Review

The Human Connectome: A Structural

Description of the Human Brain
Olaf Sporns*, Giulio Tononi, Rolf Kötter

ABSTRACT Experimental approaches to human cognition have been

significantly enhanced by the arrival of functional

T
he connection matrix of the human brain (the human neuroimaging [5], a set of techniques that can be applied to
‘‘connectome’’) represents an indispensable study a broad range of cognitive functions, with ever-
foundation for basic and applied neurobiological increasing spatial and temporal resolution. But the
research. However, the network of anatomical connections mechanistic interpretation of neuroimaging data is limited,
linking the neuronal elements of the human brain is still in part due to a severe lack of information on the structure
largely unknown. While some databases or collations of large- and dynamics of the networks that generate the observed
scale anatomical connection patterns exist for other activation patterns. A potential theoretical framework for
mammalian species, there is currently no connection matrix conceptualizing cognition as a network phenomenon is based
of the human brain, nor is there a coordinated research effort on two main organizational principles found in the cerebral
to collect, archive, and disseminate this important cortex, functional segregation, and functional integration
information. We propose a research strategy to achieve this [6,7]. Emerging network theories of cognition emphasize the
goal, and discuss its potential impact. contextual [8], distributed [9], dynamic [10], and degenerate
[11,12] nature of structure–function mappings in the brain.
Introduction To successfully map structure to function in the human brain,
To understand the functioning of a network, one must we urgently need a comprehensive, detailed structural model
know its elements and their interconnections. The purpose of of neuronal units and their connections. Connectional
this article is to discuss research strategies aimed at a models of the human brain are scarce and poorly defined
comprehensive structural description of the network of [13], and they are largely based on extrapolating anatomical
elements and connections forming the human brain. We information from other primate species such as the macaque
propose to call this dataset the human ‘‘connectome,’’ and we monkey, an approach that is problematic [14], in part,
argue that it is fundamentally important in cognitive because of our incomplete understanding of evolutionary
neuroscience and neuropsychology. The connectome will changes.
significantly increase our understanding of how functional Our proposal to assemble the human connectome has
brain states emerge from their underlying structural several components. First, we attempt to define a level of
substrate, and will provide new mechanistic insights into how scale at which a first draft of the human connectome might be
brain function is affected if this structural substrate is assembled. We consider several potential experimental and
disrupted. It will provide a unified, time-invariant, and neuroinformatics approaches for creating this first draft. We
readily available neuroinformatics resource that could be then discuss potential problems and limitations of the
used in virtually all areas of experimental and theoretical connectome, including the central issues of individual
variability and development. Finally, we sketch out the
neuroscience.
potential impact of the connectome in computational and
Recent research in neuroscience has resulted in a rapid
cognitive neuroscience.
proliferation of neuroscience datasets and in the arrival of a
new discipline, neuroinformatics [1–4]. Despite considerable
advances in experimental techniques and computational
paradigms, we still have an incomplete understanding of how
human cognitive function emerges from neuronal structure
and dynamics. Here, we focus on the relative lack of Citation: Sporns O, Tononi G, Kötter R (2005) The human connectome: A structural
information about the different types of neural elements and description of the human brain. PLoS Comput Biol 1(4): e42.
their neural connections in the human brain. While a larger Copyright: ! 2005 Sporns et al. This is an open-access article distributed under the
number of anatomical studies of the human brain have been terms of the Creative Commons Attribution License, which permits unrestricted
use, distribution, and reproduction in any medium, provided the original author
carried out at the macroscopic (cerebral lobes, surface and source are credited.
landmarks, and white matter tracts) or microscopic
Abbreviation: DTI, diffusion tensor imaging
(cytoarchitectonics, myeloarchitectonics,
chemoarchitectonics, etc.) anatomical level, there is virtually Olaf Sporns is in the Department of Psychology and Programs in Neuroscience and
Cognitive Science, Indiana University, Bloomington, Indiana, United States of
no information on the finer connectivity patterns, including America. Giulio Tononi is in the Department of Psychiatry, University of Wisconsin,
neuronal connection densities or laminar projection patterns Madison, Wisconsin, United States of America. Rolf Kötter is at C. and O. Vogt Brain
in relation to anatomically segregated cortical areas or Research Institute and Institute of Anatomy II, Heinrich Heine University,
Düsseldorf, Germany.
intraregional differentiation. Furthermore, none of the
available information is deposited in a single standardized *To whom correspondence should be addressed. E-mail: [email protected]
data format, nor can it be accessed through a public database. DOI: 10.1371/journal.pcbi.0010042

PLoS Computational Biology | www.ploscompbiol.org 0245 September 2005 | Volume 1 | Issue 4 | e42
Scales and Levels of Structural Description (approximately 1010 neurons and 1013 connections in the
cortex alone) [18,19]. Recording or tracing 1015 connections
The human genome is composed of approximately 3 3 109
is not only technically impossible, it may also be unnecessary.
base pairs, containing around 20,000–30,000 genes [15]. The
While a genomic mutation in a single base pair can have
compilation of the genome was aided by the fact that base
dramatic consequences, alterations of single synapses or cells
pairs and genes are relatively straightforward choices as basic
have not been shown to have similar macroscopic effects.
