Molecular & Biochemical Parasitology 155 (2007) 103–112

Structural polymorphism and diversifying selection on the pregnancy

malaria vaccine candidate VAR2CSA
Joseph Bockhorst a,b , Fangli Lu c,1 , Joel H. Janes d , Jon Keebler e , Benoit Gamain f ,
Philip Awadalla e , Xin-zhuan Su c , Ram Samudrala g ,
Nebojsa Jojic a , Joseph D. Smith h,d,∗
a Microsoft Research, Seattle, WA, USA
bUniversity of Wisconsin—Milwaukee, WI, USA
c Laboratory of Malaria and Vector Research, National Institute of Allergy and Infectious Diseases, National Institutes of Health, Bethesda, MD, USA
d Department of Pathobiology, University of Washington, Seattle, WA, USA
e Department of Genetics, North Carolina State University, Raleigh, NC, USA
f Unité de Biologie des Interactions Hôte-Parasite, Institut Pasteur, Paris, France
g Department of Microbiology, University of Washington, Seattle, WA, USA
h Seattle Biomedical Research Institute, Seattle, WA, USA

Received 7 April 2007; received in revised form 11 June 2007; accepted 12 June 2007
Available online 26 June 2007

Abstract
VAR2CSA is the main candidate for a pregnancy malaria vaccine, but vaccine development may be complicated by sequence polymorphism.
Here, we obtained partial or full-length var2CSA sequences from 106 parasites and applied novel computational methods and three-dimensional
modeling to investigate VAR2CSA geographic variation and selection pressure. Our analysis reveals structural patterns of VAR2CSA sequence
variation in which polymorphic sites group into segments of limited diversity. Within these segments, two or three basic types characterize a
substantial majority of the parasite samples. Comparison to the primate malaria Plasmodium reichenowi shows that these basic types have ancient
origins. Globally, var2CSA genes are comprised of a mosaic of these ancestral polymorphic segments that have recombined extensively between
var2CSA alleles. Three-dimensional modeling reveals that polymorphic segments concentrate in flexible loops at characteristic locations in the
six VAR2CSA Duffy binding-like (DBL) adhesion domains. Individual DBL domain surfaces have distinct patterns of diversifying selection,
suggesting that limited and differing portions of each DBL domain are targeted by host antibody. Since standard phylogenetic tree analysis is
inadequate for highly recombining genes like var2CSA, we developed a novel phylogenetic approach that incorporates recombination and tracks
new mutations in segment types. In the resulting tree, P. reichenowi is confirmed as an outlier and African and Asian P. falciparum isolates have
slightly diverged. These findings validate a new approach to modeling protein evolution in the presence of frequent recombination and provide a
clearer understanding of how var gene products function as immunoevasive binding ligands.
© 2007 Elsevier B.V. All rights reserved.

Keywords: Antigenic variation; Malaria; Plasmodium falciparum; var Genes

1. Introduction

Pregnancy associated malaria (PAM), a substantial cause of

Abbreviations: CSA, chondroitin sulfate A; DBL, Duffy binding-like; disease and death in pregnant women and newborns, is associ-
PfEMP1, Plasmodium falciparum erythrocyte membrane protein 1; PAM, preg- ated with a unique subset of Plasmodium falciparum-infected
nancy associated malaria erythrocytes (IE) that bind chondroitin sulfate A (CSA) in the
∗ Corresponding author at: Seattle Biomedical Research Institute, Seattle, WA,

USA. Tel.: +1 206 256 7384; fax: +1 206 256 7229.

placenta [1–3]. Pregnant women acquire protective antibodies
E-mail address: [email protected] (J.D. Smith). that cross-react with geographically diverse placental isolates
1 Present address: Sun Yat-sen University, Guangzhou, PR China. [4–6] suggesting that surface molecule(s) expressed by PAM

0166-6851/$ – see front matter © 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.molbiopara.2007.06.007
104 J. Bockhorst et al. / Molecular & Biochemical Parasitology 155 (2007) 103–112

