The macaque ventral intraparietal functional connectivity patterns reveal an anterio-posterior specialization mirroring that described in human ventral intraparietal area

The macaque ventral intraparietal area (VIP) is one of several anatomically and functionally distinguishable regions in and around the macaque intraparietal sulcus. It is located in the fundus of the intraparietal sulcus (IPS). VIP has been implicated in a diverse range of sensory, sensorimotor, and cognitive functions (Avillac et al., 2004, 2007; Bremmer, Duhamel, et al., 2002; Bremmer, Klam, et al., 2002; Duhamel et al., 1997). The region’s cytoarchitecture (Lewis & Van Essen, 2000b) and structural connectivity (Lewis & Van Essen, 2000a) have also been well defined.

Much effort has gone into defining analogue regions of the monkey’s intraparietal regions in the human brain. However, cytoarchitecture data are incomplete (Amunts et al., 2020; Niu et al., 2020) and matching functional activations to anatomical regions is not straight-forward. One strategy to delineate human functional areas is to assume homology between human and macaque brain areas when interpreting human fMRI data (Mantini et al., 2012; Vanduffel et al., 2014). This strategy has led to suggestions of human homologues of the macaque anterior intraparietal area (AIP), (Binkofski et al., 1999; Cavina-Pratesi et al., 2007; Culham et al., 2003; Frey et al., 2005; Mantini et al., 2012), area V6A (Pitzalis et al., 2013, 2015), and the lateral intraparietal area (LIP) (Denys et al., 2004; Konen & Kastner, 2008). In the case of LIP, however, it appears that macaque LIP has diversified into several neighboring but distinct regions in humans (Kastner et al., 2017). Regarding ventral intraparietal area (VIP), several incompatible suggestions have been made for a human homologue region, with currently no consensus.

We have recently reviewed the literature about macaque and human VIP and found that fMRI-based human VIP studies often use one of three distinct approaches to localize human VIP for further characterization (Foster et al., 2022). These three approaches are (#1) egomotion-consistent visual motion as opposed to local motion (Wall & Smith, 2008); (#2) conjunction of neural responses to visual, tactile, and auditory motion stimuli (Bremmer et al., 2001); and (#3) topographic mapping of tactile and visual stimuli to the face (Sereno & Huang, 2006). Each approach focuses on a subset of macaque VIP functional responses and has consistently resulted in identification of three different locations for a putative human VIP. Most locations that other human fMRI studies have been interpreted as the “human VIP” cluster around these three locations. Our review revealed that these three clusters exhibit distinct functional properties and we therefore proposed that the human homologue of macaque VIP is a complex of three distinct regions hVIP#1, hVIP#2, and hVIP#3 (with h denoting “human”); thus, just like LIP, VIP appears to have diversified and expanded in the human brain, as compared to the macaque monkey’s singular region.

The most posterior cluster hVIP#1 has a border with LIP and was reported to be active in studies that used large field optic flow and required processing of self-motion. The most lateral cluster hVIP#2 borders with AIP and was activated in studies that applied neck-proprioceptive stimulation and employed tasks that involved dissociating self- and object-motion. The most anterior cluster hVIP#3 contains topographically organized, tactile, and visual face maps, and activations were observed in the context of active head movements.

In principle, as suggested thus far, the three hVIP areas could form one large area with different functional specializations, rather than three distinct areas. Whether a brain region should be classified as distinct depends on multiple criteria, however, such as cytoarchitectonics, connectivity profile, topographic organization (when present), and functional properties (Orban, 2016; Orban et al., 2004). Each hVIP cluster falls into distinct cytoarchitectonic region(s) of published cytoarchitectonic parcellations (Amunts et al., 2020; Richter et al., 2019; Scheperjans et al., 2008). The three hVIPs were distinct also in a parcellation based on anatomical connectivity assessed with diffusion-weighted tensor imaging (DTI) (Mars et al., 2011). In addition, phase-encoded mapping in human PPC revealed retinotopic, saccade-related maps in the location of hVIP#1 (Konen & Kastner, 2008), retinotopic, smooth pursuit-related maps in the location of hVIP#2 (Konen et al., 2005), and somatotopic maps in the location of hVIP#3 (Huang et al., 2017). Thus, each of the three hVIPs appears topographically organized, though for different stimulus characteristics. Viewed together, anatomical and functional characteristics of the three VIPs are most consistent with the view that they are three distinct areas.

This proposal raises the question whether macaque area VIP is truly a single area, or whether it, too, can be subdivided into multiple regions. In fact, cytoarchitectonic mapping divides macaque VIP into a medial and a lateral part (Lewis & Van Essen, 2000b) but no differences in connectivity or physiological functions have so far been associated with this division (Lewis & Van Essen, 2000a). In contrast, functional fMRI studies in macaques have revealed functional differences along its anterior-posterior axis, with anterior VIP responding preferentially to visuo-tactile stimuli around and on the face, and posterior VIP responding preferentially to large-field motion (Guipponi et al., 2013, 2015; Wardak et al., 2016). Here, we used functional connectivity as another means to address potential regional specialization by characterizing resting-state fMRI connectivity of macaque VIP (Gamberini et al., 2020; Mars et al., 2018; Vijayakumar et al., 2018). Specifically, we divided VIP into three regions of interest (ROI) along its anterior-posterior axis and characterized the ipsilateral whole-brain connectivity of these three seed ROIs. The functional networks into which each of the anterior and the posterior ROIs was embedded were clearly distinct, in agreement with their respective functions. Thus, functional resting-state connectivity profiles support the proposal that macaque VIP exhibits an anterior-posterior functional gradient. Moreover, comparison of the functional division within macaque VIP with the functions ascribed to the three human hVIPs suggests that the distribution of function in the macaque may correspond to the areal diversification observed in the human brain.

We reused data from previous studies of Wim Vanduffel’s lab (Mantini et al., 2011, 2013; Pijnenburg et al., 2019). These data were produced in different experiments; however, the recording parameters were similar regarding the purposes and analyses of the present paper.

