoe, P., Severijnen, L.A., Hoogeveen, A., Oostra, B., and
Willemsen, R. (2000). Immunocytochemical characterization of FMRP, FXR1P and
FXR2P during embryonic development in the mouse. Gene Function & Disease 1, 28-
37.
Deflorian, G., Tiso, N., Ferretti, E., Meyer, D., Blasi, F., Bortolussi, M., and Argenton, F.
(2004). Prep1.1 has essential genetic functions in hindbrain development and
cranial neural crest cell differentiation. Development 131, 613-627.
Diez-Roux, G., Banfi, S., Sultan, M., Geffers, L., Anand, S., Rozado, D., Magen, A., Canidio,
E., Pagani, M., Peluso, I., et al. (2011). A high-resolution anatomical atlas of the
transcriptome in the mouse embryo. PLoS biology 9, e1000582.
Prediction of Transcriptional Regulatory Networks for Retinal Development 371

Ferretti, E., Villaescusa, J.C., Di Rosa, P., Fernandez-Diaz, L.C., Longobardi, E., Mazzieri,
R., Miccio, A., Micali, N., Selleri, L., Ferrari, G., et al. (2006). Hypomorphic
Mutation of the TALE Gene Prep1 (pKnox1) Causes a Major Reduction of Pbx
and Meis Proteins and a Pleiotropic Embryonic Phenotype. Mol Cell Biol 26, 5650-
5662.
Finger, J.H., Smith, C.M., Hayamizu, T.F., McCright, I.J., Eppig, J.T., Kadin, J.A., Richardson,
J.E., and Ringwald, M. (2011). The mouse Gene Expression Database (GXD): 2011
update. Nucleic Acids Research 39, D835-D841.
Frazer, K.A., Pachter, L., Poliakov, A., Rubin, E.M., and Dubchak, I. (2004). VISTA:
computational tools for comparative genomics. Nucleic Acids Research 32, W273-
W279.
French, C.R., Erickson, T., Callander, D., Berry, K.M., Koss, R., Hagey, D.W., Stout, J.,
Wuennenberg-Stapleton, K., Ngai, J., Moens, C.B., et al. (2007). Pbx homeodomain
proteins pattern both the zebrafish retina and tectum. BMC Dev Biol 7, 85.
Gray, P.A., Fu, H., Luo, P., Zhao, Q., Yu, J., Ferrari, A., Tenzen, T., Yuk, D.I., Tsung, E.F., Cai,
Z., et al. (2004). Mouse brain organization revealed through direct genome-scale TF
expression analysis. Science 306, 2255-2257.
Hatakeyama, J., Tomita, K., Inoue, T., and Kageyama, R. (2001). Roles of homeobox and
bHLH genes in specification of a retinal cell type. Development 128, 1313-1322.
Heine, P., Dohle, E., Bumsted-O'Brien, K., Engelkamp, D., and Schulte, D. (2008). Evidence
for an evolutionary conserved role of homothorax/Meis1/2 during vertebrate
retina development. Development 135, 805-811.
Hsiau, T.H., Diaconu, C., Myers, C.A., Lee, J., Cepko, C.L., and Corbo, J.C. (2007). The cis-
regulatory logic of the mammalian photoreceptor transcriptional network. PLoS
One 2, e643.
Hu, J., Wan, J., Hackler, L., Jr., Zack, D.J., and Qian, J. (2010). Computational analysis of
tissue-specific gene networks: application to murine retinal functional studies.
Bioinformatics 26, 2289-2297.
Inoue, M., Iida, A., Satoh, S., Kodama, T., and Watanabe, S. (2010). COUP-TFI and -TFII
nuclear receptors are expressed in amacrine cells and play roles in regulating
the differentiation of retinal progenitor cells. Experimental Eye Research 90, 49-
56.
Kent, W.J., Sugnet, C.W., Furey, T.S., Roskin, K.M., Pringle, T.H., Zahler, A.M.,
and Haussler, D. (2002). The human genome browser at UCSC. Genome Res 12, 996-
1006.
Kerschensteiner, D., Liu, H., Cheng, C.W., Demas, J., Cheng, S.H., Hui, C.-c., Chow, R.L.,
and Wong, R.O.L. (2008). Genetic Control of Circuit Function: Vsx1 and Irx5
Transcription Factors Regulate Contrast Adaptation in the Mouse Retina. J Neurosci
28, 2342-2352.
Kim, D.S., Matsuda, T., and Cepko, C.L. (2008a). A Core Paired-Type and POU
Homeodomain-Containing Transcription Factor Program Drives Retinal Bipolar
Cell Gene Expression. J Neurosci 28, 7748-7764.
372 Computational Biology and Applied Bioinformatics

Kim, D.S., Matsuda, T., and Cepko, C.L. (2008b). A core paired-type and POU
homeodomain-containing transcription factor program drives retinal bipolar cell
gene expression. J Neurosci 28, 7748-7764.
Knuppel, R., Dietze, P., Lehnberg, W., Frech, K., and Wingender, E. (1994). TRANSFAC
retrieval program: a network model database of eukaryotic transcription regulating
sequences and proteins. Journal of computational biology : a journal of computational
molecular cell biology 1, 191-198.
Kumar, J.P. (2009). The molecular circuitry governing retinal determination. Biochim Biophys
Acta 1789, 306-314.
Le, T.T., Wroblewski, E., Patel, S., Riesenberg, A.N., and Brown, N.L. (2006). Math5 is
required for both early retinal neuron differentiation and cell cycle progression.
Developmental Biology 295, 764-778.
Lee, H.Y., Wroblewski, E., Philips, G.T., Stair, C.N., Conley, K., Reedy, M., Mastick, G.S., and
Brown, N.L. (2005). Multiple requirements for Hes1 during early eye formation.
Developmental Biology 284, 464-478.
Leo M. Chalupa, R.W.W., ed. (2008). Eye, Retina, and Visual System of the Mouse (The MIT
Press).
Longabaugh, W.J.R., Davidson, E.H., and Bolouri, H. (2009). Visualization, documentation,
analysis, and communication of large-scale gene regulatory networks. Biochimica et
Biophysica Acta (BBA) - Gene Regulatory Mechanisms 1789, 363-374.
Martinez-de Luna, R.I., Moose, H.E., Kelly, L.E., Nekkalapudi, S., and El-Hodiri, H.M.
(2010). Regulation of retinal homeobox gene transcription by cooperative activity
among cis-elements. Gene 467, 13-24.
McEvilly, R.J., de Diaz, M.O., Schonemann, M.D., Hooshmand, F., and Rosenfeld, M.G.
(2002). Transcriptional regulation of cortical neuron migration by POU domain
factors. Science 295, 1528-1532.
Mummenhoff, J., Houweling, A.C., Peters, T., Christoffels, V.M., and Rüther, U. (2001).
Expression of Irx6 during mouse morphogenesis. Mechanisms of Development 103,
193-195.
Oliver, G., Mailhos, A., Wehr, R., Copeland, N.G., Jenkins, N.A., and Gruss, P. (1995). Six3, a
murine homologue of the sine oculis gene, demarcates the most anterior border of
the developing neural plate and is expressed during eye development. Development
121, 4045-4055.
Peng, G.-H., Ahmad, O., Ahmad, F., Liu, J., and Chen, S. (2005). The photoreceptor-
specific nuclear receptor Nr2e3 interacts with Crx and exerts opposing effects
on the transcription of rod versus cone genes. Human Molecular Genetics 14, 747-
764.
Portales-Casamar, E., Thongjuea, S., Kwon, A.T., Arenillas, D., Zhao, X., Valen, E., Yusuf, D.,
Lenhard, B., Wasserman, W.W., and Sandelin, A. (2010). JASPAR 2010: the greatly
expanded open-access database of transcription factor binding profiles. Nucleic
Acids Research 38, D105-110.
Richardson, L., Venkataraman, S., Stevenson, P., Yang, Y., Burton, N., Rao, J., Fisher, M.,
Baldock, R.A., Davidson, D.R., and Christiansen, J.H. (2010). EMAGE mouse
Prediction of Transcriptional Regulatory Networks for Retinal Development 373

embryo spatial gene expression database: 2010 update. Nucleic Acids Research 38,
D703-D709.
Rowan, S., and Cepko, C.L. (2005). A POU factor binding site upstream of the Chx10
homeobox gene is required for Chx10 expression in subsets of retinal progenitor
cells and bipolar cells. Developmental Biology 281, 240-255.
Sandelin, A., Alkema, W., Engstrom, P., Wasserman, W.W., and Lenhard, B. (2004). JASPAR:
an open-access database for eukaryotic transcription factor binding profiles. Nucleic
Acids Research 32, D91-94.
Satoh, S., Tang, K., Iida, A., Inoue, M., Kodama, T., Tsai, S.Y., Tsai, M.-J., Furuta, Y., and
Watanabe, S. (2009). The Spatial Patterning of Mouse Cone Opsin Expression Is
Regulated by Bone Morphogenetic Protein Signaling through Downstream Effector
COUP-TF Nuclear Receptors. J Neurosci 29, 12401-12411.
Satow, T., Bae, S.-K., Inoue, T., Inoue, C., Miyoshi, G., Tomita, K., Bessho, Y., Hashimoto, N.,
and Kageyama, R. (2001). The Basic Helix-Loop-Helix Gene hesr2 Promotes
Gliogenesis in Mouse Retina. J Neurosci 21, 1265-1273.
Schug, J. (2002). Using TESS to Predict Transcription Factor Binding Sites in DNA Sequence
(John Wiley & Sons, Inc.).
Shengguo Li, Z.M., Xuejie Yang, (2004). Foxn4 Controls the Genesis of Amacrine and
Horizontal Cells by Retinal Progenitors. Neuron 43, 795-807.
Sicinski, P., Donaher, J.L., Parker, S.B., Li, T., Fazeli, A., Gardner, H., Haslam,
S.Z., Bronson, R.T., Elledge, S.J., and Weinberg, R.A. (1995). Cyclin D1 provides a
link between development and oncogenesis in the retina and breast. Cell 82, 621-
630.
Sugitani, Y., Nakai, S., Minowa, O., Nishi, M., Jishage, K., Kawano, H., Mori, K., Ogawa, M.,
and Noda, T. (2002). Brn-1 and Brn-2 share crucial roles in the production and
positioning of mouse neocortical neurons. Genes Dev 16, 1760-1765.
Swaroop, A., Kim, D., and Forrest, D. (2010). Transcriptional regulation of photoreceptor
development and homeostasis in the mammalian retina. Nat Rev Neurosci 11, 563-
576.
Tang, K., Xie, X., Park, J.I., Jamrich, M., Tsai, S., and Tsai, M.J. (2010). COUP-TFs regulate eye
development by controlling factors essential for optic vesicle morphogenesis.
Development 137, 725-734.
Valouev, A., Johnson, D.S., Sundquist, A., Medina, C., Anton, E., Batzoglou, S., Myers, R.M.,
and Sidow, A. (2008). Genome-wide analysis of transcription factor binding sites
based on ChIP-Seq data. Nat Meth 5, 829-834.
Visel, A., Blow, M.J., Li, Z., Zhang, T., Akiyama, J.A., Holt, A., Plajzer-Frick, I., Shoukry, M.,
Wright, C., Chen, F., et al. (2009). ChIP-seq accurately predicts tissue-specific
activity of enhancers. Nature 457, 854-858.
Visel, A., Minovitsky, S., Dubchak, I., and Pennacchio, L.A. (2007). VISTA Enhancer
Browser—a database of tissue-specific human enhancers. Nucleic Acids Research 35,
D88-D92.
Wall, D.S., Mears, A.J., McNeill, B., Mazerolle, C., Thurig, S., Wang, Y., Kageyama, R., and
Wallace, V.A. (2009). Progenitor cell proliferation in the retina is dependent on
Notch-independent Sonic hedgehog/Hes1 activity. J Cell Biol 184, 101-112.
374 Computational Biology and Applied Bioinformatics

Young, J.E., Kasperek, E.M., Vogt, T.M., Lis, A., and Khani, S.C. (2007). Conserved
interactions of a compact highly active enhancer/promoter upstream of the
rhodopsin kinase (GRK1) gene. Genomics 90, 236-248.
 20

The Use of Functional Genomics in

Synthetic Promoter Design
Michael L. Roberts
Synpromics Ltd
United Kingdom

1. Introduction
The scope of this chapter is to examine how advances in the field of Bioinformatics can be
applied in the development of improved therapeutic strategies. In particular, we focus on
how algorithms designed to unravel complex gene regulatory networks can then be used in
the design of synthetic gene promoters that can be subsequently incorporated in novel gene
transfer vectors to promote safer and more efficient expression of therapeutic genes for the
treatment of various pathological conditions.

2. Development of synthetic promoters: a historical perspective

A synthetic promoter is a sequence of DNA that does not exist in nature and which has been
designed to control gene expression of a target gene. Cis-regulatory sequences derived from
naturally-occurring promoter elements are used to construct these synthetic promoters
using a building block approach; which can either be carried out by rational design or by
random ligation (illustrated in figure 1A). The result is a sequence of DNA composed of
several distinct cis-regulatory elements in a completely novel orientation that can act as a
promoter enhancer; typically to initiate RNA polymerase II-mediated transcription.
Construction of synthetic promoters is possible because of the modular nature of naturally-
occurring gene regulatory regions. This was cleverly demonstrated by a group that used
synthetic promoters to evaluate the role of the TATA box in the regulation of transcription
(Mogno et al., 2010). The authors looked at the role of the TATA box in dictating the
strength of gene expression. They found that the TATA box is a modular component in that
its strength of binding to the RNA polymerase II complex and the resultant strength of
transcription that it mediates is independent of the cis-regulatory element enhancers
upstream. Importantly, they also found that the TATA box does not add noise to
transcription, i.e. it acts as a simple amplifier without altering specificity of gene expression
dictated by the upstream enhancer elements. Thus implying that any combination of cis-
regulatory enhancers could be coupled to a TATA box and it would be the enhancers that
would mediate specificity without any interference from the TATA box. The implications
from this study suggest that it should be possible to construct any type of synthetic
promoter that is specifically engineered to display a highly restrictive pattern of gene
regulation.
376 Computational Biology and Applied Bioinformatics

Synthetic promoters have been used in the study of gene regulation for more than two
decades. In one of the first examples a synthetic promoter derived from the lipoprotein gene
in E. Coli was used to efficiently drive the expression of a number of tRNA genes (Masson et
al., 1986). In the years that followed a technique was developed that enabled the mutation of
prokaryotic sequences flanking the essential -10 and -35 promoter elements (illustrated in
figure 1B) and thus the efficient construction of synthetic promoters for use in bacteria
(Jacquet et al., 1986). This approach was successfully used to produce promoters with much
higher activity compared to naturally occurring sequences and it was immediately realised
that such an approach would have important applications in the biotech industry,
particularly in the enhanced production of biopharmaceuticals (Trumble et al., 1992).

A. Typical Mammalian Synthetic Promoter Layout

Antibiotic
cis cis cis cis cis Minimal Promoter Reporter Gene Resistance
polyA

Random or ordered assembly of cis- Minimal promoter Gene that allows the Allows selection Essential Post-
regulatory regions obtained from sequence containing identification of most of promoter transcriptional
endogenous promoters the TATA box efficient promoters sequences process signal

SYNTHETIC PROMOTER ELEMENT SELECTION ELEMENT

B. Typical Prokaryotic Synthetic Promoter

-35 -10 +1
NNNNNNNNNNN-TTGACA-NNNNNNNNNNNNNNNNNNN-TATAAT-NNNNNNNNNN-ATG

Variable Region: Target for Variable Region: Target for Variable Region: Target for
random mutations random mutations random mutations

Fig. 1. Typical synthetic promoter layouts for prokaryotes and eukaryotes

Most of these studies were initially undertaken with a view to establish the important
structural features of prokaryotic or eukaryotic promoters so that essential elements could
be identified. In one example, the role of the Tat protein in the regulation of HIV gene
expression was studied using synthetic promoters (Kamine et al., 1991). In this study a series
of minimal promoters containing Sp1- binding sites and a TATA box were constructed and
analysed to see if the Tat protein from HIV could activate them. The results demonstrated
that Tat could only activate the synthetic promoters containing Sp1 sites and not promoters
with the TATA box alone. The observations enabled the authors to propose that in vivo the
Tat protein is brought to the promoter site by TAR RNA and then interacts with Sp1 to drive
gene expression. In recent years more sophisticated studies using synthetic promoters have
been undertaken to evaluate the important factors driving transcription factor binding to
their corresponding cis-regulatory elements (Gertz et al., 2009a) and to thermodynamically
model trans-factor and cis-element interactions (Gertz et al., 2009b).
As alluded to above, it was soon realised that synthetic promoter technology had direct
implications in the improvement of the efficiency of gene expression. Indeed, one of the
most widely used eukaryotic promoters employed for research purposes today is actually a
The Use of Functional Genomics in Synthetic Promoter Design 377

synthetic promoter. The steroid-inducible Glucocorticoid Receptor Element (GRE) is a

naturally occurring sequence that regulates the expression of a plethora of genes that are
responsive to glucocorticoids. In a relatively early study several of these elements were
linked together in order to construct a promoter with enhanced responsiveness to these
steroids (Mader et al., 1993). This study detailed the construction of a 5 x GRE synthetic
promoter linked to the Adenovirus type 2 major late promoter TATA region that displayed
50-fold more expression levels in response to steroid hormones when compared to the
natural promoter sequence. This synthetic promoter is now a widely used constituent of a
number of reporter constructs adopted in a variety of different research applications.
Finally, synthetic promoters have also been used in prokaryotic systems to reveal that
regulation of gene expression follows boolean logic (Kinkhabwala et al., 2008). In this
prototypical study the authors found that two transcription repressors generate a NOR
logic; i.e. a OR b (on OR off), while one repressor plus one activator determines an ANDN
logic; i.e. a AND NOT b (on AND NOT off). This idea was later expanded on to demonstrate
that various combinations of synthetic promoters could combine to generate 12 out of 16
boolean logic terms (Hunziker et al., 2010). Most interestingly the results from these studies
demonstrated that if a promoter does not follow a specific logic it is more likely to be leaky,
in that it will drive gene expression under conditions where it is not expected to.
In this chapter we describe the evolution of synthetic promoter technology, its application in
the development of improved tissue-specific promoters and its potential use for the
development of effective disease-specific gene regulators; thus enabling the development of
safer and more effective gene therapies.

3. Recent advances in the design of the synthetic promoter

In recent years some efforts have been made to construct synthetic promoters for tissue
specific transcription based on the linking of short oligonucleotide promoter and enhancer
elements in a random (Li et al., 1999; Edelman et al., 2000) or ordered (Chow et al., 1997;
Ramon et al., 2010) fashion.
In what can be described as one of the first attempts to rationally design a tissue-specific
synthetic promoter, Chow et al. describe the rearrangement of the cytokeratin K18 locus to
construct a promoter mediating a highly restrictive pattern of gene expression in the lung
epithelium (Chow et al., 1997). In this study the authors describe the generation of
transgenic mice with this construct and demonstrate expression only in the lung. They also
generated CMV (Cytomegalovirus) and SV40 (Sarcoma Virus 40) promoter based constructs
and found lack of specificity and no expression in the lung epithelia. This study had
important implications for researchers developing lung-based gene therapies, i.e. if CMV,
one of the most widely used promoters, could not regulate gene expression in the lung
epithelia then it is necessary to identify (or develop) new promoters that can efficiently
regulate gene expression in this location. Indeed, it is now becoming increasingly apparent
that traditional virus-derived promoters like CMV and RSV (Rous Sarcoma Virus) will have
limited application in the development of modern gene therapeutics.
The random assembly of cis-regulatory elements has shown particular success as a means to
develop synthetic promoters. In one such approach, which aimed to identify synthetic
promoters for muscle-specific expression, duplex oligonucleotides from the binding sites of
muscle-specific and non-specific transcription factors were randomly ligated and cloned
upstream of a minimal muscle promoter driving luciferase (Li et al. 1999). Approximately
378 Computational Biology and Applied Bioinformatics

1000 plasmid clones were individually tested by transient transfection into muscle cells and
luciferase activity was determined in 96-well format by luminometry. By this approach
several highly active and muscle specific promoters were identified that displayed
comparable strength to the most commonly used viral promoters such as CMV.

Fig. 2. Typical procedure for generation of synthetic mammalian promoters (reproduced

from PNAS, Vol. 97, No. 7, pp. 3038-3043 copyright (c) 2000 by the National Academy of
Sciences, USA)
Retroviral vectors have also been used to screen for synthetic promoters in eukaryotic cells
(Edelman et al., 2000). This study was the first description of a retroviral library approach
using antibiotic resistance and FACS selection to isolate promoter sequences (illustrated in
figure 2). The libraries generated using random oligonucleotides in an effort to identify new
sequences as well as examining the effects of combinations of known elements and for
uncovering new transcriptional regulatory elements. After preparing a Ran18 promoter
library comprises random 18mer oligonucleotides, the authors analysed the sequences of the
generated synthetic promoters by searching for known transcription factor binding motifs.
They found that the highest promoter activities were associated with an increased number
of known motifs. They examined eight of the best known motifs; AP2, CEBp, gre, ebox, ets,
creb, ap1 AND sP1/maz. Interestingly, several of the promoter sequences contained none of
these motifs and the author's looked for new transcription factors.
In a similar effort employed to examine one million clones, Sutton and co-workers adopted
the FACS screening approach based on the establishment of a lentiviral vector-based library
(Dai et al., 2004). In this study duplex oligonucleotides from binding sites of endothelial cell-
specific and non-specific transcription factors were cloned in a random manner upstream of
a minimal promoter driving expression of eGFP in a HIV self-inactivating expression vector.
A pool of one million clones was then transfected into endothelial cells and the highest
The Use of Functional Genomics in Synthetic Promoter Design 379

expressers were selected by FACS sorting. Synthetic promoters were then rescued from
stable transfectants by PCR from the genomic DNA where the HIV vectors had integrated.
The results from this study also demonstrated the possibility of isolating several highly
active endothelial cell-specific synthetic promoter elements from a random screen.
Synthetic promoters active only in the liver have also been developed (Lemken et al., 2005).
In this study transcriptional units from ApoB and OTC genes were used in a controlled,
non-random construction procedure to generate a series of multimeric synthetic promoters.
Specifically, 2x, 4x, 8x and 16x repeats of the ApoB and OTC promoter elements were
ligated together and promoter activity analysed. The results indicated that the promoter
based on 4xApoB elements gave the optimal levels of gene expression and that 8x and 16x
elements gave reduced levels of expression, thus demonstrating the limitations of simply
ligating known promoter elements together in a repeat fashion to achieve enhanced
expression.
When adopting this type of methodology in the design of synthetic tissue-specific promoters
it is important to use well-designed duplex oligonucleotides. For example, each element has
to be spaced in such a way that the regulatory elements appear on the same side of the DNA
helix when reassembled, relevant minimal promoter elements have to be employed so that
the screen produces promoters capable of expressing efficiently only in the tissue of interest
and there must be some sort of mechanism, such as the addition of Sp1 sites, for the
protection against promoter silencing through methylation.
In addition to tissue-specific promoters, cell-type synthetic promoters have also been
developed. In one study, researchers designed a synthetic promoter to be active in
nonadrenergic (NA) neurones (Hwang et al., 2001). They authors randomly ligated cis-
regulatory elements that were identified from the human dopamine beta-hydroxylase
(hDBH) gene and constructed promoters with up to 50-fold higher activity than the original
promoter. Specifically, two elements from the promoter were used to generate a multimeric
synthetic promoter; PRS1 and PRS2 which are bound to by the Phox2a transcription factor.
The results demonstrated that the PRS2 was responsible for higher levels of gene expression
as it had higher affinity to Phox-2a. It was also found that eight copies of PRS2 in the same
orientation yielded maximum activity.
In a similar type of study a synthetic promoter was constructed that was specifically active
in myeloid cells (He et al., 2006). The promoter comprised myeloid-specific elements for
PU.1, C/EBPalpha, AML-1 and myeloid-associated elements for Sp1 and AP-1, which were
randomly inserted upstream of the p47-phox minimal promoter. Synthetic promoters
constructed showed very high activity. Haematopoietic Stem Cells (HSC) were initially
transduced then the expression in differentiated cells was examined; only myeloid cells
were found to express the reporter construct. To test therapeutic applicability of these
promoters apoE-/- mice were transplanted with HSC transduced with a lentiviral vector
expressing apoE from CMV and synthetic promoters. Even though transduced cells
containing CMV and synthetic promoters both corrected the artherosclerotic phenotype, the
cells derived from lentiviral vectors harbouring the synthetic promoter did so with less
variability. Thus highlighting the improved safety features when using synthetic promoters
for gene therapy applications.
In addition to tissue- and cell type-specific constitutive promoters, inducible synthetic
promoters can also be constructed. One group describe a synthetic promoter constructed by
placing the EPO enhancer region upstream of the SV40 promoter. The result is a strong
380 Computational Biology and Applied Bioinformatics

promoter that is active only under ischaemic conditions. The authors tested this promoter by
developing Neural Stem Cells (NSC) responsive to hypoxia and proposed that this system
could be used to deliver therapeutic stem cells to treat ischaemic events. The authors were
able to demonstrate that transplantation of NSC modified with a hypoxia-sensitive synthetic
promoter resulted in specific expression of the luciferase reporter gene in response to
ischaemic events in vivo (Liu et al., 2010).

