Organization of parietoprefrontal and temporoprefrontal networks in the macaque

Abstract

The connectivity among architectonically defined areas of the frontal, parietal, and temporal cortex of the macaque has been extensively mapped through tract-tracing methods. To investigate the statistical organization underlying this connectivity, and identify its underlying architecture, we performed a hierarchical cluster analysis on 69 cortical areas based on their anatomically defined inputs. We identified 10 frontal, four parietal, and five temporal hierarchically related sets of areas (clusters), defined by unique sets of inputs and typically composed of anatomically contiguous areas. Across the cortex, clusters that share functional properties were linked by dominant information processing circuits in a topographically organized manner that reflects the organization of the main fiber bundles in the cortex. This led to a dorsal-ventral subdivision of the frontal cortex, where dorsal and ventral clusters showed privileged connectivity with parietal and temporal areas, respectively. Ventrally, temporofrontal circuits encode information to discriminate objects in the environment, their value, emotional properties, and functions such as memory and spatial navigation. Dorsal parietofrontal circuits encode information for selecting, generating, and monitoring appropriate actions based on visual-spatial and somatosensory information. This organization may reflect evolutionary antecedents, in which the vertebrate pallium, which is the ancestral cortex, was defined by a ventral and lateral olfactory region and a medial hippocampal region.

NEW & NOTEWORTHY The study of cortical connectivity is crucial for understanding brain function and disease. We show that temporofrontal and parietofrontal networks in the macaque can be described in terms of circuits among clusters of areas that share similar inputs and functional properties. The resulting overall architecture described a dual subdivision of the frontal cortex, consistent with the main cortical fiber bundles and an evolutionary trend that underlies the organization of the cortex in the macaque.

INTRODUCTION

In the cerebral cortex, the neural computations underlying complex brain functions are performed by populations of neurons belonging to different architectonically defined areas, that, operating in unison, give rise to multiple interconnected brain networks (1, 2). Therefore, a description of the connectivity among areas within these networks is crucial for understanding the functional organization of the forebrain.

Early studies, based on less sensitive axonal degeneration methods, identified dominant patterns in the connectivity of the ventral and the dorsal prefrontal cortex (3–5). Subsequent to this, the introduction of increasingly sensitive axonal tracing methods (6) provided a complex dataset of connectivity. One can, however, identify several organizational principles within this complexity (7–10). For example, within the auditory, visual, and somesthetic systems, there is an orderly progression of projections within each sensory system, as well as to frontal areas, before the systems converge (7). The laminar projection pattern also defines the feed-forward and the feedback flow of information within each sensory system (11). Specifically, feed-forward connections tend to originate in layer 3 and terminate in layer 4, whereas feedback projections tend to originate in deeper layers (5 and 6) and terminate in layer 1 (12). Furthermore, feed-forward versus feedback projections can be predicted by the degree of laminar elaboration (agranular, dysgranular, and granular) of areas (9).

Several studies have addressed the statistical organization of cortex through the application of computational methods (2, 13–24), providing a partial statistical description of the connectivity between subsets of cortical areas. This work has suggested that the organization of connections between association areas reflects the underlying fiber systems through which cortical-cortical axons pass (25) as well as the evolutionary and related developmental topology of the vertebrate pallium that evolved into the mammalian cortex (26–29).

In previous studies, we characterized the intrinsic connectivity of areas within frontal and parietal cortices, as well as the organization of parietofrontal networks of the macaque brain (19, 20, 24). In the present study, we extended the analysis to much of the cortex, excluding only primary sensory areas. We included most frontal, temporal, and parietal areas for which we could identify modern tract-tracing data. Our dataset consisted of a connectivity matrix among all areas generated by examining available tract-tracing studies in the literature. We then applied a hierarchical clustering algorithm to the connectivity matrix. Our method clustered areas that have similar inputs and allowed us to identify a well-defined hierarchical organization of frontal, temporal, and parietal areas. Based on their inputs, we identified 10 frontal, four parietal, and five temporal hierarchically related sets of areas (clusters). These clusters were composed of areas that shared basic functional properties and, in most instances, were anatomically contiguous. Moreover, the analysis of the connectivity among clusters revealed the presence of distinct information processing circuits spanning the entire cortex. These circuits subserve different sensory, motor, and cognitive functions, which we consider in the discussion.

METHODS

We applied the same methods used in our previous studies (19, 20, 24). We compiled a connectivity matrix N (inputs to each area) × N (areas that send inputs) by examining the primary literature of anatomical tract-tracing studies in the frontal, parietal, and temporal cortex of macaque monkeys (Macaca fascicularis, Macaca mulatta, Macaca nemestrina, Macaca fuscata) (see appendix for the list of studies and Table 1 for the list of cortical areas). See Supplemental Table S1 (https://doi.org/10.6084/m9.figshare.15031695.v1) for the connectivity matrix.

Table 1.

List of cortical areas and abbreviations used in this study along with a comparison with other nomenclatures used in the literature

Prefrontal Areas
Subdivision used in this study	Walker (30)	Petrides and Pandya (31)	Barbas and Pandya (32)	Carmichael and Price (33)
8A (FEF, frontal eye field), 8B	8A, 8B	8Av, 8Ad, 8B	8d, 8v
9l, 9m	9	9	9l, 9m
a10	10	10	10	10m, 10o
a11	11	11	11	11m, 11l
a12l, a12o	12	47/12	12l, 12o	12l, 12o, 12m, 12r,
a13	13	13	13	13a, 13b, 13l, 13m
14m, 14o, a25	14	14, 25	14, 25	14r, 14c,
a32	32	32	32
46d, 46v	46	46, 9/46d, 9/46v,	46dr, 46dc, 46vr, 46vc
45a, 45b	45	45A, 45B
PrCo, precentral opercular area

Premotor/Motor Areas			Parietal Areas
Subdivision used in this study [Matelli et al. (34,35)]		Barbas and Pandya (32)	Subdivision used in this study [* Pandya and Seltzer (36)]
F2vr, ventrorostral dorsal premotor cortex		6DC	PF, area PF of the inferior parietal lobule*
F2 pre-CD, precentral dimple			PFop, parietal opercular region*
F4, PMvc, ventrocaudal premotor cortex		6Va	PGop, second parietal opercular region*
F5, PMvr, ventrorostral premotor cortex		6Vb	PG, area PG of the inferior parietal lobule*
F3, SMA, supplementary motor area			PFG, area PFG of the inferior parietal lobule*
F6, pre-SMA, pre-supplementary motor area		MII	PGm, medial parietal area PGm*
F7, PMdr, rostral area of the dorsal premotor cortex		6DR	PE, parietal association area*
F7-SEF, supplementary eye field			PEa, anterior parietal association area*
Cingulate cortex [Dum and Strick (37)]			PEc, caudal parietal association area*
CMAd, dorsal cingulate motor area		23	Opt, area Opt of the inferior parietal lobule*
CMAv, ventral cingulate motor area; 23c			SI, primary somatosensory cortex
24c (CMAr, rostral cingulate motor area), 24a, 24b		24	SII, secondary somatosensory cortex
			MST, middle superior temporal visual area [Maunsell and van Essen (38)]
			VIP, ventral intraparietal sulcal cortex [Maunsell and van Essen (38)]
			LIP, lateral intraparietal sulcal cortex [Andersen et al. (39)]
			MIP, medial intraparietal area [Colby et al. (40)]
			V6A, visual area 6A [Galletti et al. (41)]
			V6, visual area 6 [Galletti et al. (41)]