Ten rhesus monkeys (Macaca mulatta) participated in the study (6 females, 4 males). They sat in a sphinx position inside a plastic primate chair (Vanduffel et al., 2001), with their heads constrained by a plastic headpost. The monkeys were trained to maintain fixation at a small red fixation point on a uniform gray background within a 2° × 3° virtual window in the center of the screen, while eye movements were monitored by a pupil-corneal reflection eye tracking system (ISCAN, RK-726PCI) at 60 Hz. The monkeys were rewarded for correct eye fixation behavior, reward frequency increasing to a threshold for longer fixations. Before each scanning session, 8–11 mg/kg monocrystalline iron oxide nanoparticle (Molday ION, BioPAL) was injected via the femoral/saphenous vein to improve the contrast-to-noise ratio (CNR) and to avoid the contribution of superficial draining veins (Leite et al., 2002; Vanduffel et al., 2001). Animal care and experimental procedures were performed in accordance with the National Institute of Health’s Guide for the Care and Use of Laboratory Animal, the European legislation (Directive 2010/63/EU) and were approved by the Animal Ethics Committee of the KU Leuven.

In all studies that produced the data re-used here, in-vivo MRI scans were performed with a 3T MR Siemens Trio scanner in Leuven, Belgium. For functional measurements, gradient echo planar images (GE EPI) covering the whole brain were acquired with an eight-channel phased-array receive coil and a saddle-shaped, radial transmit-only surface coil (see Kolster et al., 2014; 40 slices; 84-by-84 in-plane; flip angle = 75°; repetition time (TR) = 2.0 s or 1.4 s depending on the monkey; echo time (TE) = 19 ms; voxel size = 1.25 mm x 1.25 mm x 1.25 mm voxels). Specific parameters varied across monkeys (Table 1). T1-weighted anatomical images were obtained during different sessions using a magnetization-prepared rapid gradient echo (MP-RAGE) sequence (TR = 2200 ms, TE = 4.06 ms, voxel size = 0.5 mm x 0.5 mm x 0.5 mm). During the anatomical scans, the animals were sedated using ketamine/xylazine (ketamine 10 mg/kg I.M., xylazine 0.5 mg/kg I.M., maintenance dose of 0.01–0.05 mg ketamine per minute I.V.). The specific acquisition parameters varied across studies and animals (longer runs and higher number of runs were acquired for monkeys achieving better fixation). This resulted in a more robust evaluation of functional connectivity in these individuals. The shortest cumulated resting-state scanning duration (81 min), the shortest duration of individual runs (10 min), and the lowest number of runs (8 runs) combined is in the standard recommendations in the field (Birn et al., 2013). The variability induced by the differences in scanning parameters is thus expected to be minimal, and if anything, it is expected to enhance the robustness of our results, as our approach highlights common connectivity patterns across all monkeys, irrespective of low-level methodological differences.

Functional time-series were realigned, coregistered to the anatomical images, and normalized to F99 macaque brain template with 1 mm isotropic resolution (Van Essen, 2004). The time-series were band-pass filtered (0.01–0.1 Hz or 0.0025 and 0.05 Hz), nuisance corrected for ventricle and white matter signal, and motion-scrubbed for movement artefacts within and across sessions and slice time, both by means of regression analysis (Vincent et al., 2007). A spatial smoothing was then applied with a 1.5 mm Full Width at Half Maximum (FWHM) Gaussian Kernel.

We selected voxels along the IPS by hand from consecutive coronal slices in the volume space of the F99 T1 anatomy for each of the anterior, middle, and posterior VIP (aVIP, mVIP, and pVIP, see illustration in Fig. 1). For aVIP and mVIP, we selected five voxels each from four consecutive coronal slices. For pVIP, we selected four voxels each from five consecutive coronal slices. These arbitrary subdivisions resulted in all three VIP regions of interest (also later referred to as sectors) to contain 20 voxels. These voxels were selected using a definition of VIP as provided by the Markov atlas (Markov et al., 2014). The difference in the number of slices assigned to the different ROIs is due to the variation in the dorso-ventral extent of area VIP along the antero-posterior orientation (Fig. 1).

Analyses were analyzed using AFNI (Cox, 1996) and FSL (FSL, RID: birnlex_2067; Jenkinson et al., 2012, http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/) embedded in MATLAB. The hemodynamic signal of the 20 voxels of each of the three VIP ROIs was used as seed signal for a seed-to-whole-brain connectivity analysis. Connectivity was calculated as separate correlations of the seed signal and each other voxel of the animal’s brain for each run of each monkey. A Fisher’s r-to-z transformation was applied to the obtained correlation matrix. We then performed, using AFNI, a one-sample t-test (against zero) on the resulting maps of all the runs for each ROI, separately for each monkey. This procedure, thus, resulted in one correlation map (z-score) for each VIP ROI and each monkey. We thresholded these maps at p < 0.001 (Froesel et al., 2024).

We first projected each individual functional connectivity map onto the brain surface using the Connectome Workbench (Marcus et al., 2013), thus projecting the data from the volume space onto the surface space. Due to the specific mathematical properties of this process, each voxel in the volume space was projected onto one or multiple surface voxels (or faces) on the 2D flatmap (Figs. 2–4, average maps; Supplementary Fig. S1, individual monkey maps). We then quantified, across all individuals, the percentage of surface voxels with signal (Fig. 5) and the average z-score of these surface voxels (Supplementary Fig. S4), in each of the different functional brain areas as defined by the Markov atlas (Markov et al., 2014).

For visualization, we selected surface voxels with a z-score above 0.05 (see Supplementary Fig. 3 for results with different thresholds). For Figures 3 and 4, and Supplementary Figure S1, we display the three VIP ROI’s maps on a single surface; non-overlapping surface voxels from the aVIP, mVIP, and pVIP were colored in blue, green, and red, respectively. Overlap between them is displayed by adding the RGB color values of those colors that overlap, resulting in cyan coloring for anterior-middle VIP overlap, yellow for middle-posterior overlap, and magenta for anterior-posterior overlap. Surface voxels containing signals from all three maps are depicted in white.

Last, we assessed the similarity of the three ROIs’ whole-brain functional connectivity maps by a Spearman’s correlation analysis. Figure 2 illustrates the similarity obtained for the whole-brain functional connectivity maps averaged across all monkeys. Supplementary Figure S2 represents the distributions of the Spearman’s correlation coefficients between the ipsilateral aVIP, mVIP, and pVIP whole-brain connectivity maps across individual monkeys.

In Foster et al. (2022), we proposed that human VIP has diversified into three functionally distinct regions. We wondered, therefore, whether the functional specialization evident in the connectivity of macaque VIP may be regarded as a precursor of the functional specialization of the human VIP complex. To approach this question, we characterized the functional resting-state connectivity pattern of the anterior, middle, and posterior third of macaque VIP (aVIP, mVIP, and pVIP, respectively) in both hemispheres of 10 macaque monkeys (see Fig. 1). Each of these VIP ROIs served as a seed for a whole-brain functional connectivity analysis.