4. Applications of synthetic promoter technology

Synthetic promoters have direct applications in large-scale industrial processes where
enzymatic pathways are used in the production of biological and chemical-based products
(reviewed in Hammer et al., 2006). One of the most important limitations in industrial-scale
processes that synthetic promoter technology addresses is the inherent genetic instability in
synthetically engineered biological systems. For instance, in prokaryotic organisms designed
to express two or more enzymes, mutations will invariably arise in very few generations
resulting in the termination of gene expression. This is because there is the lack of
evolutionary pressure keeping all the components intact. The result is that mutations accrue
over generations resulting in the deactivation of the circuit. Homologous recombination in
natural promoters driving high levels of gene expression is the main reason why this
circuitry fails (Sleight et al., 2010). Therefore, the use of synthetic promoters in these systems
should serve to lower gene expression to result in more genetic stability, allow the
avoidance of repeat sequences to prevent recombination and allow the use inducible
promoters (a feature that also reduces genetic instability). In summary, the use of synthetic
promoter technology in complex genetically engineered synthetic organisms expressing a
variety of components should serve to increase genetic stability and improve the efficiency
of the processes that the components control.
One interesting therapeutic application for synthetic promoter technology that has been
described is the generation of a class of replication-competent viruses that enable tumour
cell-specific killing by specifically replicating in cancer cells. In this study a replication
competent retrovirus was developed to selectively kill tumour cells (Logg et al., 2002). The
authors added a level of transcriptional targeting by incorporating the prostate-specific
probasin (PB) promoter into the retroviral LTR and designed more efficient synthetic
promoters based on the PB promoter to increase the efficiency of retroviral replication in
prostate cancer cells. The result was a retrovirus that could efficiency transduce and
replicate only in cancer cells. This is an attractive therapeutic strategy for the treatment of
cancer, as tumour virotherapy has actually been examined as a potential therapeutic
strategy for several decades.
Synthetic promoters that are active only in cycling endothelial cells would be another
attractive tool for the development of cancer gene therapies. The rationale being that by
targeting new blood vessels growing into tumours we would be able to develop a cancer
gene therapy that could cut off supply of nutrients to the growing cancer. In a study that
adopted this approach the cdc6 gene promoter was identified as a candidate promoter
active only in cycling cells and was coupled to the endothelin enhancer element to construct
a promoter active in dividing endothelial cells (Szymanski et al., 2006). Four endothelin
elements conjugated to the cdc6 promoter gave the optimal results in vitro. When introduced
into tumour models in vivo, the synthetic promoter was more efficient at driving gene
expression in cancerous tissues, when compared to a CMV promoter.
The Use of Functional Genomics in Synthetic Promoter Design 381

Perhaps one of the most impressive applications of synthetic promoter technology thus far
was the development of a liver-specific promoter that could be used to essentially cure
diabetes in a transgenic mouse model (Han et al., 2010). In this study a synthetic promoter
active in liver cells in response to insulin was constructed. The authors designed 3-, 6- & 9-
element promoters based on random combinations of HNF-1, E/EBP and GIRE cis-elements.
In the 3-element promoters all 27 combinations of the three were tested and the highest
activity promoters were used to generate the 6-element promoter and so on. Using this
technique promoters with activity up to 25% of CMV were identified. Finally, the optimal
promoter was chosen depending on its responsiveness to glucose. This promoter showed
highest specificity to liver cells and in response to Glucose and yielded expression levels
21% that of CMV. Adenoviral vectors containing this promoter driving expression of insulin
were injected into a mouse diabetic model. Injection with the highest dose of virus resulted
in protection against hyperglycaemia for 50 days. Importantly, injection with adenovirus
expressing insulin from a CMV promoter resulted in death of the animals due to
hypoglycaemia, thus illustrating the importance of regulated expression in gene therapy.
Importantly, the results from this study excellently illustrated why the clever design of
synthetic promoters controlling restricted gene expression will be essential in the
development of safe gene therapy.
Synthetic promoters are increasingly being used in gene therapy type of studies. In one
recent study their potential application to the gene therapy of Chronic Granulomatous
Disease (X-CGD; an X-linked disorder resulting from mutations in gp91-phox, whose
activity in myeloid cells is important in mounting an effective immune response) was
examined (Santilli et al., 2011). The authors cite a clinical trial using a retroviral vector,
which was successful at correcting the phenotype, but expression was short-lived due to
promoter inactivation. In order to address this issue a chimeric promoter was constructed
that was a fusion of Cathepsin G and c-Fes minimal promoter sequences, which are
specifically active in cells of the myeloid lineage. This promoter was used to drive the
expression of gp91-phox in myeloid cells in mice using a SIN lentiviral vector and the
results show effective restricted expression to monocytes and subsequent introduction of
gp91 results in high levels of expression in target cells and restoration of wild type
phenotype in vitro. X-CGD cells were then transduced with the lentiviral vector and grafted
into a mouse model of CGD. The vector was able to sustain long-term expression of gp91-
phox, resulting in levels of expression that could correct the phenotype. Expression was
specifically seen in granulocytes and monocytes, and not B- and T-cells.
These studies serve to highlight the potential application of synthetic promoter technology
in gene therapy. They particularly highlight the importance of achieving cell-type specific
gene expression and address the common issue of promoter shutdown that is seen when
using stronger viral promoters like those derived from the CMV and RSV. If gene therapy is
to be a success in the clinic it will be imperative to develop promoters that are highly
specific and which display a restrictive and predictable expression profile. Thus, synthetic
promoter technology represents the ideal solution to achieve this goal and its use is likely to
become an increasingly popular approach adopted by researchers developing gene
therapeutics.

5. Bioinformatic tools and synthetic promoter development

We first described how functional genomics experimentation and bioinformatics tools could
be applied in the design of synthetic promoters for therapeutic and diagnostic applications
382 Computational Biology and Applied Bioinformatics

several years ago (Roberts, 2007). Since then a number of scientists have also realised that
this approach can be broadly applied across the biotech industry (Venter et al., 2007). In this
section we discuss some of the tools that we use to analyse data obtained from large-scale
gene expression analyses, which is subsequently used in the smart design of synthetic
promoters conveying highly-specific regulation of gene expression.
To design a synthetic promoter it is essential to identify an appropriate number of cis-
regulatory elements that can specifically bind to the trans factors that enhance gene
transcription. This is where the importance of a number of bioinformatic algorithms
becomes apparent. Over the past several years a number of databases and programs have
been developed in order to identify transcription factor biding sites (TFBSs) on a variety of
genomes. Below we introduce the most extensively used resources and discuss their
application to the design of synthetic promoters, we pay particular attention to the
identification of transcription networks active in cancer and how this information can be
used to design cancer-specific promoters that can be used in the design of safer and more
effective tumour-targeted gene therapies.
There is now a growing trend for researchers to analyse microarray data in terms of ‘gene
modules’ instead of the presentation of differentially regulated gene lists. By grouping genes
into functionally related modules it is possible to identify subtle changes in gene expression
that may be biologically (if not statistically significantly) important, to more easily interpret
molecular pathways that mediate a particular response and to compare many different
microarray experiments from different disease states in an effort to uncover the
commonalities and differences in multiple clinical conditions. Therefore, we are moving into
a new era of functional genomics, where the large datasets generated by the evaluation of
global gene expression studies can be more fully interpreted by improvements in
computational methods. The advances in functional genomics made in recent years have
resulted in the identification of many more cis-regulatory elements that can be directly
related to the increased transcription of specific genes. Indeed, the ability to use
bioinformatics to unravel complex transcriptional pathways active in diseased cells can
actually serve to facilitate the process of choosing suitable cis-elements that can be used to
design synthetic promoters specifically active in complex pathologies such as cancer.
In cancer the changes in the gene expression profile are often the result of alterations in the
cell’s transcription machinery induced by aberrant activation of signalling pathways that
control growth, proliferation and migration. Such changes result in the activation of
transcription regulatory networks that are not found in normal cells and provide us with an
opportunity to design synthetic promoters that should only be active in cancerous cells. If
microarray technology is to truly result in the design of tailored therapies to individual
cancers or even patients, as has been heralded, it is important that the functional genomics
methodology that was designed for the identification of signalling and transcription
networks be applied to the design of cancer-specific promoters so that effective gene
therapeutic strategies can be formulated (Roberts & Kottaridis, 2007). The development of
bioinformatics algorithms for the analysis of microarray datasets has largely been applied in
order to unravel the transcription networks operative under different disease and
environmental conditions. To this date there has been no effort to use this type of approach
to design synthetic promoters that are operative only under these certain disease or
environmental conditions.
The Use of Functional Genomics in Synthetic Promoter Design 383

The regulation of gene expression in eukaryotes is highly complex and often occurs through
the coordinated action of multiple transcription factors. The use of trans-factor combinations
in the control of gene expression allows a cell to employ a relatively small number of
transcription factors in the regulation of disparate biological processes. As discussed herein,
a number of tools have been developed that allow us to utilise microarray data to identify
novel cis-regulatory elements. It is also possible to use this information to decipher the
transcriptional networks that are active in cells under different environmental conditions. In
yeast, the importance of the combinatorial nature of transcriptional regulation was
established by specifically examining clusters of upregulated genes for the presence of
combinations of cis-elements. By examining microarray data from yeast exposed to a variety
of conditions the authors were able to construct a network of transcription revealing the
functional associations between different regulatory elements. This approach resulted in the
identification of key motifs with many interactions, suggesting that some factors serve as
facilitator proteins assisting their gene-specific partners in their function. The idea that a
core number of transcription factors mediate such a vast array of biological responses by
adopting multiple configurations implies that it may be possible to hijack the transcriptional
programs that have gone awry in multifactorial diseases in an effort to develop disease-
specific regulatory elements. For instance, the meta-analyses of cancer datasets has
permitted the identification of gene modules, allowing for the reduction of complex cancer
signatures to small numbers of activated transcription programs and even to the
identification of common programs that are active in most types of cancer. This type of
analysis can also help to identify specific transcription factors whose deregulation plays a
key role in tumour development. In one such study, the importance of aberrant E2F activity
in cancer was reaffirmed during a search for the regulatory programs linking transcription
factors to the target genes found upregulated in specific cancer types (Rhodes et al., 2005). It
was shown that E2F target genes were disproportionately upregulated in more than half of
the gene expression profiles examined, which were obtained from a multitude of different
cancer types. It was thus proposed that integrative bioinformatics analyses have the
potential to generate new hypotheses about cancer progression.
Different bioinformatics tools, examples of which are given in table 1, may be used to screen
for cis-regulatory elements. In general, such tools function by comparing gene expression
profiles between differentially regulated genes and examining upstream sequences,
available through genome sequence resources. For the phylogenetic footprinting tools, the
untranslated regions of specific genes are compared between species and the most highly
conserved sequences are returned and proposed to be potential cis-elements. A combination
of all available approaches may be employed in order to identify regulatory sequences that
predominate in the profile of specific cell or tissue types. The most common sequences
identified are then used as the building blocks employed in the design of synthetic
promoters.
The ability to use gene expression data to identify gene modules, which mediate specific
responses to environmental stimuli (or to a diseased state) and to correlate their regulation
to the cis-regulatory elements present upstream of the genes in each module, has
transformed the way in which we interpret microarray data. For instance, by using the
modular approach it is possible to examine whether particular gene modules are active in a
variety of different cancers, or whether individual cancers require the function of unique
gene modules. This has allowed us to look for transcriptional commonalities between
384 Computational Biology and Applied Bioinformatics

different cancers, which should aid in the design of widely applicable anti-cancer
therapeutic strategies. In one early study, gene expression data from 1975 microarrays,
spanning 22 different cancers was used to identify gene modules that were activated or
deactivated in specific types of cancer (Segal et al., 2004). Using this approach the authors
found that a bone osteoblastic module was active in a number of cancers whose primary
metastatic site is known to be the bone. Thus, a common mechanism of bone metastasis
between varieties of different cancers was identified, which could be targeted in the
development of novel anticancer therapies.
It is also possible to identify the higher-level regulator that controls the expression of the
genes in each module (Segal et al., 2003). Examination of the upstream regulatory sequences
of each gene in a module may reveal the presence of common cis-regulatory elements that
are known to be the target of the module’s regulator. Therefore, by identifying specific
regulatory proteins that control the activation of gene modules in different cancers, it should
be possible to extrapolate the important cis-elements that mediate transcription in the
transformed cell. Thereby, allowing us to design and construct novel tumour-specific
promoters based on the most active cis-regulatory elements in a number of tumour-specific
gene modules. The ability to identify specific transcriptional elements in the human genome
that control the expression of functionally related genes is transforming the application of
functional genomics. Until recently the interpretation of data from microarray analysis has
been limited to the identification of genes whose function may be important in a single
pathway or response. How this related to global changes in the cellular phenotype had been
largely ignored, as the necessary tools to examine this simply did not exist. With the
advancement of bioinformatics we are now in a position to utilise all the data that is
obtained from large-scale gene expression analysis and combine it with knowledge of the
completed sequence of the human genome and with transcription factor, gene ontology and
molecular function databases, thereby more fully utilising the large datasets that are
generated by global gene expression studies.
For nearly two decades scientists have been compiling databases that catalogue the trans-
factors and cis-elements that are responsible for gene regulation (Wingender et al., 1988).
This has primarily been done in an effort to elucidate the various transcription programs
that are activated in response to different biological stimuli in a range of organisms. The
result is the emergence of useful tools that can be used to identify transcription factors and
their corresponding cis-regulatory sequences that are useful in the design of synthetic
promoters. In the remaining part of this chapter we briefly discuss each resource, indicating
the unique aspect of its functionality.
TRANSFAC is perhaps the most comprehensive TFBS database available and indexes
transcription factors and their target sequences based solely on experimental data (Matys et
al., 2003). It is maintained as a relational database, from which public releases are made
available via the web. The release consists of six flat files. At the core of the database is the
interaction of transcription factors (FACTOR) with their DNA-binding sites (SITE) through
which they regulate their target genes (GENE). Apart from genomic sites, ‘artificial’ sites
which are synthesized in the laboratory without any known connection to a gene, e.g.,
random oligonucleotides, and IUPAC consensus sequences are also stored in the SITE table.
Sites must be experimentally proven for their inclusion in the database. Experimental
evidence for the interaction with a factor is given in the SITE entry in form of the method
that was used (gel shift, footprinting analysis, etc.) and the cell from which the factor was
The Use of Functional Genomics in Synthetic Promoter Design 385

derived (factor source). The latter contains a link to the respective entry in the CELL table.
On the basis of those, method and cell, a quality value is given to describe the ‘confidence’
with which an observed DNA- binding activity could be assigned to a specific factor. From a
collection of binding sites for a factor nucleotide weight matrices are derived (MATRIX).
These matrices are used by the tool MatchTM to find potential binding sites in
uncharacterized sequences, while the program PatchTM uses the single site sequences, which
are stored in the SITE table. According to their DNA-binding domain transcription factors
are assigned to a certain class (CLASS). In addition to the more ‘planar’ CLASS table a
hierarchical factor classification system is also used.
TRANSCompel® originates from COMPEL, and functions to emphasize the key role of
specific interactions between transcription factors binding to their target cis-regulatory
elements; whilst providing specific features of gene regulation in a particular cellular
content (Kel-Margoulis et al., 2002). Information about the structure of known trans factor
and cis sequence interactions, and specific gene regulation achieved through these
interactions, is extremely useful for promoter prediction. In the TRANSCompel database,
each entry corresponds to an individual trans/cis interaction within the context of a
particular gene and thus contains information about two binding sites, two corresponding
transcription factors and experiments confirming cooperative action between transcription
factors.
ABS is a public database of known cis-regulatory binding sites identified in promoters of
orthologous vertebrate genes that have been manually collated from the scientific literature
(Blanco et al., 2006). In this database some 650 experimental binding sites from 68
transcription factors and 100 orthologous target genes in human, mouse, rat or chicken
genome sequences have been documented. This tool allows computational predictions and
promoter alignment information for each entry and is accessed through a simple and easy-
to-use web interface; facilitating data retrieval and allowing different views of the
information. One of the key features of this software is the inclusion of a customizable
generator of artificial datasets based on the known sites contained in the whole collection
and an evaluation tool to aid during the training and the assessment of various motif-
finding programs.
JASPAR is an open-access database of annotated, high-quality, matrix-based TFBS profiles
for multi-cellular eukaryotic organisms (Sandelin et al., 2004). The profiles were derived
exclusively from sets of nucleotide sequences that were experimentally demonstrated to
bind transcription factors. The database is accessible via a web-interface for browsing,
searching and subset selection. The interface also includes an online sequence analysis
utility and a suite of tools for genome-wide and comparative genome analysis of regulatory
regions.
HTPSELEX is a public database providing access to primary and derived data from high-
throughput SELEX experiments that were specifically designed in order to characterize the
binding specificity of transcription factors (Jagannathan et al., 2006). The resource is
primarily intended to serve computational biologists interested in building models of TFBSs
from large sets of cis-regulatory sequences. For each experiment detailed in the database
accurate information is provided about the protein material used, details of the wet lab
protocol, an archive of sequencing trace files, assembled clone sequences and complete sets
of in vitro selected protein-binding tags.
386 Computational Biology and Applied Bioinformatics

TRED is a database that stores both cis- and trans-regulatory elements and was designed to
facilitate easy data access and to allow for the analysis of single-gene-based and genome-
scale studies (Zhao et al., 2005). Distinguishing features of TRED include: relatively
complete genome-wide promoter annotation for human, mouse and rat; availability of gene
transcriptional regulation information including TFBSs and experimental evidence; data
accuracy is ensured by hand curation; efficient user interface for easy and flexible data
retrieval; and implementation of on-the-fly sequence analysis tools. TRED can provide good
training datasets for further genome-wide cis-regulatory element prediction and annotation;
assist detailed functional studies and facilitate the deciphering of gene regulatory networks.
Databases of known TFBSs can be used to detect the presence of protein-recognition
elements in a given promoter, but only when the binding site of the relevant DNA-binding
protein and its tolerance to mismatches in vivo is already known. Because this knowledge is
currently limited to a small subset of transcription factors, much effort has been devoted to
the discovery of regulatory motifs by comparative analysis of the DNA sequences of
promoters. By finding conserved regions between multiple promoters, motifs can be
identified with no prior knowledge of TFBSs. A number of models have emerged that
achieve this by statistical overrepresentation. These algorithms function by aligning
multiple untranslated regions from the entire genome and identifying sequences that are
statistically significantly overrepresented in comparison to what it expected by random.
YMF is a program developed to identify novel TFBSs (not necessarily associated with a
specific factor) in yeast by searching for statistically overrepresented motifs (Sinha et al.,
2003; Sinha & Tompa, 2002). More specifically, YMF enumerates all motifs in the search
space and is guaranteed to produce those motifs with the greatest z-scores.
SCORE is a computational method for identifying transcriptional cis-regulatory modules
based on the observation that they often contain, in statistically improbable concentrations,
multiple binding sites for the same transcription factor (Rebeiz et al., 2002). Using this
method the authors conducted a genome-wide inventory of predicted binding sites for the
Notch-regulated transcription factor Suppressor of Hairless, Su(H), in drosophila and found
that the fly genome contains highly non-random clusters of Su(H) sites over a broad range
of sequence intervals. They found that the most statistically significant clusters were very
heavily enriched in both known and logical targets of Su(H) binding and regulation. The
utility of the SCORE approach was validated by in vivo experiments showing that proper
expression of the novel gene Him in adult muscle precursor cells depends both on Su(H)
gene activity and sequences that include a previously unstudied cluster of four Su(H) sites,
indicating that Him is a likely direct target of Su(H).
At present these tools are mainly applied in the study of lower eukaryotes where the
genome is less complex and regulatory elements are easier to identify, extending these
algorithms to the human genome has proven somewhat more difficult. In order to redress
this issue a number of groups have shown that it is possible to mine the genome of higher
eukaryotes by searching for conserved regulatory elements adjacent to transcription start
site motifs such as TATA and CAAT boxes, e.g. as catalogued in the DBTSS resource
(Suzuki et al. 2004; Suzuki et al., 2002), or one can search for putative cis-elements in CpG
rich regions that are present in higher proportions in promoter sequences (Davuluri et al.,
2001). Alternatively, with the co-emergence of microarray technology and the complete
sequence of the human genome, it is now possible to search for potential TFBSs by
comparing the upstream non-coding regions of multiple genes that show similar expression
The Use of Functional Genomics in Synthetic Promoter Design 387

profiles under certain conditions. Gene sets for comparative analysis can be chosen based on
clustering, e.g. hierarchical and k-means (Roth et al., 1998), from simple expression ratio
(Bussemaker et al., 2001) or functional analysis of gene products (Jensen et al., 2000). This
provides scientists with the opportunity to identify promoter elements that are responsive to
certain environmental conditions, or those that play a key role in mediating the
differentiation of certain tissues or those that may be particularly active in mediating
pathologic phenotypes.
Phylogenetic footprinting, or comparative genomics, is now being applied to identify novel
promoter elements by comparing the evolutionary conserved untranslated elements
proximal to known genes from a variety of organisms. The availability of genome sequences
between species has notably advanced comparative genomics and the understanding of
evolutionary biology in general. The neutral theory of molecular evolution provides a
framework for the identification of DNA sequences in genomes of different species. Its
central hypothesis is that the vast majority of mutations in the genome are neutral with
respect to the fitness of an organism. Whilst deleterious mutations are rapidly removed by
selection, neutral mutations persist and follow a stochastic process of genetic drift through a
population. Therefore, non-neutral DNA sequences (functional DNA sequences) must be
conserved during evolution, whereas neutral mutations accumulate. Initial studies
sufficiently demonstrated that the human genome could be adequately compared to the
genomes of other organisms allowing for the efficient identification of homologous regions
in functional DNA sequences.
Subsequently, a number of bioinformatics tools have emerged that operate by comparing
non-coding regulatory sequences between the genomes of various organisms to enable the
identification of conserved TFBSs that are significantly enriched in promoters of candidate
genes or from clusters identified by microarray analysis; examples of these software suites
are discussed below. Typically these tools work by aligning the upstream sequences of
target genes between species thus identifying conserved regions that could potentially
function as cis-regulatory elements and have consequently been applied in the elucidation of
transcription regulatory networks in a variety of models.
TRAFAC is a Web-based application for analysis and display of a pair of DNA sequences
with an emphasis on the detection of conserved TFBSs (Jegga et al., 2002). A number of
programs are used to analyze the sequences and identify various genomic features (for
example, exons, repeats, conserved regions, TFBSs). Repeat elements are masked out using
RepeatMasker and the sequences are aligned using the PipMaker-BLASTZ algorithm.
MatInspector Professional or Match (BioBase) is run to scan the sequences for TFBSs. TRAFAC
then integrates analysis results from these applications and generates graphical outputs;
termed the Regulogram and Trafacgram.
CORG comprises a catalogue of conserved non-coding sequence blocks that were initially
computed based on statistically significant local suboptimal alignments of 15kb regions
upstream of the translation start sites of some 10793 pairs of orthologous genes (Dieterich et
al., 2003). The resulting conserved non-coding blocks were annotated with EST matches for
easier detection of non-coding mRNA and with hits to known TFBSs. CORG data are
accessible from the ENSEMBL web site via a DAS service as well as a specially developed
web service for query and interactive visualization of the conserved blocks and their
annotation.
388 Computational Biology and Applied Bioinformatics

