2002 Nature Publishing Group
different dendritic trees17. So, the different cytoarchitecture of different areas has consequences for processing within those areas. However, here we propose that each cytoarchitectonic area also has a unique set of extrinsic inputs and outputs, and this is crucial in determining the functions that the area can perform. We call this unique set a connectional fingerprint. The term fingerprint was first introduced by Hudspeth et al.18 to describe the distribution of cell density across cortical layers in the human primary visual cortex. Subsequently, Zilles and colleagues19 used it to describe the particular pattern of receptor architecture for each cortical region, as shown by the degree of binding for the different receptor types. FIGURE 1 presents an example of two anatomical fingerprints. These are based on a meta-analysis of data for prefrontal regions in the macaque connectivity database CoCoMac20. The areas are designated according to the numbers given by Walker21. In these fingerprints, the strength of any connection (rated as weak = 1, medium = 2, strong = 3) is shown by the radial distance. For simplicity of presentation, the fingerprints include only the local connections between prefrontal areas; more detailed fingerprints could be produced by including the afferents from and efferents to other regions of the brain. It can be seen that area 9 and 14 share some connections. For example, both receive afferents from area 25 and send efferents to area 24. However, even when they have common afferents or efferents these might differ in strength. For example, the strength of the efferents to area 13 differs for these two areas. Last, there also connections that are unique to one area or the other. This is true of the afferents to area 9 from areas 8A and 8B, or of the efferents from area 9 to area 46. It is the overall pattern that distinguishes the two areas. The hypothesis of unique connectional fingerprints has been supported by statistical analyses of cortical connectivity in primate22 and feline cortex23. For example, Young 22 compared the connectivity of visual, auditory and somatomotor regions in the macaque cortex. A strong test of the hypothesis is provided by analyses that restrict their focus to areas from a single cortical region. For example, the recent analyses by Stephan et al.20 and Ktter et al.24 used data from CoCoMac to examine the connectional organization of subregions within the prefrontal cortex. Fortunately, the connectivity of different prefrontal regions has been well studied25,26, and an extensive account of these data is available in CoCoMac. We have used these data to examine the hypothesis of unique connectional fingerprints, adapting the statistical analyses of Ktter and colleagues24. Methodological details have been reported elsewhere20,24,2729. In our formal analysis, we used two independent multivariate techniques MULTIDIMENSIONAL SCALING (MDS) and HIERARCHICAL CLUSTER ANALYSIS (HCA). These methods reveal similarities and dissimilarities between elements in a multidimensional feature space. Applied to the SPEARMAN CORRELATION MATRIX of the connectivity data, MDS and HCA provide intuitive visual representations of the relationships between cortical areas on

0.0 0.5 1.0 1.5 2.0

Dimension 1
Euclidean distances of Spearman correlation coefficients
Figure 2 | Analysis of prefrontal connectivity. a | Multidimensional scaling creates a high-dimensional metric representation in which distances between elements optimally reflect the overall similarity between their properties; in this case, the connectional patterns of the cortical areas (identified by their cytoarchitectonic numbers, as designated by Walker21). Spearman correlation coefficients between the connectivity vectors of the individual areas were computed, and MDS was then applied to these correlation coefficients, with Kruskals STRESS values to determine the goodness-of-fit. Values approaching zero denote a better fit. RSQ gives the proportion of variance of the scaled data that is explained by the distances computed by MDS. b | Hierarchical cluster analysis was applied to the same correlation coefficients between area-connectivity vectors using a Euclidean distance metric. Distances between areas are an inverse measure of the correlation between their connectional patterns. The fact that no two areas are merged at a distance of zero means that each of the prefrontal areas has a unique connectional fingerprint. The figure also shows that the definition of families of areas depends on the somewhat arbitrary choice of the threshold for similarity (see main text for details). Both parts of the figure are based on data from REF. 20.
HIERARCHICAL CLUSTER ANALYSIS
A multivariate method for solving classification problems. The object is to sort items into groups such that the degree of association is strong between members of the same cluster and weak between members of different clusters. In addition, this technique visualizes the hierarchical structure of similarity between all identified clusters.
SPEARMAN CORRELATION MATRIX
A matrix of so-called Spearman correlation coefficients, each of which represents a measure of association between two sets of rank-ordered measurements.
the basis of the similarities of their connectional fingerprints. MDS (FIG. 2a) arranges the prefrontal cortical areas in a sequence of lateral (45, 46, 8A), dorsal (8B, 9), dorsomedial (24) and orbitomedial (10, 11, 12, 13, 14, 25) areas. So, we find clusters of regions with varying degrees of resemblance. However, no two areas share the identical location, even after scaling similarities down to only two dimensions. This means that no two areas have exactly the same pattern of connections. This result is not trivial; although any sparse parcellation divides the cerebral cortex into areas with distinct characteristics, Walkers map21 is based on cytoarchitectural rather than connectional distinctions. The same message is conveyed independently by HCA. This procedure amalgamates the individual areas to groups on the basis of the similarities of their connectional fingerprints (FIG. 2b). Clearly, no two areas share the same pattern of connections. If any pair of areas did so, the distance at which the two areas merge would be zero. Compared with the MDS results, there are minor differences in the detailed arrangements (for example, areas 14 and 25 are the first to merge, but not the closest in MDS), but the groups of areas that emerge from HCA correspond exactly to the MDS arrangement. To show that the validity of our conclusions is not restricted to the prefrontal cortex, a similar analysis has been carried out for the premotor cortex27. Whereas Brodmann30 divided the prefrontal cortex into several different cytoarchitectonic regions, he defined the