structural elements. The connectome will consist of two main
Instead, there is overwhelming evidence that human cognitive
descriptors defining its network architecture: neural elements
functions depend on the activity and coactivity of large
and neural connections. Data fields for these elements specify
populations of neurons in distributed networks, including the
superordinate or subordinate structures, a normalized
corticothalamic system [20]. Furthermore, individual neurons
position within a standard coordinate system, and additional
and connections are subject to rapid plastic changes. These
parameters such as physiological/biophysical metadata that
changes include synaptic weights as well as structural
are crucial for specifying neural dynamics. The set of all N
remodeling of dendritic spines and presynaptic boutons [21],
elements constitutes the columns (targets) and rows (sources)
possibly switching synaptic connections between large
of an N2–N connection matrix A, whose aij entries represent
numbers of potential synaptic sites [22,23]. We suggest that
connections between individual elements i and j. In keeping
the vast number, high variability, and fast dynamics of
with conventions adopted by other authors [16,17], confirmed individual neurons and synapses render them inappropriate
absence of a connection is denoted by aij ¼ 0, while confirmed as basic elements for an initial draft of the connectome.
presence of a connection (irrespective of its strength or
physiological characteristics) results in aij ¼ 1. Once a Macroscale: Brain Regions and Pathways
connection is confirmed to be present, its nonzero matrix An advantage of single neurons is that the elements
element receives additional data entries cataloguing a range themselves are relatively easily demarcated and well defined.
of structural and physiological parameters, such as fiber In contrast, brain areas and neuronal populations are more
direction, connection density, strength, sign (excitatory/ difficult to delineate. No single universally accepted
inhibitory), conduction delay, potential modulatory effects parcellation scheme currently exists for human brain regions
(voltage dependence), etc. The union of the binary (e.g., areas of the cerebral cortex), posing a significant
connection matrix and connection-specific physiological data obstacle to creating a unified resource such as the
then results in a structural description that combines connectome. In the human cerebral cortex, neurons are
connection topology and biophysics. arranged in an unknown number of anatomically distinct
Compiling the connectome faces two significant challenges regions and areas, perhaps on the order of 100 [24] or more.
not shared by other natural or technological networks. First, Different subdivisions of the human brain (e.g., brain stem,
the human brain is a highly complex organ with a great thalamus, cerebellum, or cortex) may require different
number of structurally distinct, heterogeneous, yet criteria for parcellation.
interconnected components. Because a primary application Nonetheless, anatomically distinct brain regions and inter-
of the connectome will likely be a structural substrate for regional pathways represent perhaps the most feasible
understanding human cognitive function and interpreting organizational level for compiling a first draft of the human
neuroimaging studies, a first draft of the connectome may connectome. Several neuroinformatics resources recording
focus on the structural description of the corticothalamic large-scale connection patterns in the cerebral cortex of
system, including all regions of the cortex and their various mammalian species already exist, for example, for
associated thalamic nuclei. Extensions of this first draft might most cortical regions of the macaque monkey [16,17,25], cat
include additional connected regions (striatum, cerebellum, [26], and rat [27]. Computational analyses of these datasets
etc.), with the ultimate aim of compiling the connectome of have revealed a broad range of network characteristics [28],
the whole brain. including the existence of clusters of brain regions [29],
A second challenge is that basic structural elements of the hierarchical organization [30,31], small-world attributes
human brain, in terms of network nodes and connections, are [32,33], distinct functional streams [34], motifs [35], and areal
difficult to define. Different kinds of structural descriptions contributions to global network measures [36].
could target at least three rather distinct levels of A broad range of experimental approaches exist at the
organization. At opposite ends of the scale are the level of macroscale. Cerebral white matter has traditionally been
single neurons and synapses (microscale) and the level of taken as a marker of the amount of connectivity within a
anatomically distinct brain regions and inter-regional cortical system. The relative contribution of cerebral white
pathways (macroscale). Between these two levels is the level of matter has increased throughout phylogeny to such an extent
neuronal groups or populations (mesoscale). It is important that its volume and metabolic requirements may present a
to determine which level of description is the most limitation to further increases in connectional complexity
appropriate for a first draft of the human connectome. [37]. The structural organization of white matter has been
investigated by dissection, histological staining, degeneration
Microscale: Single Neurons and Synapses methods, and axonal tracing [38]. Axonal tracing methods are
Attempting to assemble the human connectome at the level the main basis for existing mammalian large-scale connection
of single neurons is unrealistic and will remain infeasible at matrices [16,25–27], and their systematic compilation is
least in the near future. With single neurons as the basic currently being refined [17,39] to extract more sophisticated
element, the size of the connectome would be several orders data, perhaps with the help of automated text analysis. In the
of magnitude larger than that of the genome, comprising an human brain, postmortem tracing with carbocyanine dyes has
estimated 1011 neurons, with 1015 connections between them provided details of connectivity within and between adjacent

PLoS Computational Biology | www.ploscompbiol.org 0246 September 2005 | Volume 1 | Issue 4 | e42
areas [40,41]. However, because of the slow transport and the sufficiently powerful or comprehensive. A systematic
length of fibers in the human brain, this method fails to reveal application of sophisticated diffusion-weighted imaging
more remote connections. Another tracing method employs combined with spatially registered high-resolution
the in vivo detection by magnetic resonance imaging of high- anatomical or spectroscopic imaging, regional activation, and
contrast rare earth ions (e.g., manganese [42]) that have been coactivation data (e.g., electroencephalography,
injected into fiber tracts or inhaled, and taken up by neurons. magnetoencephalography) obtained within the same
The invasive nature and potential toxicity of the procedure individual subject, may offer the most feasible strategy for
makes it an unlikely candidate for human connectivity mapping the human connectome at the macroscale. This first
analyses. draft of the human connectome would take the form of a
Recent noninvasive imaging methods (diffusion tensor probabilistic map of voxel-by-voxel connectivity embedded
imaging [DTI] in its several variants, commonly followed by within standard coordinates containing approximately 104–
computational tractography) have been shown to produce 105 elements and approximately 105–107 structural
results that are consistent with known pathways formed by connections. It would not only provide information on the
major fiber tracts in the human brain [43–45], although there large-scale connection patterns within the corticothalamic
continues to be some limitations in data acquisition and system, but also on parcellation of human cortex into distinct
processing algorithms [46]. To disambiguate signals produced areas based on a combination of structural and functional
by crossing or intersecting fibers, advances in diffusion data in the same individual. Since this connectivity matrix is
imaging technology may allow the resolution of multiple voxel-based, it can be cross-referenced with existing
axonal pathways within a single image voxel [47,48]. Despite reference templates and with population-based brain atlases.