Temporal Areas				Insular Areas
Subdivision used in this study		Baylis et al. (42)	Seltzer and Pandya (43)	Subdivision used in this study [Mesulam and Mufson (44)]
STGr, rostral portion of the superior temporal gyrus		TS	TS1-TS2-TS3	Ig, granular insular cortex
TE, temporal cortex		TE1, TE2,	TE1, TE2, TE3	Ia, agranular insular cortex
TEO, occipital portion of the temporal cortex		TE3	TEO	Id, dysgranular insular cortex
STSdr, dorsorostral portion of the superior temporal sulcus		PGa, TPo,
STSv, ventral portion of the superior temporal sulcus		Tem, Tea, IPa
TF, lateral and rostral parahippocampal cortex		Rosene and Pandya (45)
TH, medial and caudal parahippocampal cortex		TF-TL
a28, Ent, entorhinal cortex		TH
a35, area 35 of the perirhinal cortex		TH
a36, area 36 of the perirhinal cortex
Sub, subiculum
Hipp, hippocampus; CA, cornu ammonis
23ab, area 23ab of the posterior cingulate cortex
a30, area 30 of the retrosplenial cortex
a29, area 29 of the retrosplenial cortex
L, lateral nucleus of the amygdala
B, basolateral nucleus of the amygdala
AB, accessory basal nucleus of the amygdala
M, medial nucleus of the amygdala
Ce, central nucleus of the amygdala

*Subdivision and nomenclature of parietal areas based on Pandya and Seltzer (36).

Inputs come from studies using injections of retrograde and anterograde transport of tracers. In many cases, both were used in the same study to characterize the connectivity. We defined the strength of inputs to each area from any other area using the values reported in the examined studies. Values were sometimes reported as a cell count in a specific area or as gradient of connectivity, e.g., 0 (absence), + (weak), ++ (moderate), etc. When the strength was not reported in the study, we inspected the published data, sometimes from several papers. Accordingly, we classified inputs to each area as absent (0), weak (16), moderate (33), medium (50), strong (67), or very strong (100). Very strong (100) self-connection was also given to all areas, because of the presence of a strong local connectivity within the cortex (46, 47). To keep the variability at a minimum, the tabulated data were created by the same author, B. B. Averbeck. Previous work showed that the clustering was similar when we used a binary connectivity matrix instead of the graded connectivity (19). We did not differentiate inputs and outputs based on the layer, as this information was not always available.

We applied a hierarchical clustering algorithm to our connectivity matrix (19). Since tree-fitting algorithms are not guaranteed to identify optimal trees, we first generated 10,000 bootstrap datasets by sampling with replacement from the rows of the connectivity matrix. Then, we applied the agglomerative tree-fitting algorithm from MATLAB to generate a tree both for the original dataset and for each bootstrap dataset, obtaining a set of 10,001 candidate trees. A limitation of this algorithm is that it does not define parametrically the distances between clusters (the length of the branch in the tree) nor how well the trees fit the data. It only defines areas that cluster together.

Thus, in the next step, we applied a maximum likelihood (ML) tree-fitting algorithm to our set of 10,001 trees, generating a fit for each tree based on maximizing the log-likelihood of the data given the tree. This algorithm models the distance between pairs of nodes in the trees based on a branching Gaussian diffusion process and gives an estimate of how well the tree fits the data (19, 20, 24, 48). In other words, this algorithm allowed us to identify the tree that fit the data best, i.e., the most-likely (ML) tree.

In the next step of the analysis, we estimated how robust the clusters were in the trees. To this purpose, we first sorted the 10,001 trees as a function of their likelihood. Next, we applied an algorithm developed for fitting consensus trees (CT) to phylogenetic data to the 100 best ML trees. The algorithm is part of the PHYLIP package and can be found at the PHYLIP Home Page (washington.edu). The CT algorithm returns the number of times that each cluster occurred in the 100 best ML trees, defining how often the most common clusters were detected, i.e., how robust the clusters were. We also found that the consensus trees were essentially identical to the ML tree, providing another check on the robustness of the identified trees. Once we defined clusters for the consensus tree, we calculated the fraction of inputs to single frontal and parietofrontal clusters that came from all other clusters. This allowed us to identify the dominant connectivity between clusters in the cortex.

Finally, to evaluate whether the inputs to a given cluster originated from a wide or a limited set of other clusters, we computed the entropy of the input distributions to each cluster (20, 24). For a given cluster, the entropy, H, is given by

$$H=-\sum _{j\in inputs}{p}_j\cdot ln{(p}_j),$$

where p_j is the fraction (or probability) of inputs coming from each other cluster. Maximum entropy corresponds to a uniform distribution of inputs from all other clusters, whereas minimum (zero) entropy corresponds to a distribution where all the inputs originated from one other cluster.

RESULTS

The goal of our analysis was to identify a statistically reliable organization of the connectivity across frontal, temporal, and parietal networks. To this end we carried out a hierarchical cluster analysis on 32 frontal, 20 temporal, and 17 parietal areas based on their anatomically defined inputs (see Table 1 for the full list of areas). We first generated a dataset of inputs among areas. We then carried out a hierarchical cluster analysis, clustering together areas with similar inputs. There are no direct methods to find the optimal clusters for a given dataset, and cluster analyses often do not provide goodness-of-fit metrics. Therefore, to identify the best hierarchical structure for the data, we generated a set of candidate trees, by fitting 10,000 trees to bootstrap sampled datasets and one tree to the original dataset. We then estimated how well each tree fit the data by using a maximum likelihood clustering algorithm that provided a log-likelihood metric of fit (see methods). This method allowed us to identify the tree from our set of 10,001 candidate trees that fit the data best, i.e., the maximum likelihood (ML) tree. Furthermore, we estimated how robust the clusters were in the identified trees. To this purpose, we fit a consensus tree (CT) to the 100 trees with the highest statistical likelihood. The CT defines how often the most common clusters were detected in the 100 best ML trees. The ML tree and the CT do not have to agree, but we have found that they often do (19, 20, 24). In the following sections, we first describe a hierarchical organization of frontal, parietal, and temporal areas. We then describe the parietofrontal and temporofrontal network organization in the light of the distribution of connectivity among clusters.

Frontal Clusters

Based on inputs coming from all areas, we identified at least 10 clusters in the frontal cortex (Fig. 1). These are generally consistent with our previous analyses (19, 20, 24), although not exactly, as we have expanded frontal areas to include motor areas and continued to add connectivity to the matrix as additional studies have been published. There was a perfect correspondence between the ML tree and the CT; thus, we only show the latter.