CONSITE is a flexible suite of methods for the identification and visualization of conserved
TFBSs (Lenhard et al., 2003). The system reports those putative TFBSs that are both situated
in conserved regions and located as pairs of sites in equivalent positions in alignments
between two orthologous sequences. An underlying collection of metazoan transcription-
factor-binding profiles was assembled to facilitate the study. This approach results in a
significant improvement in the detection of TFBSs because of an increased signal-to-noise
ratio, as as the authors demonstrated with two sets of promoter sequences.
CONFAC enables the high-throughput identification of conserved TFBSs in the regulatory
regions of hundreds of genes at a time (Karanam et al., (2004). The CONFAC software
compares non-coding regulatory sequences between human and mouse genomes to enable
identification of conserved TFBSs that are significantly enriched in promoters of gene
clusters from microarray analyses compared to sets of unchanging control genes using a
Mann–Whitney statistical test. The authors analysed random gene sets and demonstrated
that using this approach, over 98% of TFBSs had false positive rates below 5%. As a proof-
of-principle, the CONFAC software was validated using gene sets from four separate
microarray studies and TFBSs were identified that are known to be functionally important
for regulation of each of the four gene sets.
VAMP is a graphical user interface for both visualization and primary level analysis of
molecular profiles obtained from functional genomics experimentation (La Rosa et al., 2006).
It can be applied to datasets generated from Comparative Genomic Hybridisation (CGH)
arrays, transcriptome arrays, Single Nucleotide Polymorphism arrays, loss of heterozygosity
analysis (LOH), and Chromatin Immunoprecipitation experiments (ChIP-on-chip). The
interface allows users to collate the results from these different types of studies and to view
it in a convenient way. Several views are available, such as the classical CGH karyotype
view or genome-wide multi-tumour comparisons. Many of the functionalities required for
the analysis of CGH data are provided by the interface; including searches for recurrent
regions of alterations, comparison to transcriptome data, correlation to clinical information,
and the ability to view gene clusters in the context of genome structure.
CisMols Analyser allows for the filtering of candidate cis-element clusters based on
phylogenetic conservation across multiple gene sets (Jegga et al., 2005). It was previously
possible to achieve this for individual orthologue gene pairs, but combining data from cis-
conservation and coordinate expression across multiple genes proved a more difficult task.
To address this issue, the authors extended an orthologue gene pair database with
additional analytical architecture to allow for the analysis and identification of maximal
numbers of compositionally similar and phylogenetically conserved cis-regulatory element
clusters from a list of user-selected genes. The system has been successfully tested with a
series of functionally related and microarray profile-based co-expressed orthologue pairs of
promoters and genes using known regulatory regions as training sets and co-expressed
genes in the olfactory and immunohematologic systems as test sets. A significant amount of
effort has been dedicated to the cataloguing of transcription factors and their corresponding
cis-elements. More recently, these databases have been compiled with the aim to utilise them
to unravel regulatory networks active in response to diverse stimuli.
PreMod was developed in an effort to identify cis-regulatory modules (CRM) active under
specific environmental conditions (Blanchette et al., 2006; Ferretti et al., 2007). Starting from
a set of predicted binding sites for more than 200 transcription factor families documented
in the Transfac database (described above), the authors describe an algorithm relying on the
The Use of Functional Genomics in Synthetic Promoter Design 389

principle that cis-regulatory modules (CRMs) generally contain several phylogenetically

conserved binding sites for a small variety of transcription factors. The method allowed the
prediction of more than 118,000 CRMs within the human genome. During this analysis, it
was revealed that CRM density varies widely across the genome, with CRM-rich regions
often being located near genes encoding transcription factors involved in development.
Interestingly, in addition to showing enrichment near the 3’ end of genes, predicted CRMs
were present in other regions more distant from genes. In this database, the tendency for
certain transcription factors to bind modules located in specific regions was documented
with respect to their target genes, and a number of transcription factors likely to be involved
in tissue-specific regulation were identified.
CisView was developed to facilitate the analysis of gene regulatory regions of the mouse
genome (Sharov et al., 2006). Its user interface is a browser and database of genome-wide
potential TFBSs that were identified using 134 position-weight matrices and 219 sequence
patterns from various sources. The output is presented with information about sequence
conservation, neighbouring genes and their structures, GO annotations, protein domains,
DNA repeats and CpG islands. The authors used this tool to analyse the distribution of
TFBSs and revealed that many TFBSs were over-represented near transcription start sites. In
the initial paper presenting the tool they also identified potential cis-regulatory modules
defined as clusters of conserved TFBSs in the entire mouse genome. Out of 739,074 CRMs
identified, 157,442 had a significantly higher regulatory potential score than semi-random
sequences. The CisView browser provides a user-friendly computer environment for
studying transcription regulation on a whole-genome scale and can also be used for
interpreting microarray experiments and identifying putative targets of transcription
factors.
BEARR is web browser software designed to assist biologists in efficiently carrying out the
analysis of microarray data from studies of specific transcription factors (Vega et al., 2004).
Batch Extraction and Analysis of cis-Regulatory Regions, or BEARR, accepts gene identifier
lists from microarray data analysis tools and facilitates identification, extraction and analysis
of regulatory regions from the large amount of data that is typically generated in these types
of studies.
VISTA is a family of computational tools that was built to assist in the comparative analysis
of DNA sequences (Dubchak & Ryaboy, 2006). These tools allow for the alignment of DNA
sequences to facilitate the visualization of conservation levels and thus allow for the
identification of highly conserved regions between species. Specifically, sequences can be
analysed by browsing through pre-computed whole-genome alignments of vertebrates and
other groups of organisms. Submission of sequences to Genome VISTA enables the user to
align them to other whole genomes; whereas submission of two or more sequences to
mVISTA allows for direct alignment. Submission of sequences to Regulatory VISTA is also
possible and enables the predication of potential TFBSs (based on conservation within
sequence alignments). All VISTA tools use standard algorithms for visualization and
conservation analysis to make comparison of results from different programs more
straightforward.
PromAn is a modular web-based tool dedicated to promoter analysis that integrates a
number of different complementary databases, methods and programs (Lardenois et al.,
2006). PromAn provides automatic analysis of a genomic region with minimal prior
390 Computational Biology and Applied Bioinformatics

knowledge of the genomic sequence. Prediction programs and experimental databases are
combined to locate the transcription start site (TSS) and the promoter region within a large
genomic input sequence. TFBSs can be predicted using several public databases and user-
defined motifs. Also, a phylogenetic footprinting strategy, combining multiple alignments of
large genomic sequences and assignment of various scores reflecting the evolutionary
selection pressure, allows for evaluation and ranking of TFBS predictions. PromAn results
can be displayed in an interactive graphical user interface. It integrates all of this
information to highlight active promoter regions, to identify among the huge number of
TFBS predictions those which are the most likely to be potentially functional and to facilitate
user refined analysis. Such an integrative approach is essential in the face of a growing
number of tools dedicated to promoter analysis in order to propose hypotheses to direct
further experimental validations.
CRSD is a comprehensive web server that can be applied in investigating complex
regulatory behaviours involving gene expression signatures, microRNA regulatory
signatures and transcription factor regulatory signatures (Liu et al., 2006). Six well-known
and large-scale databases, including the human UniGene, mature microRNAs, putative
promoter, TRANSFAC, pathway and Gene Ontology (GO) databases, were integrated to
provide the comprehensive analysis in CRSD. Two new genome-wide databases, of
microRNA and transcription factor regulatory signatures were also constructed and further
integrated into CRSD. To accomplish the microarray data analysis at one go, several
methods, including microarray data pre-treatment, statistical and clustering analysis,
iterative enrichment analysis and motif discovery, were closely integrated in the web server.
MPromDb is a database that integrates gene promoters with experimentally supported
annotation of transcription start sites, cis-regulatory elements, CpG islands and chromatin
immunoprecipitation microarray (ChIP-chip) experimental results within an intuitively
designed interface (Sun et al., 2006). Its initial release contained information on 36,407
promoters and first exons, 3,739 TFBSs and 224 transcription factors; with links to PubMed
and GenBank references. Target promoters of transcription factors that have been identified
by ChIP-chip assay are also integrated into the database and thus serving as a portal for
genome-wide promoter analysis of data generated by ChIP-chip experimental studies.
A comprehensive list of the all the databases described above with a summary of their
features and a reference to the original citation are shown in table 1.
Each of the aforementioned databases can be used when searching for potential regulatory
sequences for inclusion in the design of synthetic promoters. Indeed, these resources can be
used in order to identify cis-regulatory elements that may play a role in the formation of a
particular cellular phenotype, or those that may be important in driving differentiation in
developing organs. Synpromics, an emerging synthetic biology company recently
incorporated in the United Kingdom, has cleverly utilised these tools in developing a
proprietary method of synthetic promoter production where identified elements are
incorporated into the design of promoters that are able to specifically regulate gene
expression in a particular cellular phenotype. This method harnesses a cell’s gene expression
profile in order to facilitate the design of highly specific and efficient promoters. The result
is a range of promoters that are inducible, tissue (or cell)-specific, active in response to a
particular pathogen, chemical or biological agent and even able to mediate gene expression
only under certain pathological conditions, such as cancer. Indeed, Synpromics has
successfully generated a range of synthetic promoters that specifically drive high levels of
The Use of Functional Genomics in Synthetic Promoter Design 391

gene expression in colorectal cancer and are looking to apply these promoters in the
development of safer gene therapies (manuscript in preparation).

Resource Description Citation

DBTSS Database of transcriptional start sites Suzuki et al., (2002)
TRAFAC Conserved cis-element search tool Jegga et al., (2002)
TRANSCom Database of composite regulatory elements Kel-Margoulis et al.,
pel (2002)
TRANSFAC Eukaryotic transcription factor database Matys et al., (2003)
Phylofoot Tools for phylogenetic footprinting purposes Lenhard et al., (2003)
CORG Multi-species DNA comparison and annotation Dieterich et al., (2003)
CONSITE Explores trans-factor binding sites from two species Lenhard et al., (2003)
CONFAC Conserved TFBS finder Karanam et al., (2004)
CisMols Identifies cis-regulatory modules from inputed data Jegga et al., (2005)
TRED Catalogue of transcription regulatory elements Zhao et al., (2005)
Oncomine Repository and analysis of cancer microarray data Rhodes et al., (2005)
ABS Database of regulatory elements Blanco et al., (2006)
JASPAR Database of regulatory elements Sandelin et al., (2004)
HTPSELEX Database of composite regulatory elements Jagannathan et al.,
(2006)
PReMod Database of transcriptional regulatory modules in the Blanchette et al.,
human genome (2006)
CisView Browser of regulatory motifs and regions in the Sharov et al., (2006)
genome
BEARR Batch extraction algorithm for microarray data analysis Vega et al., 2004)
VISTA Align and compare sequences from multiple species Dubchak et al., (2006)
PromAn Promoter analysis by integrating a variety of databases Lardenois et al.,
(2006)
Table 1. Bioinformatics tools used for the identification of cis-regulatory elements
Importantly, synthetic promoters often mediate a level of gene expression with much
greater efficiency than that seen with viral promoters, such as CMV, or compared to
naturally occurring promoters within the genome. Given that the entire Biotech industry is
centred on the regulation of gene expression, it is likely that synthetic promoters will
eventually replace all naturally-occurring sequences in use today and help drive the growth
of the synthetic biology sector in the coming decades.

6. Conclusion
In summary, synthetic promoters have emerged over the past two decades as excellent tools
facilitating the identification of important structural features in naturally occurring
392 Computational Biology and Applied Bioinformatics

promoter sequences and allowing enhanced and more restrictive regulation of gene
expression. A number of early studies revealed that it was possible to combine the cis-
regulatory elements from promoters of a few tissue-specific genes and use these as building
blocks to generate shorter, more efficient tissue-specific promoters. Several simple
methodologies to achieve this emerged and have been applied in a multitude of organisms;
including plant, bacteria, yeast, viral and mammalian systems.
Recent advances in bioinformatics and the emergence of a plethora of tools specifically
designed at unravelling transcription programs has also facilitated the design of highly-
specific synthetic promoters that can drive efficient gene expression in a tightly regulated
manner. Changes in a cell’s gene expression profile can be monitored and the transcription
programs underpinning that change delineated and the corresponding cis-regulatory
modules can be used to construct synthetic promoters whose activity is restricted to
individual cell types, or to single cells subject to particular environmental conditions. This
has allowed researchers to design promoters that are active in diseased cells or in tissues
treated with a particular biological or chemical agent; or active in cells infected with distinct
pathogens.
A number of institutions, such as Synpromics, have taken advantage of these advances and
are now working to apply synthetic promoter technology to the enhanced production of
biologics for use in biopharmaceutical, greentech and agricultural applications; the
development of new gene therapies; and in the design of a novel class of molecular
diagnostics. As the synthetic biology field continues to develop into a multi-billion dollar
industry, synthetic promoter technology is likely to remain at the heart of this ever-
expanding and exciting arena.

7. References
Blanchette, M., Bataille, A. R., Chen, X., Poitras, C., Laganière, J., Lefèbvre, C., et al. (2006).
Genome-wide computational prediction of transcriptional regulatory modules
reveals new insights into human gene expression. Genome Research, Vol. 16, No. 5,
May 2006, pp. 656-668. ISSN 1088-9051
Blanco, E., Farré, D., Albà, M. M., Messeguer, X., & Guigó, R. (2006). ABS: a database of
Annotated regulatory Binding Sites from orthologous promoters. Nucleic Acids
Research, Vol. 34, January 2006, pp. D63-67. ISSN 0305-1048
Bussemaker, H. J., Li, H., & Siggia, E D. (2001). Regulatory element detection using
correlation with expression. Nature Genetics, Vol. 27, No. 2, February 2001, pp. 167-
71. ISSN 1061-4036
Chow, Y. H., O’Brodovich, H., Plumb, J., Wen, Y., Sohn, K. J., Lu, Z., et al. (1997).
Development of an epithelium-specific expression cassette with human DNA
regulatory elements for transgene expression in lung airways. Proceedings of the
National Academy of Sciences of the United States of America, Vol. 94, No. 26,
December 1997, pp. 14695-14700. ISSN 1091-6490
Dai, C., McAninch, R. E., & Sutton, R. E. (2004). Identification of synthetic endothelial cell-
specific promoters by use of a high-throughput screen. Journal of Virology, Vol. 78,
No. 12, June 2004, pp. 6209-6221. ISSN 0022-538X
Davuluri, R V, Grosse, I., & Zhang, M Q. (2001). Computational identification of promoters
and first exons in the human genome. Nature Genetics, Vol. 29, No. 4, December
2001, pp. 412-417. ISSN 1061-4036
The Use of Functional Genomics in Synthetic Promoter Design 393

Dieterich, C., Wang, H., Rateitschak, K., Luz, H., & Vingron, M. (2003). CORG: a database
for COmparative Regulatory Genomics. Nucleic Acids Research, Vol. 31, No. 1,
January 2003, pp. 55-57. ISSN 0305-1048
Dubchak, I., & Ryaboy, D. V. (2006). VISTA family of computational tools for comparative
analysis of DNA sequences and whole genomes. Methods in Molecular Biology, Vol.
338, April 2006, pp. 69-89. ISSN 1064-3745
Edelman, G. M., Meech, R., Owens, G. C., & Jones, F. S. (2000). Synthetic promoter elements
obtained by nucleotide sequence variation and selection for activity. Proceedings of
the National Academy of Sciences of the United States of America, Vol. 97, No. 7, March
2000, pp. 3038-3043. ISSN 1091-6490
Ferretti, V., Poitras, C., Bergeron, D., Coulombe, B., Robert, F., & Blanchette, M. (2007).
PReMod: a database of genome-wide mammalian cis-regulatory module
predictions. Nucleic Acids Research, Vol. 35, January 2007, pp. D122-126. ISSN 0305-
1048
Gertz, J., & Cohen, B. A. (2009). Environment-specific combinatorial cis-regulation in
synthetic promoters. Molecular Systems Biology, Vol. 5, February 2009, pp. 244. ISSN
1744-4292
Gertz, J., Siggia, Eric D, & Cohen, B. A. (2009). Analysis of combinatorial cis-regulation in
synthetic and genomic promoters. Nature, Vol. 457, No. 7226, July 2009, pp. 215-218.
ISSN: 0028-0836
Hammer, K., Mijakovic, I., & Jensen, P. R. (2006). Synthetic promoter libraries--tuning of
gene expression. Trends in Biotechnology, Vol. 24, No. 2, February 2006, pp. 53-55.
ISSN 0167-7799
Han, J., McLane, B., Kim, E.-H., Yoon, J.-W., & Jun, H.-S. (2010). Remission of Diabetes by
Insulin Gene Therapy Using a Hepatocyte-specific and Glucose-responsive
Synthetic Promoter. Molecular Therapy, Vol. 19, No. 3, March 2010, pp.470-478. ISSN
1525-0016
He, W., Qiang, M., Ma, W., Valente, A. J., Quinones, M. P., Wang, W., et al. (2006).
Development of a synthetic promoter for macrophage gene therapy. Human Gene
Therapy, Vol. 17, No. 9, September 2006, pp. 949-959. ISSN 1043-0342
Hunziker, A., Tuboly, C., Horváth, P., Krishna, S., & Semsey, S. (2010). Genetic flexibility of
regulatory networks. Proceedings of the National Academy of Sciences of the United
States of America, Vol. 107, No. 29, July 2010, pp. 12998-13003. ISSN 1091-6490
Hwang, D. Y., Carlezon, W. A., Isacson, O., & Kim, K. S. (2001). A high-efficiency synthetic
promoter that drives transgene expression selectively in noradrenergic neurons.
Human Gene Therapy, Vol. 12, No. 14, September 2001, pp.1731-1740. ISSN 1043-0342
Jacquet, M. A., Ehrlich, R., & Reiss, C. (1989). In vivo gene expression directed by synthetic
promoter constructions restricted to the -10 and -35 consensus hexamers of E. coli.
Nucleic Acids Research, Vol. 17, No. 8, April 1989, pp. 2933-2945. ISSN 0305-1048
Jagannathan, V., Roulet, E., Delorenzi, M., & Bucher, P. (2006). HTPSELEX--a database of
high-throughput SELEX libraries for TFBSs. Nucleic Acids Research, Vol. 34, January
2006, pp. D90-94. ISSN 0305-1048
Jegga, A. G., Sherwood, S. P., Carman, J. W., Pinski, A. T., Phillips, J. L., Pestian, J. P., et al.
(2002). Detection and visualization of compositionally similar cis-regulatory
element clusters in orthologous and coordinately controlled genes. Genome Research,
Vol. 12, No. 9, September 2002, pp. 1408-1417. ISSN 1088-9051
394 Computational Biology and Applied Bioinformatics

Jegga, A. G., Gupta, A., Gowrisankar, S., Deshmukh, M. A., Connolly, S., Finley, K., et al.
(2005). CisMols Analyzer: identification of compositionally similar cis-element
clusters in ortholog conserved regions of coordinately expressed genes. Nucleic
Acids Research, Vol. 33, July 2005, pp. W408-411. ISSN 0305-1048
Jensen, L. J., & Knudsen, S. (2000). Automatic discovery of regulatory patterns in promoter
regions based on whole cell expression data and functional annotation.
Bioinformatics, Vol. 16, No. 4, April 2000, pp. 326-333. ISSN 1460-2059
Kamine, J., Subramanian, T., & Chinnadurai, G. (1991). Sp1-dependent activation of a
synthetic promoter by human immunodeficiency virus type 1 Tat protein.
Proceedings of the National Academy of Sciences of the United States of America, Vol. 88,
No. 19, October 1991, pp. 8510-8514. ISSN 1091-6490
Karanam, S., & Moreno, C. S. (2004). CONFAC: automated application of comparative
genomic promoter analysis to DNA microarray datasets. Nucleic Acids Research, Vol.
32, July 2004, pp. W475-484. ISSN 0305-1048
Kel-Margoulis, Olga V, Kel, Alexander E, Reuter, Ingmar, Deineko, I. V., & Wingender,
Edgar. (2002). TRANSCompel: a database on composite regulatory elements in
eukaryotic genes. Nucleic Acids Research, Vol. 30, No. 1, January 2002, pp. 332-334.
ISSN 0305-1048
Kinkhabwala, A., & Guet, C. C. (2008). Uncovering cis regulatory codes using synthetic
promoter shuffling. PloS One, Vol. 3, No. 4, April 2008, pp. e2030. ISSN 1932-6203
La Rosa, P., Viara, E., Hupé, P., Pierron, G., Liva, S., Neuvial, P., et al. (2006). VAMP:
visualization and analysis of array-CGH, transcriptome and other molecular
profiles. Bioinformatics, Vol. 22, No. 17, September 2006, pp. 2066-2073. ISSN 1460-
2059
Lardenois, A., Chalmel, F., Bianchetti, L., Sahel, J.-A., Léveillard, T., & Poch, O. (2006).
PromAn: an integrated knowledge-based web server dedicated to promoter
analysis. Nucleic Acids Research, Vol. 34, July 2006, pp. W578-83. ISSN 0305-1048
Lemken, M.-L., Wybranietz, W.-A., Schmidt, U., Graepler, F., Armeanu, S., Bitzer, M., et al.
(2005). Expression liver-directed genes by employing synthetic transcriptional
control units. World Journal of Gastroenterology, Vol. 11, No. 34, September 2005, pp.
5295-5302. ISSN 1007-9327
Lenhard, B., Sandelin, A., Mendoza, L., Engström, P., Jareborg, N., & Wasserman, W. W.
(2003). Identification of conserved regulatory elements by comparative genome
analysis. Journal of Biology, Vol. 2, No. 2, May 2003, pp. 13. ISSN 1475-4924
Li, X., Eastman, E. M., Schwartz, R. J., & Draghia-Akli, R. (1999). Synthetic muscle
promoters: activities exceeding naturally occurring regulatory sequences. Nature
Biotechnology, Vol. 17, No. 3, January 1999, pp. 241-245. ISSN: 1087-0156
Liu, C.-C., Lin, C.-C., Chen, W.-S. E., Chen, H.-Y., Chang, P.-C., Chen, J. J. W., et al. (2006).
CRSD: a comprehensive web server for composite regulatory signature discovery.
Nucleic Acids Research, Vol. 34, July 2006, pp. W571-7. ISSN 0305-1048
Liu, M.-L., Oh, J. S., An, S. S., Pennant, W. A., Kim, H. J., Gwak, S.-J., et al. (2010). Controlled
nonviral gene delivery and expression using stable neural stem cell line transfected
with a hypoxia-inducible gene expression system. The Journal of Gene Medicine, Vol.
12, No. 12, December 2010, pp. 990-1001. ISSN 1521-2254
The Use of Functional Genomics in Synthetic Promoter Design 395

