Interspecies activation correlations reveal functional correspondences between marmoset and human brain areas

We first identified the cortical brain regions with consistent md-fMRI activity among subjects to remove the low reproducible areas from the subsequent analysis. To this end, we calculated intersubject correlation maps for the human and marmoset participants. We found high correlations between subjects in occipital, temporal, and parietal areas in both species (Fig. 1 A–F and SI Appendix, Fig. 1 A–D for the right hemisphere). Unlike humans, marmosets also showed high correlations in prefrontal cortices (PFC) including the medial PFC (mPFC; A25, A32), ventrolateral PFC (vlPFC; A45, A47), and dorsomedial PFC (dmPFC; A8b, A9). These findings are very similar to a previous human and macaque monkey study (20). A difference from this previous study is a lack of correlations in human auditory areas because our subjects watched the movie with no sound.

To identify the intra- and interspecies functional correspondence in early and high-order visual areas, we created the volume-of-interests (VOIs) based on the intersubject correlation maps using the multimodal cortical parcellation atlas (21) for humans and the Paxinos atlas for marmosets (22) (Fig. 1 G and H and SI Appendix, Fig. 1 E and F for the right VOIs). The time courses were extracted from all VOIs, and cross-correlation coefficients were calculated between all human and marmoset VOIs (Fig. 1I and Methods). The correlation matrices are shown in SI Appendix, Fig. 2. We found similar functional responses in human face- and known marmoset face-specific areas. For example, we found functional similarities between the fusiform face complex (FFC) and posterior inferotemporal areas (PIT) (21, 23–26) in humans and the V4 transitional part (V4T), fundal superior temporal area (FST), and TE3 in marmosets. The human peri-entorhinal and ectorhinal cortex (PeEc) was also correlated with the marmoset TE3. These marmoset areas are known to correspond to face patches based on blocked fMRI tasks (12, 13).

As a validation of our method, we next asked whether the previously described marmoset face patches (12, 13) could be detected by comparing the time courses across the marmoset brain to the time course in one of the human face patches, PeEc (21), also known as the anterior face patch or Brodmann area 35/36 (23, 26). Calculating the cortex-wise correlation with the time course in the human face patch revealed discrete patches of the activity in both human (red regions in Fig. 2 A and C) and marmoset occipital and temporal areas (red regions in Fig. 2 B and D). There were four main clusters in temporal areas (SI Appendix, Tables 1 and 2 for humans and marmosets, respectively). Each cluster (F1, F2, F3, and F4) in humans consisted of PeEc and TF for F1, FFC and PH, the TE2 posterior area (TE2p) for F2, LO2, and PIT, V4T for F3, and the middle temporal area (MT) and medical superior temporal area (MST) for F4 (Fig. 2C). Each cluster (F1, F2, F3, and F4) in marmosets consisted of area 36 (A36) and the entorhinal cortex (Ent) for F1, TE3 and the temporo-parieto-occipital association area for F2, FST and Pga-IPa for F3, and MT, MST, and V4T for F4 (Fig. 2C). The locations of these patches closely resembled those reported in the previous macaque (26–30) and marmoset fMRI studies (12, 13), but we also found statistically significant activation in marmoset A36. This area has not been found in previous marmoset face fMRI studies, but it cytoarchitectonically corresponds to the anterior face patch in humans (21, 23, 26, 31). These marmoset areas were also found when we used the other human face patches, FFC and PIT, as seed regions (SI Appendix, Fig. 5). This mapping of the face patches under more naturalistic viewing conditions demonstrates that it is possible to use this approach to identify marmoset areas that have the same function as specific human areas. All marmoset correlation coefficient maps with human VOIs are shown in SI Appendix, Figs. 3–6, and the peak locations of these maps are summarized in Fig. 3.