Figure 1.

Clusters of frontal areas. The number at each branch node indicates the number of times a cluster was detected in the 100 best maximum likelihood trees.

The first cluster (PMd) was composed of dorsal premotor cortex areas F2 (pre-CD), F2vr, F7 (PMdr), and F6 (pre-SMA). This cluster was detected 67 times in the best 100 trees. More posteriorly in the frontal cortex, a second cluster (M1) included the primary motor cortex (M1, F1) and the supplementary motor area (SMA, F3). This cluster occurred 80 times. The third cluster (PMv) was composed of ventral premotor cortex areas F4 and F5 located on the ventral surface of the precentral gyrus. It was detected 60 times in the best 100 trees. Another superordinate cluster to which PMd, M1, and PMv clusters belong could also be identified. This superordinate cluster occurred 78 times and represents the main premotor/motor domain of the frontal cortex, extending from the central sulcus to the border with the prefrontal areas. Moving to the medial aspect of frontal cortex, we identified a cluster (24ab/9m) that consisted of areas 24a and 24 b of the cingulate gyrus and area 9m. More dorsally, we identified a cluster (CMA), which included areas 24c and 23c of the cingulate motor cortex. Both 24ab/9m and CMA clusters were highly robust in the best trees, in which they occurred 78 and 94 times, respectively. The higher-order cluster to which they both belong was detected 51 times. The sixth cluster (PFCv) consisted mainly of the ventral prefrontal cortex areas, including two subdivisions of area 45, area 46v, and area 12l. This cluster also included the dorsal part of area F7, which embodies the supplementary eye field (SEF). The seventh cluster (PFCd) extended from the principal sulcus to the dorsal surface of prefrontal cortex. It included areas 46d, 9l, 8a and 8 b. Clusters PFCv and PFCd occurred 58 and 31 times, respectively. Moreover, the higher-order cluster to which they both belong was detected 34 times in the best 100 trees. The eighth cluster (OFC) included orbitofrontal cortex areas 10, 11, 12o, and the frontal operculum (PrCo). This cluster occurred 43 times in the best 100 trees. The ninth cluster (PFCvm) consisted of areas 14 m, 14o, 32, and 25 of the ventromedial prefrontal cortex. This cluster occurred 98 times in the best 100 trees. The last frontal cluster (la/13) we identified was composed of orbital areas 13 and the caudal orbital/agranular insula area la. This cluster was detected 100 times in the best 100 trees. Note that we included the agranular insula with frontal areas, as it has been closely associated with orbital areas in anatomical work (49, 50).

Temporal and Parietal Clusters

We next investigated the statistical organization of all temporal and parietal areas based on their inputs from all areas. We found a complex hierarchical structure, which we decomposed into nine different clusters (Fig. 2). Despite its complexity, we were able to discriminate different well-defined sets of clusters in the tree, which belonged either to the parietal cortex or the temporal cortex. Parietal areas formed a set of four clusters. The first cluster (PARv) consisted of areas PG, PFG, and PF on the cortical convexity of the inferior parietal lobule (IPL) and area AIP in the anterior ventral bank of the intraparietal sulcus (IPS). This cluster occurred 71 times in the best 100 trees. A second cluster (PARd) included areas MIP and PEa on the dorsal surface of the IPL, the dorsal area PEc, and the dorsomedial area 7m. This cluster occurred 59 times. The third cluster (PARml) included areas VIP and LIP of the IPS, areas V6 and V6A in the anterior bank of the parietooccipital sulcus, and the dorsal area Opt at the level of the parietooccipital junction. This cluster was detected 17 times in the best 100 trees. The last parietal cluster (SS/Ig) was composed of postcentral areas SI and PE, area SII in the parietal operculum, and the caudal granular insular cortex area Ig that is adjacent to SII. This cluster was detected 68 times.

Figure 2.

Clusters of temporal and parietal areas. The number at each branch node indicates the number of times a cluster was detected in the 100 best maximum likelihood trees.

At the temporal cortex level, we identified five clusters. The first cluster (TE) consisted of the ventral bank of the superior temporal sulcus (STSv) and areas TE and TEO in the inferior convexity of the temporal lobe. This cluster was detected 97 times in the best 100 trees. A second cluster (PH/RSP) included areas TF and TH of the parahippocampal (PH) gyrus, area 23ab of the posterior cingulate cortex, and areas 29 and 30 of the retrosplenial cortex. This cluster was detected 74 times. The third cluster (Amg/ST/Id) included basolateral and lateral amygdala nuclei, the anterior superior temporal gyrus including the temporal pole (excluding areas 35 and 36), the dorsal portion of the superior temporal sulcus (STSdr), and the mid-insular dysgranular cortex area Id. This cluster was detected 39 times. A fourth cluster (Hipp) was composed of the hippocampal complex CA1-CA3 and the subiculum. This cluster occurred 98 times in the best 100 trees. The last cluster (Per/Ent) was located in the medial aspect of the temporal lobe and included the perirhinal cortex, made up of areas 35 and 36, and the entorhinal cortex. This cluster occurred 49 times in the best 100 trees. Note these are unrooted trees (the orientation of the branches of a given node is arbitrary), and therefore, at a superordinate level, a cluster that included all temporal clusters and another including all parietal clusters can be identified, strengthening the presence of a distinct set of inputs to temporal versus parietal cortices. The superordinate cluster of temporal areas occurred 65 times, and the superordinate cluster of parietal areas occurred 62 times.

To further investigate the statistical organization within the temporofrontal network, as these are the data new to this study, we also carried out a hierarchical cluster analysis only on temporal areas. The resulting CT is shown in Fig. 3 and is generally consistent with the temporal subcluster from the analysis that also included parietal areas (Fig. 2). We found a sixth temporal cluster (Ins) that consisted of the insular cortex areas Id and Ig. In the previous tree (Fig. 2), the Ig area was included in the parietal SS/Ig cluster, whereas the Id area clustered together with amygdala and STS areas. This is not surprising considering that insular areas are located on the border between frontal, temporal, and parietal lobes. Thus, insular cortex areas Id, Ig, and Ia may represent an interface between the three lobes.

Figure 3.

Clusters of temporal areas. The number at each branch node indicates the number of times a cluster was detected in the 100 best maximum likelihood trees.

Inputs to Frontal and Parietotemporal Clusters

In the previous section, we showed the presence of a hierarchical relationship among sets of areas within frontal, temporal, and parietal cortices based on their inputs. Distinct sets of clusters were composed of hierarchically related sets of areas that shared similar inputs and, in most instances, were anatomically contiguous. In the following section, we performed a qualitative and quantitative analysis of the connectivity between frontal and parietotemporal clusters. Although most of the clusters were characterized by heterogeneous inputs, we identified the largest inputs to each cluster, which reflected its functional properties as well as the strength of the dominant input.