infected erythrocytes have conserved epitopes and, thus, that a 2.2. Amplification and sequencing of var2CSA sequences
PAM vaccine may be possible.
The specific targets of PAM immunity are currently being Nearly complete or full-length var2CSA sequences were
investigated, but recent evidence suggests that VAR2CSA, a amplified from 10 different parasites isolates (Table S1) rep-
member of the P. falciparum erythrocyte membrane 1 (PfEMP1) resenting South East Asia (five isolates), East or West Africa
protein family, may have an important role in PAM disease and (four isolates) or Central or South America (two isolates) using
immunity [7]. PfEMP1 proteins are clonally variant parasite previously defined PCR conditions [20]. In brief, var2CSA
adhesion ligands expressed at the surface of infected erythro- sequences were amplified in a process of trial and error
cytes [8]. Each parasite genome encodes approximately 60 in two to five overlapping parts using published degenerate
different PfEMP1 proteins, or var genes, but only expresses one primers to the unique UpsE-type 5 gene flanking sequence
PfEMP1 protein at a time [9]. The var2CSA gene is unusual [20] and gene-specific primers to highly conserved coding
for the var gene family because it is found in all parasite iso- regions. In addition, var2CSA sequences were collected from
lates and has been shown to be transcriptionally upregulated in the HB3 sequencing project [Broad Institute of Harvard and
both placental isolates [10,11] and laboratory parasites selected MIT (http://www.broad.mit.edu)] and from the P. falciparum
to bind CSA [12]. VAR2CSA contains multiple CSA binding Ghana isolate at The Wellcome Trust Sanger Institute website
Duffy binding-like (DBL) domains [13] and appears to be one at http://www.sanger.ac.uk/Projects/P falciparum/.
of the few or only PfEMP1 proteins mediating CSA binding
because parasites in which the gene is experimentally disrupted 2.3. Entropy calculations
lose the ability to adhere strongly to CSA [14,15]. Furthermore,
VAR2CSA is the target of maternal antibodies [16–19] making Protein multiple sequence alignments are widely used to infer
it the leading candidate for a pregnancy malaria vaccine. amino acid conservation within evolutionary related families.
Because the VAR2CSA protein displays extensive sequence The entropy score, a common approach for measuring sequence
and antigenic polymorphism [12,16,17,20,21], designing an variation [27], calculates amino acid variation at a single multi-
effective PAM vaccine will likely require detailed understand- ple alignment position using the formula:
ing of VAR2CSA sequence, structural and geographic variation. 20
The three-dimensional structures for DBL domains contained in Sentropy = − pi log2 (pi )
i=1
the binding regions of two different erythrocyte binding ligands
involved in erythrocyte invasion were recently solved [22,23]. where pi represents the observed frequency of residue type i in
Despite limited sequence similarity, these domains were found to the aligned column. The minimum positional entropy (0) occurs
have highly related structures suggesting that all DBL domains, at perfectly conserved positions and the maximum positional
including those in VAR2CSA, may share a similar structure. entropy (4.32 = log2 20) occurs at positions where all amino
This opens up new opportunities to investigate how VAR2CSA acids are observed with equal frequency. Following segmenta-
[16,17], and PfEMP1 proteins in general, have evolved as immu- tion (see below), we compute the post-segmentation entropy
noevasive binding ligands [24]. of each position (see supplemental methods). A site’s post-
A challenge of studying evolution of var2CSA, (and other segmentation entropy, a number between zero and the site’s
highly recombinogenic gene families), is that recombination positional entropy, is reduced if its variation strongly correlates
reduces the effectiveness of traditional phylogenetic approaches, with variations at other polymorphic sites in the same segment.
which assume evolution via point mutation [25]. An additional
practical complication is that var2CSA is large (∼10 kb), and 2.4. Segmentation
consequently most gene comparisons have relied on small gene
fragments. An important problem in computation biology is dis- Segmentation analysis investigates correlations among mul-
covering correlations among nearby positions in amino acid tiple nearby polymorphic sites [28]. We begin by computing
or DNA alignments. The immune system, for example, rec- the optimal segmentation of our VAR2CSA multiple alignment
ognizes short stretches of amino acids in pathogen proteins. using a maximum segment length of 15 amino acids and a max-
Thus, understanding patterns of sequence diversity is impor- imum of three types per segment. To create longer segments,
tant in vaccine design. In this study, we amplified full-length we subsequently tried merging adjacent segments; however, this
VAR2CSA sequences from a global collection of parasite iso- did not lead to better segmentations. We also tried different
lates and applied novel “recombination aware” computational maximum numbers of segment types, but three appeared to cap-
methods and molecular modeling to investigate the diversity of ture most of the variation in the VAR2CSA dataset (data not
VAR2CSA. shown). The output of the segmentation process consists of (i)
the segment boundaries, (ii) an identification of the total num-
2. Materials and methods ber of types found at each segment (either two or three), (iii)
a type assignment to each sequence in each segment and (iv)
2.1. Parasite isolates a probabilistic sequence model of each segment’s sequences
(see supplemental methods). We use these models to assign
Genomic DNAs in this study were prepared from culture types. Qualitatively, the segmentation procedure places segment
adapted parasite isolates that were previously published [26]. boundaries so that polymorphic sites in the same segment are
 J. Bockhorst et al. / Molecular & Biochemical Parasitology 155 (2007) 103–112 105

strongly correlated while nearby polymorphic sites in different types at that segment position thereby avoiding comparisons
segments are less (or not) correlated. To assess the significance between different segment types because these are most likely
of the correlations implicit in the segmentation, we used a 10- introduced through gene conversion/recombination between
fold cross validation methodology [29] to compare sequence var2CSA sequences. Question marks were introduced as gaps in
models produced via segmentation with sequence models that the type expanded nucleotide alignment for segment types not
assume mutations happen independently among sites (indepen- present in a specific var2CSA sequence. The resulting compari-
dent model). The total cross-validated log likelihood is much son leads to what we term ‘population trees’, which emphasize
greater under the segmentation model (−24132) than under the new mutation in ancestral recombination blocks. This approach
independent models (−31476) (p-value < 10−6 in a two-tailed treats each observed sequence at a leaf of the phylogenetic tree
paired t-test) indicating that the segmentation model more accu- as a representative of a recombining parasite population that
rately predicts unseen sequences than the independent model. contains all types at each segment. The leaf sequences have
one observed type per segment (i.e., from that parasite geno-
2.5. Phylogenetic analysis type) and the other types in the parasite population are hidden.
The interior nodes of the tree also represent populations whose
For the phylogenetic analysis we used the dnaml pro- sequences (as in standard tools) are hidden and to be inferred
gram of the PHYLIP package (http://evolution.genetics. through phylogeny.
washington.edu/phylip.html). Trees were compared using stan-
dard nucleotide alignments and type expanded nucleotide 2.6. Evolutionary analysis
alignments. The type expanded alignment takes the output
from the segmentation analysis and specifically compares To test for positive selection among amino acid encoding
nucleotide variation occurring within each of the basic segment codons we calculated estimates of rates of non-synonymous