We describe in detail the functional connectivity of aVIP, mVIP, and pVIP with the rest of the brain at different spatial scales, from the whole brain to the level of the cortical area, with the aim of inferring possible functional specificities from these patterns of functional connectivity. Thus, in a first step, we characterized the functional connectivity of aVIP, mVIP, and pVIP ROIs with the rest of the cortex and identified the cortical regions that are functionally connected to either the entire VIP, or to only some parts of VIP (Fig. 3). In a second step, we zoomed in to the patterns of VIP functional connectivity with the rest of the brain, and we describe how aVIP, mVIP, and pVIP connectivity organizes in the main cortical regions, namely the occipital, the parietal, the temporal, the premotor and motor, the limbic, the cingulate, the prefrontal, and the orbitofrontal cortex (Fig. 4). We additionally quantified, for all cortical areas defined by the Markov atlas (Markov et al., 2014), their degree of functional connectivity with aVIP, mVIP, and pVIP (Fig. 5). For our report, we focus first on those cortical areas that are dominated by functional connectivity with the anterior part of VIP, that is, aVIP or both aVIP and mVIP. We then shift our attention to the cortical areas dominated by functional connectivity with the posterior part of VIP, that is, pVIP, or both of pVIP and mVIP. We draw extrapolations on the putative functions of aVIP, mVIP, and pVIP and parallels to the human VIP areas in the discussion.

We first used the aVIP, mVIP, and pVIP ROIs to characterize the functional connectivity of the entire brain relative to the antero-posterior axis of VIP. We present averaged mapping results, with individual maps available in Supplementary Figure S1. Figure 2 displays the averaged connectivity of the three VIP ROIs in the two hemispheres. It is apparent that the three ROIs share similar connectivity patterns, with the strongest connectivity within and around the IPS as well as with the prefrontal cortex anterior to the arcuate sulcus (AS). The connectivity maps of aVIP and mVIP were highly correlated (Spearman’s correlations, middle-anterior: left hemisphere r = 0.72, p < 0.05, right hemisphere r = 0.75, p < 0.05), as were mVIP and pVIP maps (Spearman’s correlations, middle-posterior: left hemisphere r = 0.85, p < 0.05, right hemisphere r = 0.75, p < 0.05). In contrast, the correlation was lower between aVIP and pVIP (Spearman’s correlations, left hemisphere: r = 0.54, p < 0.05, right hemisphere: 0.49, p < 0.05). The pattern of similarity between neighboring ROIs with a greater difference between the anterior and posterior sectors is indicative of a gradual modulation of connectivity. In order to test whether the spatial organization of aVIP, mVIP, and pVIP functional connectivity patterns speaks in favor or against the parcellation of VIP in submodules, in spite of significant correlation in global functional connectivity, we performed a quantitative analysis of the similarities and differences between the spatial functional connectivity patterns of the three VIP subdivisions from coarse to finer spatial scales.

Figure 3 displays the three maps overlaid to allow gauging this gradient more intuitively. Connectivity with anterior regions is dominated by aVIP (Fig. 3, blue), followed by regions connected to aVIP and mVIP (Fig. 3, cyan), and, finally, regions connected to all three ROIs (Fig. 3, white). In analogy, connectivity of posterior regions is dominated by pVIP (Fig. 3, red), followed by regions connected to posterior and medial VIP and, then, regions connected to all the ROIs. This functional anterior-posterior gradient of VIP’s connectivity is in line with prior observations from awake macaque fMRI studies that observed a functional gradient between posterior and anterior VIP (Guipponi et al., 2013, 2015; Wardak et al., 2016).

In order to better characterize the functional significance of this antero-posterior aVIP, mVIP, and pVIP to whole-brain functional connectivity gradient, we visualized the cortical regions that are connected to all three ROIs, thus highlighting a core network recruiting the entire VIP; those regions that are connected to two of these ROIs; as well as those regions uniquely connected to one of the aVIP, mVIP, and pVIP ROIs, thus highlighting their functional specificity. To do so, we set a z-score threshold of 0.05 and we colored the surface voxels using an RGB color code (Fig. 3). When applying this specific threshold of z < 0.05, approximately an equal number of cortical surface voxels is either connected to VIP, or not, most probably describing both direct and second-order indirect anatomical connectivity with VIP. Varying this threshold from between 0.03 and 0.07 did not substantially change the observations reported below (see Supplementary Fig. S3).

In a first step, we focus on the regions that show functional connectivity with all three VIP ROIs. There were two such regions (Fig. 3, white, Supplementary Fig. S1 for individual monkeys’ maps). They represent 16% and 18% of the total brain surface in the left and the right hemispheres respectively, and 33% and 36% of VIPs overall functional connectivity to the rest of the cortex. The first region covers a large extent of the parietal cortex and adjacent brain regions. Specifically, it is centered on the intraparietal sulcus (IPS) and extends along a medio-lateral axis. It thus includes the posterior part of the cingulate sulcus (CiS) and medial parietal cortex. It also includes the entire IPS, extending into the inferior parietal cortex up to the lunate sulcus (LuS). Last, it includes the posterior section of the superior temporal sulcus (STS), and extends into the cortical convexity between the STS and the lateral sulcus (LS). The second region is located in the frontal cortex, anterior to the AS, and mostly extends within and dorsally to the principal sulcus (PS).

In a second step, we focus on the regions that show functional connectivity with only two of our three VIP ROIs (Fig. 3, yellow, cyan, and magenta). These represent 12% and 10% of the whole-brain surface in the left and the right hemispheres respectively, and 24% and 20% relative to all the cortical surface voxels functionally connected to VIP (pie charts in Fig. 3 for detailed % per category).

The mVIP functional connectivity map exhibits large overlap with pVIP and aVIP functional connectivity maps, likely due to the current resolution of our fMRI data as well as to partial volume effects, resulting in enhanced functional connectivity between near-by voxels. As a result, most cortical surface voxels with significant functional connectivity with mVIP are also significantly functionally connected to either pVIP or aVIP (Fig. 3). We first describe the topographical organization of surface voxels functionally connected to aVIP, or to both aVIP and mVIP. Then, we describe the topographical organization of surface voxels functionally connected to pVIP, or to both pVIP and mVIP.