Generally, we observed a gradient of inputs to frontal clusters (Fig. 4). More posteriorly in the frontal cortex, motor and premotor areas were preferentially targeted by superior parietal areas following a topographic pattern, whereas moving more anteriorly through prefrontal and cingulate areas, inputs originated mostly from temporal and medial parietal areas.

Figure 4.

Inputs to frontal clusters. Distribution of inputs to frontal clusters from temporal and parietal clusters. Light gray and dark gray bars indicate the sum of inputs to frontal clusters from temporal and parietal clusters, respectively. H, entropy of the input distributions to each frontal cluster.

M1 was the frontal cluster with the most limited set of inputs. This cluster received 67% of its inputs from areas belonging to the parietal SS/Ig cluster. A minor fraction of inputs to the M1 cluster came from the PARd parietal cluster (22%) and the PH/RSP temporal cluster (11%). The strongest inputs to the PMv cluster came from the PARv parietal cluster (37%). Furthermore, the PMv cluster represented the frontal cluster with the strongest inputs from parietal clusters (95%) and the weakest inputs from temporal clusters (Amg/ST/Id: 5%). The PMd cluster was mostly targeted by the PARd parietal cluster, whose inputs represented 42% of total inputs, and to a lesser extent by other parietal (PARml:19%, PARv: 12%, SS/Ig: 6%) and temporal (PH/RSP, 11%; Per/Ent, 9%; Hipp, 1%) clusters.

The strongest inputs to 24ab/9m, PFCd, and CMA clusters came from the PH/RSP temporal cluster (34%, 45%, and 52% of total, respectively). These three clusters differed in their inputs from the less dominant parietal and temporal clusters. Whereas minor inputs to the 24ab/9m cluster came mostly from temporal clusters (Amg/ST/Id: 24%, Per/Ent: 17%, Hipp; 11%), PFCd and CMA clusters were mostly targeted by PARml (22%) and SS/Ig (28%) parietal clusters, respectively.

The PFCv cluster received its main input from the temporal cluster TE (35%) and to a lesser extent from the Amg/ST/Id cluster (18%). PFCv was also the frontal cluster with the most heterogeneous sets of inputs (followed closely by the OFC cluster). The PFCv inputs came from all clusters except the PARd parietal cluster. The strongest inputs to OFC, Ia/13, and PFCvm clusters originated from the Amg/ST/Id temporal cluster (30, 48, and 44% of total, respectively). The OFC cluster received a similar fraction of minor inputs from both the temporal (PH/RSP: 18%, Per/Ent: 8%, TE: 11%, 37% of total) and parietal (SS/Ig: 23%, PARv: 10%, 33% of total) clusters. On the other hand, the Ia/13 cluster received a larger fraction of inputs from temporal areas (Per/Ent: 27%, TE: 8%, Hipp: 5%, PH/RSP: 5%, 45% of total) and only a small fraction of inputs from parietal areas (SS/Ig: 7%). The PFCvm cluster received all minor inputs from temporal areas (Hipp: 29%, Per/Ent: 16%, PH/RSP: 10%, TE: 1%).

Connections between frontal and parietal and temporal clusters were largely reciprocal (Fig. 5). The SS/Ig cluster received its main inputs from the M1 cluster (25%). Among all clusters, the SS/Ig cluster also showed the most heterogeneous sets of inputs, which also reflected its wide reciprocal connections with all frontal clusters except the PFCvm cluster. The main inputs to ventral (PARv) and dorsal (PARd) parietal clusters came from the ventral (PMv: 33%) and dorsal (PMd: 47%) premotor clusters, respectively. The PARml cluster received major inputs from the PFCd cluster (38%) and to a lesser extent from the PMd (20%) cluster. This distribution of inputs also reflected the strength of PARml projections to PFCd and PMd clusters (Fig. 4). Similarly, inputs to PH/RSP cluster mostly originated from the adjacent frontal 24ab/9m (29%) and PFCd (26%) clusters, which in turn received their main inputs from the PH/RSP cluster. The principal inputs to the TE cluster were also reciprocally from the PFCv cluster (56%). The principal inputs to the Amg/ST/Id cluster came from PFCvm (35%), Ia/13 (24%), and OFC (16%) clusters, which in turn also received strong inputs from the Amg/ST/Id cluster. The Per/Ent cluster received its principal frontal inputs from the Ia/13 cluster (30%), whereas secondary inputs came from cingulate/prefrontal clusters (PFCvm: 25%, OFC: 19%, PFCv: 15%, 24ab9m: 8%, PFCd: 4%). Finally, inputs to the Hipp temporal cluster came only from the PFCvm (100% of total), making the Hipp cluster the most defined cluster in terms of inputs.

Figure 5.

Inputs to parietofrontal clusters. Distribution of frontal inputs to parietal and temporal cluster. H, entropy of the input distributions to each cluster.

The connectivity between areas rostral and caudal to the central sulcus was highly reciprocal and showed substantial dorsal-ventral segregation (Fig. 6). Based on the fraction of inputs to a given cluster, we were able to identify several information processing circuits spanning the cortex. Dominant circuits were supported by stronger reciprocal connectivity among clusters, whereas secondary circuits resulted from reciprocal connectivity of different strength. This organization could also be clearly seen when shown on the macaque brain (Fig. 7).

Figure 6.

Organization of temporofrontal and parietofrontal networks. Summary of the main reciprocal connections among frontal, parietal, and temporal clusters. Arrows size indicates the fraction of inputs to a given cluster, whereas arrows color indicates the cluster from which the inputs originated.

Figure 7.

Overall view of dominant information processing circuits in the parietofrontal and temporofrontal system of the macaque cortex. Areas forming a given cluster are represented with the same color. Similar color in the frontal, parietal, and temporal clusters reflects dominant reciprocal connections (arrows) among those clusters. Areas in gray were not included in the analysis due to insufficient data, except the medial and central amygdala, which are striatal and not cortical in origin. Sulcal abbreviations: AS, arcuate sulcus; CS, central sulcus; IPS, intraparietal sulcus; SF, Sylvian fissure; STS, superior temporal sulcus. See Table 1 for the full list of areas, abbreviations used, and a comparison between our and other subdivisions and nomenclatures of cortical areas used in the literature.

Finally, we calculated the entropy of the input distributions to each frontal, temporal, and parietal cluster (Figs. 4 and 5). This allowed us to estimate whether the inputs to a given cluster originated form a wide or a limited set of other clusters. In other words, from how many other clusters did inputs originate. The higher (lower) the value of entropy, the broader (more limited) the distribution of inputs that characterized a given cluster. Considering frontal clusters, M1 had the lowest entropy (0.85 nats), whereas clusters with the highest entropy were PFCv and OFC (1.70 and 1.69 nats, respectively). These values of entropy reflected the number of clusters from which M1, PFCv, and OFC received inputs (3, 7, and 7, respectively) and the different distributions of their inputs.