the promise of diffusion imaging, a systematic atlas of DTI-
based neuroanatomy, including probabilistic data gathered Mesoscale: Minicolumns and Their Connection Patterns
from individual brains, has not yet been produced, and the A first draft of the corticothalamic connectome at the
relationship among tensor fields, fiber tracts, and neuronal macroscale might provide a comprehensive dataset
connections remains controversial. This controversy is likely comprising several hundred brain regions and thousands of
to be resolved only through a combination of anatomical pathways, but it does not incorporate information on
tracing techniques with noninvasive diffusion and functional functional subdivisions or segregated subcircuits within each
imaging. brain region. While such a macroscale description will
Perhaps the most promising avenue for compiling the provide many fundamental insights into the large-scale
human connectome originates from the notion that organization of human cortex, it is an insufficient basis for a
individual brain regions maintain individual connection complete understanding of the human brain’s functional
profiles. What defines a segregated brain region is that all its dynamics and information processing capacities. A further
structural elements share highly similar long-range step in this direction involves the characterization of
connectivity patterns, and that these patterns are dissimilar connection patterns among elementary processing units,
between brain regions. These connectivity patterns corresponding to local populations of neurons such as
determine the region’s functional properties [49], and also cortical minicolumns. Mountcastle [54,55] originally
allow their anatomical delineation and mapping. Diffusion proposed the cortical minicolumn as a basic functional unit
imaging can be used to identify borders between cortical of mammalian cerebral cortex. While details of minicolumn
areas [50], most clearly on the basis of differences in long- architecture are likely to vary across different brain regions
range connections to the thalamus [51,52]. The idea that [56–58], minicolumns generally contain approximately 80–
patterns of connectivity can be used to identify areal 100 neurons, spanning all cortical layers, with a diameter of
boundaries has also been tested in a combination of approximately 30–50 lm [55]. Minicolumns may possess
functional imaging and DTI/tractography in the human relatively stereotypic internal processing, and maintain
medial frontal cortex [53]. First, connection probabilities generic patterns of inputs and outputs with minicolumns in
from voxels within the medial frontal cortex to all other other regions [56–59].
voxels in the rest of the brain were obtained. Binarized Recent studies have provided evidence for functionally
connection patterns were then used to calculate a cross- specialized and precisely wired subnetworks of neurons
correlation matrix, which was examined for the existence of coexisting within single cortical columns [60,61]. These
distinct clusters of voxels with shared connection patterns. studies have shown that cortical columns may contain
Such clusters were taken to represent anatomically segregated subnetworks, corresponding to minicolumns,
segregated areas corresponding to human supplementary which promote greater intracolumnar functional
motor area and pre- supplementary motor area, respectively. independence and informational heterogeneity. The
Functional mapping revealed close correspondence between members of such subnetworks are selectively interconnected
DTI and functional activation patterns. While this combined with each other, indicating that connections within and
structural–functional approach is computationally intensive, between minicolumns follow more complex rules than simple
nothing prevents its application to the entire corticothalamic random patterns or Gaussian (distance-dependent)
system. We suggest that the correlated use of noninvasive connection profiles.
structural and functional imaging methods offers the most Minicolumns may be a sensible choice for neural elements
promising experimental route toward the human at the mesoscale of the connectome because they may
connectome. represent basic functional elements that are crucial for
Structural connection data obtained using this combined cortical information processing. While tracing or recording
methodology can in principle be validated by histological or all minicolumns in an individual brain is still impossible, it
tracing methods. Likely, no single method will turn out to be may be feasible to collect data on minicolumn anatomy for

PLoS Computational Biology | www.ploscompbiol.org 0247 September 2005 | Volume 1 | Issue 4 | e42
selected brain regions, which can then be ‘‘fit’’ into the elements and their connections exhibit significant variations
appropriate positions within the macroscale connection between individuals, at all levels of scale. Some individual
matrix. Functional imaging at columnar resolution has been variations may be due to genetic differences, others may be
carried out in animal experiments using high field strength the result of developmental and experiential history, gender
[62], and may be facilitated in the future through selective differences, pathologies, or responses to injury. To
imaging of fast and spatially precise capillary cerebral blood complicate matters further, the human connectome
flow response components [63]. Especially important for undergoes development through time, from early stages of
determining functional responses of brain regions are the embryo to adolescence to adult age. Incorporating
connection patterns between each region’s constituent individual variations and developmental stages is absolutely
elements. Axonal tracing methods have revealed specific crucial in rendering the connectome an effective resource.