Surface voxels functionally connected to aVIP and aVIP-mVIP were organized along a gradient (indicated by the white-cyan-blue colors) anterior to the parietal cortex, 1) starting medially from the lateral bank of the CiS, extending along the medial bank of the IPS and ending laterally in the fundus of the LS in both hemispheres; 2) in the central sulcus (CS); as well as 3) in the medial parietal cortex, along the entire extent of the CiS, though with an emphasis on the anterior cingulate cortex. The mVIP-aVIP functionally connected surface voxels (cyan) occupied 2% and 4% of the whole-brain surface in the left and the right hemispheres respectively, while functionally connected surface voxels of aVIP (blue) accounted for 7% and 12% of all cortical surface voxels.

Surface voxels functionally connected to pVIP and pVIP-mVIP are organized along a gradient (white-yellow-red surface voxels) that extends 1) into the major parts of the occipital cortex, except most of the operculum. Part of this connectivity gradient starts medially at the posterior end of the CiS, and extends along the posterior occipital sulcus (POS) and the posterior bank of the LuS; a second branch starts between the STS and the inferior occipital sulcus (IOS), and occupies a large area in the lateral occipital cortex, including the IOS and the occipitotemporal sulcus (OTS). 2) Furthermore, surface voxels connected to pVIP and pVIP-mVIP are found in the middle of the superior temporal sulcus (STS), as well as 3) around and above the principal sulcus. The surface voxels functionally connected to pVIP-mVIP (yellow) occupy 9% and 5% of the whole-brain surface in the left and the right hemispheres respectively, while the pVIP functionally connected surface voxels (red) accounted for 12% and 10% of all cortical surface voxels in both hemispheres, respectively.

Although pVIP and aVIP are non-contiguous VIP sectors, 1% of all cortical surface voxels are connected to these two sectors but not to mVIP between them (magenta). They are located in very small sectors of the temporal, parietal, and occipital cortex. Their occurrence cannot be attributed to the current resolution of the functional MRI data. These surface voxels are few relative to the other types of surface voxels. Yet, they are remarkably symmetrical across hemispheres, suggesting that they reflect not mere measurement noise but a consistent organizational pattern across hemispheres and animals. Hence, these surface voxels may contribute to multisensory functions which rely on the combination of specific neuronal signals encoded in both pVIP and aVIP.

Last, surface voxels exclusively connected to mVIP (green) occupy less than 1% of the whole brain surface in both hemispheres, and are mainly found in prefrontal and cingulate regions. Again, since they were identified in symmetrical locations in both hemispheres and remotely from the seeding ROIs, i.c., in the medial prefrontal cortex and in the ventral lateral prefrontal cortex, this suggests that mVIP might be exclusively connected to specific cortical areas, not targeted by pVIP and aVIP.

In order to better understand the functional significance of the differences in whole-brain functional connectivity of aVIP, mVIP, and pVIP at different scales, we next assessed their functional connectivity to specific brain regions and to specific brain areas. The first regional level includes orbitofrontal, prefrontal, premotor/motor, cingulate, limbic, temporal, parietal, and occipital regions (Fig. 4, same connectivity maps as Fig. 3 but with additional regional and areal broaders), (Markov et al., 2014). We calculated the percentage of surface voxels functionally connected to aVIP, mVIP, and pVIP and their combinations (Fig. 4, pie charts). At the second areal level, we further parcellated the cortex into 92 functionally and anatomically defined cortical areas (Fig. 4), using Markov atlas (Markov et al., 2014). We then calculated the ratios of surface voxels showing significant functional connectivity with each of the individual VIP seeds (Fig. 5) as well as the corresponding average z-scores (Supplementary Fig. S4), separately for each region and each area. In the following paragraph, we first discuss the cortical regions expressing a dominance of functional connectivity with aVIP, or aVIP-mVIP. Then, we discuss the cortical regions expressing a dominance of functional connectivity with pVIP, or pVIP-mVIP.

When we zoom in to connectivity in the parietal cortex, the percentage of surface voxels that were functionally connected to all three VIP seeds (white surface voxels) was the largest, compared to all other categories (Fig. 4, left hemisphere: 36%, right hemisphere: 39% of all parietal surface voxels; note that in Figure 3, percentages are computed over the entire hemisphere while in Figure 4, they are computed relative to the specific brain region); this finding is unsurprising given that VIP is located in the parietal cortex. Parietal surface voxels connected to all three VIP ROIs were located in areas LIP, MIP, PIP, 7A, and V6A (Figs. 4 and 5; Supplementary Fig. S4). The latter areas did not show a preferential functional connectivity pattern to any of the three VIP ROIs (Kruskal-Wallis test, p > 0.05).

Twenty-two percent and 28% of the parietal surface voxels in the left and right hemisphere, respectively, were functionally connected to aVIP (Fig. 4, blue surface voxels). Only a small fraction of surface voxels were functionally connected to both aVIP and mVIP (Fig. 4, cyan surface voxels, left hemisphere: 7%, right hemisphere: 7% of all parietal surface voxels). These surface voxels were located in areas 1, 2, 3, 5, 7m, 7op, 7B, AIP, and S2. They were characterized by stronger functional connectivity with aVIP than mVIP and pVIP when pooled across hemispheres (Fig. 5, Kruskal-Wallis test with Benjamini-Hochberg correction, p < 0.05).

A yet smaller percentage of parietal surface voxels were functionally connected to aVIP-pVIP (magenta surface voxels), to mVIP-pVIP (yellow surface voxels), or only to pVIP (red surface voxels). These latter types of surface voxels were either very few or located in non-symmetrical location across hemispheres. They were situated in area DP (Markov et al., 2014) also known as Opt (Pandya & Seltzer, 1982) which, overall, had stronger functional connectivity with pVIP than the other two RIOs (Kruskal-Wallis test Benjamini-Hochberg procedure, p < 0.05).

Overall, parietal surface voxels thus exhibited a posterior to anterior functional connectivity to aVIP, mVIP, and pVIP, such that the most posterior surface voxels were functionally connected to all three VIP ROIs; these were bordered by aVIP-mVIP functionally connected surface voxels, while surface voxels uniquely connected to aVIP were located more anteriorly.