Next, we sought to examine which marmoset areas correspond to body- and scene-specific areas in humans. To do so, we seeded the human TE2p (21) and parieto-occipital sulcus area 1 [POS1 (21, 32)], also known as the retrosplenial cortex (33), to identify marmoset body- and scene-specific areas, respectively. Body-specific areas have already been described in one previous blocked fMRI study in marmosets (13), and these areas are located adjacent to face patches, as in humans and macaque monkeys (29, 34, 35). Scene-specific areas have not yet been identified in marmosets.

The time course in the human body-specific area (TE2p) was highly correlated with parts of face-specific areas (FFC and PIT) and the posterior superior temporal sulcus in humans, which is known as one of the body patches (35) (blue regions in Fig. 2 A and C and SI Appendix, Table 1). The time course in the human body patch was also highly correlated with the areas adjacent to the face patches in marmosets (blue regions in Fig. 2B and D and SI Appendix, Table 2).

The time course in the human scene-specific area (POS1) was highly correlated with A31, the dorsal visual transitional area, the PG posterior part, the parahippocampal area (PHA), the prostriate area (ProSt), the ventromedial visual area (VMV), V1, V2, and V3 in humans (green regions in Fig. 2 A and B and SI Appendix, Table 1), and area 19 of medial part (A19M), ProSt, TH (also known as PHA), V1, V2, and V3 in marmosets (green regions: Fig. 2 C and D and SI Appendix, Table 2). The locations of scene-specific areas in humans closely resembled those reported in previous human studies (36–39). The activation areas in marmosets were broadly cytoarchitectonically consistent with human activation areas.

To identify cortical areas with similar BOLD time courses in humans and marmosets, we computed for each human VOI the correlation of its time course with all marmoset cortical voxels and then plotted the peak locations of these correlation maps on the marmoset brain surface (Fig. 3). Although human and marmoset regions do not always correspond directly based on cytoarchitecture, we could clearly identify regions with similar time courses between the two species. Human face-specific areas (FFC and PIT), for example, have a similar time course to marmoset area V4T (red point in Fig. 3B), and the human scene-specific area POS1 has a similar time course as the PHA TH in marmosets (light blue point in Fig. 3B).

To evaluate what features in the movie activate marmoset regions functionally corresponding to human face and body patches, we first identified, for each repetiton time (TR) interval (1.5 s) of the movie, whether a marmoset, other animals (humans, owls, capybaras, dogs, cats, pigeons, roosters, frogs, scorpions, and ants), or no animals were visible. These three pseudoevent-related designs were convolved with a hemodynamic response function using FSL FEAT (Fig. 4A). We then calculated the correlation coefficients between these predicted designs and the time courses in each face and body patch. We found that the human face patches were more active when they watched not only marmosets but also the other animals compared to no animals (Fig. 4B; P < 0.05 for cluster F1, P < 0.01 for clusters F2, F3, and F4; ANOVA with Bonferroni post hoc correction). This tendency was also found for the human body patches (Fig. 4B; P < 0.01 for cluster B1; paired t test with Bonferroni post hoc correction). Unlike humans, marmoset face patches were more active when they watched marmosets (Fig. 4C; P < 0.05 for cluster F2, P < 0.01 for cluster F4; ANOVA with Bonferroni post hoc correction) than when they watched the other animals. As in humans, the marmoset body patches were activated when they watched both marmosets and the other animals compared to no animals (Fig. 4B; P < 0.05; paired t test with Bonferroni post hoc correction).

As shown in Fig. 1, both human and marmoset parietal areas also had high reproducibility between scans. We next asked which marmoset areas correspond to each human parietal area using the same method as high-order visual areas. However, for most human parietal areas, we could not find correlations with parietal areas in marmosets (SI Appendix, Figs. 7–8). Therefore, we next identified human parietal areas that correlated with each marmoset area surrounding the intraparietal sulcus (IPS) (the anterior intraparietal parietal area [AIP], lateral intraparietal parietal area [LIP], medial intraparietal area [MIP], PE, PG, occipito-parietal transitional area [OPt], and the ventral parietal area [VIP]). The analysis showed that the peak locations of the human correlation maps were located in the area PG inferior part (PGi) for the marmoset anterior/ventral parietal regions (AIP, OPt, PE, and PG) and in the human area PG superior part (PGs) for the marmoset parietal regions LIP, MIP, and VIP (Fig. 5).