Fig. 1. Positional entropy values across a VAR2CSA amino acid alignment. Positional entropy values were calculated from a multiple alignment of 106 complete
and partial VAR2CSA extracellular sequences. Entropy values were derived from all available sequences at each alignment position (range 11–55 sequences). The
N-terminal segment (NTS), DBL and interdomain regions (ID) are labeled. The DBL semi-conserved homology blocks B, D, F and H are shown to scale.
106 J. Bockhorst et al. / Molecular & Biochemical Parasitology 155 (2007) 103–112

(dN) and synonymous substitution (dS) among codons. We cal- for the respective side chains and main chains and finds the
culated these ratios, henceforth called ω, for each codon of optimal combination using an all-atom scoring function [33].
the whole gene to ask whether particular codons were evolv- These approaches are described in further detail in [34,35].
ing under positive selection in order to ask whether a codon The positional entropies and polymorphic segments determined
was under positive selection (ω > 1) using a model of natural from the alignment of 18 fuller-length var2CSA sequences
selection that allows for variable ω among codons. We used were mapped onto the IT4var DBL models after correcting for
a Bayesian approach [30] to test for departures from neutrality indels.
among codons that allows for recombination to occur among lin-
eages throughout the sequence (and hence, independent multiple 3. Results and discussion
genealogies).
3.1. VAR2CSA DBL domains are structured into variable
2.7. Molecular modeling and semi-conserved blocks

The three-dimensional conformations of the IT4 VAR2CSA Whereas VAR2CSA is the primary candidate for a PAM
DBL domains were modeled on the EBA-175 region vaccine, relatively limited sequence information exists and
II [23] using the PROTINFO structure prediction server most of this is partial gene fragments concentrated in only
(http://protinfo.compbio.washington.edu). Modeling was per- a small part of the protein. To acquire information about
formed using the comparative modeling protocol, which has how VAR2CSA has diversified, we amplified full or nearly
been shown to work well in the CASP protein structure pre- full-length var2CSA sequences from 10 parasite isolates from
diction experiments [31,32]. Initial models were constructed around the world, and collected all of the partial or complete
using a minimum perturbation approach that aims to preserve var2CSA sequences present in GenBank or at P. falciparum
as much information as possible from the template structure genome sequencing projects (Table S1). In total, we com-
solved by X-ray diffraction (the template with the Protein pared 106 var2CSA sequences including the complete or nearly
Data Bank identifier 1zro was used). Variable side chains and complete extracellular binding region from 18 P. falciparum iso-
main chains were then built using a graph-theory clique-finding lates, 87 partial var2CSA sequences, and a partial VAR2CSA
approach that explores a variety of possible conformations sequence from the chimpanzee malaria Plasmodium reichenowi.

Fig. 2. Segmentation structure of the VAR2CSA DBL2X domain. A segmentation analysis was performed to identify local correlations in amino acid polymorphism.
Regions under diversifying selection (dN/dS values > 1) are underlined. Within each segment, different types are shaded blue, yellow or green. Segment boundaries
are indicated by unshaded positions or shading differences. Brown amino acids denote mutations from the consensus segment type. The B, D, F and H semi-conserved
blocks are labeled. Bar heights indicate the amount of positional entropy prior to (black) or post-segmentation (red). Entropy calculations are based on a total of 106
VAR2CSA sequences and sequence tags. Parasite isolates are shaded by geographic origin; Asian (top), African (middle) and Central or South American (bottom).
Pr denotes P. reichenowi.
 J. Bockhorst et al. / Molecular & Biochemical Parasitology 155 (2007) 103–112 107