premotor areas as a single area that is, area 6. However, on the basis of staining with cytochrome oxidase, it is possible to distinguish several subregions31,32. FIGURE 3 shows the anatomical fingerprints for areas F3 and F5 (REF. 27). These correspond to the supplementary motor area (SMA) proper and the ventral premotor area, respectively. As in FIG. 1, for simplicity, these fingerprints include only the local connections between the motor areas. FIGURE 4a presents the results of MDS. Again, no two areas share the same space. The analysis distinguishes between motor cortex (F1), the medial premotor cortex (SMA and pre-SMA, F3 and F6), the dorsolateral premotor cortex (F2 and F7) and the ventrolateral premotor cortex (F4 and F5). FIGURE 4b shows the results of HCA. The same subdivisions result, but with the added information that there is a relationship between the dorsolateral and medial sectors, and between the ventrolateral sector and the motor cortex. In the analysis of prefrontal and premotor areas, the pattern of connections has been studied using boundaries defined by either cytoarchitecture (prefrontal) or cytochrome oxidase staining (premotor). Ktter et al.27 have used a formal method for comparing the classification of areas by their connectivity with the classification of areas on the basis of other criteria, such as receptor architecture. The same method could now be used to compare formally the classification of prefrontal areas by connectivity (FIG. 2a,b) with the classification obtained for these areas on the basis of cytoarchitecture33.
somatosensory motor system and frontolimbic complex. More recently, Hilgetag et al.42 introduced optimal set analysis (OSA), a cluster analysis that is based on an evolutionary algorithm, to determine clusters of cortical areas on the basis of their anatomical connections. Similar clusters have been shown using data from 43 STRYCHNINE NEURONOGRAPHY. Applying several independent statistical approaches to these data on functional interactions, Stephan et al.43 showed that these interactions are not equally distributed. Instead, they are clustered into three main groups of areas sensorimotor, visual and orbito-temporo-insular clusters. We have shown that, in the prefrontal cortex (FIG. 2b) and premotor cortex (FIG. 4b), it is possible to detect clusters of areas with a similar, although not identical, pattern of connections. However, it is important to note that there is no objective criterion for defining the size of a family. As shown in FIG. 2b, the threshold for defining families of areas is arbitrary. For example, if one chooses d2 (dashed line in FIG. 2b) as a similarity threshold, one finds exactly the same three groups of areas that were identified by MDS (FIG. 2a). On the other hand, if a stricter threshold is chosen (d1 in FIG. 3), each of these three groups is broken up into two smaller clusters. We make the assumption that, for the purpose of functional localization, the more dissimilar their pattern of connections, the easier it will be to distinguish between the functions of areas. Until now, the standard for functional localization has been the double dissociation44. A lesion in area X should have an effect on task A but not on task B, whereas a lesion in area Y should have an effect on task B but not on task A. So, removal of superior temporal area 22 impairs the performance of auditory but not visual discriminations, and removal of inferior temporal area 21 impairs the performance of visual but not auditory discriminations5,45. Similarly, removal of parieto-occipital cortex impairs the ability to choose spatial locations on the basis of a landmark, but has much less effect on the performance of visual discriminations46,47; and removal of the inferotemporal cortex impairs the performance of visual discriminations, but has much less effect on the performance of the landmark task47. Removal of dorsal prefrontal area 46 leads to a very severe impairment on spatial delayed-response tasks, but does not impair the performance of visual discriminations4,8,48. Correspondingly, lesions in inferotemporal area 21 impair the performance of visual discriminations, but do not impair performance on delayed-response tasks5,49. These dissociations occur between areas with a very different pattern of connections that is, between areas belonging to different large families in parallel systems. For the above examples, these are the visual and auditory streams, the dorsal and ventral visual stream, and the ventral visual stream and the extension of the dorsal stream into the prefrontal cortex50. It has been more difficult to find double dissociations within streams, although they can be found. For example, Buckley et al.51 were able to show that lesions in inferotemporal area 21 impair the performance of colour discriminations but

Afferents of F3 FFF6 F3 F2
Afferents of F5 FFF6 F3 F2
Efferents of F3 FFF6 F3 F2
Efferents of F5 FFF6 F3 F2
Figure 3 | Diagram of anatomical fingerprints for two premotor areas F3 and F5. The upper row shows the afferent and the lower row the efferent connections of the supplementary motor area (F3), the ventral premotor area (F5) and other motor areas that are defined on the basis of staining with cytochrome oxidase31,32. The strength of any connection (rated as absent = 0, weak = 1, medium/ambiguous = 2, strong = 3) is shown by the radial distance. Based on data from REF. 27.