The functional responses in the human body-specific patch (TE2p) were in good agreement with those in the marmoset area located adjacent to MD. This area is already known as a marmoset body-specific patch based on a previous electrocorticography and fMRI study (13). In addition, we also found that the human TE2p correlated with the marmoset area adjacent to the anterior part of AD, which had not been found yet. This finding is broadly consistent with previous findings in macaques and humans of body-selective areas located adjacent to face patches (29, 34, 50–53).

In humans, previous fMRI studies (29, 50, 51, 53) have described three visual cortical regions that are more active during the presentation of scenes or isolated houses compared with the presentation of other visual stimuli such as faces, objects, body parts, or scrambled scenes. Typically, these human brain regions are located in POS1, PHA, and transverse occipital sulcus. In this study, we found that functional responses in human POS1 and PHA corresponded to those in the marmoset retrosplenial cortex (A19M) and in the parahippocampal area (TH), suggesting a homologous neural architecture for scene-selective regions in the visual cortex for humans and marmosets.

We identified areas in the marmoset that functionally corresponded to face-, body-, and scene-specific areas in humans. Next, we sought to examine which marmoset areas show functional correspondences to other visual areas in humans. Our results demonstrate that the area MT and its surrounding areas in humans (FST, LO1, LO2, LO3, MST, and V4T) corresponded to a region at the border between FST, MST, and V4T. The area MT in both New- and Old-World monkeys has direction-selective neurons (54) and is considered to play an important role in motion perception, direction selectivity, and speed turning (55–57). The major output of MT is in the cortical areas surrounding it (i.e., FST, MST, and V4T) (see ref. 58 for macaques and ref. 59 for marmosets). Therefore, the movie may not be suitable to identify small functional differences between these regions. The movie we used was selected to identify face, body, and scene patches. To better identify motion-related areas, it would be ideal to use a movie in which stimuli move in various directions, speeds, and rotations, but this is beyond the scope of this study.

Although we could not identify functional differences between most marmoset parietal areas, the anterior/ventral (AIP, OPt, PE, and PG) and parietal regions surrounding the IPS (LIP, MIP, and VIP) in marmosets were correlated with PGi and PGs, respectively. As described in numerous studies, human PG is a part of the default mode network (DMN), while the areas surrounding the IPS (e.g., area LIP) is a part of the attention network (ATN) (21, 60, 61). In addition, the time courses of BOLD signals in human PG are anticorrelated with those in task positive regions (e.g., frontal eye field and LIP) even under resting state (62). In marmosets, one of the core regions of the DMN is the cortex surrounding the IPS (6, 63, 64). A functional imaging study in marmosets showed that these areas were deactivated during a visual stimulation task relative to the period when a black screen was presented (64). On the other hand, the area surrounding the marmoset IPS have also been linked to the attention-like frontoparietal network by several RS-fMRI studies (6, 65). It is also known that electrical microstimulation in the areas surrounding the IPS evoke saccadic eye movements (65) and that single neurons in the region show neural correlates for the gap effect in saccadic eye movement tasks (66). As such, it seems that parietal areas surrounding the IPS in the marmoset are involved in both default mode and ATNs. Taken together with our finding that the marmoset parietal area corresponded to the parietal regions in the human DMN, the DMN might be predominantly engaged during movie viewing in the marmoset. Indeed, our results showed that the human PG was correlated with the posterior cingulate cortex, dorsolateral prefrontal cortex, and parts of the anterior cingulate cortex and temporal areas (SI Appendix, Fig. 9) and that these areas overlap with the human DMN (21, 60, 61), which is deactivated during attention-demanding tasks (62, 67). Note that standardized cytoarchitectonic borders (as in the Paxinos atlas we used) may not always be accurate for each subject. Indeed, a recent paper showed that the location of marmoset LIP exhibits substantial variability across subjects (68).