Overall, the 11 most complete VAR2CSA sequences, represent- DBL5 domains, and the DBL3 domain was not highly variable at
ing parasite isolates from Asia, Africa and Central America, these locations (Fig. S1). These locations are also highly variable
average 78% amino acid identity (range 75–83%). The result- in alignments of all DBL sequences, suggesting that the DBL
ing VAR2CSA alignment length was 2859 amino acids in fold is relatively insensitive to sequence and length variability
which the maximum coverage per site in the alignment is 55 in these regions [24].
sequences (Table S1). Unlike DBL domains, which have a variable/conserved block
VAR2CSA is made up of six receptor-like DBL domains structure, no global patterns of variation apply to interdomain
interspersed by variably sized interdomain (ID) regions. Pre- regions. Interdomain regions can be classified into larger regions
vious sequence analysis revealed that DBL domains could be (ID1, ID2 and ID4), which contain more than 100 amino acids,
organized into 10 variable and 10 semi-conserved blocks (named or smaller regions (ID3, ID5, ID6), which contain less than 100
A-J) [36]. The semi-conserved blocks correspond to structural amino acids (Fig. 1). Similar to DBL domains, sequence vari-
scaffolding in solved DBL structures [22,23] and can be used as ability in the ID2 and ID4 regions concentrates in variable blocks
a frame of reference between DBL sequences. As the simplest (Fig. 1), suggesting these interdomain regions may also be pre-
measure of VAR2CSA diversification we assessed the positional serving a specific three-dimensional fold. By comparison, ID1
entropy or amino acid conservation at each position in the align- has relatively few invariant residues, and these are not concen-
ment (Fig. 1). Ignoring gaps, 1321 (46.2%) of the alignment trated but rather are distributed throughout the ∼200 amino acid
positions are polymorphic, and 1174 (41.1%) have a positional region. Of the smaller interdomain regions, ID5 and ID6 are
entropy > 0.32, (the entropy for a position with one mutation highly polymorphic while ID3 is extremely conserved, and may
in 17 sequences). However, there exist substantial local cor- fold as part of the DBL domain [17]. Curiously, although the
relations that decrease this complexity (discussed below). As DBL6 and ID6 are membrane proximal and may have been
expected, variable blocks have a greater concentration of both expected to be less exposed, they are the most polymorphic
high entropy positions and gaps in the alignment (Figs. 1 and 2). (Table S2, Fig. S1).
Although the six DBL domains in VAR2CSA differ in amino
acid conservation between 61 and 88% (Table S2), they all 3.2. VAR2CSA mutations are not independent and gene
tend to have variable blocks adjacent to the B, D, F and H diversification is associated with a high rate of segmental
semi-conserved blocks. However, the distribution of variabil- gene recombination/gene conversion
ity differs between domains. For instance, there were variable
blocks on both sides of semi-conserved block B in the DBL1, To investigate multiple-site patterns of VAR2CSA polymor-
DBL2 and DBL6 domains, but only one side in the DBL4 and phism, we performed a segmentation analysis. Unlike positional

Fig. 3. Phylogenetic relationship between VAR2CSA sequences. (a) Example comparing a standard alignment to a type expanded alignment for three DBL2 segments.
In the type expanded alignment, only segments of the same type are directly compared. Segment types absent from a specific VAR2CSA sequence are represented by
question marks. Population trees are constructed by submitting the type expanded alignment to standard phylogenetic tools. A standard tree (b) and type expanded
population tree (c) created from nucleotides of the 17 most complete VAR2CSA sequences between the beginning of the protein and the interdomain 2 (alignment
length = 3405 nucleotides).
108 J. Bockhorst et al. / Molecular & Biochemical Parasitology 155 (2007) 103–112

entropy, which treats sites as independent, segmentation anal- Table 1

ysis uses a statistical model that assumes the protein contains Variation at polymorphic segments
a number of ‘correlated segments’ (or simply ‘segments’) in No. of mismatches from Frequency among polymorphic
which mutations are linked. Each segment may assume a num- consensus types segmentsa
ber of basic types, where for simplicity a type can be thought 0 4370 (70.3%)
of as a (possibly mutated) consensus sequence (see supporting 1 1367 (22.0%)
materials). Under the segmentation model the total VAR2CSA 2 308 (5.0%)
entropy is decreased because many of the polymorphic sites are 3 86 (1.4%)
4 50 (0.8%)
correlated (compare black and red bars, Figs. 2 and S1). The 5 10 (0.1%)
VAR2CSA alignment has 263 segments (average segment has 6 13 (0.0%)
6.1 amino acids) that in total cover 1613 (56.4%) of the total 7 6
positions and the majority (90.3%) of the strongly polymorphic 8 3
(entropy > 0.32) positions, indicating the presence of substantial 9 3
local correlations in amino acid polymorphism (Figs. 2 and S1). a Between 106 var2CSA sequences, a total of 6216 polymorphic segments

Correlated segments were identified in both variable and semi- compared.

conserved blocks.
In general, we observe two broad classes of differences
among types in the same segment. Some types are relatively ment. Although within a segment, there is generally substantial
closely related to each other, suggesting they likely arose by divergence between types (21% identity across types at polymor-
point mutation. For example, the yellow (consensus KFLAG- phic sites compared to an overall 63% identity at polymorphic
CLIVS) and blue (consensus KFLAGCLIAA) types in the sites), the observed types are remarkably conserved between
final segment in variable block 3 of DBL2 (Fig. 2). This is, isolates. For instance, between the 105 var2CSA sequences and
however, the minority case as more typically types are quite P. reichenowi, 70% of the polymorphic segments (4370/6216)
different from each other (compare to the green type (consen- exactly match one of the consensus types and an additional 22%
sus LLKE-WIIAA) in this same segment), suggesting that some differ by only one amino acid from consensus (Table 1). Fur-
mechanism other than point mutation brought them into align- thermore, many of these basic types are present both across