Connectional families

STRYCHNINE NEURONOGRAPHY
A method in which (potentially polysynaptic) anatomical connections are identified by applying strychnine to one area and then recording spikes in other areas.
In functional systems, different areas share some of their inputs and outputs. For example, Selemon and Goldman-Rakic 34 have pointed out that there are similarities in the pattern of outputs of the parietal lobe area 7a and the lateral intraparietal area (LIP), and the prefrontal area 46 with which they are interconnected. They form part of the same distributed system. It is presumably the common connectivity patterns that lead to the functional co-activation of areas within this distributed system. For example, studies using 2-deoxyglucose35, single-unit recording36, positron emission tomography (PET)37 and functional MRI (fMRI)38 show co-activation of the dorsal prefrontal cortex and intraparietal cortex during a spatial working memory task. We suggest the term family for a cluster of areas that share a similar pattern of connections. We take the term from Zilles et al.39, who noted that related areas within the motor system, such as the SMA and the pre-SMA, can be grouped into neurochemically similar families on the basis of receptor mapping. A formal proof of the existence of connectional families was provided by Young22,40 for the macaque brain and by Scannell et al.23,41 for the feline brain. Young22 used MDS to distinguish between visual areas in the dorsal and ventral visual stream, and between the auditory system,
VOLUME 3 | AUGUST 2002 | 9
2 STRESS = 0.03 RSQ = 0.99

F1 F6 F3

Dimension 2

F4 F5 F7 F2

Figure 4 | Analysis of premotor connectivity. Areas are defined on the basis of staining with cytochrome oxidase31,32. a | Analysis using multidimensional scaling (see legend to FIG. 2a). b | Analysis using hierarchical cluster analysis (see legend to FIG. 2b). The fact that no two areas are merged at a distance of zero means that each of the motor areas has a unique connectional fingerprint. Both analyses are based on data from REF. 27.
not the recognition of objects, whereas lesions in the perirhinal cortex had the opposite effect. In general, the more connections two areas share, the more difficult it is to find double dissociations between them. Young et al.11 have discussed at length the problems associated with making functional interpretations on the basis of double dissociations, but have shown formally how knowledge about connectivity can help in interpreting the effects of lesions.

Proportions of functional cell types
A crucial question is the extent to which differences in the patterns of activity of cell firing in different areas are determined by differences in extrinsic connectivity of these areas. Unfortunately, we do not have adequate information on intrinsic connectivity, which also differs between areas. However, the extrinsic connections must set a limit for the processing that can occur within the area. This idea has been previously expressed by Young22, who stated that the place of an area in the cortical macro-circuitry might determine in large part the areas functional properties. The analysis is clearly easier the nearer we are to the sensory inputs. For example, reviewing physiological studies of visual areas V4 and V5, DeYoe and Van Essen52 found that 85% of cells in V5 showed direction selectivity, whereas only 5% of cells in V4 did so. By contrast, 50% of the cells in V4 showed neuronal activity that was selective to wavelength, whereas no cells had been found in V5 that did so. In both areas, there was orientationselective activity, accounting for 75% of the cells in V5 and 50% of the cells in V4. The difference in the pattern of activity can be directly related to the visual inputs to these areas from the magnocellular and parvocellular pathways53. However, even in the case of early visual processing, there is considerable complexity in trying to
account for the processing of form or motion in terms of the detailed anatomy of the pathways53,54. Clearly, the problem of relating the physiology to the extrinsic connections becomes much more intractable the further one is from peripheral sensory inputs. If one compares any two areas, both will have a very large number of inputs, and the two areas might be connected through one or two synaptic relays22. Consider, for example, the premotor cortex, the SMA and the motor cortex. These structures lie within the somatomotor system that was defined by Young22 on the basis of multivariate statistics. They share many inputs and outputs, and are interconnected5557. It is therefore not surprising that cells can be found in these three areas that fire in association with the same task events. For example, one can find similar cell types in the SMA and the motor cortex5861, in the premotor cortex and the motor cortex62,63, in the pre-SMA and the SMA64,65, and in the premotor cortex and the SMA66,67. However, the proportions of cells with activity that is related to particular tasks or task components differ between these areas. For example, Shen and Alexander62,63 compared the activity of cells in the premotor and motor cortices. They distinguished between cells that fired in relation to the target location and cells that fired in relation to the direction of the movement. There were more target cells in the premotor than in the motor cortex, and more direction cells in the motor cortex. The crucial question is whether it can be shown that differences in the proportions of functional cell types between areas relate to their different inputs or outputs. FIGURE 5 presents a worked example from Mushiake et al.66. The comparison was made between the ventral premotor cortex (F5), the SMA (F3) and the motor cortex (F1). The monkeys were trained on two tasks. In the first, they performed a sequence of three movements