Logg, C. R., Logg, A., Matusik, R. J., Bochner, B. H., & Kasahara, N. (2002). Tissue-specific
transcriptional targeting of a replication-competent retroviral vector. Journal of
Virology, Vol. 76, No. 24, December 2002, pp. 12783-12791. ISSN 0022-538X
Mader, S., & White, J. H. (1993). A steroid-inducible promoter for the controlled
overexpression of cloned genes in eukaryotic cells. Proceedings of the National
Academy of Sciences of the United States of America, Vol. 90, No. 12, June 1993, pp.
5603-5607. ISSN 1091-6490
Masson, J. M., & Miller, J. H. (1986). Expression of synthetic suppressor tRNA genes under
the control of a synthetic promoter. Gene, Vol. 47, No. 2-3, February 1986, pp.179-83.
ISSN 0378-1119
Matys, V., Fricke, E., Geffers, R., Gössling, E., Haubrock, M., Hehl, R., et al. (2003).
TRANSFAC: transcriptional regulation, from patterns to profiles. Nucleic Acids
Research, Vol. 31, No. 1, January 2003, pp. 374-378. ISSN 0305-1048
Mogno, I., Vallania, F., Mitra, R. D., & Cohen, B. A. (2010). TATA is a modular component of
synthetic promoters. Genome Research, Vol. 20, No. 10, October 2010, pp. 1391-1397.
ISSN 1088-9051
Ramon, A., & Smith, H. O. (2010). Single-step linker-based combinatorial assembly of
promoter and gene cassettes for pathway engineering. Biotechnology Letters, Vol 33,
No. 3, March 2010, pp. 549-555. ISSN 0141-5492
Rebeiz, M., Reeves, N. L., & Posakony, J. W. (2002). SCORE: a computational approach to
the identification of cis-regulatory modules and target genes in whole-genome
sequence data. Site clustering over random expectation. Proceedings of the National
Academy of Sciences of the United States of America, Vol. 99, No. 15, July 2002, pp.
9888-9893. ISSN 1091-6490
Roberts, M. L. (2007). A method for the construction of cancer-specific promoters using
functional genomics. WIPO WO/2008/107725.
Roberts, M. L., & Kottaridis S. D. (2007). Interpreting microarray data: towards the complete
bioinformatics toolkit for cancer. Cancer Genomics and Proteomics, Vol 4, No. 4,
July-August 2007, pp 301-308. ISSN: 1109-6535.
Roth, F. P., Hughes, J. D., Estep, P. W., & Church, G. M. (1998). Finding DNA regulatory
motifs within unaligned noncoding sequences clustered by whole-genome mRNA
quantitation. Nature Biotechnology, Vol. 16, No. 10, October 1998, pp. 939-945. ISSN:
1087-0156
Sandelin, A., Alkema, W., Engström, P., Wasserman, W. W., & Lenhard, B. (2004). JASPAR:
an open-access database for eukaryotic transcription factor binding profiles. Nucleic
Acids Research, Vol. 32, January 2004, pp. D91-94. ISSN 0305-1048
Santilli, G., Almarza, E., Brendel, C., Choi, U., Beilin, C., Blundell, M. P., et al. (2011).
Biochemical correction of X-CGD by a novel chimeric promoter regulating high
levels of transgene expression in myeloid cells. Molecular Therapy, Vol. 19, No. 1,
January 2011, pp. 122-132. ISSN 1525-0016
Segal, E., Friedman, N., Koller, D., & Regev, A. (2004). A module map showing conditional
activity of expression modules in cancer. Nature Genetics, Vol. 36, No. 10, October
2004, pp. 1090-1098. ISSN 1061-4036
Segal, E., Shapira, M., Regev, A., Peʼer, D., Botstein, D., Koller, D., et al. (2003). Module
networks: identifying regulatory modules and their condition-specific regulators
396 Computational Biology and Applied Bioinformatics

from gene expression data. Nature Genetics, Vol. 34, No. 2, June 2003, pp. 166-176.
ISSN 1061-4036
Sharov, A. A., Dudekula, D. B., & Ko, M. S. H. (2006). CisView: a browser and database of
cis-regulatory modules predicted in the mouse genome. DNA Research, Vol. 13, No.
3, June 2006, pp. 123-134. ISSN 1340-2838
Sinha, S., & Tompa, M. (2002). Discovery of novel TFBSs by statistical overrepresentation.
Nucleic Acids Research, Vol. 30, No. 24, December 2002, pp. 5549-5560. ISSN 0305-
1048
Sinha, S., & Tompa, M. (2003). YMF: A program for discovery of novel TFBSs by statistical
overrepresentation. Nucleic Acids Research, Vol. 31, No. 13, July 2003, pp. 3586-3588.
ISSN 0305-1048
Sleight, S. C., Bartley, B. A., Lieviant, J. A., & Sauro, H. M. (2010). Designing and engineering
evolutionary robust genetic circuits. Journal of Biological Engineering, Vol. 4, No. 1,
November 2010, pp. 12. ISSN 1754-1611
Sun, H., Palaniswamy, S. K., Pohar, T. T., Jin, V. X., Huang, T. H.-M., & Davuluri, Ramana V.
(2006). MPromDb: an integrated resource for annotation and visualization of
mammalian gene promoters and ChIP-chip experimental data. Nucleic Acids
Research, Vol. 34, January 2006, pp. D98-103. ISSN 0305-1048
Suzuki, Y., Yamashita, R., Nakai, K., & Sugano, S. (2002). DBTSS: DataBase of human
Transcriptional Start Sites and full-length cDNAs. Nucleic Acids Research, Vol. 30,
No. 1, January 2002, pp. 328-331. ISSN 0305-1048
Suzuki, Y., Yamashita, R., Sugano, S., & Nakai, K. (2004). DBTSS, DataBase of
Transcriptional Start Sites: progress report 2004. Nucleic Acids Research, Vol. 32,
January 2004, pp. D78-81. ISSN 0305-1048
Szymanski, P., Anwer, K., & Sullivan, S. M. (2006). Development and characterization of a
synthetic promoter for selective expression in proliferating endothelial cells. The
Journal of Gene Medicine, Vol. 8, No. 4, April 2006, pp. 514-523. ISSN 1521-2254
Trumble, W. R., Sherf, B. A., Reasoner, J. L., Seward, P. D., Denovan, B. A., Douthart, R. J., et
al. (1992). Protein expression from an Escherichia coli/Bacillus subtilis
multifunctional shuttle plasmid with synthetic promoter sequences. Protein
Expression and Purification, Vol. 3, No. 3, June 1992, pp. 169-177. ISSN 1046-5928
Vega, V. B., Bangarusamy, D. K., Miller, L. D., Liu, E. T., & Lin, C.-Y. (2004). BEARR: Batch
Extraction and Analysis of cis-Regulatory Regions. Nucleic Acids Research, Vol 32,
July 2004, pp. W257-260. ISSN 0305-1048
Venter, M. (2007). Synthetic promoters: genetic control through cis engineering. Trends in
Plant Science, Vol. 12, No. 3, March 2007, pp. 118-124. ISSN 1360-1385
Wingender, E. (1988). Compilation of transcription regulating proteins. Nucleic Acids
Research, Vol. 16, No. 5, March 1988, pp. 1879-1902. ISSN 0305-1048
Zhao, F., Xuan, Z., Liu, L., & Zhang, Michael Q. (2005). TRED: a Transcriptional Regulatory
Element Database and a platform for in silico gene regulation studies. Nucleic Acids
Research, Vol. 33, January 2005, pp. D103-107. ISSN 0305-1048
 21

Analysis of Transcriptomic and Proteomic Data

in Immune-Mediated Diseases
Sergey Bruskin1, Alex Ishkin2, Yuri Nikolsky2,
Tatiana Nikolskaya2 and Eleonora Piruzian1
1Vavilov Institute of General Genetics, Russian Academy of Sciences, Moscow
2Thomson Reuters, Professional, Saint Joseph, MI
1Russia
2USA

1. Introduction
Psoriasis (a skin disease) and Crohn’s disease (a disease of the intestinal epithelium) are
multifactorial diseases caused by abnormalities in genetic machinery regulation. Both
pathologies disturb the immune system, and the pathological processes are triggered by
environmental factors. In the case of psoriasis, these factors are psychoemotional stresses,
infections (group A streptococci and Staphylococcus aureus), drugs (lithium-containing,
antimalarial, and antituberculous agents and Novocain), smoking, and mechanical damages
(the so-called Koebner phenomenon) [Bowcock A et al., 2004]. Psoriasis vulgaris is one of
the most prevalent chronic inflammatory skin diseases affecting approximately 2% of
individuals in Western societies, and found worldwide in all populations. Psoriasis is a
complex disease affecting cellular, gene and protein levels and presented as skin lesions. The
skin lesions are characterized by abnormal keratinocyte differentiation, hyperproliferation
of keratinocytes, and infiltration of inflammatory cells [Boehncke WH et al. 1996; Ortonne
JP, 1996]. The factors triggering Crohn’s disease include psychoemotional stresses, infections
(Mycobacterium avium ssp. paratuberculosis and invasive Escherichia coli variants), drugs
(antibiotics and nonsteroid antiimflammatory agents), smoking, and nutritional regimen
[Sartor R., 2006]. Crohn’s disease known only since the 1920s [Crohn B et al., 1932] and now
affecting up to 0.15% of the northwest European and North American population [Binder
V., 2005].
Both psoriasis and Crohn’s disease are now regarded as incurable, and the goal of their
therapy is to extend the remission periods and decrease the severity of the disease. These
two diseases are tightly related at the genetic level, as over five genetic loci are involved in
the development of both psoriasis and Crohn’s disease.
The mechanisms of both psoriasis and Crohn’s disease are complex and involve genetic and
environmental factors. As we gain more knowledge about molecular pathways implicated
in diseases, novel therapies emerge (such as etanercept and infliximab that target TNF-α or
CD11a- mediated pathways [Pastore S et al., 2008; Gisondi P et al., 2007]).
We have studied earlier the components of AP-1 transcription factor as psoriasis candidate
genes. This study was performed by bioinformatics analysis of the transcription data using
the GEO DataSets database (http://www.ncbi.nlm.nih.gov/geo/) [Piruzian ES et al., 2007].
398 Computational Biology and Applied Bioinformatics

The same approach was used by other researchers to detect potential therapeutic targets for
psoriasis [Yao Y., et al., 2008]. In next step, we performed a comparative analysis of the
molecular processes involved in the pathogenesis of two diseases, psoriasis and Crohn’s
disease [Piruzian ES et al., 2009].
Despite the fact that psoriasis and Crohn`s disease affect completely different body systems
(skin and intestine), they are much more similar that it may seem at first glance. Both skin
and intestinal epithelium are barrier organs, that are the first to resist the environmental
factors, including microorganisms. Both pathologies are immune-mediated inflammatory
diseases, that is also marked by the same drug therapies. Finally, they have a lot of common
susceptibility loci (Fig. 1).

Fig. 1. Localization of various linkage regions for barrier diseases on human chromosomes
map [Schreiber S et al., 2005])
In recent years, microarray mRNA expression profiling [Oestreicher JL et al., 2001; Bowcock
AM et al., 2001; Zhou X et al., 2003; Quekenborn-Trinquet V et al., 2005] of lesional psoriatic
skin revealed over 1,300 differentially expressed genes. Enrichment analysis (EA) showed
that these genes encode proteins involved in regeneration, hyperkeratosis, metabolic
function, immune response, and inflammation and revealed a number of modulating
signaling pathways. These efforts may help to develop new-generation drugs. However,
enrichment analysis limits our understanding of altered molecular interactions in psoriasis
as it provides a relative ranking based on ontology terms resulting in the representation of
fragmented and disconnected perturbed pathways. Furthermore, analysis of gene
expression alone is not sufficient for understanding the whole variety of pathological
changes at different levels of cellular organization. Indeed, new methodologies have been
applied to the analysis of OMICs data in complex diseases that include algorithm-based
Analysis of Transcriptomic and Proteomic Data in Immune-Mediated Diseases 399

biological network analysis [Nikolskaya T, et al., 2009; Nikolsky Y et al., 2005; Bhavnani SK
et al., 2009; Ideker T et al., 2008; Chuang HY et al., 2007] and meta-analysis of multiple
datasets of different types [Cox B et al., 2005; Wise LH et al., 1999; Ghosh D et al., 2003;
Warnat P et al., 2005; Hack CJ, 2004; Menezes R et al., 2009]. Here, we applied several
techniques of network and meta-analysis to reveal the similarities and differences between
transcriptomics- and proteomics-level perturbations in psoriasis lesions. We particularly
focused on revealing novel regulatory pathways playing a role in psoriasis development
and progression.

2. Transcriptomic and proteomic data, network analysis

Data preparation. The data deposited with the public database of microarray experiments,
GEO (http://www.ncbi.nlm.nih.gov/geo/), were analyzed. The expression data on
psoriasis were contained in entry GDS1391, and on Crohn’s disease, in entry GDS1330. Since
these data were obtained using different microarrays and experimental schemes, analysis
was individually performed for each disease with subsequent comparison of the lists of
genes with altered expression for each case.
Two sets were selected from the overall data on psoriasis, namely, four experiments with
gene expression in psoriatic skin lesions, and four, with gene expression in the healthy skin
of the same patients. The selected data for Crohn’s disease were also represented by two
sets: 10 experiments on expression in intestinal epithelial lesions, and 11, on expression in
the intestinal tissue of healthy individuals. The data were prepared for analysis using the
GeneSpring GX (http://www.chem.agilent.com/scripts/pds.asp?lpage=27881) software
package. This processing comprised discarding of the genes with poorly detectable
expression and normalization of the remaining data. In addition to the values of expression,
the so-called absent call flags were added for psoriasis cases; these flags characterize the
significance of the difference in expression of a particular gene from the background noise.
The genes displaying the flag value of A (i.e., absent, which means that the expression of a
particular gene in experiment is undetectable) in over 50% of experiments were discarded
from further analysis. This information was unavailable for Crohn’s disease; therefore, this
step was omitted. The results were normalized by the median gene expression in the
corresponding experiment to make them comparable with one another.
Detection of the genes with altered expression. Differentially expressed genes were sought
using Welch’s t-test [Welch B.L., 1947]. This test does not require that the distribution
variances for the compared samples be equal; therefore, it is more convenient for analyzing
expression data than a simple t-test. FDR algorithm [Benjamini Y et al., 1995] with a
significance threshold of 0.1 was used to control the type I errors in finding differentially
expressed genes; in this case, the threshold determined the expected rate of false positive
predictions in the final set of genes after statistical control.
Detection of common biological processes. The resulting gene lists were compared, and the
molecular processes mediated by the genes displaying altered expressions in both diseases
were sought using the MetaCore (GeneGo Inc., www.genego.com) program. The
significance of the biological processes where the genes displaying altered expressions in
both diseases was assessed according to the degree to which overlapping between the list of
differentially expressed genes and the list of genes ascribed to the process exceeded random
overlapping. Hypergeometric distribution [Draghici S et al., 2007] was used as a model of
400 Computational Biology and Applied Bioinformatics

random overlapping between the gene lists. The measure of signifi- cance for the input gene
list, the p value, in this distribution is calculated as

min ( n , R )
pVal ( r , n , R , N ) = ∑ P ( i , n, R , N )
i = max ( r , R + n − N )

R ! n !( N − R ) !( N − n ) !
=
N!
min ( n , R )
1
× ∑ i ! ( R − i ) ! ( n − i ) ! ( N − R − n + i ) !'
i = max ( r , R + n − N )

where N is the number of genes in the MetaCore database; R, the number of genes ascribed
to a particular process; n, the size of the input gene list; and r, the number of genes from the
input list related to this process.
Three ontologies of biological processes were used in this work: GO (www.gene-
ontology.org) and two ontologies included in the MetaCore, Canonical pathways and
GeneGo process networks. The processes contained in the MetaCore ontologies are gene
networks, which reflect the interaction of proteins involved in a particular biological
regulatory or metabolic pathway. The processes for all three ontologies were prioritized by
the negative logarithm of p value.
The common molecular biological pathways were determined based on the analysis of
significant biological processes and expressions of the genes involved in these processes.
The MetaCore contains the algorithms providing for detection in the total network of gene
interactions the particular regulatory pathways and subnetworks saturated with the objects
of research interest, in this case, the genes with altered expression. The resulting
assumptions on the pattern of common biological pathways were visualized as a gene
network using the MetaCore.
Skin biopsies. Acquisition of the human tissue was approved by the Vavilov Institute of
General Genetics of Russian Academy of Sciences review board and the study was
conducted after patient's consent and according to the Declaration of Helsinki Principles. A
total of 6 paired nonlesional and lesional (all were plaque-type) skin biopsies from 3
psoriatic patients were profiled using 2D electrophoresis. All the donors who gave biopsy
tissue (both healthy controls and individuals with psoriasis) provided a written informed
consent for the tissue to be taken and used in this study. Clinical data for all patients are
listed in Table 3. Full-thickness punch biopsies were taken from uninvolved skin (at least 2
cm distant from any psoriatic lesion; 6 mm diameter) and from the involved margin of a
psoriatic plaque (6 mm diameter) from every patient.
Sample preparation, two-dimensional electrophoresis, gel image analysis and
massspectrometry was carried out using the standard procedure [Gravel P & Golaz O, 1996;
Mortz E, et al., 2001].
Microarray data analysis. We used recently published data set [Yao Y, et al., 2008] from
GEO data base (http://www.ncbi.nlm.nih.gov/geo/; accession number GSE14095). We
compared 28 pairs of samples (in each pair there was a sample of lesional skin and a sample
of healthy skin from the same patient). Values for each sample were normalized by sample
median value in order to unify distributions of expression signals. For assessment of
differential expression we used paired Welch ttest with FDR correction [Benjamini Y et al.,
Analysis of Transcriptomic and Proteomic Data in Immune-Mediated Diseases 401

1995]. Probe set was considered as differentially expressed if its average fold change
exceeded 2.5 and FDR corrected p-value was less than 0.01.
Overconnection analysis. All network-based analyses were conducted with MetaCore
software suite http://www.genego.com. This software employs a dense and manually
curated database of interactions between biological objects and variety of tools for
functional analysis of high-throughput data. We defined a gene as overconnected with the
gene set of interest if the corresponding node had more direct interactions with the nodes of
interest than it would be expected by chance. Significance of overconnection was estimated
using hypergeometric distribution with parameters r - number of interactions between
examined node and the list of interest; R - degree of examined node, n - sum of interactions
involving genes of interest and N - total number of interactions in the database:

min ( n , R )
pVal ( r , n , R , N ) = ∑ P ( i , n, R , N )
i = max ( r , R + n − N )

Hidden nodes analysis. In addition to direct interacting objects, we also used objects that may
not interact directly with objects of interest but are important upstream regulators of those
[Dezso Z et al., 2009]. The approach is generally the same as described above, but the shortest
paths instead of direct links are taken into account. As we were interested in transcriptional
regulation, we defined a transcriptional activation shortest path as the preferred shortest path
from any object in the MetaCore database to the transcription factor target object from the data
set. We added an additional condition to include the uneven number of inhibiting interactions
in the path (that's required for the path to have activating effect). If the number of such paths
containing examined gene and leading to one of objects of interest were higher than expected
by chance, this gene was considered as significant hidden regulator. The significance of a
node's importance was estimated using hypergeometric distribution with parameters r -
number of shortest paths between containing currently examined gene; R - total number of
shortest paths leading to a gene of interest through transcriptional factor, n - total number of
transcription activation shortest paths containing examined gene and N - total number of
transcription activation shortest paths in the database.
Rank aggregation. Both topology significance approaches produced lists of genes
significantly linked to a gene or protein set of interest, ranked by corresponding p-values.
To combine results of these two approaches, we used a weighted rank aggregation method
described in [Pihur V et al., 2009]. Weighted Spearman distance was used as distance
measure and the genetic algorithm was employed to select the optimal aggregated list of
size 20. This part of work was accomplished in R 2.8.1 http://www.r-project.org.
Network analysis. In addition to topology analysis, we examined overexpressed genes and
proteins using various algorithms for selecting connected biologically meaningful
subnetworks enriched with objects of interest. Significance of enrichment is estimated using
hypergeometric distribution. We first used an algorithm intended to find regulatory pathways
that are presumably activated under pathological conditions. It defines a set of transcription
factors that are directly regulating genes of interest and a set of receptors whose ligands are in
the list of interest and then constructs series of networks; one for each receptor. Each network
contains all shortest paths from a receptor to the selected transcriptional factors and their
targets. This approach allows us to reveal the most important areas of regulatory machinery
affected under the investigated pathological condition. Networks are sorted by enrichment p-
402 Computational Biology and Applied Bioinformatics

value. The second applied algorithm used was aimed to define the most influential
transcription factors. It considers a transcriptional factor from the data base and gradually
expands the subnetwork around it until it reaches a predefined threshold size (we used
networks of 50 nodes). Networks are sorted by enrichment p-value.

3. A comparative analysis of the molecular genetic processes in the

pathogenesis of psoriasis and Crohn’s disease
Constructing List of the Genes with Altered Expression in Both Pathologies We detected the
lists of differentially expressed genes separately in each dataset and compared these lists at
the system level. This approach to analysis was dictated by the properties of expression data
in general (a high noise level and a large volume of analyzed data) and individual
properties of datasets selected for analysis, which were obtained using different microarrays
with different numbers of probes. That is why the datasets were incomparable in a direct
fashion. The dataset on psoriasis initially contained information on the expression levels of
12626 probes from eight experiments (four specimens of skin lesions, and four of the healthy
skin from the same patients). After discarding the probes with poorly detectable expression
(see Materials and Methods), the set was reduced to 5076 probes. The list of the probes with
statistically significant differences in expression between the lesion and healthy tissue
contained 410 items at a significance level of 0.1.
The dataset on Crohn’s disease contained information on the expression level of 24016
probes from 21 experiments (11 specimens of epithelial lesions and 10 specimens of healthy
epithelium). The list of probes displaying statistically significant differences in expression
between the lesion and healthy tissue contained 3850 probes at a significance level of 0.1.
This pronounced difference in the sizes of gene lists result from the fact that the algorithm
used for controlling type I errors (FDR) depends on the input set. The larger the initial gene
list, the larger number of genes will pass the FDR control at a similar p-value distribution; in
our case, the number of analyzed probes in the dataset for Crohn’s disease is five times
larger than that in the dataset for psoriasis.
The lists of differentially expressed genes were input into the MetaCore program. Because
microarrays contained not only gene probes, but also a large number of ESTs with
unidentified functions, the size of gene lists at this stage changed because not all the probes
had the corresponding gene in the MetaCore database and because some probes
corresponded to more than one gene. The lists of recognized genes comprised 425 and 2033
items for psoriasis and Crohn’s disease, respectively.
The common part for the compared lists comprised 49 genes, which is a significant
overlapping (p value = 4.94 × 10 –2 ). The significance was estimated using Fisher’s test. The
complete set contained 9017 genes present in both studied datasets (this set was identified
by comparing the complete lists of genes for both microarrays in MetaCore). The lists of
genes with altered expression were reduced to the subset of genes present in both datasets.
Thus, these particular 49 genes were selected for further analysis (Table 1).
It was of interest to determine the molecular processes with which the genes common to
psoriasis and Crohn’s disease are associated. Table 2 consolidates the most probable cell
processes with involvement of the genes listed in Table 1, as determined by the MetaCore
software tools. These processes (Table 2) fall into two main groups—related to inflammation
and cell cycle. Indeed, the pathological lesions in both psoriasis and Crohn’s disease
are inflammatory foci. The cell cycle is also considerably affected in both pathologies.
Analysis of Transcriptomic and Proteomic Data in Immune-Mediated Diseases 403

An increased proliferation of keratinocytes is observed in the psoriatic skin, an

inflammatory focus.

GNA15 SFPQ IFI35 IER2 OAS2 RFK UBE2L6

CBX3 CG018 CSNK1D SYNCRIP PSME2 CTSC CASP4
GPM6B UGT1A4 STAT3 S100A8 FOXC1 SOSTDC1 ETS2
UGT1A6 VKORC1 TRIM22 RARG TRAK2 SERPINB5 MECP2
IFI44 H2AFY TXNDC1 ARMET ZNF207 KIAA1033 QPCT
DEGS1 MIB1 IRF9 DDOST DNAJC7 RBPMS JUNB
LONRF1 HMGN1 MRPL9 FGFR2 CDC42EP1 S100A9 PHGDH
Table 1. Genes displaying altered expression in both psoriasis and Crohn's disease.

Process p value
Inflammation: interferon signaling pathways 2.19E-03
Signal transduction: Wnt signaling pathways 1.20E-02
Regulation of translation initiation 5.66E-02
Morphogenesis of blood vessels 9.76E-02
DNA repair 1.17E-01
Inflammation: amphoterin signaling pathways 1.19E-01
Proteolysis determined by the cell cycle and apoptosis 1.29E-01
Interleukin regulation of the cell cycle in G1-S phase 1.29E-01
Signal transduction: androgen receptor signaling pathways 1.34E-01
Table 2. Cell processes common to psoriasis and Crohn's disease.
For a more detailed description of the inflammatory response and cell cycle in the parts of
them most tightly related to the genes listed in Table 1, we constructed gene networks,
which are fragments of the larger gene networks describing the inflammatory response (Fig.
2) and cell cycle control (Fig. 3). Figure 2 shows that the inflammatory response is initiated
by such well-known cytokines as TNF-α, IFN-γ, IL-2, IL-6, IL-17, and IL-23. Then protein
kinases activate the transcription factors AP-1, STAT3, C/EBP, NF-κB, ISGF3, and others.
Figure 3 shows that the key cell cycle regulators that changed gene expression are the
transcription factors AP-1, c-Myc, and STAT3. It is also evident that the genes encoding AP-
1 transcription factor components are involved in both the inflammatory response and cell
cycle control. It is known that the genes depending on AP-1 play an important role in
regulation of proliferation, morphogenesis, apoptosis, and cell differentiation. Induction of
cell differentiation activates transcription of the genes encoding the components of AP-1
complex [Turpaev K.T., 2006]. We assume that the genes of AP-1 transcription factor are the
candidate genes involved in the pathogenesis of both psoriasis and Crohn’s disease;
moreover, this hypothesis is particularly based on the bioinformatics analysis of microarray
data. Therefore, it was interesting to compare our data with the available information about
the chromosome localization of the loci associated with psoriasis and Crohn’s disease.
404 Computational Biology and Applied Bioinformatics

Fig. 2. Detail of the gene network describing inflammatory response. Green arrows indicate
activation of the corresponding network elements, from the level of cytokines to
transcription factors; light blue arrows, the activation of effector genes by transcription
factors; and red circles, genes from the list.