Fig. 4. Three-dimensional models of polymorphism in the VAR2CSA DBL2 and DBL3 domains. Domains were modeled based upon the EBA-175 F1 and F2 DBL
domains [23]. Panels A–D correspond to VAR2CSA DBL2 and panels E–H to VAR2CSA DBL3. (A, E) The amino acid diversity or positional entropies (range:
0–2.1) determined from a multiple alignment of 18 full-length VAR2CSA sequences are shown using a temperature scale from blue to red (low to high entropy).
(B, F) Mapping of variable sequence blocks determined by segmentation analysis. Polymorphic segments are colored and those under strong diversifying selection
are numbered. Semi-conserved homology blocks B, D, F and H are indicated. Conserved residues are shown in grey. (C, D, G and H) Space-fill representation of
positional entropy, panels (D) and (H) are rotated 180◦ relative to (C) and (G), respectively.
 J. Bockhorst et al. / Molecular & Biochemical Parasitology 155 (2007) 103–112 109

broad geographical regions and in P. reichenowi (Fig. S1) even extensive recombination of VAR2CSA would require hundreds
though the primate malaria is estimated to have diverged from of trees that would be challenging to comprehend. Second, and
P. falciparum ∼6–10 Mya [37,38]. Thus, we conclude that gene more importantly, the tree topologies would be determined pri-
recombination/gene conversion of ancient and slowly mutating marily by large scale differences (i.e., type differences) between
basic segment types likely creates the observed var2CSA com- sequences even though such differences likely arose prior to
binatorial diversity and mosaicism (Figs. 3 and S1). In addition, P. falciparum continental separation and, thus, are not part of
evidence suggests that recombination sites typically fall between patterns of variation we wish to discover. Instead, we propose
(and not within) correlated segments. a novel phylogenetic analysis approach that tracks gene rela-
Using this data set, we investigated geographic relationships tionships based upon new mutations that have occurred since
among VAR2CSA sequences to discover any geographi- the basic segment types evolved. This approach differs from
cally linked patterns of variation that may impact vaccine the multiple tree approach both because it produces a single
design. Standard phylogenetic tree approaches, however, assume tree and because recent evolution events determine the tree
sequences evolve through mutation and are not appropriate for topology. Our ‘population tree’ approach assumes that entire
studying highly recombinogenic gene families [25]. Indeed, populations of genes, each with all types for all segments
standard VAR2CSA trees suggest some P. falciparum sequences present, evolve and spread independently. Fortunately, popu-
are more distant from each other than from P. reichenowi lation trees can be obtained by submitting a ‘type expanded’
(Fig. 3b). Alternative approaches that construct separate trees in alignment as input to standard phylogenetic programs (Fig. 3).
each non-recombinant block are also not helpful [39]. First the By avoiding alignments of different segment types, expanded

Table 2
Diversifying selection on VAR2CSA DBL domains
VAR2CSA region Codon boundaries Mean omega Mean prob.a VAR2CSA region Codon boundaries Mean omega Mean prob.a

NTS 1–66 1.00 0.49 DBL4 1717–2002 0.46 0.18

DBL1 67–357 1.01 0.37 B BLOCK 1742–1757 0.04 0.00
VB1b 68–74 1.54 0.85 VB1 1764–1774 1.48 0.84
VB2 84–98 2.01 0.89 VB2 1796–1813 1.01 0.50
B BLOCK 96–114 0.96 0.59 D BLOCK 1814–1839 0.03 0.00
VB3 115–123 0.71 0.38 F BLOCK 1863–1873 0.55 0.20
VB4 139–143 0.41 0.04 VB3 1872–1886 2.58 0.99
D BLOCK 152–180 1.20 0.29 VB4 1903–1915 0.86 0.28
VB5c 170–215 3.15 0.92 H BLOCK 1930–1960 0.27 0.12
F BLOCK 217–237 0.34 0.17 VB5 1954–1965 0.93 0.46
VB6 247–262 0.80 0.38 DBL5 2152–2431 0.58 0.30
H BLOCK 276–315 0.20 0.03 VB1 2167–2176 0.39 0.04
VB7 313–331 1.04 0.50 B BLOCK 2178–2196 0.06 0.00
DBL2 570–917 1.05 0.54 VB2 2226–2241 0.87 0.38
VB1 573–581 1.38 0.85 D BLOCK 2241–2259 0.13 0.04
VB2 587–592 1.85 0.88 VB3 2260–2267 0.70 0.24
B BLOCK 600–612 0.46 0.29 VB4 2286–2291 1.70 0.76
VB3 612–649 1.80 0.96 F BLOCK 2292–2309 0.37 0.23
VB4 658–678 1.81 0.92 H BLOCK 2325–2359 0.57 0.31
D BLOCK 680–699 0.22 0.11 VB5 2349–2371 1.16 0.73
VB5 728–736 1.27 0.67 VB6 2387–2394 1.09 0.68
F BLOCK 743–763 0.21 0.11 VB7 2407–2419 1.16 0.73
VB6 769–805 1.16 0.74 DBL6 2493–2752 4.46 0.65
H BLOCK 811–846 0.59 0.29 VB1 2499–2529 1.11 0.65
VB7 849–878 2.13 0.99 B BLOCK 2529–2546 1.34 0.46
VB8 885–907 1.42 0.50 VB2 2542–2550 2.52 0.74
DBL3 1333–1647 0.52 0.25 VB3 2578–2586 0.96 0.44
VB1 1335–1343 0.74 0.35 D BLOCK 2587–2605 0.37 0.08
B BLOCK 1365–1378 0.07 0.00 VB4 2606–2636 2.42 0.98
VB2 1428–1455 1.19 0.62 F BLOCK 2630–2649 1.08 0.41
D BLOCK 1463–1480 0.07 0.01 VB5 2650–2680 6.17 0.97
VB3 1501–1517 1.21 0.74 H BLOCK 2678–2709 9.07 0.76
F BLOCK 1520–1539 1.00 0.60 VB6 2690–2752 12.46 0.98
VB4 1537–1555 1.22 0.72
H BLOCK 1562–1596 0.26 0.03
VB5 1597–1621 0.85 0.35
a The posterior probability that a codon is subject to diversifying selection averaged across codons for the defined region.
b VB denotes variable block.
c Bolded regions have at least one codon inferred to be subject to significant positive selection.
110 J. Bockhorst et al. / Molecular & Biochemical Parasitology 155 (2007) 103–112