Premotor area

Supplementary motor area

Primary motor cortex

% cells
Figure 5 | Cell firing associated with a pair of visually and memory-guided tasks. Distribution of cells in ventral premotor area F5, supplementary motor area F3 and motor cortex (F1), classified according to the degree to which they were active in association with the visually guided sequence (VS) or the same sequence performed from memory (MS). 1 = exclusively related to VS, 2 = much more related to VS, 3 = more related to VS, 4 = equally related to VS and MS, 5 = more related to MS, 6 = much more related to MS, 7 = exclusively related to MS. Reproduced, with permission, from REF. The American Physiological Society.
as instructed by visual cues. In the second, they performed the same sequence from memory. The figure shows the percentage of cells that fall into one of seven categories, with the actual number of cells above each histogram; the analysis is for the movement period, but similar results were found for the pre-movement period. Category 4 was given to cells that fired equally on the two tasks, category 1 to cells that fired exclusively on the visually guided task, and category 7 to cells that fired exclusively on the memory guided task. The other categories are for activity that was associated more with one task than the other. As shown previously, MDS of connectivity data places the motor cortex (F1) on its own among the motor areas (FIG. 5). In the motor cortex, almost all the cells fired equally on the two tasks (category 4). In other studies, it has also been shown that, when monkeys perform sequences of movement, cells in the motor cortex tend to fire in association with the execution of individual movements61,67. FIGURE 5 also shows that, even in the premotor cortex and the SMA, many cells also fire equally (category 4) irrespective of whether the sequence is guided by visual cues or performed from memory. However, the overall pattern differs between the areas. MDS (FIG. 4a) and HCA of connectivity data (FIG. 4b) distinguish between the ventrolateral premotor cortex (F5) and the SMA (F3). In area F5, the greater proportion of cells fired in association with sequences guided by visual cues (categories 13), whereas in area F3 the greater proportion of cells fired in association with sequences performed from memory (categories 57). There was a statistically significant difference between the overall pattern for the motor cortex and those for other areas, and between the patterns seen for F5 and F3 (REF. 66). Other studies have also shown that many cells in F3 fire when monkeys perform sequences from memory, with many cells firing differently according to the specific sequence that is performed61,67.

The question is whether the differences in the overall pattern relate to differences in the pattern of inputs. The anterior intraparietal area (AIP) projects to F5, but not to F3 or F1 (for a review, see REF. 68). Many cells in AIP are visual in the sense that they fire when monkeys observe objects that they are going to grasp6972. F5 also receives a heavy input from thalamic nucleus X (REFS 7375), which in turn receives input from the dentate nucleus of the cerebellum75. Van Donkelaar et al.76 recorded in nucleus X when monkeys were performing visually guided or internally generated movements, and found that most of the cells fired exclusively or preferentially in association with the visually guided task. In the study by Mushiake et al.66, there was also a tendency for cells in the dorsal premotor cortex (F7) to fire in association with the visually guided task. Furthermore, in progressing anteriorly from area 4 through dorsal area 6, there was a progressive increase in the proportion of cells that fired in association with the presentation of visual cues77. There are projections to the anterior part of dorsal area 6 (F7) from the middle intraparietal area (MIP) in the dorsal bank of the intraparietal sulcus78, and there are visual responses in MIP79. There is also a decrease in the proportion of cells that fire at the time of movement as one progresses anteriorly from area 4 through area 6 (REF. 77). Motor cortex and the posterior part of the dorsal premotor cortex (area F2) receive the somatic input from parietal area 5 (REF. 68). These findings indicate a possible relationship between anatomical inputs and the electrophysiological data.

Functional fingerprints

The above examples provide informal support for the proposed relationship between connectional fingerprints and the functional properties of areas. To provide more formal support, we have analysed the data of Humphrey and Tanji80. These data give the response properties of cells in the motor cortex, the SMA, the
VOLUME 3 | AUGUST 2002 | 1

% Set-related only

PL5 SMA PMd PMa M1
2 STRESS = 0.04 RSQ = 0.99
PL5 Set-related Dimension 2 PMa SMA PMd Movement/muscle Motor coupling Proprioceptive/ cutaneous Auditory/ visual M1

% Motor coupling

% Auditory/ visual

% Movement/ muscle

% Proprioceptive/ cutaneous
Figure 6 | Functional fingerprints for five motor areas. a | For each area, the radial plots give the proportions of cells that show set-related discharge only, that are responsive to auditory/visual stimulation, that are responsive to proprioceptive/cutaneous stimulation, that show specificity in their movement/muscle field, or that are coupled to motor variables. M1, motor cortex area 4; PL5, superior parietal area 5; PMa, post-arcuate premotor cortex (part of the ventral premotor area); PMd, dorsal premotor cortex; SMA, supplementary motor cortex. b | Multidimensional scaling (MDS) of these data. The unfolding model was used, which treats proximities between cortical areas and neuronal response properties as a submatrix of the regular MDS matrix128. M1 cells are the most strongly related to motor coupling, whereas premotor areas preferentially show set-related activity. Activity in parietal area 5 is mainly movement or muscle related. RSQ gives the proportion of variance of the scaled data that is explained by the distances computed by MDS. Both parts of the figure are based on data from REF. 80.