In the parietal cortex, SS/Ig was the cluster with the highest entropy (2.02 nats) and the broadest distribution of inputs (it received inputs from all frontal clusters except the PFCvm), whereas the one with the lowest entropy was the PARv (1.81 nats). In the temporal cortex, the cluster with the highest entropy was the PH/RSP cluster, whereas the one with the lowest entropy was the Hipp (0.00 nats), as it received inputs only from the PFCvm cluster. We also found that inputs to each frontal cluster came, on average, from the 64.4% (5.8 out of 9 clusters) of parietofrontal clusters. Specifically, inputs came from 70% of temporal (from 3.5 out of 5 clusters) and 58% of parietal (from 2.3 out of 4 clusters) clusters. Each parietal and temporal cluster received, on average, inputs from the 77.5% (7.75 out of 10) and 52% (5.2 out of 10) of frontal clusters, respectively. Similarly, the average entropy of frontal, parietal, and temporal clusters was 1.43, 1.76, and 1.22 nats, respectively, suggesting a slight difference in the organization of inputs to the three cortical lobes.

DISCUSSION

We performed a statistical analysis of the connectivity among architectonically defined areas of the frontal, parietal, and temporal cortex in the macaque, based on their anatomical inputs. We showed that areas that have similar inputs can be grouped into specific clusters. Furthermore, each cluster was composed of anatomically adjacent areas, giving rise to a higher-order parcellation of the cerebral cortex that follows the hierarchical relationship among sets of areas. The analysis of the connectivity among clusters revealed the presence of a network of cortical connections that, in most instances, were reciprocal (Fig. 6). Within these reciprocal connections, we identified a series of parietofrontal and temporofrontal circuits. These circuits preferentially linked frontal areas with distinct parietal or temporal areas in a topographically organized manner (Figs. 6 and 7). Different principles can be derived from this organization. Generally, the gradient of inputs to frontal clusters described a dual, dorsal-ventral, subdivision of the frontal cortex where more posterior and dorsolateral (motor) clusters showed a privileged connectivity with the parietal cortex, and medial and ventrolateral clusters showed distinct connections with the temporal cortex. This differential connectivity of the dorsal and ventral prefrontal regions was apparent in early studies based on less sensitive silver staining techniques (4). In this context, posterior cingulate areas seem to be at the interface between the main parietofrontal and temporofrontal connections, linking together parahippocampal areas in the ventromedial aspect of the temporal lobe and the PFCd cluster in the frontal cortex.

Evolutionary and Developmental Factors That Give Rise to the Statistical Organization

The dominant dorsal-ventral organization of cortex seen in our statistical analyses likely reflects its evolutionary and developmental origin. Early comparative architectonic work showed that the reptilian pallium, which is homologous to the mammalian cortex, is composed of a lateral region that processes olfactory information and a medial hippocampal region that processes spatial information (51). Between the lateral and medial regions was a region later referred to as the dorsal pallium, which receives sensory inputs from the thalamus (52). This organization, principally as it applies to the prefrontal cortex, was later independently reproposed, on the basis of cytoarchitectonic work in human and macaque cortex, as the dual-origin hypothesis (26). This hypothesis suggested that the ancestral pallium was composed of the insular/pyriform cortex, which processed olfactory information, and the rostral cingulate cortex adjacent to the corpus collosum (approximately area 24a and part of area 25), which has a hippocampal origin and processed spatial information. In primates, the pyriform cortex forms a part of the primary olfactory cortex. This hypothesis further suggested that subsequent evolutionary expansion of the pallium led to an increase in the prefrontal cortex (expanding dorsally from an origin in 24a and expanding laterally from an origin in the pyriform cortex). The most recently evolved prefrontal regions, represented in primates, correspond to dorsal and ventral area 46 in the macaque. This places the principal sulcus as the meeting point between these two trends. In the human cortex, the inferior frontal sulcus is the meeting point between the two trends. From a cytoarchitectonic point of view, the pyriform cortex is composed of two or three layers (allocortex) and the rostral cingulate (areas 24a/25) is composed of less than six layers (mesocortex) (53), reflecting the ancestrally less developed cytoarchitectonic organization. As one progresses dorsally from area 24a/25 or laterally from the pyriform cortex, the cortex becomes increasingly dysgranular, as it shows more laminar structure. Areas 46 and 8 are the most highly differentiated of the prefrontal 6 layer granular cortex. This increase in granularity also occurs as one progresses anteriorly in the orbitofrontal cortex, toward the frontal pole area 10 (54). Primary sensory areas also have a well-defined six-layer structure. This puts dysgranular cortical areas (with a less developed layer 4) between agranular (without layer 4) and granular (with a well-defined layer 4) cortical areas. Consistent with this trend, granular areas become progressively more differentiated with increasing distance from dysgranular areas (55). Furthermore, within frontal cortical areas, a pattern of laminar connectivity can be defined that is related to the degree of laminar differentiation (i.e., agranular versus granular). Projections from agranular and dysgranular, limbic, areas to granular areas (eulaminate cortex) are feedback, terminating in mostly superficial cortical layers. In contrast to this, connections from eulaminate to limbic areas are feed forward, terminating mostly in deeper layers. Lateral projections linking areas with the same laminar organization originate in superficial (2 and 3) and deep (5 and 6) layers and terminate in all layers (1–6) (9, 56).

Recent work has shown that these cytoarchitectonic gradients are also reflected in gene expression profiles, which follow the same gradients. Specifically, modern imaging and genetic work has shown that the agranular areas, which are also low in myelin, have unique patterns of gene expression relative to the granular areas (57). There are also gradients in molecular plasticity markers that match these cytoarchitectonic gradients (58). More recent developmental gene expression work has supported and extended this hypothesis. This work has shown that the pallium (ancestral cortex) of most vertebrates can be divided into three or perhaps four regions, with approximate correspondence to the regions originally proposed on the basis of cytoarchitectonics (27, 55, 59–61). The medial, dorsal, and lateral pallium, identified by developmental gene expression maps, correspond to the hippocampal, dorsal, and pyriform areas. The lateral pallium has been further divided into a more restricted lateral pallium, which corresponds to the insula and claustrum, and a ventral pallium, which corresponds to the primary olfactory cortex and the basolateral amygdala (61). There is some debate about whether a dorsal pallium exists in some fish species and amphibians, but the medial, ventral, and lateral pallium can be identified (62). Thus, the ventral/lateral olfactory and medial hippocampal pallium, separated by a dorsal sensory-motor pallium that receives sensory input from the thalamus, is fundamental to the organization of the vertebrate pallium, with the medial and ventral/lateral pallium likely present in all vertebrates, and the dorsal pallium present since the emergence of sauropsids (lizards and birds) and mammals, and perhaps earlier.