Zooming in on the cingulate cortex revealed wide-spread functional connectivity to VIP (left hemisphere, 48%; right hemisphere 57%, of all cingulate surface voxels). Posterior cingulate regions were either functionally connected to all three VIP ROIs, to both pVIP-mVIP, or only to pVIP (Fig. 4). This pattern of functional connectivity is very similar to the one we report in the present study in the occipital cortex and mostly concerns area 31 (Figs. 4 and 5). More anteriorly, functional connectivity shifts to aVIP-mVIP and more rostrally to aVIP. This functional connectivity pattern covers areas 23 and 24d and extends dorsally toward the medial part of the premotor cortex (Fig. 4). More anteriorly, functional connectivity is evident with pVIP and pVIP-mVIP bilaterally, and with mVIP in the left hemisphere. This pattern corresponds to a medial extension of the functional connectivity characterizing adjacent prefrontal cortex. Overall, area 31 is the most densely functionally connected cingulate region to the VIP ROIs, followed by areas 24b, 24c, 24d, and 23. Areas 25 and 24a, on the other hand, are only weakly functionally connected to VIP (Fig. 5, ratio <20%).

Zooming in on the premotor and motor regions reveals a dominance of surface voxels uniquely functionally connected to aVIP in both hemispheres (Fig. 4, left hemisphere, 20%; right hemisphere: 40% of premotor/motor surface voxels). In these regions, F1 to F7 except area ProM were functionally connected to aVIP in both hemispheres (Fig. 5). However, area F1 stood out as an area in which the ratio of surface voxels functionally connected to aVIP were significantly higher than the ratio of surface voxels functionally connected to either mVIP or pVIP in both hemispheres (Kruskal-Wallis test with Benjamini-Hochberg procedure, p < 0.05). The other premotor/motor areas had higher ratios of surface voxels functionally connected to pVIP or mVIP. This was particularly marked for areas F4 and F7 (Figs. 4 and 5). In F2, F4, and F5 of both hemispheres, surface voxels connected to mVIP and pVIP form a weak mVIP-aVIP to aVIP (white-cyan-blue) gradient posteriorly to the white surface voxels (i.e., those functionally connected to all three VIP ROIs) centered between the AS and the PS (Fig. 4). In the medial part of the premotor/motor cortex, F3 and F6 were functionally connected mostly to aVIP (Kruskal-Wallis test with Benjamini-Hochberg procedure, p < 0.05). In these areas, the ratios of surface voxels correlated with aVIP were higher in the right than in the left hemisphere.

Only aVIP shows significant functional connectivity with the insula (relative to mVIP and pVIP when data from both hemispheres are pooled, Kruskal-Wallis test with Benjamini-Hochberg correction, p < 0.05, Fig. 5. If the left or right hemispheres are considered separately, both are non-significant). This corresponds to the ventral-most end of the cortex-wide mVIP-aVIP to aVIP (white-cyan-blue) gradient (Fig. 4). Other areas of the limbic system are largely uncorrelated with any of the VIP seeds (Figs. 4 and 5). The aVIP surface voxels make up 6% and 14% of the entire limbic cortex in the left and the right hemispheres, respectively, indicating relatively weak functional connectivity of VIP to this region.

The mVIP-pVIP to pVIP (white-yellow-red) gradient posterior to the seed ROIs was mainly located in the occipital and temporal regions (Fig. 4). In the occipital cortex, the pVIP (red) and mVIP-pVIP (yellow) surface voxels occupied 33% and 21% in the left hemisphere, and 29% and 14% in the right hemisphere, respectively (Fig. 4). While functional connectivity to pVIP was predominant, significant functional connectivity to mVIP and to a lesser extent to aVIP was also evident (Fig. 5). Furthermore, ratios of surface voxels that functionally connected to all three VIPs increased from V1 (left: 26%; right: 20%), to V2 (30%, 27%), to V3 (37%, 32%), to V3A (66%, 62%) as well as from V4 (42%, 34%), to V4t (47%, 64%), and to V6 (57%, 60%).

Zooming in on the temporal cortex revealed that pVIP (red) and mVIP-pVIP (yellow) surface voxels occupy 16% and 6% of the temporal region of the left hemisphere, respectively, and 10% and 5% in the right hemisphere (Fig. 4). The mVIP-pVIP to pVIP (white-yellow-red) global gradient contained two branches, and a small anterior part of the second (lateral) branch was located in the temporal region, especially in TEO and TEOm (Fig. 4). These areas were close to the lateral end of the mVIP-pVIP to pVIP gradient in STS, including FST, PGa and IPa. The connectivity pattern in the aforementioned five areas suggests a weak trend of functional connectivity to pVIP (Fig. 5), but this apparent trend was not statistically significant (Kruskal-Wallis test with Benjamini-Hochberg procedure, p > 0.05). The ratio of surface voxels functionally connected to pVIP decreased more laterally in TEam/p and TEpd. Medial to these areas were white surface voxels functionally significantly connected to all three VIP ROIs, including MT, MST, STPc, and Tpt (Fig. 4). This was also near the location where the aVIP-mVIP-pVIP to pVIP-mVIP to pVIP gradient and the aVIP-mVIP-pVIP to aVIP-mVIP to aVIP gradient touch each other. As a result, MST and STPc had very high ratios of functionally connected voxels to all three VIPs, while MT and Tpt were dominated by functionally connected surface voxels of pVIP and aVIP, respectively (Kruskal-Wallis test with Benjamini-Hochberg procedure, p < 0.05, Fig. 5).

Zooming in on the prefrontal cortex revealed dense connectivity to all three VIP ROIs, which is remarkable given its remote distance from the IPS. It represented the second largest functionally connected region to all VIP ROIs, covering the AS and the PS and accounting for 51% and 31% of the entire prefrontal region in the right and left hemispheres respectively (Fig. 4). In the left hemisphere, areas 9, 9/46d, 45A, 46d, and 46v were strongly functionally connected to pVIP, with weaker functional connectivity to mVIP and aVIP (Fig. 5). In the right hemisphere, functional connectivity was equally strong between these areas and the three VIP ROIs (Fig. 5). In the left hemisphere, an additional mVIP-pVIP to pVIP (white-yellow-red) gradient was evident anteriorly. The pVIP (red) connected surface voxels and the mVIP-pVIP (yellow) surface voxels occupied 15% and 23% of the prefrontal region. In the right hemisphere, there were no obvious mVIP-pVIP to pVIP (white-yellow red) or mVIP-aVIP to aVIP (white-cyan-blue) gradients but a “mosaic composition”, associating functional connectivity from different VIP ROI combination, at the edge of the white surface voxels region. This mosaic composition was larger at the ventral side (Fig. 4). The ratios of the aVIP (6%) and the mVIP-aVIP (8%) connected surface voxels were, however, higher than the pVIP (2%) and the mVIP-pVIP (4%) surface voxels within the mosaic prefrontal cortex composition. This contrasted with the left hemisphere, where pVIP and mVIP-pVIP surface voxels made up the mVIP-pVIP to pVIP gradient, and where the ratios of the aVIP and the mVIP-aVIP connected surface voxels accounted for less than 2% of the prefrontal region (Fig. 4). Another mosaic organization was evident at the junction of the prefrontal, cingulate, and premotor/motor regions, bilaterally, including areas 8B, 9, 24c, and F6. In the left hemisphere, this spot was dominated by pVIP (red), mVIP-pVIP (yellow), and aVIP-pVIP (magenta) surface voxels, while the right hemisphere exhibited a mosaic composition (Fig. 4).