SET-RELATED ACTIVITY

Neuronal activity that reflects the behavioural set of the animal, which can include information about a planned movement or about the state of readiness of the animal.
dorsal premotor cortex and the arcuate premotor area, which forms part of the ventral premotor cortex. The analysis includes the superior parietal cortex (area 5), which makes direct connections with the motor cortex68. Humphrey and Tanji80 analysed published studies to rate cells in these five areas in terms of their discharge properties. There are five scales: first, the proportion of cells that show SET-RELATED ACTIVITY only; second, the proportion that are responsive to auditory/visual stimulation; third, the proportion that are responsive to proprioceptive/cutaneous stimulation; fourth, the proportion that show specificity in their movement/muscle field; and fifth, the relative coupling to motor variables. FIGURE 6a shows the data for each area in the form of functional fingerprints. We use this term to describe a polar plot in which the data are the proportions of active neurons with given response properties. These response properties can be assessed over a series of different tasks (such as memory-guided or visually guided sequences) or in terms of other properties of cellular discharge (such as set-related or muscle/movement fields). The same format is used for these plots as for the anatomical fingerprints, the difference being that the rating is for the proportion of cells with specific properties, rather than the strength of the connections. Parietal area 5 and the motor cortex are similar in their fingerprints, and they differ in their fingerprints from the three premotor areas. Because the ratings of Humphrey and Tanji80 resulted from a meta-analysis of the literature, and not from a single study in which all five measures were obtained, we should be cautious in drawing any other firm conclusions from the fingerprints.

Relating functional to anatomical fingerprints
One of the goals in neuroscience is to establish comprehensive and quantitative structurefunction relationships across all levels of brain organization. A large number of variables would be required for a complete model of brain architecture and dynamics, and measuring these variables simultaneously and in real time is impossible. It is therefore essential that relatively simple metrics for brain architecture and dynamics be found to allow a sufficiently accurate description of the brain as a dynamic system97. We believe that connectional and functional fingerprints, as defined here, provide useful measures for this purpose. Anatomical connectivity fundamentally constrains effective connectivity that is, how distinct brain structures causally influence each other at both the synaptic and population levels98. Effective connectivity is fundamental for the principles of brain dynamics, including population rates and synchrony97, and is the most likely direct determinant of functional fingerprints, as defined here. So, connectional fingerprints are linked with functional fingerprints through effective connectivity. We do not claim to have provided a formal proof of the relationship between anatomical and functional fingerprints. This would require the investigation of large sets of connectional and functional data for various cortical areas. A pioneering step in this direction has been made by Scannell and co-workers99. They predicted the likely responses of cells in the anterior ectosylvian visual area of the cat to moving gratings and plaid stimuli. Their prediction was made on the basis of a multivariate analysis of a large database that comprised the entire network of feline cortico-cortical connections23.
VOLUME 3 | AUGUST 2002 | 3
Subsequently, they confirmed their predictions by single-unit recordings. Their approach has recently been continued by Burns and Young100, who also used a large database, and mathematically analysed on this basis the connectivity of hippocampus-related structures in the rat. They found good agreement between the connectional organization and known physiological properties of neurons in the various areas. What we now need are formal analyses of connectional and functional fingerprints for the same cortical areas and in the same species. One methodological problem is that these two sets of data can be on different scales: connectivity data are on an ordinal scale, whereas electrophysiological data are on a ratio scale. Recently, a mixture of correlation analyses and multivariate techniques has been suggested to deal with this type of problem 27. Briefly, the principle of this approach is to compute similarity profiles in each data modality by applying scale-dependent cross-correlation techniques to the feature vectors of all areas. The structure of the resulting correlation matrices can then be compared qualitatively by multivariate classification techniques such as MDS or HCA. In addition, MULTIPLE CORRESPONDENCE ANALYSIS (MCA) can be used to determine the overall classification of areas on the basis of the combined data sets. In addition, this method can show the correspondence between variables from different data modalities. Further advances towards a quantitative assessment of the direct relationship between data sets can be derived from the GENERAL LINEAR MODEL that is, multivariate analysis of covariance or canonical correlation analysis101. Alternatively, INFORMATION THEORY could be used to quantify directly the degree of mutual information between different data sets102. So, it should be feasible to test the hypothesis that we have put forward in this paper. We suggest that it would be most practicable to attempt the task for the visual areas, and to collect the functional data by brain imaging. This would be done best in the macaque, in which both detailed information on anatomical connections and sophisticated fMRI technology are available85,103. The functional data could also be collected for the human brain, although this would introduce the added problem that assumptions have to be made about the correspondence of connections in the human and macaque brain. We have very little direct information, other than from studies of anterograde degeneration104, about the connections of the human brain105. It is not yet clear to what extent DIFFUSION106108 WEIGHTED IMAGING will be able to discriminate the fine details of anatomical connections that can be observed using tracing methods. For the time being, despite the uncertainty as to how well identical cytoarchitecture predicts functional equivalence, the inputs and outputs of areas that are activated in imaging experiments on human subjects must be inferred from the connections of areas with similar cytoarchitecture in the macaque brain109. This means that one must be able to identify the cytoarchitectonic area in which there is activation in the human brain. The problem can be solved by producing a probability atlas that incorporates data on the cytoarchitecture of a group of brains, and therefore provides the probability that an activation is localized in any particular area110,111. Information of this sort is available for only some regions, such as the primary somatosensory cortex112 and Brocas area113, and it will be a while before data are available for the entire brain. These problems are perhaps of less concern for studies of the early visual areas. Much progress has been made in establishing homologies between visual areas in the human and macaque brain114117, and it is unlikely that there are significant differences between the connectivity of these areas in these species. However, the most reliable method would be to compare in macaque monkeys functional data from MRI103,118 with connectional data. The connections of visual areas have been much researched and are well described40,119. It is feasible to present a wide variety of visual stimuli in the same fMRI experiment. Stimuli could be presented that make demands not only on the low-level processing of aspects such as colour, orientation and motion, but also on higher-level processing, as in the perception of objects120,121, faces122, aftereffects123 and illusions124. The practicality of presenting a wide range of visual stimuli during a single fMRI session has already been shown. The results of such studies can be seen in Orban et al.125, Sunaert et al.126 and Moore and Engel127. So, by using the statistical methods described above, it should be possible to determine functional fingerprints of the various visual areas, and to relate them to connectional fingerprints, such as those as documented in the CoCoMac database20.