This basic organizational plan gives shape to the architecture that we see in our statistical analysis of the macaque connectivity data. Brain organization and connectivity do not have to correspond, but our data suggest that they do. Although the CA portion of the hippocampus has retreated to the medial temporal lobe in primates, bringing it topographically close to pyriform cortex, this is not its ancestral location. Furthermore, the cingulate fasciculus connects the hippocampus and adjacent parahippocampal and retrosplenial areas (PH/RSP) with 24ab and PFCvm. The PFCvm and 24ab clusters are approximately at the ancestral location of the hippocampus. These are also the identified agranular frontal node in the dual-origin hypothesis (26, 53, 63). The olfactory system, which is less developed in macaques because of the domination of the visual system, connects the pyriform cortex with portions of area 13, the agranular insula, and the entorhinal cortex. Therefore, the ancestral medial hippocampal system is reflected in the PH/RSP/24ab spatial system, and the ancestral ventral/lateral olfactory system is reflected in the caudal-orbital, agranular insula, entorhinal, and perirhinal areas. The perirhinal and entorhinal areas have become more visual, also reflecting the domination of the visual system in primates (64–66). It is possible that the amygdala, which is part of the developmental ventral pallium, clusters separately because it receives predominantly visual input in the primate, as opposed to olfactory input in nonprimate species including rodents (67, 68). Furthermore, we did not consider the peri-amygdaloid cortex that receives inputs from the pyriform cortex (67).

Dorsal cortex has undergone the largest expansion in mammals, and it has continued to expand in primates (28, 69). This expansion is reflected in the expanding ring or nested structure of sensory and motor cortex. M1 and S1 are a single structure in monotremes and marsupials and did not split into independent cortical areas until placental mammals (29). M1 and SS/Ig clusters may correspond to most of the nonvisual cortex in early mammals, later restricted to a smaller cortical sector following the expansion of parietal and frontal areas. It is also possible that the granular insular cortex of primates corresponds to the parietal ventral area of early mammals (70). At the next nested level beyond the somatomotor cortex are, dorsally, the dorsal-parietal premotor circuits (PARd and PMd) and, ventrally, the inferior parietal ventral premotor circuits (PARv and PMv). At the next level beyond that, the dorsal-parietal, parahippocampal, and retrosplenial areas are connected to the dorsal-lateral prefrontal cortex (PARml and PH/RSP to dlPFC), and the inferior temporal visual areas are connected to the ventral-lateral prefrontal cortex (TE to PFCv). Thus, the ancestral dorsal pallium has massively expanded in primates, giving rise to much of what we refer to as the cortex. This expansion in neural tissue is likely reflected behaviorally in the dexterity in the reach and grasp system in primates, as well as the orienting system that controls eye movements and attention to regions of space, discussed further in the next section. Interestingly, the PH/RSP, SS/Ig, and Amg/ST/Id clusters have the broadest projections to prefrontal cortex (Fig. 6), and these clusters may also be related to the medial hippocampal, dorsal, and ventral/lateral pallial structures, respectively.

There is also a close correspondence between the clusters we have identified, the corresponding parietofrontal and temporofrontal connectivity, and the main fiber bundles in the cortex (25). Each of the clusters is approximately the termination of a major fiber bundle, and the dominant connections are defined by the two ends of each fiber bundle. The cingulate fasciculus connects the posterior cingulate and retrosplenial areas to the anterior cingulate and dorsal-lateral prefrontal areas (PH/RSP to 24ab/9m, PFCd, CMA). The superior longitudinal fasciculus (dorsal portion) connects the dorsal parietal areas to the dorsal premotor areas (PARd to PMd). The superior longitudinal fasciculus (medial portion) and the occipitofrontal fasciculus connect the medial-lateral parietal areas to the dorsal-lateral prefrontal cortex (PARml to PFCd). The superior longitudinal fasciculus (ventral portion) connects the ventral inferior parietal areas to the ventral premotor areas (PARv to PMv). The uncinate fasciculus connects the temporal pole and medial temporal areas to the ventral-medial prefrontal areas (Amg/ST/Id to PFCvm) and also carries the fibers that connect the inferotemporal visual cortex to the ventral lateral prefrontal cortex (TE to PFCv). One hypothesis is that as the dorsal pallium expanded, post-rolandic sensory and pre-rolandic executive/motor areas were pushed farther apart by the expanding somatomotor cortex. The white matter bundles connect areas that used to be closer together, or intermingled within a much smaller dorsal pallium, as they were pushed apart by the expansion and specialization of cortical areas. This hypothesis is supported by the development and evolution of white matter tracts. In the human brain, for example, early limbic areas are connected by curved tracts (e.g., the uncinate fasciculus) due to the expansion of eulaminate areas, which develop late and are connected by straight tracts (e.g., the superior longitudinal fasciculus) (71).

Functional Organization of Identified Clusters

Early studies by Ungerleider and Mishkin (72) described a dorsal-ventral organization of the primate cortex in terms of two distinct visual pathways that originate from early visual areas. Specifically, there is a ventral pathway, from V1 to V4, continuing to the inferotemporal lobe, which represents mostly object identity, and a dorsal pathway, from V1 to area MT/V5, continuing to the parietal cortex, which represents visual-spatial information (72). This model was later revised by Goodale and Milner (73) who proposed that the ventral and dorsal pathways were responsible for visual perception and visual control of action, respectively. According to their view, the ventral stream enables visual representation of the environment, attributing meaning to objects and events. The dorsal stream allows the control of actions needed to interact with those objects, underlying transformations used to perform actions.

Subsequent studies have shown interconnection between the ventral and dorsal streams (74–78). Based on this connectivity, models have been proposed to explain and expand the functional relationship between perception and action (78, 79). It has also been proposed that behavior is the result of continuous parallel processing of potential actions based on a sensory representation of the world, where only the most appropriate action, at a given moment, is selected (80, 81). This selection would take place within a distributed interacting network that embraces the dorsal parietofrontal system, the ventral temporofrontal system, and the basal ganglia (82–84).

We have recently put forward a related hypothesis that outlines broad functional roles for systems that we refer to as the ventral and dorsal circuits, based upon their connectivity through the striatum (85). The ventral system is the circuitry that connects the medial temporal lobe structures including the amygdala, perirhinal cortex, and entorhinal cortex to PFCvm, caudal OFC, and the ventral striatum. This circuit is strongly interconnected with the hypothalamus. Therefore, it links information about objects in the environment, provided by visual information flowing through the medial temporal lobe structures (i.e., amygdala and perirhinal cortex), with motivational circuitry in the hypothalamus that represents the internal physiological state. This system identifies current behavioral goals, for example, identifying sources of food when one is hungry. The dorsal circuit is composed of the dorsal-parietal and lateral prefrontal circuit, which connects to the dorsal striatum. The dorsal circuitry codes spatial information useful to direct behavior on the fly to obtain goal objects in the environment coded by the ventral system.

Neurophysiology studies have shown that frontal, parietal, and temporal clusters connected within a given circuit are composed of neurons whose activity encodes similar aspects of behavior, consistent with the fact that they receive similar inputs from other cortical areas. Therefore, it is interesting to explore and try to summarize the functional properties within each circuit.