patterns of horizontal connections between individual cells Anatomical and imaging studies have revealed significant
and cell populations within a brain region, which are often interindividual variability in the size and location of brain
found to preferentially link cells with similar response areas, as well as in the relationship between macrostructures
characteristics [64,65], resulting in clustered intra-areal (e.g., the cortical gyrification pattern) and microstructures
connectivity. Such patterns might also be accessible in the (e.g., cytoarchitectonics and cytochemistry). Statistical
human brain [40], and regional variations in such patterns analysis of variations in macroscopic surface features of the
may provide important clues regarding the way in which human brain, for example its sulcal geometry [67],
information is processed within each region. demonstrates the extent to which even large-scale features of
The axonal tracing approach delivers minicolumn maps for cortical morphology vary between individuals, possibly as a
each distinct brain region, including information on their result of genetic differences [68]. An anatomical study of
functional segregation and local (intra-areal) Broca’s area in ten postmortem human brains revealed
interconnectivity. While parcellated brain regions at the significant variations in size as well as in the area’s relation to
macroscale can be identified across individuals, we have no sulcal landmarks [69]. Applying structural magnetic
means to resolve the locations of corresponding minicolumns resonance imaging to map the boundaries of the planum
across different brains. The mapping of smaller-than- temporale has demonstrated significant variations in its size
macroscopic units, therefore, requires coordinate- and position across 50 individuals [70]. Functional
independent mapping approaches [66], which preserve the neuroimaging studies have also revealed significant
anatomical relationships and basic physiological properties interindividual differences, for example, in the location and
of these units. Thus, macroscale and minicolumn descriptions extent of area MT/V5 [71] and other visual cortical areas [72].
deliver complementary datasets that need cross-level These functional differences are presumably due to
integration to achieve a single mesoscale version of the variations in underlying structural (cytoarchitectonic and
connectome. Effectively, minicolumn maps need to be connectional) substrates.
mapped onto brain region–specific voxel sets rather than The presence of significant interindividual variability in
individual voxels, with voxel sets providing spatial embedding structural connection patterns, even at the macroscale level,
and probabilistic long-range connections, and minicolumn and the fundamentally probabilistic nature of connectivity
maps providing local connectivity, processing, and coding datasets provided in the connectome may be viewed as
information. The minicolumn description provides a fundamental weaknesses of the proposal, undermining its
functionally heterogeneous architecture that is unique to comprehensive goal of a definitive structural description of
each parcellated brain region, with specific (probabilistic) the human brain. However, we should consider the fact that
patterns of intra-areal and interareal minicolumn there is also clear interindividual variability in the human
connections. A crucial task will be to convert long-range, genome. Nevertheless, the first draft with a DNA sequence
nondirected voxel-by-voxel connectivity into directed, obtained from cells from only a few individuals [15] has
functionally heterogeneous long-range minicolumn proven immensely useful for gaining insights into general
interconnectivity. This requires intermediate descriptive organizational features of the human genome. Mapping of
levels such as stripes, bands, and blobs in the early visual interindividual variability in the connectome is a necessary
system, for which specific connection patterns have been further step, but does not detract from the potential insights
demonstrated across areal boundaries. When accomplished, gained from a first draft that does not yet systematically
this cross-level integration will result in a mesoscale incorporate these differences.
connection matrix of the human brain that might comprise
as many as 107–108 structural elements (comparable in size to
the 2005 World Wide Web), with minicolumn elements Steps Toward the Human Connectome:
directly cross-referenced to voxels in the macroscale map. Its Compilation, Assembly, and Integration
Based on a combination of functional and diffusion-
weighted imaging, we outline the following five-step strategy
Individual Variability and Development for compiling a first draft of the human connectome at the
The large-scale connectivity structure of the brain above macroscale.