Conclusions

MULTIPLE CORRESPONDENCE ANALYSIS
A method that aims to explain the relationships between multiple variables that are identified on identical or different measurement scales, and may include categorical data.

GENERAL LINEAR MODEL

A general mathematical framework from which many commonly used statistical procedures (for example, analysis of variance) are derived.

INFORMATION THEORY

A scientific discipline that is concerned with mathematical laws underlying systems that transmit, store and process information. It also deals with the quantitative measurement of various types of information.
DIFFUSION-WEIGHTED IMAGING
A magnetic resonance imaging method that makes use of the variability in the random movement of water molecules in nervous tissue, which is restricted by cell bodies, blood vessels, axon bundles and other structures. Two opposite magnetic field gradients are applied. The magnetic spins will be de-phased by the first gradient and, because of water diffusion, the second gradient will not completely re-phase them. As the directionality of diffusion is highly ordered in white matter, the spatial orientation of the bundles can be reconstructed.
Here, we have spelled out what many neuroscientists probably believe already. However, we hope that we have clarified the necessary stages of the argument, made it clear where we have relevant evidence, and indicated what evidence should be collected to establish the last stage of the argument. The paper has made five claims. First, that each cytoarchitectonic area has a unique connectional fingerprint; we have provided worked examples for prefrontal and premotor areas using the CoCoMac database. Second, that there are families of areas that share a resemblance in their connections; again, we have provided examples for prefrontal and premotor areas using CoCoMac. Third, that the proportion of cells that fire in association with different tasks or task events differs between areas; areas have their own functional fingerprints. We have provided examples for the premotor areas. Fourth, that the differences between these functional fingerprints are determined by the extrinsic and intrinsic connections of these areas. Last, that imaging will be a useful tool for detecting functional fingerprints. Carrying out fMRI studies on the many areas of the visual system could allow a formal test of the relationship between functional and anatomical fingerprints.

54. 55.

9. 10. 11.

21. 22.