In the parietofrontal system, we identified four dominant circuits, related to the ancestral and developmental dorsal pallium. The SS/Ig-MI circuit encodes kinetic and kinematic limb variables required for generating and controlling the skeletomotor system. These variables include limb position (86, 87), movement direction, amplitude, speed, and force (88–95). Within this circuit, areas in the parietal cluster SS/Ig also encode somatosensory, visual, and proprioceptive information (96–100) that contributes to the control of movements by providing somatosensory and visual feedback information to the frontal motor cluster.

The dorsal parietofrontal circuit defined by the connectivity between PARd and PMd clusters is at the core of visually guided control of limb movements, primarily concerning arm reaching movements. Areas in the PARd cluster encode spatial visual and somatic information necessary for reaching movements (86, 90, 101–104). This information is integrated in the PMd cluster, which encodes the selection, preparation, and execution of limb movements (105–109); their online correction (110); and inhibition (111–113). The PMd cluster also encodes decision-making variables in the reach system (114–119).

Connections between the PARv and PMv clusters define the parietofrontal circuit involved in somatosensory and visually guided control of purposeful hand actions, such as object manipulation and grasping (120–126) and, to a lesser extent, in the control of arm reaching and mouth movements (127–129). This circuit also embodies frontal (area F5) and parietal (area PFG) nodes of the mirror system (130, 131).

The parietofrontal circuit linking PARml and PFCd clusters has a key role in the visuomotor transformation that subserves several aspects of eye movements, including oculomotor intention and control, selective visual attention, decision-making, and spatial working memory (132–140).

The functional and anatomical organization of parietofrontal cortex of primates has been also described in terms of a map of complex, behaviorally relevant, actions. Early studies showed that long-duration stimulation of specific domains in the precentral gyrus and parietal areas evoked distinct sets of actions that reflected the natural repertoire of movements (141, 142). These actions included, but were not limited to, exploratory gaze, reaching, grasping, hand-to-mouth interaction, and defensive or aggressive behaviors. This functional map of the cortex has been further described in terms of connections between parietal and frontal domains that evoke similar actions, identified in multiple species of primates (29, 143–148). The resulting connectivity within the “action map” broadly overlaps with our parietofrontal network both anatomically and functionally.

In the temporofrontal system, connections between the TE and PFCv clusters underlie the visual information processing circuit that encodes the visual and abstract representation of objects. Areas in the TE cluster are involved in face recognition (149–151) and in the analysis of object properties such as shape, color, and texture (152–155). The PFCv cluster is involved in working memory and learning processes based on object features in space (156–160) and appears to play a role in the top-down control of inferotemporal cortex functions (161, 162) and object based attention (163). The PFCv cluster also contains the spatially incongruous but functionally similar SEF, a key component of the neural system for controlling and monitoring eye movements (164–166).

Reciprocal connections between the Amg/ST/Id cluster and PFCvm, OFC, and Ia/13 clusters, the developmental and ancestral ventral pallium, give rise to a network involved in several behavioral functions. Converging evidence from lesion and neurophysiological studies in monkeys has shown the role of these connections in mediating emotional responses (167–171), in encoding and learning the subjective value of immediate and future reward-guided choices (172–179), and in social aspects of behavior and decision-making (180–183). Also, a dominant circuit emerged from connections among PFCvm and Hipp clusters, related to the developmental and ancestral medial pallium. This circuit is at the core of the neural system underlying associative memory, learning, and spatial navigation in monkeys (184–190).

Dominant circuits also linked PH/RSP, CMA, 24ab/9m, and PFCd clusters. These circuits are also related to the medial pallium. Within these circuits, clusters located in and near the cingulate gyrus were composed of areas involved in the control of eye and arm movements (191–195) in value-based decision-making (191, 196) and in social cognition (197, 198). Areas in the PH/RSP cluster might be involved in both spatial and episodic-like memory processes (199, 200). Thus, these circuits seem to integrate both cognitive and executive functions.

This overall organization of the connectivity among clusters closely reflects the functional and anatomical architecture of visual information pathways across the primate cortex. Specifically, it has been proposed that the dorsal visual system of primates, historically mostly confined to a parietofrontal network, is actually composed of multiple pathways linking parietal areas with both frontal and temporal regions (75). In this framework, parietal clusters convey visual information to frontal clusters through three distinct pathways that support eye movements and working memory (to PFCd), arm reaching movements (to PMd), and hand actions (to PMv). A fourth pathway, from areas in the posterior parietal lobule to posterior cingulate and medial temporal areas (corresponding to our PH/RSP and Hipp clusters), has been described as carrying visual information supporting spatial navigation. This suggests that the medial temporal clusters PH/RSP and Hipp would represent an interface between the dorsal and ventral visual processing pathways, where both the spatial (from the dorsal system) information and the object (from the ventral system) information are integrated and subsequently conveyed to prefrontal regions.

The cortical organization we have described in monkeys also relates to networks defined in the human brain by noninvasive, MRI-based, techniques. Detailed homology between human and macaque cortical areas is an ongoing source of debate (201), specifically within parietal (202), frontal (203), and temporal (204) associative regions that show the greatest expansions in humans (205–208). However, a growing body of evidence suggests similarities between patterns of corticocortical connectivity in humans and macaques (209–211). Resting-state functional connectivity studies in humans (210, 212) show strong connectivity between PMv and more anterior and ventral areas located in the inferior parietal lobule (IPL), including human parietal regions hlp2 and 7m, corresponding to areas PFG and AIP in the macaque. In contrast to the macaque, the central portion of the human IPL seems to be connected to lateral frontal pole area FPl (212, 213). In the superior parietal lobule (SPL), a cluster centered on areas 5L and hlp3, and active during visually guided reaching movements (214), shows strong connectivity with the PMd. This cluster has been proposed to be homologous to macaque area MIP, probably also incorporating part of the macaque area PE (211, 212). Two further clusters have been identified in the human SPL. One is composed of areas IPS1, IPS2, and IPS3 that, based on modulation during visual attention and eye movements (215, 216), may correspond to macaque area LIP. The second cluster, centered on area 7PC, has been suggested as equivalent to area VIP in the macaque. Consistent with what is observed in monkeys, these clusters show functional connectivity with the dlPFC. In humans, FEF shows the strongest connectivity with areas IPS1, IPS2, and IPS3. Strong functional connectivity has been found between area 4 (M1), the caudal portion of area 6 (that may correspond to monkey area F3), and the somatosensory cortex (areas 3, 1 and 2) and area 5L (monkey area PE). This functional connectivity reflects closely the connectivity between our SS-M1 clusters and, in humans, has been defined as the somatomotor network (210). In the human inferotemporal cortex, rostroventral areas show functional connectivity with the orbitofrontal cortex, whereas more posterior regions are connected to the ventrolateral prefrontal cortex (210). These networks have been described as part of the limbic and the ventral attention networks, respectively (210, 217). The main differences emerge when considering the expansion of the temporal cortex (208) and reorganization of the main fiber bundles connecting the temporal, parietal, and frontal cortex, probably related to specific human functions (e.g., language, semantic memory). An example is the reorganization of the arcuate fasciculus. In monkeys, the arcuate fasciculus connects the superior temporal gyrus to the dlPFC and the PMd. In humans, on the other hand, it connects Brodmann areas 21, 22, 41, and 42 in the posterior temporal cortex to areas 44 and 45 in the vlPFC (218).