the synaptic level represents a relatively invariant Step 1 is to perform diffusion-weighted imaging followed
characteristic of our species. Once the elements and by probabilistic tractography of thalamocortical tracts as well
connections in the human brain are recorded, this dataset as corticocortical interareal pathways, using correlations in
will remain stable, essentially forever. However, as in the case connectivity profiles to assist in parcellating human cortical
of the genome, the precise combinations and patterns of regions. The end result is a voxel-wise probabilistic all-to-all

PLoS Computational Biology | www.ploscompbiol.org 0248 September 2005 | Volume 1 | Issue 4 | e42
structural connectivity matrix for the human brain. Step 2 is Conclusions: The Potential Impact
to perform a correlation analysis of spatially registered, of the Connectome
equally resolved resting activity and/or multistimulus/
How can the connectome be used to map brain structure to
multitask activation data (functional magnetic resonance
function? The step from structure to function is essential for
imaging and/or magnetoencephalography) recorded in the
understanding how cognitive processes emerge from their
same person [73], emphasizing strong functional
relationships that are consistent across tasks [74]. The end morphological substrates. Our central motivating hypothesis
result is a voxel-wise all-to-all functional connectivity matrix is that the pattern of elements and connections as captured in
for the human brain. Step 3 is to perform a cluster analysis of the connectome places specific constraints on brain
correspondences between the structural and functional dynamics, and thus shapes the operations and processes of
connectivity matrix obtained under steps 1 and 2, with the human cognition. In turn, data recording the activity of the
goal of identifying regions of consistent structure–function human brain in combination with the structural model
relationships in the human brain, possibly involving indirect provided by the connectome will help to discern causal
projections [75]. Step 4 is to compare the results obtained by interactions in large-scale brain networks (e.g., [76–78]). We
cluster analysis (step 3) with macaque data in order to identify emphasize that structure–function relationships are not
correspondences (e.g., in visuomotor pathways) and directly evident from the connectional dataset itself. Rather,
deviations (e.g., in structures such as the fasciculus arcuatus). their elucidation will require further intense empirical and
Step 5 is to validate the strongest predictions generated by computational study. Depending on sensory input, global
assembling the final combined structural–functional brain state, or learning, the same structural network can
connectivity matrix using custom-designed stimuli and support a wide range of dynamic and cognitive states. This
perturbational techniques such as transcranial magnetic should not, however, discourage the effort to assemble the
stimulation. connectome. Similar difficulties in mapping structure to
The following three steps represent additional stages function exist in the case of the genome, although they
designed to further test and verify the human connectome, generally were not foreseen when the project was initiated.
including population analyses of individual variability and Highly complex transcriptional regulatory networks,
pathology. Step 6 is to perform a population analysis of signaling pathways, mechanical forces, and elaborate
healthy subjects and spatial registration to standard brain mechanisms of gene expression all play essential roles in
coordinates for probabilistic statements about data from translating base sequences into functioning cells, tissues, and
steps 1–5. Step 7 is to compare population data on clustered organisms. Both genome and connectome constitute complex
brain regions to histologically identified regions in a networks [79], whose functions are not fully understood even
probabilistic human brain atlas to assess correspondence. if their structural substrates are fully catalogued.
Step 8 is to compare population data between healthy An obvious and related use of the human connectome
individuals and patient groups with specific pathologies, to would be providing structural information that can be
identify connectional differences. implemented as part of large-scale computational models
Ultimately, the connectome will likely describe [80,81]. If the connectome is sufficiently comprehensive and
connectivity patterns at multiple levels of scale, for example, accessible, it could also provide a set of structural
by incorporating linkages between the macroscale of brain benchmarks that might facilitate the comparison and
regions and pathways in more elementary mesoscale integration of the numerous specialized and structurally
functional units such as minicolumns and their patterns of based models that have already been proposed in
interconnectivity. As experimental techniques mature, the computational neuroscience. Drawing upon human
connectome will gradually evolve through different stages of connectional datasets would help ground modeling efforts
assembly as it is refined, updated, cross-validated, and ‘‘filled aimed specifically at brain mechanisms of human cognitive
in.’’ Standardization of parcellation schemes, elimination of function (e.g., language). Other computational applications of
unreliable data, and incorporation of additional structural the connectome are topological analyses of network structure
levels above and below the one chosen for the initial draft will [28], perturbational studies aiming at mapping structural
drive this effort. An additional driving force is the continued disruption to functional defects, and synthetic brain imaging
innovation in data acquisition and analysis techniques, [82–84].
particularly in diffusion-weighted imaging and tractography, The human connectome could potentially have a major
which will result in progressive revision, refinement, and impact on our understanding of brain damage and
extension of the connectome. To become a useful research subsequent recovery. The effects of developmental variations
tool, the connectome must be linked to other databases or abnormalities, traumatic brain injury, or
(compiled in parallel efforts) that record additional neurodegenerative disease can all be captured as specific
information mapped across the human brain, such as structural variants of the human connectome. The functional
receptor distributions or gene expression patterns. We consequences of network perturbations will allow a better
expect that assembling even the first draft of the connectome understanding of structural causes of dysfunction, and may
will require significant experimental and computational permit the design of strategies for recovery based on network
resources over an extended period of time. Compilation, analysis. Understanding the basic network causes of brain
assembly, and integration efforts are likely to be extensive diseases may also open new avenues for therapy and
tasks, requiring large-scale collaboration, coordinated data prevention by harnessing inherent network mechanisms that
collection and dissemination, and the establishment of new ensure robustness and compensation.
computational methods, data standards, and mechanisms for There are many structural and functional aspects that the
controlled validation and quality assurance. human connectome, as envisioned in this article, does not

PLoS Computational Biology | www.ploscompbiol.org 0249 September 2005 | Volume 1 | Issue 4 | e42
contain or address. For example, it does not explicitly 22. Stepanyants A, Hof PR, Chklovskii DB (2002) Geometry and structural
plasticity of synaptic connectivity. Neuron 34: 275–288.
capture or catalogue the rich variety of neuronal 23. Kalisman N, Silberberg G, Markram H (2005) The neocortical microcircuit
morphologies, the diversity of physiological and biochemical as a tabula rasa. Proc Natl Acad Sci U S A 102: 880–885.