VOLUME 3 | AUGUST 2002 | 5
receiving areas of the thalamus during visually-triggered and internally-generated limb movements. J. Neurophysiol. 82, 934945 (1999). Johnson, P. B., Ferraina, S., Bianchi, L. & Caminiti, R. Cortical networks for visual reaching: physiological and anatomical organization of frontal and parietal lobe arm regions. Cereb. Cortex 6, 102119 (1996). Matelli, M., Govoni, P., Galletti, C., Kutz, D. F. & Luppino, G. Superior area 6 afferents from the superior parietal lobule in the macaque monkey. J. Comp. Neurol. 402, 327352 (1998). Colby, C. L. & Duhamel, J. R. Heterogeneity of extrastriate visual areas and multiple parietal areas in the macaque monkey. Neuropsychologia 29, 517537 (1991). Humphrey, D. R. & Tanji, J. in Motor Control: Concepts and Issues (eds Humphrey, D. R. & Freund, H.-J.) 413443 (Wiley, New York, 1991). A meta-analysis of studies on the features to which cells respond in three cortical motor areas and in parietal area 5. The data form the basis for the functional fingerprints that are presented in figure 6a of the present article. Moffet, A., Ettlinger, G., Morton, H. B. & Piercy, M. F. Tactile discrimination performance in the monkey: the effect of ablation of various subdivisions of posterior parietal cortex. Cortex 3, 5996 (1967). Heeger, D. J., Boynton, G. M., Demb, J. B., Seidemann, E. & Newsome, W. T. Motion opponency in visual cortex. J. Neurosci. 19, 71627174 (1999). Heeger, D. J., Huk, A. C., Geisler, W. S. & Albrecht, D. G. Spikes versus BOLD: what does neuroimaging tell us about neuronal activity? Nature Neurosci. 3, 631633 (2000). Rees, G., Friston, K. & Koch, C. A direct quantitative relationship between the functional properties of human and macaque V5. Nature Neurosci. 3, 716723 (2000). Logothetis, N. K., Pauls, J., Augath, M., Trinath, T. & Oeltermann, A. Neurophysiological investigation of the basis of the fMRI signal. Nature 412, 150157 (2001). In this study, electrophysiological activity and the BOLD signal were measured simultaneously in macaques. This paper provides the most direct evidence on the relationship between these two measures. Bandettini, P. A. & Ungerleider, L. G. From neuron to BOLD: new connections. Nature Neurosci. 4, 864866 (2001). Chawla, D., Lumer, E. & Friston, K. J. The relationship between synchronisation among neuronal populations and their mean activity levels. Neural Comput. 11, 13891411 (1999). Chawla, D., Lumer, E. D. & Friston, K. J. Relating macroscopic measures of brain activity to fast, dynamic neuronal interactions. Neural Comput. 12, 28052821 (2000). Georgopoulos, A. P., Kalaska, J. F., Caminiti, R. & Massey, J. T. Spatial coding of movement: a hypothesis concerning the coding of movement direction by motor cortical populations. Exp. Brain Res. 7, 327336 (1983). Georgopoulos, A. P., Schwartz, A. B. & Kettner, R. A. Neuronal population coding of movement direction. Science 233, 14161419 (1988). Chen, L. L. & Wise, S. P. Conditional oculomotor learning: population vectors in the supplementary eye field. J. Neurophysiol. 78, 11661169 (1997). Wise, S. P. & Murray, E. A. Arbitrary associations between antecedents and actions. Trends Neurosci. 23, 271276 (2000). Paus, T., Koski, L., Zografos, C. & Westbury, C. Regional differences in the effects of task difficulty and motor output on blood flow response in the human anterior cingulate cortex: a review of 107 PET activation studies. Neuroreport 9, 3747 (1998). 94. Koski, L. & Paus, T. Functional connectivity of the anterior cingulate cortex within the human frontal lobe: a brainmapping meta-analysis. Exp. Brain Res. 133, 5565 (2000). 95. Duncan, J. & Owen, A. M. Common regions of the human frontal lobe recruited by diverse cognitive demands. Trends Neurosci. 23, 475483 (2000). 96. Van Horn, J. D. et al. The Functional Magnetic Resonance Imaging Data Center (fMRIDC): the challenges and rewards of large-scale databases in imaging studies. Phil. Trans. R. Soc. Lond. B 356, 13231339 (2001). 97. Friston, K. J. The labile brain. I. Neuronal transients and nonlinear coupling. Phil. Trans. R. Soc. Lond. B 355, 215236 (2000). 98. Friston, K. J. Functional and effective connectivity in neuroimaging: a synthesis. Hum. Brain Mapp. 2, 5678 (1995). 99. Scannell, J. W. et al. Visual motion processing in the anterior ectosylvian sulcus of the cat. J. Neurophysiol. 76, 895907 (1996). Scannell et al. used multivariate analysis of a large database to predict the response properties of cells in a higher visual area in the cat. They confirmed the predictions by making direct electrophysiological recordings in this area. 100. Burns, G. A. P. C. & Young, M. P. Analysis of the connectional organization of neural systems associated with the hippocampus in rats. Phil. Trans. R. Soc. Lond. B 355, 5570 (2000). 101. Chatfield, C. & Collins, A. J. An Introduction to Multivariate Analysis (Chapman & Hall, New York, 1991). 102. Cover, T. M. & Thomas, J. A. Elements of Information Theory (John Wiley, New York, 1991). 