Conclusions

Our analysis showed that frontal, parietal, and temporal areas of the macaque cortex can be grouped into clusters of cytoarchitectonic areas that share similar inputs and basic functional properties. Consistent with our previous studies on the frontal and parietal cortex (19, 20, 24), areas within each cluster were anatomically adjacent. Clustering areas allowed us to identify the hierarchical organization of circuits linking dorsal and ventral parietal and temporal areas with their dorsal and ventral prefrontal counterparts. Within these networks, we identified distinct information processing circuits that reflect the main fiber bundles linking areas within the cortex, as well as their developmental and evolutionary antecedents. Functionally, ventral circuits defined by the temporofrontal network encode information necessary to discriminate objects in the environment and their values. Dorsal circuits, which form the parietofrontal network, are involved in the control of spatial goal-oriented behaviors including selecting, generating, and monitoring appropriate actions based on visual, auditory, and somatosensory information (85). These analyses extend the dorsal-ventral circuit organization of the visual cortex to a broader set of areas encompassing most of the cortex beyond early sensory areas (72) and begin to embed these ideas within a broader evolutionary and developmental framework.

SUPPLEMENTAL DATA

Supplemental Table S1: https://doi.org/10.6084/m9.figshare.15031695.v1.

GRANTS

This work was supported by the intramural research program of the National Institute of Mental Health (ZIA MH002928).

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

B.B.A. conceived and designed research; B.B.A. performed experiments; F.G. and B.B.A. analyzed data; F.G. and B.B.A. interpreted results of experiments; F.G. and B.B.A. prepared figures; F.G. drafted manuscript; F.G. and B.B.A. edited and revised manuscript; F.G. and B.B.A. approved final version of manuscript.

APPENDIX

See Table A1 for a list of connectional studies used for compiling the connectivity matrix.

Table A1.

List of connectional studies used for compiling the connectivity matrix

Authors	Year	Reference
Frontal cortex
Barbas H	1988	(219)
Barbas H	1993	(220)
Barbas H and Mesulam MM	1985	(221)
Barbas H and Pandya DN	1987	(32)
Barbas H et al.	1999	(222)
Borra E et al.	2008	(223)
Borra E et al.	2019	(224)
Carmichael ST and Price JL	1995	(49)
Carmichael ST et al.	1994	(225)
Cavada et al	2000	(226)
Eradath MK et al.	2015	(227)
Gerbella M et al.	2013	(228)
Gerbella M et al.	2016	(229)
Hatanaka N et al.	2001	(230)
Huerta MF and Pons TP	1990	(231)
Joyce and Barbas	2018	(54)
Kurata K	1991	(232)
Leichnetz GR	1986	(233)
Lu M-T et al.	1994	(234)
Luppino G et al.	1993	(235)
Luppino G et al.	1999	(236)
Luppino G et al.	2001	(237)
Marconi B et al.	2001	(238)
Markov et al.	2014	(239)
Matelli M et al.	1986	(240)
Matelli M et al.	1991	(34)
Muakkassa KF and Strick PL	1979	(241)
Petrides M and Pandya DN	2006	(242)
Petrides M and Pandya DN	2007	(243)
Preuss TM and Goldman-Rakic PS	1989	(244)
Rosa P et al.	2019	(245)
Saleem KS et al.	2008	(246)
Saleem KS et al.	2014	(247)
Stanton GB et al.	1995	(248)
Tanne-Gariépy J et al.	2002	(249)
Wang Y et al.	2005	(250)
Parietal cortex
Bakola S et al.	2010	(251)
Bakola S et al.	2013	(252)
Blatt GJ et al.	1990	(253)
Borra E et al.	2008	(223)
Boussaoud D et al.	1990	(254)
Petrides M and Pandya DN	1984	(255)
Caminiti R et al.	1999	(256)
Cavada C and Goldman-Rakic PS	1989	(257)
Cerkevich CM et al.	2014	(258)
Cipolloni PB and Pandya DN	1999	(259)
Gamberini M et al.	2009	(260)
Gharbawie OA et al.	2011	(147)
Hihara S et al.	2006	(261)
Hilgetag et al	2016	(10)
Leichnetz GR	2001	(262)
Lewis JW and Van Essen DC	2000	(263)
Maioli MG et al.	1998	(264)
Marconi B et al.	2001	(238)
Markov et al.	2014	(239)
Medalla and Barbas	2006	(56)
Morecraft RJ et al.	2004	(265)
Passarelli L et al.	2011	(266)
Pons TP and Kaas JH.	1986	(267)
Rozzi S et al.	2008	(268)
Shipp S et al.	1998	(269)
Temporal cortex
Aggleton JP et al.	2015	(270)
Baizer JS et al.	1991	(271)
Barbas H and Glatt GJ	1995	(272)
Barbas H and De Olmos J	1990	(273)
Carmichael ST and Price JL,	1995	(50)
Ghashghaei HT and Barbas H	2002	(274)
Ghashghaei HT et al.	2007	(275)
Insausti R and Munoz M	2001	(276)
Insausti R et al.	1987	(277)
Hackett M et al.	1998	(278)
Hilgetag et al	2016	(10)
Kondo H et al.	2003	(279)
Kondo H et al.	2005	(280)
Lavenex P et al.	2002	(281)
Lavenex P et al.	2004	(282)
Markov et al.	2014	(239)
Muñoz-López M and Insausti R	2005	(283)
Petrides M and Pandya DN	2007	(243)
Pritchard TC et al.	2000	(284)
Romanski LM et al. JCN	1999	(285)
Romanski LM et al.	1999	(286)
Sakata et al	2019	(287)
Seltzer B and Pandya DN	1989	(288)
Stefanacci L and Amaral DG	2000	(289)
Stefanacci L and Amaral DG	2002	(290)
Suzuki WA and Amaral DG	1994	(291)
Suzuki WA and Amaral DG	1994	(292)
Webster MJ et al.	1994	(293)
Insular cortex
Mufson EJ and Mesulam MM	1982	(294)
Mesulam MM and Mufson EJ	1982	(44)
Cingulate cortex
Arikuni T et al.	1994	(295)
Bates JF and Goldman-Rakic PS	1993	(296)
Hatanaka N et al.	2003	(297)
Kobayashi Y and Amaral DG	2003	(298)
Kobayashi Y and Amaral DG	2007	(299)
Markov et al.	2014	(239)
Morecraft RJ and Van Hoesen GW	1992	(300)
Morecraft RJ and Van Hoesen WG	1992	(301)
Morecraft RJ and Van Hoesen WG	1993	(301)
Morecraft RJ and Van Hoesen WG	1998	(302)
Morecraft RJ et al.	2004	(265)
Morecraft RJ et al.	2012	(303)
Morris R et al.	1999	(304)
Sakata et al	2019	(287)
Vogt BA and Pandya DN	1987	(305)