neural subtypes, glial cells, or brain vascularization. Its first 24. Van Essen DC, Drury HA, Joshi S, Miller MI (1998) Functional and
structural mapping of human cerebral cortex: Solutions are in the surfaces.
draft does not capture synaptic plasticity and remodeling, Proc Natl Acad U S A 95: 788–795.
nonsynaptic communication, hormonal regulation, or 25. Felleman DJ, Van Essen DC (1991) Distributed hierarchical processing in
the primate cerebral cortex. Cereb Cortex 1: 1–47.
degenerative processes. We do not view these issues as 26. Scannell JW, Burns GAPC, Hilgetag CC, O’Neil MA, Young MP (1999) The
shortcomings that fundamentally undermine the usefulness connectional organization of the cortico-thalamic system of the cat. Cereb
or potential impact of the connectome. Rather, they illustrate Cortex 9: 277–299.
27. Burns GAPC, Young MP (2000) Analysis of the connectional organisation of
the open-ended, integrative nature of the proposed research neural systems associated with the hippocampus in rats. Phil Trans R Soc
effort. The connectome will represent a foundational Lond B Biol Sci 355: 55–70.
28. Sporns O, Chialvo D, Kaiser M, Hilgetag CC (2004) Organization,
resource and a central reference point for a broad range of development and function of complex brain networks. Trends Cogn Sci 8:
specialized databases [85], thus making federating these 418–425.
databases more effective. Most importantly, the connectome 29. Hilgetag CC, Burns GAPC, O’Neill MA, Scannell JW, Young MP (2000)
Anatomical connectivity defines the organization of clusters of cortical
will provide an important tool for mechanistic modeling and areas in the macaque monkey and the cat. Phil Trans R Soc Lond B Biol Sci
interpretation of human functional brain data. & 355: 91–110.
30. Hilgetag CC, O’Neill MA, Young MP (2000) Hierarchical organisation of
macaque and cat cortical sensory systems explored with a novel network
Acknowledgments processor. Phil Trans R Soc Lond B Biol Sci 355: 71–90.
31. Hilgetag CC, O’Neill MA, Young MP (1996) Indeterminate organization of
OS was supported by National Institutes of Health National Institute the visual system. Science 271: 776–777.
on Drug Abuse grant 1R21DA15647–01. RK was supported by the 32. Sporns O, Tononi G, Edelman GM (2000) Theoretical neuroanatomy:
Deutsche Forschungsgemeinschaft (Graduate School 320 and KO Relating anatomical and functional connectivity in graphs and cortical
1560/6–2). Core ideas were developed at a workshop sponsored by the connection matrices. Cereb Cortex 10: 127–141.
33. Sporns O, Zwi J (2004) The small world of the cerebral cortex.
James S. McDonnell Foundation. The authors thank M. Arbib, K.
Neuroinformatics 2: 145–162.
Friston, R. McIntosh, B. Horwitz, and T. Behrens for comments and 34. Young MP (1992) Objective analysis of the topological organization of the
suggestions. primate cortical visual system. Nature 358: 152–155.
35. Sporns O, Kötter R (2004) Motifs in brain networks. PLoS Biol 2: DOI: 10.
1371/journal.pbio.0020369
References 36. Kötter R, Stephan KE (2003) Network participation indices: Characterizing
1. Kötter R (2001) Neuroscience databases: Tools for exploring brain component roles for information processing in neural networks. Neural
structure function relationships. Phil Trans R Soc Lond B Biol Sci 356: Netw 16: 1261–1275.
1111–1120. 37. Chklovskii DB, Schikorski T, Stevens CF (2002) Wiring optimization in
2. Amari S, Arbib MA, Kötter R (2003) Special issue: Neuroinformatics. cortical circuits. Neuron 34: 341–347.
Neural Netw 16: 1235–1398. 38. Köbbert C, Apps R, Bechmann I, Lanciego JL, Mey J, et al. (2000) Current
3. Koslow SH, Subramanian S (2005) Databasing the brain: From data to concepts in neuroanatomical tracing. Prog Neurobiol 62: 327–351.
knowledge. Hoboken (New Jersey): Wiley. 466 p. 39. Kötter R (2004) Online retrieval, processing, and visualization of primate
4. Insel TR, Volkow ND, Li TK, Battey JF, Landis SC (2003) Neuroscience connectivity data from the CoCoMac database. Neuroinformatics 2: 127–
networks: Data sharing in an information age. PLoS Biol 1: DOI: 10.1371/ 144.
journal.pbio.0000017 40. Galuske RA, Schlote W, Bratzke H, Singer W (2000) Interhemispheric
5. Frackowiak RSJ, Friston KJ, Frith C, Dolan R, Price CJ, et al., editors (2003) asymmetries of the modular structure in human temporal cortex. Science
Human brain function, 2nd ed. New York: Academic Press. 1,144 p. 289: 1946–1949.
6. Tononi G, Edelman GM, Sporns O (1998) Complexity and the integration 41. Tardif E, Clarke S (2001) Intrinsic connectivity of human auditory areas: A
of information in the brain. Trends Cogn Sci 2: 44–52. tracing study with DiI. Eur J Neurosci 13: 1045–1050.