103. Logothetis, N. K., Guggenberger, H., Peled, S. & Pauls, J. Functional imaging of the monkey brain. Nature Neurosci. 2, 555562 (1999). 104. Di Virgilio, G. & Clarke, S. Direct interhemispheric visual inputs to human speech areas. Hum. Brain Mapp. 5, 347354 (1997). 105. Crick, F. & Jones, E. Backwardness of human neuroanatomy. Nature 361, 109110 (1993). 106. Conturo, T. E. et al. Tracking neuronal fiber pathways in the living human brain. Proc. Natl Acad. Sci. USA 96, 1042210427 (1999). 107. Poupon, C. et al. Regularization of diffused-based direction maps for the tracking of brain white matter fascicles. Neuroimage 12, 184195 (2000). 108. Parker, G. J. M. et al. In vivo tracing of anatomical fibre tracts in the macaque and human brain using diffusion tensor imaging and fast marching tractography. Neuroimage 15, 797809 (2002). 109. Petrides, M. & Pandya, D. N. Dorsolateral prefrontal cortex: comparative cytoarchitectonic analysis in the human and the macaque brain and corticocortical connection patterns. Eur. J. Neurosci. 11,10111036 (1999). 110. Roland, P. E. & Zilles, K. The developing European computerized human brain database for all imaging modalities. Neuroimage 4, 3947 (1996). 111. Mazziotta, J. C. et al. in Brain Mapping: the Systems (eds Toga, A. & Mazziotta, J. C.) 132158 (Academic, New York, 2000). 112. Geyer, S., Schormann, T., Mohlberg, H. & Zilles, K. Areas 3a, 3b, and 1 of human primary somatosensory cortex. II. Spatial normalization to standard anatomical space. Neuroimage 11, 684696 (2000). 113. Amunts, K. et al. Brocas region revisited: cytoarchitecture and intersubject variability. J. Comp. Neurol. 412, 319341 (1999). 114. Watson, J. D. C., Frackowiak, R. S. J. & Zeki, S. in Functional Organization of the Human Visual Cortex (eds Gulyas, B., Ottoson, D. & Roland, P. E.) 317328 (Pergamon, Oxford, UK, 1993). 115. Shipp, S., Watson, J. D. G., Frackowiak, R. S. J. & Zeki, S. Retinotopic maps in human prestriate visual cortex: the demarcation of area V2 and V3. Neuroimage 2, 125133 (1995). 116. Tootell, R. B. H. & Taylor, J. B. Anatomical evidence for MT and additional cortical visual areas in humans. Cereb. Cortex 1, 3955 (1995). 117. Van Essen, D. C. et al. Mapping visual cortex in monkeys and humans using surface-based atlases. Vision Res. 41, 13591378 (2001). 118. Vanduffel, W. J. M. et al. Visual motion processing investigated using contrast agent-enhanced fMRI in awake behaving monkeys. Neuron 32, 565577 (2001). 119. Felleman, D. J. & Van Essen, D. C. Distributed hierarchical processing in primate cerebral cortex. Cereb. Cortex 1, 147 (1990). 120. Grill-Spector, K. et al. A sequence of object-processing stages revealed by fMRI in the human occipital lobe. Hum. Brain Mapp. 6, 316328 (1998). 121. Sereno, M. E., Trinath, T., Augath, M. & Logothetis, N. K. Three-dimensional shape representation in monkey cortex. Neuron 33, 635652 (2002). 122. Halgren, E. et al. Location of human face-selective cortex with respect to retinotopic areas. Hum. Brain Mapp. 7, 2927 (1999). 123. Tootell, R. B. H. et al. Visual motion aftereffect in human cortical area MT revealed by functional magnetic resonance imaging. Nature 375, 139141 (1995). 124. Seghier, M. et al. Moving illusory contours activate primary visual cortex: an fMRI study. Cereb. Cortex 10, 663670 (2000). 125. Orban, G. A., Sunaert, S., Todd, J. T., Van Hecke, P. & Marchal, G. Human cortical regions involved in extracting depth from motion. Neuron 24, 929940 (1999). 126. Sunaert, S., Van Hecke, P., Marchal, G. & Orban, G. A. Attention to speed of motion, speed discrimination and task difficulty: an fMRI study. Neuroimage 11, 612623 (2000). 127. Moore, C. & Engel, S. A. Neural response to perception of volume in the lateral occipital complex. Neuron 29, 277286 (2001). References 125127 show the feasibility of using fMRI to present a wide range of visual stimulus types. The results of such studies could be used to construct functional fingerprints of different areas. 128. Borg, I. & Groenen, P. Modern Multidimensional Scaling (Springer, New York, 1997).

86. 87.

Acknowledgements
This work was supported by the Wellcome Trust (R.E.P.), the Brain Research Trust (K.E.S.) and the Deutsche Forschungsgemeinschaft (R.K.). We are grateful to C. Hilgetag and K. Friston for their comments on the manuscript before submission, and to A. Duggins and W. Penny for helpful statistical discussions.

Online links

FURTHER INFORMATION CoCoMac: http://www.cocomac.org/ Encyclopedia of Life Sciences: http://www.els.net/ brain imaging: localization of brain functions | brain imaging: observing ongoing neural activity | computed tomography | magnetic resonance imaging MIT Encyclopedia of Cognitive Sciences: http://cognet.mit.edu/MITECS/ cerebral cortex | cortical localization, history of | Lashley, Karl Spencer (18901958) | magnetic resonance imaging | positron emission tomography Access to this interactive links box is free online.