We also observed consistent activation in marmoset frontal areas, in particular, the mPFC (A25, A32), vlPFC (A45, A47), and dmPFC (A8b, A9) (Fig. 1 D–F). This distribution closely resembled a recently identified network for social interaction processing (40, 69). We found that the macaque mPFC, vlPFC, and dmPFC are more activated when the monkey watched a social video (in which two monkeys were interacting with each other) compared to a nonsocial video (in which two monkeys were separately acting in their own environment). As such, consistent activation in the marmoset frontal areas might reflect social interaction processing during movie viewing. Although humans possess a similar social interaction network (70, 71), the social interactions of marmosets might not have been enough to consistently activate this network in our human participants.

A total of 13 healthy volunteers (nine males and four females, 22 to 56 y) and eight common marmosets (six males and two females, 20 to 42 mo, 300 to 436 g) participated in this study. All surgical and experimental procedures for marmosets were in accordance with the Canadian Council of Animal Care policy and a protocol approved by the Animal Care Committee of the University of Western Ontario Council on Animal Care. All animal experiments complied with the Animal Research: Reporting In Vivo Experiments guidelines. Human volunteers were informed about the experimental procedures and provided informed written consent. This study was approved by the Ethics Committee of the University of Western Ontario.

All marmosets underwent surgery to implant a head chamber to immobilize the head during MRI acquisition as described in previous reports (7, 72) and were trained to acclimatize to the MRI scanning environment and head-fixation system. Both humans and marmosets watched a 15-min naturalistic movie with no sound. The movie was edited from the British Broadcasting Corporation episode “Urban Jungles” from the “Hidden Kingdoms” series. The episode was condensed to 15 min by removing the parts that were not related to marmosets living in Rio de Janeiro and removing some scenes that seemed to be less interesting to marmosets based on some pilot data (essentially scenes that did not show any marmosets for a while).

Human volunteers lay in a supine position and watched the movie presented via a rear projection system (Avotech SV-6011, Avotec Incorporated) through a surface mirror affixed to head coil (6.9° field of view from the center to the side of the screen). Each marmoset was fixed to the animal holder using a neck plate and a tail plate. The marmoset was then head-fixed in a sphinx position using fixation pins in the MRI room to minimize the time in which the awake animal was head fixed (7). In marmosets, the eyes were monitored using an MRI-compatible camera (Model 12M-i, MRC Systems GmbH) to make sure that the animals stayed awake. For both species, the movie was presented via Powerpoint on a MacBook Pro.

Human images were acquired using a 68-cm head-only 7 T MRI scanner (Siemens Magnetom 7T MRI Plus) with an AC-84 Mark II gradient coil, an in-house 8-channel parallel transmit, and a 32-channel receive coil (73). Marmoset images were acquired using a 9.4 T 31-cm horizontal bore magnet (Varian/Agilent) and Bruker BioSpec Avance III HD console with the software package Paravision-6 (Bruker BioSpin Corp), a custom-built high-performance 15-cm-diameter gradient coil with 400-mT/m maximum gradient strength (74), and the five-channel receive coil (7). Full details of acquisition and analysis procedures can be found in SI Appendix.

To estimate the intersubject variability of stimulus-driven activity, we calculated the voxel-by-voxel correlation across subjects within each species after preprocessing the fMRI datasets such as motion correction, distortion correction, and normalization to the template. The time courses were extracted from each VOI (26 and 25 VOIs for humans and marmosets, respectively), and intra- and interspecies correlation matrices were created by cross-correlating the time courses among two regions using Matlab (MathWorks). To identify marmoset areas that have similar functional activations as human visual-related regions, we extracted the human time course in each 26 VOI, and we created the correlation coefficient maps with these time courses. These correlation coefficient maps were Fisher Z-transformed for each of the two species and were thresholded at z = 2.3. Full details of acquisition and analysis procedures can be found in SI Appendix.