The connectivity pattern of the orbitofrontal regions resembled that of the limbic regions, with weak functional connectivity with VIP. Indeed, orbitofrontal surfave-voxels showed significant functional connectivity with pVIP, specifically surfave-voxels located in area 11 and the medial portion of area 12 of both hemispheres. The pVIP surface voxels accounted for 8% and 3% in the orbitofrontal regions of the left and right hemispheres (Fig. 4).

We divided macaque VIP into three isovolumetric ROIs along its anterior-posterior axis to identify their whole-brain functional connectivity patterns and to characterize their degree of similarity. Our results confirm previous findings of strong functional connectivity between the fundus of the IPS and the prefrontal cortex anterior to the AS and around the AS from all three ROIs (Lewis & Van Essen, 2000a). In addition, and similarly to what we have described in humans (Foster et al., 2022), whole-brain functional connectivity of the posterior part of VIP was different from the whole-brain functional connectivity of its anterior part, thus defining an antero-posterior functional gradient in this region. Below, these results are discussed in light of the anatomical and physiological properties of macaque VIP as well as the functional specialization of the putative VIP homologue in humans.

Resting-state functional connectivity reflects both anatomical connections (Card & Gharbawie, 2022; Matsui et al., 2011; Tang et al., 2021; van den Heuvel et al., 2009; Wang et al., 2013) and indirect structural connections (Honey et al., 2009; Vincent et al., 2007). In our study, all known VIP anatomical connections (Lewis and Van Essen, 2000b) are also identified in its functional connectivity (Fig. 5). However, we observed additional functional connections that have not been identified anatomically, which thus most probably indicate indirect functional connectivity. A previous tracer study classified VIP connection strength in each cortical area into three levels (Lewis & Van Essen, 2000b, table 3). The areas in these three different levels corresponded to 34% ± 17% (mean +/- s.d.), 39% ± 19%, and 51% ± 24% surface voxels respectively that are functionally connected to VIP, on average (across areas and hemispheres). While the functional connectivity patterns reported in this study overall matched anatomical connectivity strength, the amount of connectivity did not always match between the two types of connectivity. For example, about 10% of surface voxels of area 24a and TEa/ma are functionally connected to VIP (Fig. 5), while 60% of surface voxels of areas 8 m and V3A are functionally connected to VIP. Yet both areas 24a and TEa/ma as well as areas 8 m and V3A show only weak anatomical connectivity with VIP. Such variability also existed in the group of areas showing the strongest anatomical connectivity with VIP, for example, approximately 20% of F1’s surface voxels are functionally connected to VIP (Fig. 5), while 80% of surface voxels in the posterior intraparietal area (PIP) are functionally connected to VIP. This is in line with a previous study comparing diffusion tensor imaging and resting-state functional connectivity showing that indirect structural connections contribute to unique functional connections without structural connections and introduce variability in functional connection strength (Honey et al., 2009).

Several areas exhibited functional connectivity to VIP, although no corresponding anatomical connectivity has so far been reported. Connectivity with orbital-frontal regions was not reported in the seminal anatomical connectivity study by Lewis and Van Essen (2000b, table 3); yet, in the resting-state data, we found functional connectivity between VIP and area 11 and 12. The extent of this functional connectivity was similar to that observed between VIP and area 24a and TEa/ma (ca. 10% of the surface voxels, Fig. 5), two cortical regions for which weak anatomical connections with VIP have been identified. If we set an arbitrary criterion of 10% of an area’s surface voxels exhibiting connectivity as evidence for functional connectivity, we identified several other cortical brain areas that are functionally connected to area VIP while no previous report of corresponding anatomical connectivity exists. These includes area 9 (including 9/46v and 9/46d from the Markov map) and area 44 in the prefrontal regions, area 24c and area 31 in the cingulate regions, and area V1 in the occipital region; multiple areas in the temporal cortex, including area TEO, TEpd, and TEpv in area TE, areas TH and TF, as well as the caudal part of the parabelt (PBc), including medial belt, core belt region, and lateral belt of the auditory cortex; as well as limbic regions including the temporal pole, parainsula, insula, and gustatory cortex.

The vast number of VIP’s anatomical connections (Lewis & Van Essen, 2000a) can be classified into eight functional categories (Foster et al., 2022) based on their contribution to the functions ascribed to VIP (Fig. 6). We used these anatomically connected areas and functional networks as a template to test whether the functional specialization of the differential connectivity patterns observed for our three macaque VIP subdivisions match with the three putative VIP homologue clusters we have suggested as the expanded human VIP homologue in the human brain (Foster et al., 2022). Human fMRI studies have applied a specific set of tasks to identify VIP based on a specific function that had previously been reported by research on macaque VIP. We could match the human fMRI task designs associated with each cluster with the eight functions attributed to macaque VIP. According to this matching procedure, the most posterior human VIP homologue area, hVIP#1, is associated with motion detection and eye movements, the more lateral hVIP#2 is associated with proprioception and eye movements, and the most anterior hVIP#3 is associated with tactile processing and peripersonal space representation. To compare the functional connectivity patterns of the VIP subdivisions between humans and macaques, we performed a meta-analysis of the literature and matched the macaque atlas-defined areas to the three human VIP clusters. We then identified to which macaque VIP ROI each macaque area has the strongest correlation with, and compared these results with the human cluster functional network (Fig. 6). As in our analysis no area had significantly stronger connectivity to the mVIP than to the aVIP or pVIP (Fig. 5; Supplementary Fig. S2), we concentrated our comparison on the latter ROIs (Fig. 6).