alignments highlight new mutation. The tree built with the type acid binding site. Therefore, potential host interaction sites in
expanded alignment correctly identifies P. reichenowi as an out- VAR2CSA DBL domains are surrounded by amino acids that
lier and has more branch-like structure than the tree built with are under strong selection for amino acid diversification, pre-
standard alignments (Fig. 3). Moreover, bootstrap analysis indi- sumably for immune evasion.
cates that two sequences from the same geographical region
are more likely to group in the same sub-tree than would be 4. Concluding remarks
expected were there no geographical effect (p-value < 10−5 ).
Greater depth of VAR2CSA sequence coverage in multiple Our findings support a model of VAR2CSA evolution in
geographic regions will be important to determine if geo- which protein diversification is concentrated at flexible loops
graphic considerations should be used in VAR2CSA vaccine in the DBL fold and other surface exposed residues. Unex-
selection. pectedly, there are extensive local correlations in VAR2CSA
polymorphism and the same variable loop types were found
3.3. Diversifying selection is highly biased on the DBL in geographically diverse parasite isolates. Comparison to P.
surface and differs in extent between the six VAR2CSA DBL reichenowi suggests the polymorphic segments have ancient ori-
domains gins highlighting the importance of gene recombination/gene
conversion in var2CSA diversification. While VAR2CSA does
Although PfEMP1 proteins have a critical role in parasite not appear to recombine extensively with other var genes [20], it
immune evasion and pathogenesis [8], there has been lim- is tempting to speculate that a similar mechanism may exist for
ited investigation of diversity selection and this has primarily the broader family of recombining var genes because polymor-
been focused on small protein regions [17,21]. Here, we ana- phic blocks are sometimes shared between otherwise distinct
lyzed selection pressures over all six VAR2CSA DBL domains var sequences [40]. This mechanism may provide significant
and modeled the three-dimensional distribution of polymor- flexibility in DBL domains to bind and sequester infected
phic sites in the VAR2CSA DBL2 and DBL3 domains. This erythrocytes from blood circulation and potentially introduces
analysis supports the hypothesis that the semi-conserved B, D, variability near previously defined DBL-host interaction sites
F and H blocks correspond to structural or scaffolding ele- [22,23].
ments in DBL domains, and that variable blocks are either A question raised by these findings is how do pregnant women
polymorphic loops that connect the scaffolding elements or sur- develop a broad antibody response to an antigen as polymorphic
face exposed residues on the DBL scaffolding itself (Fig. 4). as VAR2CSA? While the specific targets of maternal antibodies
Notably, most polymorphic segments appear to be under strong are only beginning to be defined [16,17], immune investigations
diversifying or balancing selection (dN/dS ratios, or ω val- suggest that CSA binding parasite lines display both common
ues, significantly >1.0, Table 2), consistent with B cell epitope and diverse epitopes [5,41–44]. Although it has been postu-
mapping showing that polymorphic segments are subject to lated that antibody cross-recognition of placental isolates may
antibody pressure [16,17]. Conversely, semi-conserved blocks be due to highly conserved epitopes, a distinct possibility sug-
are more constrained, although individual residues exposed on gested by this analysis is that antibody cross-reactivity may be
the DBL scaffold can be polymorphic and have large dN/dS caused by overlapping polymorphism between geographically
ratios. diverse parasite isolates. During typical infections, parasites
Multi-domain comparisons show that VAR2CSA polymor- switch between PfEMP1 proteins to evade immunity [8]. For
phism is highly biased to one surface of the DBL2 fold and placental adherent isolates, the options for switching appear
differs in total surface distribution between the DBL2 and DBL3 limited [14,15]. One possibility is that exposure to multiple dif-
domains (Fig. 4) [17]. For instance, polymorphic segments 7 ferent parasite genotypes during pregnancy may broaden the
and 8 are highly diverse in the DBL2 domain, but the equivalent maternal antibody response to diverse VAR2CSA alleles. Curi-
residues are nearly invariant in the DBL3 domain. Moreover, ously, although the VAR2CSA protein is highly polymorphic,
the pattern and extent of diversifying selection differs between a surprising amount of predicted DBL surfaces are invariant
individual domains (Table 2). These findings imply either that and diversifying selection differs between the six DBL domains
invariant surfaces are poorly targeted by antibody or may be less in VAR2CSA. These results suggest that only a limited por-
accessible in the native protein. tion of each DBL domain is actively seen by the host immune
DBL domains have been found to engage different host recep- system and may indicate the presence of previously unrecog-
tors using distinct mechanisms and different binding sites on nized domain interactions within or between VAR2CSA and
opposite sides of the DBL fold [22,23]. Although the precise other proteins at the IE surface. If invariant residues or surfaces
CSA binding contact residues have not been defined in the are exposed in the native protein, then vaccine efforts will need to
VAR2CSA DBL domains, it may be significant that variable redirect antibody responses to these less polymorphic sites. Con-
blocks 1, 2 and 3 in the DBL2 domain are located in the equiv- versely, if PAM immunity targets polymorphic segments, then
alent EBA-175::sialic acid binding region and variable blocks information from segmentation analysis can now be applied in
4 and 6 are located in the equivalent Pk␣-DBL::Duffy antigen vaccine strategies to broaden antibody reactivity. The segmen-
binding region (Fig. 4). Both locations were also predicted to tation methodology may have wider application for uncovering
have variable loops in the DBL3 domain, although there was sequence patterns in rapidly evolving and recombining gene
significantly less variability adjacent to the equivalent sialic families or highly polymorphic vaccine targets.
 J. Bockhorst et al. / Molecular & Biochemical Parasitology 155 (2007) 103–112 111