42. Saleem KS, Pauls JM, Augath M, Trinath T, Prause BA, et al. (2002)
7. Friston KJ (2002) Beyond phrenology: What can neuroimaging tell us about
Magnetic resonance imaging of neuronal connections in the macaque
distributed circuitry? Annu Rev Neurosci 25: 221–250.
monkey. Neuron 34: 685–700.
8. McIntosh AR (2000) Towards a network theory of cognition. Neural Netw
43. Conturo TE, Lori NF, Cull TS, Akbudak E, Snyder AZ, et al. (1999) Tracking
13: 861–870.
neuronal fiber pathways in the living human brain. Proc Natl Acad Sci U S
9. Bressler SL (1995) Large-scale cortical networks and cognition. Brain Res
A 96: 10422–10427.
Brain Res Rev 20: 288–304.
44. LeBihan D (2003) Looking into the functional architecture of the brain
10. Bressler SL, Kelso JAS (2001) Cortical coordination dynamics and
with diffusion MRI. Nature Rev Neurosci 4: 469–480.
cognition. Trends Cogn Sci 5: 26–36. 45. Catani M, Jones DK, Donato R, Ffytche DH (2003) Occipito-temporal
11. Tononi G, Sporns O, Edelman GM (1999) Measures of degeneracy and connections in the human brain. Brain 126: 2093–2107.
redundancy in biological networks. Proc Natl Acad Sci U S A 96: 3257– 46. Dougherty RF, Ben-Shachar M, Bammer R, Brewer AA, Wandell BA (2005)
3262. Functional organization of human occipital-callosal fiber tracts. Proc Natl
12. Price CJ, Friston KJ (2002) Degeneracy and cognitive anatomy. Trends Acad Sci USA 102: 7350–7355.
Cogn Sci 6: 416–421. 47. Tuch DS, Reese TG, Wiegell MR, Wedeen VJ (2003) Diffusion MRI of
13. Crick F, Jones E (1993) The backwardness of human neuroanatomy. Nature complex neural architecture. Neuron 40: 885–895.
361: 109–110. 48. Tuch DS, Wisco JJ, Khatchaturian MH, Ekstrom LB, Kötter R, et al. (2005)
14. Tootell RBH, Tsao D, Vanduffel W (2003) Neuroimaging weighs in: Humans Q-Ball imaging of macaque white matter architecture. Phil Trans R Soc B
meet macaques in ‘‘primate’’ visual cortex. J Neurosci 23: 3981–3989. Biol Sci 360: 869–879.
15. Venter JC, Adams MD, Myers EW, Li PW, Mural RJ, et al. (2001) The 49. Passingham RE, Stephan KE, Kötter R (2002) The anatomical basis of
sequence of the human genome. Science: 1304–1351. functional localization in the cortex. Nat Rev Neurosci 3: 606–616.
16. Young MP (1993) The organization of neural systems in the primate 50. Behrens TEJ, Johansen-Berg H (2005) Relating connectional architecture to
cerebral cortex. Proc R Soc Lond B Biol Sci 252: 13–18. grey matter function using diffusion imaging. Phil Trans R Soc B Biol Sci
17. Stephan KE, Kamper L, Bozkurt A, Burns GA, Young MP, et al. (2001) 360: 903–911.
Advanced database methodology for the collation of the connectivity data 51. Jones EG (1998) Brodmann’s areas and cerebral cortex. In: Adelman G,
on the macaque brain (CoCoMac). Philos Trans R Soc Lond B Biol Sci 356: Smith B, editors. Encyclopedia of neuroscience. New York: Elsevier. pp.
1159–1186. 294–296, 339–341.
18. Murre JMJ, Sturdy DPF (1995) The connectivity of the brain: Multi-level 52. Behrens TEJ, Johansen-Berg H, Woolrich MW, Smith SM, Wheeler-
quantitative analysis. Biol Cybern 73: 529–545. Kingshott CAM, et al. (2003) Non-invasive mapping of connections between
19. Braitenberg V, Schüz A (1991) Anatomy of the cortex: Statistics and human thalamus and cortex using diffusion imaging. Nat Neurosci 6: 750–
geometry. Berlin: Springer. 249 p. 757.
20. Mountcastle VB (1998) Perceptual neuroscience: The cerebral cortex. 53. Johansen-Berg H, Behrens TEJ, Robson MD, Drobnjak I, Rushworth MFS, et
Boston: Harvard University Press. 512 p. al. (2004) Changes in connectivity profiles define functionally distinct
21. Engert F, Bonhoeffer T (1999) Dendritic spine changes associated with regions in human medial frontal cortex. Proc Natl Acad Sci U S A 101:
hippocampal long-term synaptic plasticity. Nature 399: 66–70. 13335–13340.