Fig. 3. Detail of the gene network describing cell cycle control. Green arrows indicate
activation of the corresponding elements; red arrows, inhibition; and red circles, genes from
the list.
Analysis of Transcriptomic and Proteomic Data in Immune-Mediated Diseases 405

4. Integrated network analysis of transcriptomic and proteomic data in

psoriasis
Differentially abundant proteins. Protein abundance was determined by densitometric
quantification of the protein spots on 2D-electophoresis gel (Figure 4) followed by MALDI-
TOF mass spectrometry. Total of 10 proteins were over-abundant at least 2-fold in lesional
skin compared with uninvolved skin: Keratin 14, Keratin 16, Keratin 17, Squamous cell
carcinoma antigen, Squamous cell carcinoma antigen-2, Enolase 1, Superoxide dismutase
[Mn], Galectin-7, S100 calcium-binding protein A9 and S100 calcium-binding protein A7.

Fig. 4. Representative silver-stained 2DE gel images of lesional and uninvolved skin biopsy
lysates. a) - gel image of lesional skin biopsy lysate; b) - gel image of uninvolved skin biopsy
lysate. Spots corresponding to proteins overexpressed in lesions are marked with red
rectangles and numbered. Spot 1 correspond to 3 proteins of keratin family, spot 2 - SCCA2,
spot 3 - SCCA1, spot 4 - enolase 1, spot 5 - SOD2, spot 6 - galectin-7. S100A7 is found in
spots 7 and 8 and S100A9 corresponds to 9 th and 10 th spots.
Several of these proteins were previously reported to be over-abundant in psoriatic plaques
[Leigh IM et al., 1995; Madsen P et al., 1991; Vorum H et al., 1996; Takeda A et al.., 2002]. The
proteins belonged to a diverse set of pathways and processes.
We attempted to connect the proteins into a network using a collection of over 300,000
manually curated protein interactions and several variants of "shortest path" algorithms
applied in MetaCore suite [Nikolsky Y et al, 2009] (Figure 5). The genes encoding
overabundant proteins were found to be regulated by several common transcription factors
(TFs) including members of the NFkB and AP-1 complexes, STAT1, STAT3, c-Myc and SP1.
Moreover, the upstream pathways activating these TFs were initiated by the overabundant
S100A9 through its receptor RAGE [Ghavami S et al., 2008] and signal transduction kinases
(JAK2, ERK, p38 MAPK). This network also included a positive feedback loop as S100A9
expression was determined to be controlled by NF-kB [Schreiber J et al., 2006]. The topology
of this proteomics-derived network was confirmed by several transcriptomics studies
406 Computational Biology and Applied Bioinformatics

[Tsuruta D, 2009; Sano S et al., 2008; Ghoreschi K et al., 2003; Piruzian ES et al., 2009;
Gandarillas A & Watt FM, 1997; Arnold I & Watt FM, 2001] which showed overexpression
of these TFs in psoriasis lesions. Transiently expressed TFs normally have low protein level
and, therefore, usually fail to be detected by proteomics methods.

Fig. 5. Network illustrating regulatory pathways leading to transcription activation of

proteomics markers. Red circles denote upregulated proteins.
RAGE receptor is clearly the key regulator on this network and plays the major role in
orchestrating observed changes of protein abundance. This protein is abundant in both
keratinocytes and leukocytes, though normally its expression is low [Lohwasser C et al.,
2006]. RAGE participates in a range of processes in these cell types, including inflammation.
It is being investigated as a drug target for treatment of various inflammatory disorders
[Santilli F et al., 2009]. Thus, we may propose that RAGE can also play significant role in
psoriasis.
We used Affymetrix gene expression data set from the recent study [Yao Y et al., 2008]
involving 33 psoriasis patients. Originally, more than 1300 probe sets were found to be
upregulated in lesions as compared with unlesional skin of the same people. We identified
451 genes overexpressed in lesional skin under more stringent statistical criteria (28 samples
of lesional skin were matched with their nonlesional counterparts from the same patients in
order to exclude individual expression variations, genes with fold change >2.5 and FDR-
adjusted p-value < 0.01 were considered as upregulated). The genes encoding 7 out of 10
proteomic markers were overexpressed, well consistent with proteomics data. Expression of
Enolase 1, Keratin 14 and Galectin 7 was not altered.
Despite good consistency between the proteomics and expression datasets, the two orders of
magnitude difference in list size make direct correlation analysis difficult. Therefore, we
Analysis of Transcriptomic and Proteomic Data in Immune-Mediated Diseases 407

applied interactome methods for the analysis of common upstream regulation of the two
datasets at the level of transcription factors. First, we defined the sets of the most influential
transcription factors using two recently developed methods of interactome analysis
[Nikolsky Y et al., 2008] and the "hidden nodes" algorithm [Nikolskaya T et al., 2009]. The
former method ranks TFs based on their one-step overconnectivity with the dataset of
interest compared to randomly expected number of interactions. The latter approach takes
into account direct and more distant regulation, calculating the p-values for local
subnetworks by an aggregation algorithm [Nikolskaya T et al., 2009]. We calculated and
ranked the top 20 TFs for each data type and added several TFs identified by network
analysis approaches (data not shown). The TFs common for both data types were taken as
set of 'important pathological signal transducers' (Figure 6). Noticeably, they closely
resemble the set of TFs regulating the protein network on Figure 5.

Fig. 6. Common transcriptional factors important for regulation of objects at both

transcriptomics and proteomic levels. Objects in MetaCore database representing
transcriptional factors found to be important regulators of pathology-related genes. Red
circles denote that corresponding gene is upregulated in psoriatic lesion.
In the next step, we applied "hidden nodes" algorithm to identify the most influential
receptors that could trigger maximal possible transcriptional response. In total, we found
226 membrane receptors significantly involved into regulation of 462 differentially
expressed genes ('hidden nodes' p-value < 0.05). Assuming that topological significance
alone does not necessarily prove that all receptors are involved in real signaling or are even
expressed in the sample; we filtered this list by expression performance. The receptors used
were those whose encoding genes or corresponding ligands were overexpressed greater
than 2.5 fold. We assumed that the pathways initiated by over-expressed receptors and
ligands are more likely to be activated in psoriasis. Here we assumed that expression
alterations and protein abundance are at least collinear. An additional criterion was that the
candidate receptors had to participate in the same signaling pathways with at least one of
the common TFs. No receptor was rejected based on this criterion. In total, 44 receptors passed
408 Computational Biology and Applied Bioinformatics

the transcription cut-off. Of these 24 receptor genes were overexpressed; 23 had overexpressed
ligands and 3 cases had overexpression of both ligands and receptors (IL2RB, IL8RA and
CCR5; see Figures 7 and 8). Interestingly, for several receptors, more than one ligand was
overexpressed (Figure 7). Several receptors are composed of several subunits, only one of
which was upregulated (for example, IL-2 receptor has only gamma subunit gene significantly
upregulated). Out of 44 receptors we identified by topology analysis, 21 were previously
reported as psoriasis markers (they are listed in Table 3 with corresponding references). The
other 23 receptors were not reported to be linked to psoriasis or known to be implicated in
other inflammatory diseases. These receptors belong to different cellular processes
(development, cell adhesion, chemotaxis, apoptosis and immune response) (Table 6).

Gene Connection to psoriasis Gene Connection to psoriasis

EPHA2 No AGER Yes [Foell D et al., 2003]
EPHB2 No CCR1 Yes [Horuk R, 2005]
FCER1G No CCR2 Yes [Vestergaard C et al., 2004]
INSR No CCR3 Yes [Rottman JB et al., 2001]
LTBR No CCR5 Yes [de Groot M et al., 2007]
PLAUR No CD2 Yes [Ellis CN & Krueger GG., 2001]
TNFRSF10A No CD27 Yes [De Rie MA et al., 1996]
TNFRSF10B No CD36 Yes [Prens E et al., 1996]
CD44 Possible [Reichrath J et al, CD3D Yes [Haider AS, et al., 2007]
1997]
CSF2RB Possible [Kelly R et al., 1993] EGFR Yes [Castelijns FA et al., 1999]
CXCR4 Possible [Gu J et al., 2002] IL17RA Yes [Johansen C et al., 2009]
FZD4 Possible[Reischl J et al., 2007] IL1R1 Yes [Debets R et al., 1997]
GABBR1 Possible[Shiina T et al., 2009] IL8RA Yes [Schulz BS et al., 1993]
IL10RA Possible [Asadullah K et al., IL8RB Yes [Schulz BS et al., 1993]
1998]
IL13RA1 Possible [Cancino-Diaz JC et ITGAL Yes [Guttman-Yassky E et al., 2008]
al., 2002]
IL2RB Possible [Pietrzak A et al., ITGB2 Yes [Sjogren F et al., 1999]
2008]
IL2RG Possible [Pietrzak A et al., LRP1 Yes [Curry JL et al., 2003]
2008]
IL4R Possible [Martin R, 2003] PTPRC Yes [Vissers WH et al., 2004]
LILRB2 Possible [Penna G et al., 2005] SDC3 Yes [Patterson AM et al., 2008]
LRP2 Possible [Fu X et al., 2009] SELE Yes [Wakita H &Takigawa M, 1994]
LRP8 Possible [Fu X et al., 2009] SELPLG Yes [Chu A et al., 1999]
ROR2 Possible [Reischl J et al., 2007] TLR4 Yes [Seung NR et al., 2007]
Table 3. Receptors identified in our study and not yet studied in connection to psoriasis
('Possible' term was used if protein name co-occurred with psoriasis in articles, but no clear
evidence of its implication was shown. In some cases, ligands are associated with psoriasis
(i.e, IL-10)).
Analysis of Transcriptomic and Proteomic Data in Immune-Mediated Diseases 409

Meta-analysis of multiple OMICs data types and studies is becoming an important research
tool in understanding complex diseases. Several methods were developed for correlation
analysis between the datasets of different type, such as mRNA and proteomics [Hack CJ,
2004; Le Naour F et al., 2001; Steiling K et al., 2009; Conway JP & Kinter M, 2005; Di Pietro C
et al., 2009]. However, there are many technological challenges to resolve, including
mismatching protein IDs and mRNA probes, fundamental differences in OMICs
technologies, differences in experimental set-ups in studies done by different groups etc
[Mijalski T et al., 2005]. Moreover, biological reasons such as differences in RNA and protein
degradation processes also contribute to variability of different data types. As a result,
transcriptome and proteome datasets usually show only weak positive correlation although
were considered as complimentary. More recent studies focused on functional similarities
and differences observed for different levels of cellular organization and reflected in
different types of OMICs data [Habermann JK et al., 2007; Chen YR et al., 2006; Shachaf CM
et al., 2008; Zhao C et al., 2009]. For example, common interacting objects were found for
distinct altered transcripts and proteins in type 2 diabetes [Gerling IC et al., 2006]. In one
leukemia study [Zheng PZ et al., 2005] authors found that distinct alterations at
transcriptomics and proteomic levels reflect different sides of the same deregulated cellular
processes.

Fig. 7. Candidate receptors with their respective upregulated ligands. Initial steps of
pathways presumably activated in lesions (ligands, overexpressed at transcriptional level
and their corresponding receptors) Red circles denote that corresponding gene is
upregulated in psoriatic lesion.
The overall concordance between mRNA and protein expression landscapes was addressed
in earlier studies, although the data types were compared mostly at the gene/protein level
with limited functional analysis [Cox B et al., 2005; Mijalski T et al., 2005]. Later, ontology
enrichment co-examination of transcriptomics and proteomic data has shown that the two
data types affect similar biological processes and are complimentary [Chen YR, et al., 2006;
Zheng PZ et al., 2005; Zhao C et al., 2009]. However, the key issue of biological causality and
functional consequences of distinct regulation events at both mRNA and protein levels of
cellular organization were not yet specifically addressed. These issues cannot be resolved by
410 Computational Biology and Applied Bioinformatics

low resolution functional methods like enrichment analysis. Instead, one has to apply more
precise computational methods such as topology and biological networks, which take into
consideration directed binary interactions and multi-step pathways connecting objects
between the datasets of different types regardless of their direct overlap at gene/protein
level [Ideker T & Sharan R, 2008; Chuang HY et al., 2007]. For example, topology methods
such as "hidden nodes" [Dezso Z et al., 2009; Nikolsky Y et al., 2008] can identify and rank
the upstream regulatory genes responsible for expression and protein level alterations while
network tools help to uncover functional modules most affected in the datasets, identify the
most influential genes/proteins within the modules and suggest how specific modules
contribution to clinical phenotype [Nikolsky Y et al., 2005; Gerling IC et al., 2006].
In this study, we observed substantial direct overlap between transcriptomics and
proteomics data, as 7 out of 10 over-abundant proteins in psoriasis lesions were encoded by
differentially over-expressed genes. However, the two orders of magnitude difference in
dataset size (462 genes versus 10 proteins) made the standard correlation methods
inapplicable. Besides, proteomics datasets display a systematic bias in function of abundant
proteins, favoring "effector" proteins such as structural, inflammatory, core metabolism
proteins but not the transiently expressed and fast degradable signaling proteins. Therefore,
we applied topological network methods to identify common regulators for two datasets
such as the most influential transcription factors and receptors. We have identified some key
regulators of the "proteomics" set among differentially expressed genes, including
transcription factors, membrane receptors and extracellular ligands, thus reconstructing
upstream signaling pathways in psoriasis. In particular, we identified 24 receptors
previously not linked to psoriasis.

Fig. 8. Upregulated candidate receptors with their respective ligands. Initial steps of
pathways presumably activated in lesions (receptors, overexpressed at transcriptional level
and their corresponding ligands) Red circles denote that corresponding gene is upregulated
in psoriatic lesion.
Analysis of Transcriptomic and Proteomic Data in Immune-Mediated Diseases 411

Importantly, many ligands and receptors defined as putative starts of signaling pathways
were activated by transcription factors at the same pathways, clearly indicating on positive
regulatory loops activated in psoriasis. The versatility and the variety of signaling pathways
activated in psoriasis is also impressive, which is evident from differentially overexpression
of 44 membrane receptors and ligands in skin lesions. This complexity and redundancy of
psoriasis signaling likely contributes to the inefficiency of current treatments, even novel
therapies such as monoclonal antibodies against TNF-α and IL-23. Thus, the key regulator,
RAGE receptor, triggers multiple signaling pathways which stay activated even when
certain immunological pathways are blocked. Our study suggests that combination therapy
targeting multiple pathways may be more efficient for psoriasis (particularly considering
feasibility for topical formulations). In addition, the 24 receptors we identified by topology
analysis and previously not linked with psoriasis can be tested as potential novel targets for
disease therapy. The functional machinery of psoriasis is still not complete and additional
studies can be helpful in "filling the gaps" of our understanding of its molecular
mechanisms. For instance, kinase activity is still unaccounted for, as signaling kinases are
activated only transiently and are often missed in gene expression studies. Topological
analysis methods such as "hidden nodes" [Dezso Z et al., 2004] may help to reconstruct
regulatory events missing in the data. Also, the emerging phosphoproteomics methodology
may prove to become a helpful and complimentary OMICs technology. The network
analysis methodology is not dependent on the type of data analyzed and or any
gene/protein content overlap between the studies and is well applicable for functional
integration of multiple data types.

3. Conclusion
Thus, we succeeded in comparing the molecular processes characteristic of psoriasis and
Crohn’s disease and detecting the candidate genes involved in the processes common for
both pathologies and critical for their development. Identification of the proteins encoded
by these genes is an important aspect of the research performed, because the proteins are
particular targets for elaborating new approaches to treating psoriasis and Crohn’s disease.
Our data obtained by analyzing expression of the candidate genes for psoriasis and Crohn’s
disease can enhance the search for new biological targets for the corresponding therapeutics.
In order to gain insight into molecular machinery underlying the disease, we conducted a
comprehensive meta-analysis of proteomics and transcriptomics of psoriatic lesions from
independent studies. Network-based analysis revealed similarities in regulation at both
proteomics and transcriptomics level. We identified a group of transcription factors
responsible for overexpression of psoriasis genes and a number of previously unknown
signaling pathways that may play a role in this process. We also evaluated functional
synergy between transcriptomics and proteomics results.
We have successfully applied network-based methods to integrate and explore two distinct
high-throughput disease data sets of different origin and size. Through identification of
common regulatory machinery that is likely to cause overexpression of genes and proteins,
we came to the signaling pathways that might contribute to the altered state of regulatory
network in psoriatic lesion. Our approach allows easy integrative investigation of different
data types and produces biologically meaningful results, leading to new potential therapy
targets. We have demonstrated that pathology can be caused and maintained by a great
amount of various cascades, many previously not described as implicated in psoriasis;
therefore, combined therapies targeting multiple pathways might be effective in treatment.
412 Computational Biology and Applied Bioinformatics

4. Acknowledgment
This work was supported by grants 16.512.11.2049 from Ministry of Education and Science
of the Russian Federation, grant from Russian Academy of Sciences ("Fundamental Sciences
to Medicine" program), and grant P1309 from Ministry of Education and Science of the
Russian Federation (Federal Special Program "Scientific and Educational Human Resources
of Innovative Russia" for 2009 - 2013)

5. Reference
Arnold I, Watt FM: c-Myc activation in transgenic mouse epidermis results in mobilization
of stem cells and differentiation of their progeny. Curr Biol 2001, 11(8):558-68.
Asadullah K, et al.: IL-10 is a key cytokine in psoriasis. Proof of principle by IL-10 therapy: a
new therapeutic approach. J Clin Invest 1998, 101(4):783-94.
Benjamini Y., Hochberg Y. 1995. Controlling the false discovery rate: A practical and
powerful approach to multiple testing. J. Royal Stat. Soc. Ser. B, Methodol. 57, 289–
300.
Bhavnani SK, et al.: Network analysis of genes regulated in renal diseases: implications for a
molecular-based classification. BMC Bioinformatics 2009, 10(Suppl 9):S3.
Binder V. 2004. Epidemiology of IBD during the twentieth century: An integrated view. Best
Pract. Res. Clin. Gastroenterol. 18, 463–479.
Boehncke WH, et al.: Pulling the trigger on psoriasis. Nature 1996, 379(6568):777.
Bowcock A., Cookson W. 2004. The genetics of psoriasis, psoriatic arthritis and atopic
dermatitis. Hum. Mol. Genet. 13, 43–55.
Bowcock AM, et al.: Insights into psoriasis and other inflammatory diseases from large-scale
gene expression studies. Hum Mol Genet 2001, 10(17):1793-805.
Cancino-Diaz JC, et al.: Interleukin-13 receptor in psoriatic keratinocytes: overexpression of
the mRNA and underexpression of the protein. J Invest Dermatol 2002, 119(5):1114-
20.
Castelijns FA, et al.: The epidermal phenotype during initiation of the psoriatic lesion in the
symptomless margin of relapsing psoriasis. J Am Acad Dermatol 1999, 40(6 Pt
1):901-9.
Chen YR, et al.: Quantitative proteomic and genomic profiling reveals metastasis-related
protein expression patterns in gastric cancer cells. J Proteome Res 2006, 5(10):2727-
42.
Chu A, et al.: Tissue specificity of E- and P-selectin ligands in Th1- mediated chronic
inflammation. J Immunol 1999, 163(9):5086-93.
Chuang HY, et al.: Network-based classification of breast cancer metastasis. Mol Syst Biol
2007, 3:140.
Conway JP, Kinter M: Proteomic and transcriptomic analyses of macrophages with an
increased resistance to oxidized low density lipoprotein (oxLDL)-induced
cytotoxicity generated by chronic exposure to oxLDL. Mol Cell Proteomics 2005,
4(10):1522-40.
Cox B, Kislinger T, Emili A: Integrating gene and protein expression data: pattern analysis
and profile mining. Methods 2005, 35(3):303-14.
Crohn B., Ginzburg L., Oppenheimer G. 1932. Regional ileitis: A pathological and clinical
entity. J. Actinomycetes. Med. Assoc. 251, P73–P79 (1984).
Analysis of Transcriptomic and Proteomic Data in Immune-Mediated Diseases 413

Curry JL, et al.: Innate immune-related receptors in normal and psoriatic skin. Arch Pathol
Lab Med 2003, 127(2):178-86.
de Groot M, et al.: Expression of the chemokine receptor CCR5 in psoriasis and results of a
randomized placebo controlled trial with a CCR5 inhibitor. Arch Dermatol Res
2007, 299(7):305-13.
De Rie MA, et al.: Expression of the T-cell activation antigens CD27 and CD28 in normal and
psoriatic skin. Clin Exp Dermatol 1996, 21(2):104-11.
Debets R, et al.: The IL-1 system in psoriatic skin: IL-1 antagonist sphere of influence in
lesional psoriatic epidermis. J Immunol 1997, 158(6):2955-63.
Dezso Z, et al.: Identifying disease-specific genes based on their topological significance in
protein networks. BMC Syst Biol 2009, 3:36
Di Pietro C, et al.: The apoptotic machinery as a biological complex system: analysis of its
omics and evolution, identification of candidate genes for fourteen major types of
cancer, and experimental validation in CML and neuroblastoma. BMC Med
Genomics 2009, 2(1):20.
Draghici S., Khatri P., Tarca A.L., Amin K., Done A., Voichita C., Georgescu C., Romero R.
2007. A systems biology approach for pathway level analysis. Genome Res. 17,
1537–1545.
Ellis CN, Krueger GG: Treatment of chronic plaque psoriasis by selective targeting of
memory effector T lymphocytes. N Engl J Med 2001, 345(4):248-55.
Foell D, et al.: Expression of the pro-inflammatory protein S100A12 (ENRAGE) in
rheumatoid and psoriatic arthritis. Rheumatology (Oxford) 2003, 42(11):1383-9.
Fu X, et al.: Estimating accuracy of RNA-Seq and microarrays with proteomics. BMC
Genomics 2009, 10(1):161.
Gandarillas A, Watt FM: c-Myc promotes differentiation of human epidermal stem cells.
Genes Dev 1997, 11(21):2869-82.
Gerling IC, et al.: New data analysis and mining approaches identify unique proteome and
transcriptome markers of susceptibility to autoimmune diabetes. Mol Cell
Proteomics 2006, 5(2):293-305.
Ghavami S, et al.: S100A8/A9 at low concentration promotes tumor cell growth via RAGE
ligation and MAP kinase-dependent pathway. Journal of leukocyte biology 2008,
83(6):1484-1492
Ghoreschi K, Mrowietz U, Rocken M: A molecule solves psoriasis? Systemic therapies for
psoriasis inducing interleukin 4 and Th2 responses. J Mol Med 2003, 81(8):471-80.
Ghosh D, et al.: Statistical issues and methods for meta-analysis of microarray data: a case
study in prostate cancer. Funct Integr Genomics 2003, 3(4):180-8.
Gisondi P, Girolomoni G: Biologic therapies in psoriasis: a new therapeutic approach.
Autoimmun Rev 2007, 6(8):515-9.
Gravel P, Golaz O: Two-Dimensional PAGE Using Carrier Ampholyte pH Gradients in the
First Dimension. The Protein Protocols Handbook 1996:127-132.
Gu J, et al.: A 588-gene microarray analysis of the peripheral blood mononuclear cells of
spondyloarthropathy patients. Rheumatology (Oxford) 2002, 41(7):759-66
Guttman-Yassky E, et al.: Blockade of CD11a by efalizumab in psoriasis patients induces a
unique state of T-cell hyporesponsiveness. J Invest Dermatol 2008, 128(5):1182-91.
Habermann JK, et al.: Stage-specific alterations of the genome, transcriptome, and proteome
during colorectal carcinogenesis. Genes Chromosomes Cancer 2007, 46(1):10-26.
414 Computational Biology and Applied Bioinformatics