Acknowledgements to the placental receptor chondroitin sulphate A. EMBO Rep 2005;6:

775–81.
[16] Barfod L, Bernasconi NL, Dahlback M, et al. Human pregnancy-associated
This research was supported by the Bill & Melinda
malaria-specific B cells target polymorphic, conformational epitopes in
Gates Foundation (J.D.S.), European Malaria Vaccine Initia- VAR2CSA. Mol Microbiol 2006.
tive (B.G.) (grant no. 01/2005), and by the Intramural Research [17] Dahlback M, Rask TS, Andersen PH, et al. Epitope mapping and topo-
Program of the NIH, National Institute of Allergy and Infec- graphic analysis of VAR2CSA DBL3X involved in P. falciparum placental
tious Diseases. VAR2CSA sequence data for the HB3 and sequestration. PLoS Pathog 2006;2:e124.
[18] Salanti A, Dahlback M, Turner L, et al. Evidence for the involve-
Ghana isolates were obtained from the Plasmodium falci-
ment of VAR2CSA in pregnancy-associated malaria. J Exp Med
parum HB3 Sequencing Project [Broad Institute of Harvard and 2004;200:1197–203.
MIT (http://www.broad.mit.edu)] and from the P. falciparum [19] Tuikue Ndam NG, Salanti A, Le-Hesran JY, et al. Dynamics of anti-
Ghana isolate at The Wellcome Trust Sanger Institute website VAR2CSA immunoglobulin G response in a cohort of senegalese pregnant
at http://www.sanger.ac.uk/Projects/P falciparum/. Computa- women. J Infect Dis 2006;193:713–20.
[20] Kraemer SM, Smith JD. Evidence for the importance of genetic structuring
tional support provided by Technical Computing at Microsoft
to the structural and functional specialization of the Plasmodium falciparum
Research. var gene family. Mol Microbiol 2003;50:1527–38.
[21] Trimnell AR, Kraemer SM, Mukherjee S, et al. Global genetic diversity
Appendix A. Supplementary data and evolution of var genes associated with placental and severe childhood
malaria. Mol Biochem Parasitol 2006;148:169–80.
[22] Singh SK, Hora R, Belrhali H, et al. Structural basis for Duffy
Supplementary data associated with this article can be found,
recognition by the malaria parasite Duffy-binding-like domain. Nature
in the online version, at doi:10.1016/j.molbiopara.2007.06.007. 2006;439:741–4.
[23] Tolia NH, Enemark EJ, Sim BK, et al. Structural basis for the EBA-175 ery-
References throcyte invasion pathway of the malaria parasite Plasmodium falciparum.
Cell 2005;122:183–93.
[1] Beeson JG, Brown GV, Molyneux ME, et al. Plasmodium falciparum [24] Howell DP, Samudrala R, Smith JD. Disguising itself–insights into Plas-
isolates from infected pregnant women and children are associated with modium falciparum binding and immune evasion from the DBL crystal
distinct adhesive and antigenic properties. J Infect Dis 1999;180:464– structure. Mol Biochem Parasitol 2006;148:1–9.
72. [25] Awadalla P. The evolutionary genomics of pathogen recombination. Nat
[2] Brabin BJ, Romagosa C, Abdelgalil S, et al. The sick placenta—the role Rev Genet 2003;4:50–60.
of malaria. Placenta 2004;25:359–78. [26] Wootton JC, Feng X, Ferdig MT, et al. Genetic diversity and chloro-
[3] Fried M, Duffy PE. Adherence of Plasmodium falciparum to chondroitin quine selective sweeps in Plasmodium falciparum. Nature 2002;418:
sulfate A in the human placenta. Science 1996;272:1502–4. 320–3.
[4] Duffy PE, Fried M. Antibodies that inhibit Plasmodium falciparum adhe- [27] Valdar WS. Scoring residue conservation. Proteins 2002;48:227–41.
sion to chondroitin sulfate A are associated with increased birth weight and [28] Bockhorst J, Jojic N. Discovering patterns in biological sequences by opti-
the gestational age of newborns. Infect Immun 2003;71:6620–3. mal segmentation. Proceedings of the 23rd International Conference on
[5] Fried M, Nosten F, Brockman A, et al. Maternal antibodies block malaria. Uncertainty in Artificial Intelligence, in press.
Nature 1998;395:851–2. [29] Mitchell T. Machine learning. 1st ed. McGraw-Hill; 1997.
[6] Staalsoe T, Shulman CE, Bulmer JN, et al. Variant surface antigen-specific [30] Wilson DJ, McVean G. Estimating diversifying selection and functional
IgG and protection against clinical consequences of pregnancy-associated constraint in the presence of recombination. Genetics 2006;172:1411–