PLoS Computational Biology | www.ploscompbiol.org 0250 September 2005 | Volume 1 | Issue 4 | e42
54. Mountcastle VB (1978) An organizing principle for cerebral function. In: 71. Watson JD, Myers R, Frackowiak RS, Hajnal JV, Woods RP, et al. (1993) Area
Edelman GM, Mountcastle VB, editors. The mindful brain. Cambridge: MIT V5 of the human brain: Evidence from a combined study using positron
Press. pp. 7–50 emission tomography and magnetic resonance imaging. Cereb Cortex 3:
55. Mountcastle VB (1997) The columnar organization of the neocortex. Brain 79–94.
120: 701–722. 72. Hasnain MK, Fox PT, Woldorff MG (1998) Intersubject variability of
56. Buxhoeveden DP, Casanova MF (2002) The minicolumn hypothesis in functional areas in the human visual cortex. Hum Brain Mapp 6: 301–315.
neuroscience. Brain 125: 935–951. 73. Lahaye PJ, Poline JB, Flandin G, Dodel S, Garnero L (2003) Functional
57. Rockland KS, Ichinohe N (2004) Some thoughts on cortical minicolumns. connectivity: Studying nonlinear, delayed interactions between BOLD
Exp Brain Res 158: 265–277. signals. Neuroimage 20: 962–974.
58. Jones EG (2000) Microcolumns in the cerebral cortex. Proc Natl Acad Sci U 74. Bartels A, Zeki S (2005) Brain dynamics during natural viewing
S A 97: 5019–5021. conditions—A new guide for mapping connectivity in vivo. Neuroimage 24:
59. Douglas RJ, Martin KA (1991) A functional microcircuit for cat visual 339–349.
cortex. J Physiol 440: 735–769. 75. Stephan KE, Hilgetag CC, Burns GAPC, O’Neill MA, Young MP, et al. (2000)
60. Yoshimura Y, Dantzker JLM, Callaway EM (2005) Excitatory cortical Computational analysis of functional connectivity between areas of
neurons form fine-scale functional networks. Nature 433: 868–873. primate cerebral cortex. Phil Trans R Soc Lond B Biol Sci 355: 111–126.
61. Ohki K, Chung S, Ch’ng YH, Kara P, Reid C (2005) Functional imaging with 76. Kötter R, Sommer F (2000) Global relationship between anatomical
cellular resolution reveals precise microarchitecture in visual cortex. connectivity and activity propagation in the cerebral cortex. Phil Trans R
Nature 433: 597–603. Soc Lond B Biol Sci 355: 127–134.
62. Zhao F, Wang P, Hendrich K, Kim SG (2005) Spatial specificity of cerebral
77. McIntosh AR, Gonzalez-Lima F (1994) Structural equation modeling and its
blood volume-weighted fMRI responses at columnar resolution.
application to network analysis in functional brain imaging. Hum Brain
Neuroimage 27: 416–424.
Mapp 2: 2–22.
63. Vanzetta I, Slovin H, Omer DB, Grinvald A (2004) Columnar resolution of
78. Friston KJ, Harrison L, Penny W (2003) Dynamic causal modeling.
blood volume and oximetry functional maps in the behaving monkey:
Neuroimage 19: 1273–1302.
Implications for fMRI. Neuron 42: 843–854.
64. Callaway EM, Katz LC (1990) Emergence and refinement of clustered 79. Bota M, Dong HW, Swanson LW (2003) From gene networks to brain
horizontal connections in cat striate cortex. J Neurosci 10: 1134–1153. networks. Nat Neuroscience 6: 795–799.
65. Amirikian B, Georgopoulos AP (2003) Modular organization of 80. Horwitz B, Sporns O (1994) Neural modeling and functional neuroimaging.
directionally tuned cells in the motor cortex: Is there a short-range order? Hum Brain Mapp 1: 269–283.
Proc Natl Acad Sci U S A 100: 12474–12479. 81. Horwitz B, Tagamets MA, McIntosh AR (1999) Neural modeling, functional
66. Kötter R, Wanke E (2005) Mapping brains without coordinates. Phil Trans brain imaging, and cognition. Trends Cogn Sci 3: 91–98.
R Soc Lond B Biol Sci 360: 751–766. 82. Arbib MA, Bischoff A, Fagg AH, Grafton S T (1994) Synthetic PET:
67. Thompson PM, Schwartz C, Lin RT, Khan AA, Toga AW (1996) Three- Analyzing large-scale properties of neural networks. Hum Brain Mapp 2:
dimensional statistical analysis of sulcal variability in the human brain. J 225–233.
Neurosci 16: 4261–4274. 83. Horwitz B (2004) Relating fMRI and PET signals to neural activity by means
68. Thompson P, Cannon TD, Narr KL, van Erp T, Poutanen VP, et al. (2001) of large-scale neural models. Neuroinformatics 2: 251–266.
Genetic influences on brain structure. Nat Neurosci 4: 1–6. 84. Chadderdon G, Sporns O (2005) A large-scale neurocomputational model
69. Amunts K, Schleicher A, Bürgel U, Mohlberg H, Uylings HBM, et al. (1999) of task-oriented behavior selection and working memory in prefrontal
Broca’s region revisited: Cytoarchitecture and intersubject variability. J cortex. J Cogn Neurosci. In press.
Comp Neurol 412: 319–341. 85. Bota M, Arbib MA (2004) Integrating databases and expert systems for the
70. Westbury CF, Zatorre RJ, Evans AC (1999) Quantifying variability in the analysis of brain structures: connections, similarities, and homologies.
planum temporale: A probability map. Cereb Cortex 9: 392–405. Neuroinformatics 2: 19–58.

PLoS Computational Biology | www.ploscompbiol.org 0251 September 2005 | Volume 1 | Issue 4 | e42