Hack CJ: Integrated transcriptome and proteome data: the challenges ahead. Brief Funct
Genomic Proteomic 2004, 3(3):212-9.
Haider AS, et al.: Novel insight into the agonistic mechanism of alefacept in vivo:
differentially expressed genes may serve as biomarkers of response in psoriasis
patients. J Immunol 2007, 178(11):7442-9.
Horuk R: BX471: a CCR1 antagonist with anti-inflammatory activity in man. Mini Rev Med
Chem 2005, 5(9):791-804.
Ideker T, Sharan R: Protein networks in disease. Genome Res 2008, 18(4):644-52.
Johansen C, et al.: Characterization of the interleukin-17 isoforms and receptors in lesional
psoriatic skin. Br J Dermatol 2009, 160(2):319-24.
Kelly R, Marsden RA, Bevan D: Exacerbation of psoriasis with GM-CSF therapy. Br J
Dermatol 1993, 128(4):468-9.
Le Naour F, et al.: Profiling changes in gene expression during differentiation and
maturation of monocyte-derived dendritic cells using both oligonucleotide
microarrays and proteomics. J Biol Chem 2001, 276(21):17920-31.
Leigh IM, et al.: Keratins (K16 and K17) as markers of keratinocyte hyperproliferation in
psoriasis in vivo and in vitro. The British journal of dermatology 1995, 133(4):501-
511.
Lohwasser C, et al.: The receptor for advanced glycation end products is highly expressed in
the skin and upregulated by advanced glycation end products and tumor necrosis
factor-alpha. J Invest Dermatol 2006, 126(2):291-9.
Madsen P, et al.: Molecular cloning, occurrence, and expression of a novel partially secreted
protein "psoriasin" that is highly up-regulated in psoriatic skin. The Journal of
investigative dermatology 1991, 97(4):701-712.
Martin R: Interleukin 4 treatment of psoriasis: are pleiotropic cytokines suitable therapies for
autoimmune diseases? Trends Pharmacol Sci 2003, 24(12):613-6.
Menezes R, et al.: Integrated analysis of DNA copy number and gene expression microarray
data using gene sets. BMC Bioinformatics 2009, 10(1):203.
Mijalski T, et al.: Identification of coexpressed gene clusters in a comparative analysis of
transcriptome and proteome in mouse tissues. Proceedings of the National
Academy of Sciences of the United States of America 2005, 102(24):8621-8626
Mortz E, et al.: Improved silver staining protocols for high sensitivity protein identification
using matrix-assisted laser desorption/ionization-time of flight analysis.
Proteomics 2001, 1(11):1359-63
Nikolskaya T, et al.: Network analysis of human glaucomatous optic nerve head astrocytes.
BMC Med Genomics 2009, 2:24.
Nikolsky Y, et al.: Genome-wide functional synergy between amplified and mutated genes
in human breast cancer. Cancer research 2008, 68(22):9532-9540.
Nikolsky Y, et al.: Functional analysis of OMICs data and small molecule compounds in an
integrated "knowledge-based" platform. Methods in molecular biology (Clifton,
N.J.) 2009, 563:177-196.
Nikolsky Y, Nikolskaya T, Bugrim A: Biological networks and analysis of experimental data
in drug discovery. DrugDiscov Today 2005, 10(9):653-62.
Oestreicher JL, et al.: Molecular classification of psoriasis diseaseassociated genes through
pharmacogenomic expression profiling. Pharmacogenomics J 2001, 1(4):272-87.
Ortonne JP: Aetiology and pathogenesis of psoriasis. Br J Dermatol 1996, 135(Suppl 49):1-5.
Analysis of Transcriptomic and Proteomic Data in Immune-Mediated Diseases 415

Pastore S, et al.: Biological drugs targeting the immune response in the therapy of psoriasis.
Biologics 2008, 2(4):687-97.
Patterson AM, et al.: Differential expression of syndecans and glypicans in chronically
inflamed synovium. Ann Rheum Dis 2008, 67(5):592-601.
Penna G, et al.: Expression of the inhibitory receptor ILT3 on dendritic cells is dispensable
for induction of CD4+Foxp3+ regulatory T cells by 1,25-dihydroxyvitamin D3.
Blood 2005, 106(10):3490-7.
Pietrzak A, et al.: Genes and structure of selected cytokines involved in pathogenesis of
psoriasis. Folia Histochem Cytobiol 2008, 46(1):11-21.
Pihur V, Datta S: RankAggreg, an R package for weighted rank aggregation. BMC
Bioinformatics 2009, 10:62.
Piruzian E.S., Nikolskaya T.A., Abdeev R.M., Bruskin S.A. 2007. Transcription factor AP-1
components as candidate genes involved in the development of psoriatic process.
Mol. Biol. 41, 1069–1080.
Piruzian ES, et al.: [The comparative analysis of psoriasis and Crohn disease molecular-
genetical processes under pathological conditions]. Mol Biol (Mosk) 2009, 43(1):175-
9.
Prens E, et al.: Adhesion molecules and IL-1 costimulate T lymphocytes in the autologous
MECLR in psoriasis. Arch Dermatol Res 1996, 288(2):68-73.
Quekenborn-Trinquet V, et al.: Gene expression profiles in psoriasis: analysis of impact of
body site location and clinical severity. Br J Dermatol 2005, 152(3):489-504.
Reichrath J, et al.: Expression of integrin subunits and CD44 isoforms in psoriatic skin and
effects of topical calcitriol application. J Cutan Pathol 1997, 24(8):499-506.
Reischl J, et al.: Increased expression of Wnt5a in psoriatic plaques. J Invest Dermatol 2007,
127(1):163-9.
Rottman JB, et al.: Potential role of the chemokine receptors CXCR3, CCR4, and the integrin
alphaEbeta7 in the pathogenesis of psoriasis vulgaris. Lab Invest 2001, 81(3):335-47.
Sano S, Chan KS, DiGiovanni J: Impact of Stat3 activation upon skin biology: a dichotomy of
its role between homeostasis and diseases. J Dermatol Sci 2008, 50(1):1-14.
Santilli F, et al.: Soluble forms of RAGE in human diseases: clinical and therapeutical
implications. Curr Med Chem 2009, 16(8):940-52.
Sartor R. 2006. Mechanisms of disease: Pathogenesis of Crohn’s disease and ulcerative
colitis. Nature Clin. Pract. Gastroenterol. Hepatol. 3, 390–407.
Schreiber J, et al.: Coordinated binding of NF-kappaB family members in the response of
human cells to lipopolysaccharide. Proc Natl Acad Sci USA 2006, 103(15):5899-904
Schreiber S., Rosenstiel P., Albrecht M., Hampe J., Krawczak M. 2005. Genetics of Crohn
disease, an archetypal inflammatory barrier disease. Nature Rev. Genet. 6, 376–388.
Schulz BS, et al.: Increased expression of epidermal IL-8 receptor in psoriasis. Down-
regulation by FK-506 in vitro. J Immunol 1993, 151(8):4399-406.
Seung NR, et al.: Comparison of expression of heat-shock protein 60, Toll-like receptors 2
and 4, and T-cell receptor gammadelta in plaque and guttate psoriasis. J Cutan
Pathol 2007, 34(12):903-11.
Shachaf CM, et al.: Genomic and proteomic analysis reveals a threshold level of MYC
required for tumor maintenance. Cancer Res 2008, 68(13):5132-42.
Shiina T, et al.: The HLA genomic loci map: expression, interaction, diversity and disease. J
Hum Genet 2009, 54(1):15-39.
416 Computational Biology and Applied Bioinformatics

Sjogren F, et al.: Expression and function of beta 2 integrin CD11B/CD18 on leukocytes from
patients with psoriasis. Acta Derm Venereol 1999, 79(2):105-10.
Steiling K, et al.: Comparison of proteomic and transcriptomic profiles in the bronchial
airway epithelium of current and never smokers. PLoS One 2009, 4(4):e5043.
Takeda A, et al.: Overexpression of serpin squamous cell carcinoma antigens in psoriatic
skin. J Invest Dermatol 2002, 118(1):147-54.
Tsuruta D: NF-kappaB links keratinocytes and lymphocytes in the pathogenesis of psoriasis.
Recent Pat Inflamm Allergy Drug Discov 2009, 3(1):40-8.
Turpaev K.T. 2006. Role of transcription factor AP-1in integration of intracellular signal
pathways. Mol. Biol. 40, 945–961.
Vestergaard C, et al.: Expression of CCR2 on monocytes and macrophages in chronically
inflamed skin in atopic dermatitis and psoriasis. Acta Derm Venereol 2004,
84(5):353-8.
Vissers WH, et al.: Memory effector (CD45RO+) and cytotoxic (CD8+) T cells appear early in
the margin zone of spreading psoriatic lesions in contrast to cells expressing natural
killer receptors, which appear late. Br J Dermatol 2004, 150(5):852-9.
Vorum H, et al.: Expression and divalent cation binding properties of the novel chemotactic
inflammatory protein psoriasin. Electrophoresis 1996, 17(11):1787-96.
Wakita H, Takigawa M: E-selectin and vascular cell adhesion molecule-1 are critical for
initial trafficking of helper-inducer/memory T cells in psoriatic plaques. Arch
Dermatol 1994, 130(4):457-63.
Warnat P, Eils R, Brors B: Cross-platform analysis of cancer microarray data improves gene
expression based classification of phenotypes. BMC Bioinformatics 2005, 6:265.
Welch B.L. 1947. The generalization of “student’s” problem when several different
population variances are involved. Biometrika. 34, 28–35.
Wise LH, Lanchbury JS, Lewis CM: Meta-analysis of genome searches. Ann Hum Genet
1999, 63(Pt 3):263-72.
Yao Y., Richman L., Morehouse C., de los Reyes M., Higgs B.W., Boutrin A., White B., Coyle
A., Krueger J., Kiener P.A., Jallal B. 2008. Type I interferon: Potential therapeutic
target for psoriasis? PLoS ONE. 3, e2737.
Zhao C, et al.: Identification of Novel Functional Differences in Monocyte Subsets Using
Proteomic and Transcriptomic Methods. Journal of Proteome Research 2009,
8(8):4028-4038
Zheng PZ, et al.: Systems analysis of transcriptome and proteome in retinoic acid/arsenic
trioxide-induced cell differentiation/apoptosis of promyelocytic leukemia. Proc
Natl Acad Sci USA 2005, 102(21):7653-8.
Zhou X, et al.: Novel mechanisms of T-cell and dendritic cell activation revealed by profiling
of psoriasis on the 63,100-element oligonucleotide array. Physiol Genomics 2003,
13(1):69-78.
 22

Emergence of the Diversified Short ORFeome

by Mass Spectrometry-Based Proteomics
Hiroko Ao-Kondo, Hiroko Kozuka-Hata and Masaaki Oyama
Medical Proteomics Laboratory, Institute of Medical Science, University of Tokyo
Japan

1. Introduction
In proteomics analyses, protein identification by mass spectrometry (MS) is usually
performed using protein sequence databases such as RefSeq (NCBI;
http://www.ncbi.nlm.nih.gov/RefSeq/), UniProt (http://www.uniprot.org/) or IPI
(http://www.ebi.ac.uk/IPI/IPIhelp.html). Because these databases usually target the
longest (main) open reading frame (ORF) in the corresponding mRNA sequence, whether
shorter ORFs on the same mRNA are actually translated still shrouds in mystery. In the first
place, it had been considered that almost all eukaryotic mRNAs contains only one ORF and
functions as monocistronic mRNAs. It is now known, however, that some eukaryotic
mRNAs had multiple ORFs, which are recognized as polycistronic mRNAs. One of the well-
known extra ORFs is an upstream ORF (uORF) and it functions as regulators of mRNA
translation (Diba et al., 2001; Geballe & Morris, 1994; Morris & Geballe, 2000; Vilela &
McCarthy, 2003; Zhang & Dietrich, 2005). For getting clues to the mystery of diversified
short ORFs, full-length mRNA sequence databases with complete 5‘-untranslated regions
(5‘-UTRs) were essentially needed (Morris & Geballe, 2000; Suzuki et al., 2001).
The oligo-capping method was developed to construct full-length cDNA libraries
(Maruyama & Sugano, 1994) and the corresponding sequence were stored into the database
called DBTSS (DataBase of Transcriptional Start Site; http://dbtss.hgc.jp/) (Suzuki et al.,
1997, 2002, 2004; Tsuchihara et al., 2009; Wakaguri et al., 2008; Yamashita et al., 2006).
Comparing the dataset of DBTSS with the corresponding RefSeq entries, it was found that
about 50 % of the RefSeq entries had at least one upstream ATG (uATG) except the
functional ATG initiator codon (Yamashita et al., 2003). Although it had been suggested that
upstream AUGs (uAUGs) and uORFs play important roles for translation of the main ORF,
none of the proteins from these uORFs was detected in biological experiments in vivo. Our
previous proteomics analysis focused on small proteins revealed the first evidence of the
existence of four novel small proteins translated from uORFs in vivo using highly sensitive
nanoflow liquid chromatography (LC) coupled with the electrospray ionization-tandem
mass spectrometry (ESI-MS/MS) system (Oyama et al., 2004). Large-scale analysis based on
in-depth separation by two-dimensional LC also led to the identification of additional eight
novel small proteins not only from uORFs but also from downstream ORFs and one of them
was found to be translated from a non-AUG initiator codon (Oyama et al., 2007). Finding of
these novel small proteins indicate the possibility of diverse control mechanisms of
translation initiation.
418 Computational Biology and Applied Bioinformatics

In this chapter, we first introduce widely-recognized mechanism of translation initiation and

functional roles of uORF in translational regulation. We then review how we identified
novel small proteins with MS and lastly discuss the progress of bioinformatical analyses for
elucidating the diversification of short coding regions defined by the transcriptome.

2. Translational regulation by short ORFs

It is well known that 5‘-UTRs of some mRNAs contain functional elements for translational
regulation defined by uAUG and uORF. In this section, we show how uAUG and uORF
have biological consequences for protein synthesis on eukaryotic mRNAs.

2.1 Outline of translation initiation

Initiation of translation on eukaryotic mRNAs occurs roughly as follows (Fig. 1) (Kozak,
1989, 1991, 1999).
1. A small (40S) ribosomal subunit binds near the 5‘-end of mRNA, i.e. the cap structure.
2. The 40S subunit migrates linearly downstream of the 5‘-UTR until it encounters the
optimum AUG initiator codon.
3. A large (60S) ribosomal subunit joins the paused 40S subunit.
4. The complete ribosomal complex (40S + 60S) starts protein synthesis.

Fig. 1. The proposed procedure for initiation of translation in eukaryotes. The black region
indicates the main ORF of the mRNA.
Emergence of the Diversified Short
ORFeome by Mass Spectrometry-Based Proteomics 419

In addition to the above mechanism, initiation of translation without the step of ribosome
scanning is also known. It is called “internal initiation”, which depends on some particular
structure on an mRNA termed internal ribosome entry site (IRES).

2.2 The relationship between uORF and main ORF

In case that an mRNA contains a uORF, two models for the initiation of translation are
suggested (Fig. 2) (Hatzigeorgiou, 2002). One is called ”leaky scanning” and the other
is ”reinitiation”. If the first AUG codon is in an unfavorable sequence context defined by
Kozak (see the section 3.2), a small ribosomal subunit (40S) ignores the first AUG and
initiates translation from a more favorable AUG codon downstream located. This
phenomenon is known as ”leaky scanning” (Fig. 2-(A)). In case that a complete ribosomal
complex translates a main ORF after termination of translation of the uORF on the same
mRNA, it is termed ”reinitiation” (Fig. 2-(B)).

Fig. 2. The irregular models of ribosome scanning on eukaryotic mRNAs.

(A) Leaky scanning and (B) Reinitiation. Gray regions indicate uORFs on the mRNA,
whereas black ones represent the main ORFs.
The relations between two ORFs are classified into three types as follows; (1) A distant type;
in-frame/out-of-frame, (2) A contiguous type; in-frame and (3) An overlapped type; in-
frame/out-of-frame (Fig. 3). In-frame means that a uORF and the main ORF are on the same
frame of the mRNA sequence, whereas out-of-frame means that they are on the different
frame. According to the previous analysis of the accumulated 5‘-end sequence data, the
average size of uORF was estimated at 31 amino acids and 20 % of ORFs were categorized
into Type (3) (Yamashita et al., 2003).
420 Computational Biology and Applied Bioinformatics

Fig. 3. The location of a uORF and the main ORF on the mRNA.
(1) A distant type, (2) A contiguous type and (3) An overlapped type. Types (1) and (3) have
two subtypes based on the frames of two ORFs. One is defined by the same reading frame
(in-frame) and the other is by the different one (out-of-frame). Gray and black regions
indicate uORFs and the main ORFs on mRNAs, respectively, whereas a blue one represents
an overlap.
These different relations might bring about different events in initiating translation. In
eukaryotes, it has a tendency to increase an efficiency of reinitiation if the distance between
a uORF and the main ORF is long (Kozak, 1991; Meijer & Thomas, 2002; Morris & Geballe,
2000). Therefore, the ORFs classified as Types (2) and (3) would be difficult to be regulated
by reinitiation. It is also said that reinitiation occurs only when the length of uORF is short
(Kozak, 1991), whereas the sequence context of an inter-ORF‘s region, that of upstream of
uORF, uORF itself and even the main ORF can also affect reinitiation (Morris & Geballe,
2000). On the contrary, the ORFs of Type (3) might easily cause leaky scanning (Geballe &
Morris, 1994; Yamashita et al., 2003). As a special case, when a termination codon of the
uORF is near the AUG initiator codon of the downstream ORF, within about 50 nucleotides,
ribosomes could scan backwards and reinitiate translation from the AUG codon of the
downstream ORF (Peabody et al., 1986).

2.3 The role of short ORFs in translation regulation

The 5‘-UTR elements such as uAUGs and uORFs are well known as important regulators for
translation initiation. In case of some genes that have multiple uORFs, considerably different
effects can be generated on the translation of the main ORF depending on which
combination of uORFs is translated. Some uORFs seem to promote reinitiation of the main
ORFs and the others seem to inhibit it. It is supposed that these effects are caused by the
nucleotide sequences of the 3‘ ends of the uORFs, that of uORFs or protein products
encoded by uORFs. Such differential enhancement of translation are considered to be one of
the responses of adaptation to the environment (Altmann & Trachsel, 1993; Diba et al., 2001;
Emergence of the Diversified Short
ORFeome by Mass Spectrometry-Based Proteomics 421

Geballe & Morris, 1994; Hatzigeorgiou, 2002; Iacono et al., 2005; Meijer & Thomas, 2002;
Morris & Geballe, 2000; Vilela & McCarthy, 2003; Wang & Rothnagel, 2004; Zhang &
Dietrich, 2005). In addition to that, various factors or events are known to influence on the
translational inhibition of the main ORF; the presence of arginine, a stalling of a ribosomal
complex at the termination or an interaction between a ribosomal complex and the peptide
encoded by the uORF, which indicates that down-regulated controls by uORFs are general
(Diba et al., 2001; Geballe & Morris, 1994; Iacono et al., 2005; Meijer & Thomas, 2002; Morris
& Geballe, 2000; Vilela & McCarthy, 2003; Zhang & Dietrich, 2005).
As for downstream ORFs, there is also a report that a peptide encoded in the 3‘-UTR may be
expressed (Rastinejad & Blau, 1993). However, whether and how the peptides control the
translation initiation of the main ORF is still unknown.

3. Variability of translation start sites

How a ribosomal complex (40S + 60S) recognizes an initiator codon on the mRNA is a
matter of vital importance for defining the proteome. Here we present a part of already
proposed elements for regulation of translation initiation.

3.1 The first-AUG rule

Traditionally, the first-AUG rule is widely recognized for initiation of translation (Kozak,
1987, 1989, 1991). It states that ribosomes start translation from the first-AUG on the
corresponding mRNA. Although this rule is not absolute, 90-95 % of vertebrate ORFs was
established by the first AUG codon on the mRNA (Kozak, 1987, 1989, 1991). Our previous
proteomics analysis of small proteins also indicated that about 84 % of proteins in RefSeq
were translated from the first AUG of the corresponding mRNAs (Oyama et al., 2004). On
the other hand, there are also many negative reports concerning the rule; 29 % of cDNA
contained at least one ATG codon in their 5‘-UTR (Suzuki et al., 2000); 41 % of transcripts
had more than one uAUG and 24 % of genes had more than two uAUGs (Peri & Pandey,
2001); about 50 % of the RefSeq entries had at least one uAUG (Yamashita et al., 2003); about
44 % of 5‘-UTRs had uAUGs and uORFs (Iacono et al., 2005). There are also some reports
that the first AUG is skipped if it is too close to the cap structure, within 12 (Kozak, 1991) to
14 (Sedman et al., 1990) nucleotides (see the section 3.3). In this chapter, we cited a variety of
statistical data on the UTRs. Because they are based on different versions or generations of
sequence databases, the data vary widely (Meijer & Thomas, 2002), which is the point to be
properly considered.

3.2 Kozak’s consensus sequence

The strongest bias for initiation of translation in vertebrates is the sequence context called
“Kozak’s sequence”, known as GCCA/GCCATGG (Kozak, 1987). The nucleotides in
positions -3 (A or G) and +4 (G) are highly conserved and greatly effective for a ribosomal
complex to start translation (Kozak, 1987, 2002; Matsui et al., 2007; Suzuki et al., 2001; Wang
& Rothnagel, 2004). The context of an AUG codon in position -3 is the most highly
conserved and functionally the most important; it is regarded as strong or optimal only
when this position matches A or G, and that in position +4 is also highly conserved (Kozak,
2002). Some reports mentioned that only 0.86 % (Kozak, 1987) to 6 % (Iacono et al., 2005) of
functional initiator codons lacked Kozak’s sequence in positions -3 and +4, whereas 37 %
422 Computational Biology and Applied Bioinformatics

(Suzuki et al., 2000) to 46 % (Kozak, 1987) of uATGs would be skipped because of

unfavorable Kozak’s sequence in both of the positions. On the contrary, another report
mentioned that most initiator codons were not in close agreement with Kozak’s consensus
sequence (Peri & Pandey, 2001).

3.3 The length of the 5'-UTR

The length of 5'-UTR is also effective when translation occurs from an AUG codon near the 5’
end of the mRNA (Kozak, 1991; Sedman et al., 1990). About half of ribosomes skip an AUG
codon even in an optimal context if the length of 5‘-UTR is less than 12 nucleotides (mentioned
in the section 3.1) and this type of leaky scanning can be reduced if the length of 5‘-UTR is
more than or equal to 20 nucleotides (Kozak, 1991). In the traditional analysis based on
incomplete 5‘-UTR sequences, the distance from the 5' end to the AUG initiator codon in
vertebrate mRNAs was generally from 20 and 100 nucleotides (Kozak, 1987). The previous
analysis using RefSeq human mRNA sequences indicated that 85 % of 5‘-UTR sequences less
than 100 nucleotides contain no uAUGs (Peri & Pandey, 2001). The evidence convinced us that
the first-AUG rule was widely supported in eukaryotes. In the recent analysis based on full-
length 5‘-UTR sequences, it is 125 nucleotides long on average (Suzuki et al., 2000) and
transcriptional start sites (TSSs) vary widely (Carninci et al., 2006; Kimura et al., 2006; Suzuki
et al., 2001). The average scattered length of 5'-UTR was more than 61.7 nucleotides, with a
standard deviation of 19.5 nucleotides (Suzuki et al., 2001) and 52 % of the human RefSeq
genes contained 3.1 TSS clusters on average (Kimura et al., 2006), which has an over 500
nucleotides interval (Fig. 4). In protein-coding genes, differentially regulated alternative TSSs
are common (Carninci et al., 2006). Because the diversity of transcription initiation greatly
affects the length of the 5'-UTR, there remain some doubts whether the length of the 5'-UTR
contributes to the efficiency of translation initiation. There is also a report that the degree of
leaky scanning is not affected by the length of 5‘-UTR (Wang & Rothnagel, 2004).

Fig. 4. The schematic representation of the 5‘ ends of the TSSs.

Each TSS cluster consists of at least one TSS and has an over 500 nucleotides interval.