Macaque aVIP is functionally connected to the supplementary eye field (SEF), primary motor cortex (M1), F4, F5, 3a, 2, 2V, 5, 7m, AIP, secondary somatosensory cortex (S2), 7b, PFG, and temporoparietal area (Tpt) (Fig. 6, circled in blue). This network is specialized in the processing of eye movements, vestibular information, touch, proprioception, audition, and peripersonal space. It exhibits strong overlap with the functional specialization of human hVIP#2 and hVIP#3 for eye movement, tactile, proprioception, and peripersonal space. However, because tactile and proprioceptive information are often processed in the same areas, it is difficult to associate anterior macaque VIP selectively with either hVIP#2 or hVIP#3. In contrast, areas that contribute to the processing of peripersonal space functionally segregate hVIP#2 from hVIP#3, as only hVIP#3 is associated with this function. These areas include F5, AIP, and PFG that belong to the anterior VIP network. These areas also include V6A and the middle intraparietal area (MIP), both of which are strongly connected to both anterior and posterior VIP ROIs, possibly because they are very near to the seeds. Taken together, this points toward a functional homology between macaque anterior VIP and hVIP#3, and to a lesser extent hVIP#2.

Macaque pVIP is functionally connected to V2, V3, V3A, V6, medial temporal area (MT), medial superior temporal area (MST), fundus of the superior temporal sulcus area (FST), LIP, 7a, frontal eye field (FEF), and lateral prefrontal cortex (lPFC) (Fig. 6, circled in red). Areas in this network contribute mainly to motion detection and eye movements, which match with the functional specialization of human hVIP#1.

In sum, the functional specialization of the macaque anterior VIP functional connectivity network matches that of the lateral and the anterior human homologue clusters, while the posterior VIP functional connectivity network matches the posterior human homologue cluster. These results are in line with our hypothesis that functional specializations also exist within the macaque VIP along the anterior-posterior axis, and that this specialization within area macaque VIP can be viewed as a precursor of the more obvious separation into three distinct areas in the human brain. Tactile and proprioceptive functions are encoded in macaque aVIP, whereas motion detection and eye movement-related functions are located in macaque pVIP. These connectivity-based results confirm prior observations about functional specialization within macaque VIP that were based on functional localizers in awake macaques (Guipponi et al., 2013, 2015).

In the present study, anterior, middle, and posterior ROIs were composed of equivalent numbers of voxels. pVIP appears to uniquely match human hVIP#1, whereas anterior macaque VIP matches both human hVIP#2 and hVIP#3, while macaque mVIP may not have distinctive properties relative to macaque aVIP and pVIP. This might suggest a stronger functional specialization and expansion of middle and anterior relative to posterior human VIP, in agreement with the fact that cortical expansion in humans relative to macaques is more marked around the temporo-parietal junction (TPJ) and the inferior PPC than in medial PPC, that is, the areas in which these human VIP homologues are located (Chaplin et al., 2013; Hill et al., 2010; Mantini et al., 2013; Van Essen & Dierker, 2007). This match of cortical expansion with the emergence of hVIP#2 and hVIP#3 may also be why there is predominant colocalization of higher-order cognitive functions in hVIP#2 and hVIP#3 relative to hVIP#1, for instance for peripersonal space representation and embodiment (Foster et al., 2022).

VIP functional connectivity networks include both areas that have been described to be anatomically connected to area VIP, and areas that are activated through indirect functional connectivity. We refer to these latter areas as the broader VIP functional connectivity networks. The macaque peripersonal space network includes four sub-networks (Cléry et al., 2015; Foster et al., 2022): MIP/V6A for reaching (Battaglia-Mayer et al., 2003; Fattori et al., 2017; Gamberini et al., 2009), 7b/F5/AIP for grasping (Menz et al., 2015; Michaels & Scherberger, 2018; Salimi et al., 1999), F4/VIP for self-defense (Cooke et al., 2003; Graziano & Cooke, 2006), looming and impact prediction to the body (Cléry et al., 2017), near space coding (Cléry & Ben Hamed, 2018; Cléry et al., 2018), and FEF/LIP for oculomotor control (Ben Hamed & Duhamel, 2002; Brincat et al., 2018; Hart & Huk, 2020; Wardak et al., 2011, 2012) (Fig. 6). In addition, recent research demonstrated a strong social component in the functionality of peripersonal space (Cléry & Ben Hamed, 2018, 2021; Cléry et al., 2015, 2017; Serino, 2019), suggesting that the neural mechanisms underlying body ownership and peripersonal space are linked (Blanke, 2012; Blanke et al., 2015; Makin et al., 2008), and potentially also contribute to social processing (Orban et al., 2021). Although mirror self-recognition is generally not considered to be an intrinsic cognitive ability in monkeys (Gallup, 1982), it has recently been shown that macaques are capable of learning precise visual-proprioceptive associations of mirror images (L. Chang et al., 2017). Accumulating evidence in secondary somatosensory area S2 (Bretas et al., 2021) and in the insula (Evrard et al., 2012; R. J. Schneider et al., 1993) address the neural basis for body ownership in macaques. Interestingly, while S2 is anatomically connected to VIP, the insula is only functionally connected to VIP (suggesting an indirect anatomical connection). VIP could thus play an important role in body ownership of macaques, similarly to what has been described in the human hVIP#3.

In addition to the insula, there are other areas in the limbic, cingulate, orbitofrontal, and prefrontal regions that are functionally, but not anatomically, connected to VIP. In the limbic regions, the gustatory cortex (Kaskan et al., 2019) adds one more sensory modality to the multisensory properties of VIP, that has not been investigated to date. The temporal pole was found to contribute to the processing of familiar faces (Landi et al., 2021), while orbitofrontal area 12 processes information related to faces and emotion (Murphy & Bachevalier, 2020), as well as reversal learning (Rudebeck & Murray, 2011; Sallet et al., 2020). Related to this function, anterior cingulate cortex (Vogt et al., 1987) in the cingulate regions codes values and uncertainty needed for action selection (Monosov et al., 2020) via a network including orbitofrontal areas 11, 13 (Burke et al., 2014; Murray et al., 2015) and prefrontal areas 9, 46, 9/46 (Oemisch et al., 2015). More specifically, only area 24c (anterior cingulate sulcus) within area 24 is part of the broader VIP functional connectivity network, which signals reward allocation including during cooperative behavior (S. W. C. Chang et al., 2013). Partially overlapping with the value network, prefrontal area 44 controls vocalization (Korponay et al., 2020; Petrides et al., 2005) together with anterior/mid cingulate cortices and pre-supplementary/supplementary motor areas (Loh et al., 2017). Together, these considerations point toward a putative role of area VIP in a diverse set of higher-order social and cognitive functions that is yet to be explored.