Plasmodium falciparum malaria. Lancet 2004;363:283–9. 25.
[7] Smith JD, Deitsch KW. Pregnancy-associated malaria and the prospects [31] Hung LH, Samudrala R. PROTINFO: secondary and tertiary protein struc-
for syndrome-specific antimalaria vaccines. J Exp Med 2004;200: ture prediction. Nucleic Acids Res 2003;31:3296–9.
1093–7. [32] Hung LH, Ngan SC, Liu T, et al. PROTINFO: new algorithms for enhanced
[8] Miller LH, Baruch DI, Marsh K, et al. The pathogenic basis of malaria. protein structure predictions. Nucleic Acids Res 2005;33:W77–80.
Nature 2002;415:673–9. [33] Samudrala R, Moult J. An all-atom distance-dependent conditional prob-
[9] Frank M, Deitsch K. Activation, silencing and mutually exclusive expres- ability discriminatory function for protein structure prediction. J Mol Biol
sion within the var gene family of Plasmodium falciparum. Int J Parasitol 1998;275:895–916.
2006;36:975–85. [34] Samudrala R, Moult J. A graph-theoretic algorithm for comparative mod-
[10] Duffy MF, Caragounis A, Noviyanti R, et al. Transcribed var genes eling of protein structure. J Mol Biol 1998;279:287–302.
associated with placental malaria in Malawian women. Infect Immun [35] Samudrala R, Levitt M. A comprehensive analysis of 40 blind protein
2006;74:4875–83. structure predictions. BMC Struct Biol 2002;2:3.
[11] Tuikue Ndam NG, Salanti A, Bertin G, et al. High level of var2csa tran- [36] Smith JD, Subramanian G, Gamain B, et al. Classification of adhesive
scription by Plasmodium falciparum isolated from the placenta. J Infect domains in the Plasmodium falciparum erythrocyte membrane protein 1
Dis 2005;192:331–5. family. Mol Biochem Parasitol 2000;110:293–310.
[12] Salanti A, Staalsoe T, Lavstsen T, et al. Selective upregulation of a single [37] Escalante AA, Barrio E, Ayala FJ. Evolutionary origin of human and pri-
distinctly structured var gene in chondroitin sulphate A-adhering Plasmod- mate malarias: evidence from the circumsporozoite protein gene. Mol Biol
ium falciparum involved in pregnancy-associated malaria. Mol Microbiol Evol 1995;12:616–26.
2003;49:179–91. [38] Escalante AA, Ayala FJ. Evolutionary origin of Plasmodium and other
[13] Gamain B, Trimnell AR, Scheidig C, et al. Identification of multiple chon- Apicomplexa based on rRNA genes. Proc Natl Acad Sci USA 1995;92:
droitin sulfate A (CSA)-binding domains in the var2CSA gene transcribed 5793–7.
in CSA-binding parasites. J Infect Dis 2005;191:1010–3. [39] Kosakovsky Pond SL, Posada D, Gravenor MB, et al. Automated phyloge-
[14] Duffy MF, Maier AG, Byrne TJ, et al. VAR2CSA is the principal ligand for netic detection of recombination using a genetic algorithm. Mol Biol Evol
chondroitin sulfate A in two allogeneic isolates of Plasmodium falciparum. 2006;23:1891–901.
Mol Biochem Parasitol 2006;148:117–24. [40] Ward CP, Clottey GT, Dorris M, et al. Analysis of Plasmodium falciparum
[15] Viebig NK, Gamain B, Scheidig C, et al. A single member of the PfEMP-1/var genes suggests that recombination rearranges constrained
Plasmodium falciparum var multigene family determines cytoadhesion sequences. Mol Biochem Parasitol 1999;102:167–77.
112 J. Bockhorst et al. / Molecular & Biochemical Parasitology 155 (2007) 103–112

[41] Beeson JG, Mann EJ, Elliott SR, et al. Antibodies to variant surface antigens [43] Elliott SR, Duffy MF, Byrne TJ, et al. Cross-reactive surface epitopes
of Plasmodium falciparum-infected erythrocytes and adhesion inhibitory on chondroitin sulfate A-adherent Plasmodium falciparum-infected ery-
antibodies are associated with placental malaria and have overlapping and throcytes are associated with transcription of var2csa. Infect Immun
distinct targets. J Infect Dis 2004;189:540–51. 2005;73:2848–56.
[42] Beeson JG, Mann EJ, Byrne TJ, et al. Antigenic differences and conser- [44] Tuikue Ndam NG, Fievet N, Bertin G, et al. Variable adhesion abilities
vation among placental Plasmodium falciparum-infected erythrocytes and and overlapping antigenic properties in placental Plasmodium falciparum
acquisition of variant-specific and cross-reactive antibodies. J Infect Dis isolates. J Infect Dis 2004;190:2001–9.
2006;193:721–30.