3.4 non-AUG initiator codon

In the general translation model, a non-AUG codon is considered to be ignored by
ribosomes unless a downstream AUG codon is in a relatively weak context (Geballe &
Morris, 1994; Kozak, 1999). In case that an upstream non-AUG codon, such as ACG, CUG or
GUG, satisfies Kozak’s consensus sequence, it possibly functions as an initiator of
translation in addition to the first AUG initiator codon (Kozak, 1999, 2002). Besides Kozak’s
consensus sequence, downstream stem-and-loop and highly structured GC-rich context in
the 5‘-UTR could enhance translation initiation from a non-AUG codon (Kozak, 1991, 2002).
Emergence of the Diversified Short
ORFeome by Mass Spectrometry-Based Proteomics 423

4. Protein identification by MS
The recent progress of proteomic methodologies based on highly sensitive liquid
chromatography-tandem mass spectrometry (LC-MS/MS) technology have enabled us to
identify hundreds or thousands of proteins in a single analysis.
We succeeded in the discovery of novel small proteins translated from short ORFs using direct
nanoflow LC-MS/MS system (Oyama et al., 2004, 2007). Among 54 proteins less than 100
amino acids that were identified by retrieving several sequence databases with a
representative search engine, Mascot (Matrix Science; http://www.matrixscience.com/),
four ones were turned out to be encoded in 5‘-UTRs (Oyama et al., 2004). This showed the first
direct evidence of peptide products from the uORFs actually translated in human cells. In the
subsequent analysis using more sophisticated two-dimensional LC system, we also discovered
eight novel small proteins (Oyama et al., 2007), five of which were encoded in the 5‘-UTR and
three were encoded in the 3‘-UTR of the corresponding mRNA. Even based on the
accumulated DBTSS data, two ORFs had no putative AUG codon, which indicated the
possibility that they were translated from non-AUG initiator codon. In the article above, 197
proteins less than 20 kDa were identified by Mascot. The procedure for identifying novel
proteins by MS is described as follows.

Fig. 5. The procedure for preparing samples for proteomic analyses of small proteins.
424 Computational Biology and Applied Bioinformatics

4.1 Materials and methods

The proteins included in cultured cell lysates were first separated according to their size.
Small protein-enriched fraction through acid extraction and SDS-PAGE were treated with
enzymes. In case of SDS-PAGE, the digested peptides were extracted from the gel. The
samples were desalted and concentrated to introduce into the MS system. The schematic
procedure is shown in Fig. 5.

4.2 Protein identification

The samples were analyzed using nanoflow LC-MS/MS system. The purified peptides were
eluted with a linear gradient of acetonitrile and sprayed into the high-resolution tandem
mass spectrometer. Acquired tandem mass (MS/MS) spectra were then converted to text
files and processed against sequence databases using Mascot. Based on the principle that
each peptide has a MS/MS spectrum with unique characteristics, the search engine
compares measured data on precursor/product ions with those theoretically calculated
from protein sequence data (Fig. 6). The MS/MS spectrum file contains mass to charge ratio
(m/z) values of precursor and product ions along with their intensity. The measured
spectrum lists are searched against sequence databases to identify the corresponding
peptide in a statistical manner. The theoretical spectrum lists are totally dependent on the
contents of sequence databases themselves.

Fig. 6. The principle of protein identification.

A search engine compares measured MS/MS spectrum lists with theoretical ones. The
acquired MS/MS spectra are converted into a text file that is composed of precursor ion data
and product ion data, in a format defined by the search engine. Product ion data usually
consist of multiple pairs of m/z and its intensity. The theoretical m/z values are calculated
virtually.
Emergence of the Diversified Short
ORFeome by Mass Spectrometry-Based Proteomics 425

4.3 Finding of novel small proteins

For exploring novel small proteins, two types of sequence databases were used; one was an
artificial database computationally translated from the cDNA sequences in all the reading
frames and the other was an already established protein database. In order to process the
comparison of the large-scale protein identification data from the two kinds of databases,
several Perl scripts have been developed based on the definition that candidates of novel
small proteins were identified only in the cDNA database(s) (Fig. 7). In a result datasheet
using RefSeq sequences, each protein was annotated with NM numbers for the cDNA
database and with NP numbers for the protein database. The Perl scripts then exchanged
NM to NP numbers and evaluated them.

Fig. 7. The algorithm to compare the lists of search results using RefSeq cDNA and protein
databases.
The proteins identified from the cDNA database are annotated with NM numbers, whereas,
those from the protein database are with NP numbers. To compare these results, it is needed
to exchange NM to NP numbers. The NP numbers annotated only from the cDNA database
are considered to be candidates of novel proteins.

5. Bioinformatics approach
In order to forward MS-based identification of novel coding regions of mRNAs, MS systems,
sequence databases and bioinformatics methodologies are required to improve together.
426 Computational Biology and Applied Bioinformatics

Regarding bioinformatics, two aspects seem to be demanded; one is for retrieving target
proteins from an enormous size of database searching results, the other is for constructing
platforms to predict novel coding sequences (CDSs).

5.1 Contribution of sequence databases & bioinformatics to MS-based proteomics

The recent advances in MS-based proteomics technology have enabled us to perform large-
scale protein identification with high sensitivity. The accumulation of well-established
sequence databases also made a great contribution to efficient identification in proteomics
analyses. One of the representative databases is a specialized 5‘-end cDNA database like
DBTSS and the other is a series of whole genome sequence databases for various species. To
investigate the mechanisms in transcriptional control, DBTSS has lately attracted considerable
attention because it contains accumulated information on the transcriptional regulation of each
gene (Suzuki et al., 2002, 2004; Tsuchihara et al., 2009; Wakaguri et al., 2008; Yamashita et al.,
2006). Based on the accumulated data, the diverse distribution of TSSs was clearly indicated
(Kimura et al., 2006; Suzuki et al., 2000, 2001). On the other hand, many whole genome
sequencing projects are progressing all over the world (GOLD: Genomes Online Database;
http://www.genomesonline.org/). Complement and maintenance of sequence databases for
various species must help to find more novel proteins across the species. For example, there
are several reports that conducted bioinformatical approaches to explore novel functional
uORFs by comparing the 5'-UTR regions of orthologs based on multiple sequence alignments
(Zhang & Dietrich, 2005), using ORF Finder (http://bioinformatics.org/sms/orf find.html)
and a machine learning technique, inductive logic programming (ILP) with biological
background knowledge (Selpi et al., 2006), or applying comparative genomics and a heuristic
rule-based expert system (Cvijovic et al., 2007). Using advanced sequence databases, new
protein CDSs were added as a result of the prediction by various algorithms (e.g.
Hatzigeorgiou, 2002; Ota et al., 2004). Based on the well-established cDNA databases, MS
could evaluate whether these CDSs are actually translated in a high-throughput manner.
Construction of more detailed sequence databases will lead to detection of more novel small
proteins in the presumed 5'-UTRs (Oyama et al., 2004). To make good use of those exhaustive
sequence databases, bioinformatical techniques, especially data mining tools such as search
engines to retrieve target proteins from an enormous size of database search results, are
obviously indispensable.

5.2 Contribution of MS-based proteomics to sequence databases & bioinformatics

In addition to the technological progress of MS, sequence databases and data mining tools,
development of other bioinformatical techniques called prediction tools, are also important.
Ad-hoc algorithms for predicting new CDSs, as mentioned above, could be improved by
using MS-based novel protein data. Those novel ones can be applied to play a role in a
collection of supervised training data for machine learning, pattern recognition or rule-
based manual approach. There is an interesting bioinformatical report which hypothesized
that a uORF in the transcript down-regulates transcription of the corresponding RNA via
RNA decay mechanisms (Matsui et al., 2007). They obtained human and mouse transcripts
from RefSeq and UniGene (http://www.ncbi.nlm.nih.gov/unigene) and classified the
transcripts into Level 0 (not containing uORF) and Level 1-3 (containing uORF). Then, they
prepared the data of expression intensities and half-lives of mRNA transcripts mainly from
SymAtlas (now linked to BioGPS; http://biogps.gnf.org/#goto=welcome) and Genome
Research website (http://genome.cshlp.org/). Although they suggested that not only the
Emergence of the Diversified Short
ORFeome by Mass Spectrometry-Based Proteomics 427

expression level but also the half-life of transcripts was obviously declined in the latter
group, they did not demonstrate any interaction between uORFs and transcripts.
Advanced MS instruments can not only evaluate whether uORFs are actually translated but
also quantify time-course changes of their expression levels. Stable isotope labeling with
amino acids in cell culture (SILAC) technology enables us to quantify the changes regarding
all the proteins in vivo (Oyama et al., 2009). Based on time-course changes of specific
peptides, we could also hypothesize some regulatory interactions. In combination with the
measurement of the dynamics of the corresponding mRNAs using microarray or reverse
transcription-polymerase chain reaction (RT-PCR), transcriptional regulation by short ORFs
will be analyzed at the system level.

6. Conclusion
Although the roles of 5‘-UTR elements, especially uORFs, had been well discussed as
translational regulators for the main ORFs in the biological context, whether the proteins
encoded by the uORFs were translated had not been approached for a long time. We first
unraveled the mystery by demonstrating the existence of novel protein products defined by
these ORFs using advanced proteomics technology. Thanks to the progress of nanoLC-
MS/MS-based shotgun proteomics strategies, thousands of proteins can now be identified
from protein mixtures such as cell lysates. Some of the presumed UTRs are no longer
“untranslated“, and other noncoding transcripts are no longer “noncoding“. One of the
novel small proteins revealed in our analysis was indeed defined by a short transcript
variant generated by utilization of the downstream alternative promoters (Oyama et al.,
2007). Alternative uses of diverse transcription initiation, splicing and translation start sites
could increase the complexity of short protein-coding regions and MS -based annotation of
these novel small proteins will enable us to perform a more detailed analysis of the real
outline of the proteome, along with the translational regulation by the diversified short
ORFeome systematically.

7. References
Altmann, M. & Trachsel, H. (1993). Regulation of translation initiation and modulation of
cellular physiology. Trends in Biochemical Sciences, Vol. 18, No. 11, pp. 429-432,
Online ISSN 0167-7640; 0376-5067, Print ISSN 0968-0004.
Carninci, P., Sandelin, A., Lenhard, B., Katayama, S., Shimokawa, K., Ponjavic, J., Semple,
C.A.M., Taylor, M.S., Engstrom, P.G., Frith, M.C., Forrest, A.R.R., Alkema, W.B.,
Tan, S.L., Plessy, C., Kodzius, R., Ravasi, T., Kasukawa, T., Fukuda, S., Kanamori-
Katayama, M., Kitazume, Y., Kawaji, H., Kai, C., Nakamura, M., Konno, H.,
Nakano, K., Mottagui-Tabar, S., Arner, P., Chesi, A., Gustincich, S., Persichetti, F.,
Suzuki, H., Grimmond, S.M., Wells, C.A., Orlando, V., Wahlestedt, C., Liu, E.T.,
Harbers, M., Kawai, J., Bajic, V.B., Hume, D.A. & Hayashizaki, Y. (2006). Genome-
wide analysis of mammalian promoter architecture and evolution. Nature Genetics,
Vol. 38, No. 6, pp. 626-635, Online ISSN 1546-1718, Print ISSN 1061-4036.
Cvijovic, M., Dalevi, D., Bilsland, E., Kemp, G.J.L. & Sunnerhagen, P. (2007). Identification of
putative regulatory upstream ORFs in the yeast genome using heuristics and
evolutionary conservation. BMC Bioinformatics, Vol. 8, Article No. 295, Online ISSN
1471-2105.
428 Computational Biology and Applied Bioinformatics

Diba, F., Watson, C.S. & Gametchu, B. (2001). 5’UTR Sequences of the Glucocorticoid
Receptor 1A Transcript Encode a Peptide Associated With Translational Regulation
of the Glucocorticoid Receptor. Journal of Cellular Biochemistry, Vol. 81, No. 1, pp.
149-161, Online ISSN 1097-4644, Print ISSN 0730-2312.
Geballe, A.P. & Morris, D.R. (1994). Initiation codons within 5’-leaders of mRNAs as
regulators of translation. Trends in Biochemical Sciences, Vol. 19, No. 4, pp. 159-164,
Online ISSN 0167-7640; 0376-5067, Print ISSN 0968-0004.
Hatzigeorgiou, A.G. (2002). Translation initiation start prediction in human cDNAs with
high accuracy. Bioinformatics, Vol. 18, No. 2, pp. 343-350, Online ISSN 1460-2059,
Print ISSN 1367-4803.
Iacono, M., Mignone, F. & Pesole, G. (2005). uAUG and uORFs in human and rodent
5'untranslated mRNAs. Gene, Vol. 349, pp. 97-105, Online ISSN 1879-0038, Print
ISSN 0378-1119.
Kimura, K., Wakamatsu, A., Suzuki, Y., Ota, T., Nishikawa, T., Yamashita, R., Yamamoto, J.,
Sekine, M., Tsuritani, K., Wakaguri, H., Ishii, S., Sugiyama, T., Saito, K., Isono, Y.,
Irie, R., Kushida, N., Yoneyama, T., Otsuka, R., Kanda, K., Yokoi, T., Kondo, H.,
Wagatsuma, M., Murakawa, K., Ishida, S., Ishibashi, T., Takahashi-Fujii, A., Tanase,
T., Nagai, K., Kikuchi, H., Nakai, K., Isogai, T. & Sugano, S. (2006). Diversification
of transcriptional modulation: Large-scale identification and characterization of
putative alternative promoters of human genes. Genome Research, Vol. 16, No. 1, pp.
55-65, Online ISSN 1549-5469, Print ISSN 1088-9051.
Kozak, M. (1987). An analysis of 5'-noncoding sequences from 699 vertebrate messenger
RNAs. Nucleic Acids Research, Vol. 15, No. 20, pp. 8125-8148, Online ISSN 1362-4962,
Print ISSN 0305-1048.
Kozak, M. (1989). The Scanning Model for Translation: An Update. The Journal of Cell Biology,
Vol. 108, No. 2, pp. 229-241, Online ISSN 1540-8140, Print ISSN 0021-9525.
Kozak, M. (1991). Structural Features in Eukaryotic mRNAs That Modulate the Initiation of
Translation. The Journal of Biological Chemistry, Vol. 266, No. 30, pp. 19867-19870,
Online ISSN 1083-351X, Print ISSN 0021-9258.
Kozak, M. (1999). Initiation of translation in prokaryotes and eukaryotes. Gene, Vol. 234, No.
2, pp. 187-208, Online ISSN 1879-0038, Print ISSN 0378-1119.
Kozak, M. (2002). Pushing the limits of the scanning mechanism for initiation of translation.
Gene, Vol. 299, No. 1-2, pp. 1-34, Online ISSN 1879-0038, Print ISSN 0378-1119.
Maruyama, K. & Sugano, S. (1994). Oligo-capping: a simple method to replace the cap
structure of eukaryotic mRNAs with oligoribonucleotides. Gene, Vol. 138, No. 1-2,
pp. 171-174, Online ISSN 1879-0038, Print ISSN 0378-1119.
Matsui, M., Yachie, N., Okada, Y., Saito, R. & Tomita, M. (2007). Bioinformatic analysis of
post-transcriptional regulation by uORF in human and mouse. FEBS Letters, Vol.
581, No. 22, pp. 4184-4188, Online ISSN 1873-3468, Print ISSN 0014-5793.
Meijer, H.A. & Thomas, A.A.M. (2002). Control of eukaryotic protein synthesis by upstream
open reading frames in the 5‘-untranslated region of an mRNA. Biochemical Journal,
Vol. 367, No. 1, pp. 1-11, Online ISSN 1470-8728, Print ISSN 0264-6021.
Morris, D.R. & Geballe, A.P. (2000). Upstream Open Reading Frames as Regulators of
mRNA Translation. Molecular and Cellular Biology, Vol. 20, No. 23, pp. 8635-8642,
Online ISSN 1098-5549, Print ISSN 0270-7306.
Ota, T., Suzuki, Y., Nishikawa, T., Otsuki, T., Sugiyama, T., Irie, R., Wakamatsu, A., Hayashi,
K., Sato, H., Nagai, K., Kimura, K., Makita, H., Sekine, M., Obayashi, M., Nishi, T.,
Shibahara, T., Tanaka, T., Ishii, S., Yamamoto, J., Saito, K., Kawai, Y., Isono, Y.,
Emergence of the Diversified Short
ORFeome by Mass Spectrometry-Based Proteomics 429

Nakamura, Y., Nagahari, K., Murakami, K., Yasuda, T., Iwayanagi, T., Wagatsuma,
M., Shiratori, A., Sudo, H., Hosoiri, T., Kaku, Y., Kodaira, H., Kondo, H., Sugawara,
M., Takahashi, M., Kanda, K., Yokoi, T., Furuya, T., Kikkawa, E., Omura, Y., Abe, K.,
Kamihara, K., Katsuta, N., Sato, K., Tanikawa, M., Yamazaki, M., Ninomiya, K.,
Ishibashi, T., Yamashita, H., Murakawa, K., Fujimori, K., Tanai, H., Kimata, M.,
Watanabe, M., Hiraoka, S., Chiba, Y., Ishida, S., Ono, Y., Takiguchi, S., Watanabe, S.,
Yosida, M., Hotuta, T., Kusano, J., Kanehori, K., Takahashi-Fujii, A., Hara, H., Tanase,
T., Nomura, Y., Togiya, S., Komai, F., Hara, R., Takeuchi, K., Arita, M., Imose, N.,
Musashino, K., Yuuki, H., Oshima, A., Sasaki, N., Aotsuka, S., Yoshikawa, Y.,
Matsunawa, H., Ichihara, T., Shiohata, N., Sano, S., Moriya, S., Momiyama,H., Satoh,
N., Takami, S., Terashima, Y., Suzuki, O., Nakagawa, S., Senoh, A., Mizoguchi, H.,
Goto, Y., Shimizu, F., Wakebe, H., Hishigaki, H., Watanabe, T., Sugiyama, A.,
Takemoto, M., Kawakami, B., Yamazaki, M., Watanabe, K., Kumagai, A., Itakura, S.,
Fukuzumi, Y., Fujimori, Y., Komiyama, M., Tashiro, H., Tanigami, A., Fujiwara, T.,
Ono, T., Yamada, K., Fujii, Y., Ozaki, K., Hirao, M., Ohmori, Y., Kawabata, A., Hikiji,
T., Kobatake, N., Inagaki, H., Ikema, Y., Okamoto, S., Okitani, R., Kawakami, T.,
Noguchi, S., Itoh, T., Shigeta, K., Senba, T., Matsumura, K., Nakajima, Y., Mizuno, T.,
Morinaga, M., Sasaki, M., Togashi, T., Oyama, M., Hata, H., Watanabe, M., Komatsu,
T., Mizushima-Sugano, J., Satoh, T., Shirai, Y., Takahashi, Y., Nakagawa, K.,
Okumura, K., Nagase, T., Nomura, N., Kikuchi, H., Masuho, Y., Yamashita, R., Nakai,
K., Yada, T., Nakamura, Y., Ohara, O., Isogai, T. & Sugano, S. (2004). Complete
sequencing and characterization of 21,243 full-length human cDNAs. Nature Genetics,
Vol. 36, No. 1, pp. 40-45, Online ISSN 1546-1718, Print ISSN 1061-4036.
Oyama, M., Itagaki, C., Hata, H., Suzuki, Y., Izumi, T., Natsume, T., Isobe, T. & Sugano, S.
(2004). Analysis of Small Human Proteins Reveals the Translation of Upstream
Open Reading Frames of mRNAs. Genome Research, Vol. 14, No. 10B, pp. 2048-2052,
Online ISSN 1549-5469, Print ISSN 1088-9051.
Oyama, M., Kozuka-Hata, H., Suzuki, Y., Semba, K., Yamamoto, T. & Sugano, S. (2007).
Diversity of Translation Start Sites May Define Increased Complexity of the Human
Short ORFeome. Molecular & Cellular Proteomics, Vol. 6, No. 6, pp. 1000-1006, Online
ISSN 1535-9484, Print ISSN 1535-9476.
Oyama, M., Kozuka-Hata, H., Tasaki, S., Semba, K., Hattori, S., Sugano, S., Inoue, J. &
Yamamoto, T. (2009). Temporal Perturbation of Tyrosine Phosphoproteome
Dynamics Reveals the System-wide Regulatory Networks. Molecular & Cellular
Proteomics, Vol. 8, No. 2, pp. 226-231, Online ISSN 1535-9484, Print ISSN 1535-9476.
Peabody, D.S., Subramani, S. & Berg, P. (1986). Effect of Upstream Reading Frames on
Translation Efficiency in Simian Virus 40 Recombinants. Molecular and Cellular
Biology, Vol. 6, No. 7, pp. 2704-2711, Online ISSN 1098-5549, Print ISSN 0270-7306.
Peri, S. & Pandey, A. (2001). A reassessment of the translation initiation codon in
vertebrates. Trends in Genetics, Vol. 17, No. 12, pp. 685-687, Print ISSN 0168-9525.
Rastinejad, F. & Blau, H.M. (1993). Genetic Complementation Reveals a Novel Regulatory
Role for 3’ Untranslated Regions in Growth and Differentiation. Cell, Vol. 72, No. 6,
pp. 903-917, Online ISSN 1097-4172, Print ISSN 0092-8674.
Sedman, S.A., Gelembiuk, G.W. & Mertz, J.E. (1990). Translation Initiation at a Downstream
AUG Occurs with Increased Efficiency When the Upstream AUG Is Located Very
Close to the 5' Cap. Journal of Virology, Vol. 64, No. 1, pp. 453-457, Online ISSN
1098-5514, Print ISSN 0022-538X.
430 Computational Biology and Applied Bioinformatics

Selpi, Bryant, C.H., Kemp, G.J.L. & Cvijovic, M. (2006). A First Step towards Learning which
uORFs Regulate Gene Expression. Journal of Integrative Bioinformatics, Vol. 3, No. 2,
ID. 31, Online ISSN 1613-4516.
Suzuki, Y., Yoshitomo-Nakagawa, K., Maruyama, K., Suyama, A. & Sugano, S. (1997).
Construction and characterization of a full length-enriched and a 5′-end-enriched
cDNA library. Gene, Vol. 200, No. 1-2, pp. 149-156, Online ISSN 1879-0038, Print
ISSN 0378-1119.
Suzuki, Y., Ishihara, D., Sasaki, M., Nakagawa, H., Hata, H., Tsunoda, T., Watanabe, M.,
Komatsu, T., Ota, T., Isogai, T., Suyama, A. & Sugano, S. (2000). Statistical Analysis
of the 5’Untranslated Region of Human mRNA Using “Oligo-Capped” cDNA
Libraries. Genomics, Vol. 64, No. 3, pp. 286-297, Online ISSN 1089-8646, Print ISSN
0888-7543.
Suzuki, Y., Taira, H., Tsunoda, T., Mizushima-Sugano, J., Sese, J., Hata, H., Ota, T., Isogai, T.,
Tanaka, T., Morishita, S., Okubo, K., Sakaki, Y., Nakamura, Y., Suyama, A. &
Sugano, S. (2001). Diverse transcriptional initiation revealed by fine, large-scale
mapping of mRNA start sites. EMBO reports, Vol. 2, No. 5, pp. 388-393, Online ISSN
1469-3178, Print ISSN 1469-221X.
Suzuki, Y., Yamashita, R., Nakai, K. & Sugano, S. (2002). DBTSS: DataBase of human
Transcriptional Start Sites and full-length cDNAs. Nucleic Acids Research, Vol. 30,
No. 1, pp. 328-331, Online ISSN 1362-4962, Print ISSN 0305-1048.
Suzuki, Y., Yamashita, R., Sugano, S. & Nakai, K. (2004). DBTSS: DataBase of Transcriptional
Start Sites: progress report 2004. Nucleic Acids Research, Vol. 32 (suppl 1), Database
issue D78-D81, Online ISSN 1362-4962, Print ISSN 0305-1048.
Tsuchihara, K., Suzuki, Y., Wakaguri, H., Irie, T., Tanimoto, K., Hashimoto, S., Matsushima,
K., Mizushima-Sugano, J., Yamashita, R., Nakai, K., Bentley, D., Esumi, H. &
Sugano, S. (2009). Massive transcriptional start site analysis of human genes in
hypoxia cells. Nucleic Acids Research, Vol. 37, No. 7, pp. 2249-2263, Online ISSN
1362-4962, Print ISSN 0305-1048.
Vilela, C. & McCarthy, J.E.G. (2003). Regulation of fungal gene expression via short open
reading frames in the mRNA 5’untranslated region. Molecular Microbiology, Vol. 49,
No. 4, pp. 859-867, Online ISSN 1365-2958, Print ISSN 0950-382X.
Wakaguri, H., Yamashita, R., Suzuki, Y., Sugano, S. & Nakai, K. (2008). DBTSS: database of
transcription start sites, progress report 2008. Nucleic Acids Research, Vol. 36 (suppl
1), Database issue D97-D101, Online ISSN 1362-4962, Print ISSN 0305-1048.
Wang, X-Q. & Rothnagel, J.A. (2004). 5’-Untranslated regions with multiple upstream AUG
codons can support low-level translation via leaky scanning and reinitiation.
Nucleic Acids Research, Vol. 32, No. 4, pp. 1382-1391, Online ISSN 1362-4962, Print
ISSN 0305-1048.
Yamashita, R., Suzuki, Y., Nakai, K. & Sugano, S. (2003). Small open reading frames in 5′
untranslated regions of mRNAs. Comptes Rendus Biologies, Vol. 326, No. 10-11, pp.
987-991, Online ISSN 1768-3238, Print ISSN 1631-0691.
Yamashita, R., Suzuki, Y., Wakaguri, H., Tsuritani, K., Nakai, K. & Sugano, S. (2006). DBTSS:
DataBase of Human Transcription Start Sites, progress report 2006. Nucleic Acids
Research, Vol. 34 (suppl 1), Database issue D86-D89, Online ISSN 1362-4962, Print
ISSN 0305-1048.
Zhang, Z. & Dietrich, F.S. (2005). Identification and characterization of upstream open
reading frames (uORF) in the 5’ untranslated regions (UTR) of genes in
Saccharomyces cerevisiae. Current Genetics, Vol. 48, No. 2, pp. 77-87 Online ISSN
1432-0983, Print ISSN 0172-8083.
 23

Acrylamide Binding to Its Cellular Targets:

Insights from Computational Studies
Emmanuela Ferreira de Lima1 and Paolo Carloni2
1Scuola Internazionale Superiore di Studi Avanzati – SISSA - Statistical and Biological
Physics Sector, via Bonomea, Trieste
2German Research School for Simulation Sciences,

Jülich Research Center and RWTH-Aachen University,

1Italy
2Germany

To the memory of Ernesto Illy (1925 – 2008), who suggested this project.