The strength of the dataset on which we based our study is the high number of animals (n = 10) and the fact that the investigated animals were awake during recording. The anesthetized state mostly reduces functional connectivity (Grandjean et al., 2014; Hutchison et al., 2014; Thomas et al., 2021; Vincent et al., 2007), while at the same time, some specific networks like visual and default networks actually show anesthesia-related increases in connectivity (Xu et al., 2018). Thus, awake state measurements provide a better understanding of the physiological functional networks.

The high number of individuals involved in this study enables group-level statistics for subareal ROI to whole-brain single-voxel connectivity analysis. The high spatial resolution functional connectivity reveals details within regions, for example, VIP-M1 connections concentrates laterally in M1 (Fig. 4) which corresponds to VIP’s specific role in face/head processing. Furthermore, detailed connectivity patterns also show correspondence across areas, like the medial and lateral branches of the pVIP to pVIP-mVIP gradient in the occipital regions. These functional connection details could guide future experiments that aim for parallel recordings across VIP and other areas to probe their inter-areal communications (Bastos et al., 2015; Dann et al., 2016; M. Schneider et al., 2021; Semedo et al., 2019). On the other hand, the newly identified broader VIP functional connectivity network sheds light on higher cognitive and social networks (Fujii et al., 2008; Jezzini et al., 2015; Serino, 2019; Villa et al., 2022), in which the role of VIP is yet to be explored.

Finally, the resting-state functional connectivity serves as a common research method for cross-species comparison between human and macaques (Barron et al., 2020; Garin et al., 2022; Gorges et al., 2017; Xu et al., 2020), as well as other non-human primates like marmosets (Garin et al., 2022; Hori et al., 2021). The abundant cytoarchitecture (Lewis & Van Essen, 2000b) and structural connectivity (Lewis & Van Essen, 2000a) data in macaque PPC compensate the incomplete data in humans (Amunts et al., 2020; Niu et al., 2020) by assuming homology between the two species. Conversely, the rich fMRI studies in humans and the possibility to perform complex cognitive tasks also impact the understanding of brain functions and design of future experiments in non-human primates. Specific to this study, the three homologue human VIP clusters based on fMRI findings motivated the search of macaque VIP functional specialization based on intra-areal differential functional connectivity. Future research in humans can also probe the functional connectivity between the three hVIP clusters and the rest of the cortex for a more detailed cross-species comparison, while in macaque, electrophysiology recordings along the anterior-posterior axis within IPS can help delineate the functional specializations at the neuronal level.

Our finding that the functions of macaque VIP subdivision functional connectivity networks match the functional specializations of human homologue VIP clusters must be viewed in the light of some technical limitations. This study is uniquely based on the analysis of awake monkey resting-state data in combination with an atlas-based approach by selecting voxels along the IPS fundus that covered VIP from the Markov map (Fig. 1). Confronting the observed functional connectivity between VIP and the rest of the brain with standardized and validated functional localizers would be very valuable to confirm the proposed functional gradient. Indeed, such localizers are expected to capture functional properties more consistently and precisely than atlas-based anatomical localization (Russ et al., 2021), especially when individual variation in VIP functional activations have been reported (Guipponi et al., 2013), as it has been shown that subject-specific ROIs are important for an enhanced functional resolution (Nieto-Castañón & Fedorenko, 2012). However, designing valid localizers for areas involved in multisensory and higher cognitive functions could be more challenging than for areas with low-level functional properties, such as primary somatosensory area S1 (Thomas et al., 2021). In addition, while the present study includes 10 monkeys and, thus, provides evidence of cross-subject reproducibility of the observations, running functional localizers on such a large dataset is challenging in any given lab and will require concerted effort from multiple labs worldwide for several years (M. Milham et al., 2020, 2022; M. P. Milham et al., 2018).

Data and code are available upon request.

Conceptualization: W.-A.S. and S.B.H.; Data Curation: W.V., W.-A.S., and M.F.; Formal Analysis: W.-A.S. and S.B.H.; Funding Acquisition: S.B.H.; Investigation: W.-A.S., and S.B.H.; Methodology: W.-A.S., S.C., M.F., and S.B.H.; Resources: W.V.; Supervision: S.B.H.; Validation: S.B.H.; Visualization: W.-A.S., C.S.; Writing—Original draft: W.-A.S. and S.B.H.; and Writing—review & editing: W.-A.S., S.C., M.F., W.V., T.H., and S.B.H.

S.B.H. was funded by the French National Research Agency (ANR) ANR-18-CE92-0048-01 grant and the LABEX CORTEX funding (ANR-11-LABX-0042) from the Université de Lyon, within the program Investissements d’Avenir (ANR-11-IDEX-0007) operated by the French National Research Agency (ANR). W.V received funding from KU Leuven C14/21/111, and Fonds Wetenschappelijk Onderzoek-Vlaanderen (FWO-Flanders) G0E0520N, G0C1920N.

The authors declare no conflict of interest.

Animal care and experimental procedures were performed in accordance with the National Institute of Health’s Guide for the Care and Use of Laboratory Animals, the European legislation (Directive 2010/63/EU) and were approved by the Animal Ethics Committee of the KU Leuven. Weatherall reports were used as reference for animal housing and handling. All animals were group-housed in cages sized 16–32 m³, which encourages social interactions and locomotor behavior. The environment was enriched by foraging devices and toys. The animals were fed daily with standard primate chow supplemented with fruits, vegetables, bread, peanuts, cashew nuts, raisins, and dry apricots. The animals were exposed to natural light and additional artificial light for 12 h every day. On training and experimental days, the animals were allowed unlimited access to fluid through their performance during the experiments. Using operant conditioning techniques with positive reinforcers, the animals received fluid rewards for every correctly performed trial. During non-working days, they received water in their living quarters. Throughout the study, the animals’ psychological and veterinary welfare was monitored daily by the veterinarians, the animal facility staff, and the lab’s scientists, all specialized in working with non-human primates. The animals were healthy at the conclusion of our study. M1 and M2 are currently still employed in other studies.

We also thank Thomas Perret, Johan Pacquit, and Marco Bimbi for help with the hardware computational resources.

Supplementary material for this article is available with the online version here: https://doi.org/10.1162/imag_a_00491.