1. Introduction
Acrylamide (AC, CH2=CH–CONH2, Chart 1), present in heated foodstuffs [Rice, 2005], has
been classified by the International Agency for Research on Cancer as ‘‘probably carcinogenic
to humans” (group 2A) (IARC, 1994). The Scientific Committee on Toxicity, Ecotoxicity and
the Environment (CSTEE) in the European Union (EU) demonstrated that AC exposure to
humans should be controlled as low as possible because of its inherently toxic properties
[Zhang et. al., 2009; Dearfield et. al., 1995]. AC is a low molecular-weight, odorless, and
colorless compound. It is readily soluble in water and can rapidly polymerize from a monomer
state to a polymer form [Klaunig, 2008]. AC is biotransformed in vivo to its epoxide
glycidamide (GDE) by cytochrome P450 2E1 (CYP2E1) [Ghanayem et. al., 2005]. GDE has
genotoxic properties in both in vitro and in vivo test systems [Kurebayashi & Ohno, 2008].
In spite of the possible carcinogenic nature of AC, no consistent effect of AC exposure on
cancer incidence in humans could be identified [Rice, 2005]. This strikingly contrasts with AC
subministration in both mice and rats, which may cause tumors at multiple sites [Besaratinia &
Pfeifer, 2005; 2007]. A plausible hypothesis is this might be caused, at least partially, by the fact
that AC interacts differently with the mouse and human proteins. AC may interact with
cysteine residues not engaged in S-S bridges. Indeed, the double bond of conjugated vinyl
compounds has strong affinity with SH groups [Friedman, 2003; Carere, 2006].

CH2 C
C NH 2
H
Fig. 1. AC structure.
432 Computational Biology and Applied Bioinformatics

There are six human proteins which are known to bind to AC. For all of them, structural
information is available. These are five enzymes [Howland et. al., 1980; Sakamoto &
Hashimoto, 1985] and the serum albumin protein [Ferguson et al., 2010] (Tab. 1). All of them
are dimers. Inspection of the X-ray structures of all of these targets shows Cys residues may
be present in the enzymatic active sites and/or solvent exposed (Tab. 1).

2. Objective
This work has two main goals. On one hand, using bioinformatics, molecular docking, and
molecular simulation procedures, we aim at predicting the structural determinants of AC in
complex with its cellular targets reported in Tab. 1. On the other hand, we aim at detecting
the difference (if any) of binding these molecules to the correspondent proteins in rodents.
These differences might contribute to the carcinogenic features of these molecules in rodents
as opposed to humans.

Sequence
PDB ID of Cys not engaged in
Protein Identity
human proteins disulfur bridges
(mouse)
Cys216, Cys104,
Topoisomerase IIα 1ZXM 88% Cys392, Cys405,
Cys170, Cys300
Cys74, Cys283*,
Creatine Kinase (CK) 3B6R 96% Cys254, Cys141,
Cys146
Cys72, Cys239,
Cys289, Cys338,
Aldolase 1QO5 95%
Cys201, Cys149,
Cys177, Cys134
Serum albumin 1AO6 72% Cys34
glyceraldehyde-3-phosphate Cys152*, Cys156,
1ZNQ 93%
dehydrogenase (GAPDH) Cys247
Cys118, Cys356,
Enolase 2PSN 97% Cys336, Cys338,
Cys398, Cys388
Table 1. Structural information of AC targets. For human proteins, for which the
experimental structures are available, the PDB codes are included. For mouse proteins, as
there is no structure available, the sequence identities with human proteins are reported.
The cysteines present in the active sites of the enzymes are shown with an asterisk while the
solvent-exposed ones are highlighted in bold.

3. Methodology
3.1. Molecular docking
Docking methods attempt to find the “best” matching between two molecules, typically a
receptor and a ligand [Halperin et. al., 2002]. Hence, they make the prediction of ligand
conformation and orientation within a targeted binding site [Kitchen et. al., 2004; Halperin
Acrylamide Binding to Its Cellular Targets: Insights from Computational Studies 433

et. al., 2002]. This prediction is carried out by performing a conformational space search
based on an ad hoc potential energy function. [Halperin et. al., 2002].
The accuracy of the method can be investigated by docking ligands into the protein from
which they were extracted (self-docking) and by docking them into its target protein in a
different conformation (usually taken from different protein/ligand complexes) (cross-
docking) [Kawatkar et. al., 2009; Sutherland et. al., 2007].
The increasing availability of protein three-dimensional structures combined with
continuing computational advances have made docking a very important tool for small-
molecule lead discovery [Campbell et. al., 2003].
Here we use the Genetic Optimisation for Ligand Docking (GOLD) program [Jones et. al.,
1995; 1997]. GOLD is an automated ligand docking program that uses a genetic algorithm to
explore the full range of ligand's conformational flexibility with partial flexibility of the
protein. It satisfies the fundamental requirement that the ligand must displace loosely
bound water on binding [Jones et. al., 1995; 1997].
GOLD uses a genetic algorithm (GA) to optimize a variety of structural parameters: (a)
dihedrals of ligand's rotatable bonds; (b) ligand's ring geometries; (c) dihedrals of protein
OH groups and NH3+ groups; and (d) the position of the ligand in the binding site [Verdonk
et. at., 2003].
In GOLD, one minimizes a molecular mechanics–like function with four terms, the so called
GOLD Fitness (Eq. 1) [Jones et al., 1995; 1997]:

GOLD Fitness = Shb_ext + Svdw_ext + Shb_int + Svdw_int (1)

Shb_ext decreases with the number of protein–ligand hydrogen-bonds; Svdw_ext decreases with
the van der Waals interactions; Shb_int decreases with the intramolecular hydrogen bonds in
the ligand; Svdw_int increases with the intramolecular strain in the ligand [Verdonk et. al.,
2003]. The scoring function is taken as the negative of the GOLD Fitness [Jones, 1995; 1997].
The larger the scoring function of a pose, the higher its rank.
We have performed GOLD-based docking for the human proteins in Tab. 1. These are:
human topoisomerase IIα [Sciandrello et al, 2010]; human-brain creatine kinase [Lü Z-R. et
al, 2009]; human aldolase [Dobryszycki et al., 1999a; 1999b]; human serum albumin
[Ferguson et al., 2010]; human glyceraldehyde-3-phosphate dehydrogenase [Sakamoto &
Hashimoto, 1985] and human enolase [Howland et. al., 1980]. Since alkylation by acrylamide
is limited to cysteine SH groups, the best cys residues putatively binding AC will be those
ones with the shortest distances between S atom and AC.
For rodents, there is no experimental structure available. Fortunately, the sequence identity
(SI) between the mouse proteins and the human enzymes is always greater than 88%, except
for serum albumin, for which SI= 72% (Tab. 1 and Fig. S1, SI). The structure of the latter was
built by homology modeling using the server HHpred [Soding, 2005; Soding et. al., 2005;
Hildebrand et. al., 2009].
The active site was defined here as a sphere with radius of 10 Å centered at the sulfur atoms
of the Cys residues of Tab. 1. The cavity detection algorithm, LIGSITE [Hendlich et. al.,
1997], was used to restrict the region of interest to concave, solvent-accessible surfaces. To be
sure that most of the high-affinity binding modes were explored, the genetic algorithm was
run 200 times for each complex examined.
The poses were ranked using the GOLD Fitness [Jones et al., 1995; 1997] as well as in terms
of the number of contacts between AC and the target cysteine residues.
434 Computational Biology and Applied Bioinformatics

4. Results and discussion

Here we report the results for AC docking to all its known human targets [Friedman, 2003].
We also discuss the putative binding of AC to the correspondent mouse proteins, for which
binding has not been shown. We consider here Cys residues not involved in S-S bridges
(Target cysteines hereafter, see Tab. 1).
Table 2 shows the docking score values for the best AC poses in all targets, the values are in
Kcal/mol. It can be seen that the best score obtained was for creatine kinase, -46.2, this result
shows the great protein-ligand interaction provided by the docking calculations, this result
is in good agreement with experiments, which have shown that CK is alkylated by AC.

Protein Docking score

Topoisomerase IIα -23.3
Creatine Kinase (CK) -46.2
Aldolase -21.3
Serum albumin -22.6
glyceraldehyde-3-phosphate
-28.1
dehydrogenase (GAPDH)
Enolase -28.3
Table 2. Docking score values. The lower the score, the better the binding.
In the topoisomerase enzyme (Fig. 2), the cysteine(s) that react with AC were not identified
experimentally. The X-ray structure [Wei et. al., 2005] features six target cysteines in each
subunit, four of which are solvent-exposed (Tab. 1). Our docking procedure suggests that, in
the most likely pose, AC interacts with Cys 405 (Fig. 3). In all figures, distances are reported
in Angstroms.

Fig. 2. 3D structure of topoisomerase enzyme [Sciandrello et al, 2010]. The four exposed
cysteines (Cys104, Cys216, Cys392 and Cys405) are named and indicated in licorice
representation.
Acrylamide Binding to Its Cellular Targets: Insights from Computational Studies 435

Fig. 3. AC interactions with Lys287, Met303 and Cys405 residues inside topoisomerase
enzyme.
The SI between human and mouse proteins is as high as 88% (Fig. S2, SI). In particular, all of
the residues surrounding the target cysteines are conserved. Inspection of the structure and
of the sequence alignment in Fig. S2 strongly suggests that the chemical environment of the
target cysteines in the mouse proteins is basically the same as in the human ones. The same
argument applies to all of the enzymes considered here. Hence, AC might bind to mouse
topoisomerase in a similar way as it does to the human protein.
The human enzyme creatine kinase (Fig. 4) has five target cysteines in each subunit (Tab. 1).
One of them (Cys283) is located in each of the active sites (Fig. 5) and it is known
experimentally to bind to AC [Matsuoka et. al., 1996; Meng et. al., 2001]. Consistently with
experiment, we find that in the best pose AC interacts with Cys283 (Fig. 5). AC is not found
to bind to any other target cysteine.
The SI between human and mouse proteins is considerably high (96%) and hence also in this
case AC might bind to the mouse creatine kinase in a similar fashion.
The human enzyme enolase (Fig. 6) has 6 target cysteines in each binding site. Our docking
procedure suggests that Cys 388 may bind to AC (Fig. 7). Because of the large SI between
human and mouse protein (97%), AC might bind to the correspondent Cys residue (Cys 388,
see Fig. S3, SI) also in the mouse protein.
The human enzyme glyceraldehyde-3-phosphate dehydrogenase (GAPDH) (Fig. 8) has
three target cysteines in each dimer. One of them is present in each active sites, Cys152 (Fig.
9). Experiments show that Cys152 in fact interacts with AC [Campian et. Al., 2002]. Our
docking is consistent with the experimental evidence (Fig. 9).
The protein human aldolase (Fig. 10) have eight and one target cysteine (Cys239, about 100
times more reactive than remaining exposed groups) in each subunit. Human serum
albumin (Fig. 11) has 35 cysteine residues with 17 disulphide bonds. Sulphydryl residue in
position 34 is left free to react with thiols of the environment [Candiano et al., 2009]. Our
docking procedure could identify binding poses for AC in neither proteins. This suggests
that conformational rearrangements, which are not taken into account in the docking
procedure, might allow AC to bind to one or more Cys residues. For mouse aldolase, the SI
436 Computational Biology and Applied Bioinformatics

Fig. 4. 3D structure of creatine kinase enzyme [Lü Z-R. et al, 2009]. The Cys283, which is
located in the active side is named and indicated in licorice representation.

Fig. 5. AC interactions with Leu201, Thr59 and Cys283 residues inside creatine kinase active
site.
Acrylamide Binding to Its Cellular Targets: Insights from Computational Studies 437

Fig. 6. 3D structure of enolase enzyme [Howland et. al., 1980]. The Cys388, which binds AC
is named and indicated in licorice representation.

Fig. 7. AC interactions with Val142, Lys421, Ser140 and Cys388 residues, inside enolase
enzyme.
438 Computational Biology and Applied Bioinformatics

Fig. 8. 3D structure of GAPDH enzyme [Sakamoto & Hashimoto, 1985]. The Cys152, which
binds AC is named and indicated in licorice representation.

Fig. 9. AC interactions with Gly212, Thr211, Ser151 and Cys152 residues inside GAPDH
active site.
Acrylamide Binding to Its Cellular Targets: Insights from Computational Studies 439

is as high as 95%. This suggests similar consideration for this species. Instead, for mouse
serum albumin, SI = 72% (see Fig. S1, SI). In the structure of the mouse serum albumin,
which was built by homology modeling, there are two target cysteines (at position 58 and
603). Therefore, we may expect rather different binding in the two species. Because of the
limitations of the docking procedure with homology models, we did not proceed to
investigate AC binding poses to this protein [Leach, 2001; McGovern & Shoichet, 2003].

Fig. 10. 3D structure of aldolase enzyme [Dobryszycki et al., 1999a; 1999b]. All the cysteine
residues are in licorice representation with indication of the most reactive one, Cys239.

Fig. 11. 3D structure of serum albumin [Ferguson et al., 2010]. The only cysteine free to react,
Cys34 is indicated in licorice representation.
440 Computational Biology and Applied Bioinformatics

5. Conclusions
By means of molecular docking methodology, we have studied the interactions between AC
and its human targets. The investigation is complemented by a study of AC interactions
with the mouse protein, for which binding has not been reported so far. The calculations are
consistent with the available biochemical data and they provide novel information on
putative cysteines to which AC could be bound in both mouse and human protein. In the
case of one protein, serum albumin, binding is likely to occur at different locations in the
protein. Hence, this difference could contribute to the experimentally known differences in
toxic properties of AC in humans and mice.

6. Acknowledgment
One of the authors (Lima, EF) is recipient of a grant from the ‘Fondazione Ernesto Illy’,
Trieste-Italy.

7. References
Besaratinia, A., & Pfeifer, G. P. (2005). DNA adduction and mutagenic properties of
acrylamide. Mutatational Research, Vol. 580, pp. 31–40.
Besaratinia A., & Pfeifer G. P. (2007). A review of mechanisms of acrylamide carcinogenicity.
Carcinogenesis, Vol. 28, No. 3, pp. 519–528.
Campbell S. J., Gold N. D., Jackson R. M. & Westhead D. R. (2003). Ligand binding:
functional site location, similarity and docking, Current Opinion in Structural
Biology, Vol. 13, pp. 389–395.
Campian E. C., Cai J., & Benz F. W. (2002). Acrylonitrile irreversibly inactivates
glyceraldehyde-3-phosphate dehydrogenase by alkylating the catalytically active
cysteine 149. Chemico-Biological Interactions, Vol. 140, pp. 279–291.
Candiano G., Petretto A., Bruschi M., Santucci L., Dimuccio V., Prunotto M., Gusmano R.,
Urbani A., Ghiggeri G. M. (2009). (Review) The oxido-redox potential of albumin
Methodological approach and relevance to human diseases, journal of proteomics,
73, 188-195.
Carere A. (2006). Genotoxicity and carcinogenicity of acrylamide: a critical review, Annali
dell’Istituto Superiore di Sanità, Vol. 42, No. 2, pp. 144-155.
Dearfield K. L. , Douglas G. R., Ehling U. H., Moore M. M., Sega G. A., & Brusick D. J. (1995).
Acrylamide: a review of its genotoxicity and an assessment of heritable genetic risk.
Mutation Research, Vol. 330, pp. 71-99.
Dobryszycki P., Rymarczuk M., BuIaj G., & Kochman M. (1999a). Effect of acrylamide on
aldolase structure.I. Induction of intermediate states. Biochimica et Biophysica Acta,
Vol. 1431, pp. 338-350.
Dobryszycki P., Rymarczuk M., Gapinski J., & Kochman M. (1999b). Effect of acrylamide on
aldolase structure. II. Characterization of aldolase unfolding intermediates.
Biochimica et Biophysica Acta, Vol. 1431, pp. 351-362.
Ferguson S. A., Garey J., Smith M. E., Twaddle N. C., Doerge D. R., & Paule M. G. (2010).
Preweaning behaviors, developmental landmarks, and acrylamide and
glycidamide levels after pre- and postnatal acrylamide treatment in rats.
Neurotoxicology and Teratology, Vol. 32, pp. 373–382.
Acrylamide Binding to Its Cellular Targets: Insights from Computational Studies 441

Friedman M. (2003). Chemistry, Biochemistry, and Safety of Acrylamide. A Review. Journal

of Agricultural and Food Chemistry, Vol. 51, No. 16, pp. 4504-4526.
Ghanayem B. I., McDaniel L. P., Churchwell M. I., Twaddle N. C., Snyder R., Fennell T. R., &
Doerge D. R. (2005). Role of CYP2E1 in the Epoxidation of Acrylamide to
Glycidamide and Formation of DNA and Hemoglobin Adducts. Toxicological
Science, Vol. 88, No. 2, pp. 311–318.
Halperin I., Ma B., Wolfson H., & Nussinov R. (2002). Principles of Docking: An Overview
of Search Algorithms and a Guide to Scoring Functions. PROTEINS: Structure,
Function, and Genetics, Vol. 47, pp. 409-443.
Hendlich M., Rippmann F., & Barnickel G. (1997). LIGSITE: Automatic and efficient
detection of potential small molecule binding sites in proteins. Journal of Molecular
Graphics and Modelling, Vol. 15, No. 6, pp. 359-363.
Hildebrand A., Remmert M., Biegert A., & Soding J. (2009). Fast and accurate automatic
structure prediction with HHpred. Proteins, Vol. 77, pp. 128–132.
Howland R. D., Vyas I. L., Lowndes H. E., & Argentieri T. M. (1980). The etiology of toxic
peripheral neuropathies: in vitro effects of acrylamide and 2,5-hexanedione on
brain enolase and other glycolytic enzymes. Brain Research, Vol. 202, pp. 131-142.
Jones G., Willett P., & Glen R. C. (1995). Molecular recognition of receptor sites using a
genetic algorithm with a description of desolvation. Journal of Molecular Biology,
Vol. 245, pp. 43-53.
Jones G., Willett P., Glen R. C., Leach A. R., & Taylor R. (1997). Development and Validation
of a Genetic Algorithm for Flexible Docking. Journal of Molecular Biology, Vol. 267,
pp. 727-748.
Kawatkar S., Wang H., Czerminski R., & McCarthy D. J. (2009). Virtual fragment
screening: an exploration of various docking and scoring protocols for fragments
using Glide, Journal of Computer-Aided Molecular Design, Vol. 23, pp. 527–539.
Kitchen D. B., Decornez H., Furr J. R., & Bajorath J. (2004). Docking and scoring in
virtual screening for drug discovery: methods and applications, Nature
Reviews – Drug Discovery, Vol. 3, pp. 935.
Klaunig J. E. (2008). Acrylamide Carcinogenicity. Journal of Agricultural and Food Chemistry,
Vol. 56, pp. 5984–5988.
Kurebayashi H., & Ohno Y. (2006). Metabolism of acrylamide to glycidamide and their
cytotoxicity in isolated rat hepatocytes: protective effects of GSH precursors.
Archives of Toxicology, Vol. 80, pp. 820–828.
Leach, A. R. (2nd Edition). (2001). Molecular modelling, principles and applications, Harlow,
Essex Pearson Education Limited, ISBN 0-582-38210-6, Harlow, England.
Lü Z-R., Zou H-C., Jin P. S., Park D., Shi L., Ho O-S., Park Y-D., Bhak J., & Zou F. (2009). The
effects of acrylamide on brain creatine kinase: Inhibition kinetics and
computational docking simulation. International Journal of Biological Macromolecules,
Vol. 44, pp. 128–132.
Matsuoka M., Matsumura H., & Igisu H. (1996). Creatine kinase activities in brain and
blood: possible neurotoxic indicator of acrylamide intoxication. Occupational and
Environmental Medicine, Vol. 53, pp.468-471.
McGovern S. L. & Shoichet Brian K. (2003). Information decay in molecular docking screens
against Holo, Apo, and modeled conformations of enzymes. Journal of Medicinal
Chemistry, Vol. 46, pp. 2895-2907.
442 Computational Biology and Applied Bioinformatics

Meng F-G., Zhou H-W., & Zhou H-M. (2001). Effects of acrylamide on creatine kinase from
rabbit muscle. The International Journal of Biochemistry & Cell Biology, Vol. 33, pp.
1064–1070.
Rice, J. (2005). The carcinogenicity of acrylamide. Mutatational Research, Vol. 580, pp. 3–20.
Sakamoto J., & Hashimoto K. (1985). Effect of acrylamide and related compounds on
glycolytic enzymes in rat sciatic nerve in vivo. Archives of Toxicology, Vol. 57, pp.
282-284.
Sciandrello G., Mauro M., Caradonna F., Catanzaro I., Saverini M., & Barbata G. (2010).
Acrylamide catalytically inhibits topoisomerase II in V79 cells. Toxicology in Vitro,
Vol. 24, pp. 830–834.
Soding J. (2005). Protein homology detection by HMM–HMM comparison. Bioinformatics,
Vol. 21, No. 7, pp. 951–960.
Soding J., Biegert A., & Lupas A. N. (2005). The HHpred interactive server for protein
homology detection and structure prediction. Nucleic Acids Research, Vol. 33, pp.
W244–W248.
Sutherland J. J., Nandigam R. K., Erickson J. A., & Vieth M. (2007). Lessons in Molecular
Recognition. 2. Assessing and Improving Cross-Docking Accuracy, Journal of
Chemical Information and Modeling, Vol. 47, pp. 2293-2302.
Verdonk M. L., Cole J. C., Hartshorn M. J., Murray C. W., & Taylor R. D. (2003).
Improved Protein–Ligand Docking Using GOLD, Proteins: Structure, Function, and
Genetics, Vol. 52, pp. 609–623.
Wei H., Ruthenburg A. J., Bechis S. K., & Verdine G. L. (2005). Nucleotide-dependent
Domain Movement in the ATPase Domain of a Human Type IIA DNA
Topoisomerase. Journal of Biological Chemistry, Vol. 280, pp. 37041-37047.
Zhang Y., Ren Y., & Zhang Y. (2009). New Research Developments on Acrylamide:
Analytical Chemistry, Formation Mechanism, and Mitigation Recipes. Chemical
Reviews, Vol. 109, pp. 4375–4397.