Overall view of scene-selective regions in human and monkey visual cortex. Both species fixated the center of a screen during block-designed presentation of identical scene vs. face localizing stimuli. In the human data (panels A–E), relatively higher activity to scenes vs. faces is shown in red/yellow vs. blue/cyan, respectively (minimum = p < 10 ^ -10; maximum = p < 10 ^ -30). The human data is a group average of both the functional and the anatomical data (n=10), in cortical surface format. The right hemisphere is illustrated; data from the left hemisphere were similar. Panels A and B show the medial and lateral-posterior views of folded cortex, respectively. Panels C and D show corresponding views of inflated cortical surfaces, and panel E shows the flattened view. For comparison, panel F shows the comparable flattened activity map from a macaque monkey, viewing the same stimuli (minimum = p < 10 ^ -5; maximum = p < 10 ^ -10). In both species, presumptively corresponding scene-selective regions are named in white (preceded by ‘m’ in the macaque map). As a reference, black asterisks indicate FFA in humans (panel E) and its counterpart in macaque (panel F); these are already known to correspond to each other (Tsao et al., 2003; Rajimehr et al., 2009).

Evidence for a middle fusiform sulcus in the averaged human cortical surface maps. Panel A shows the averaged surface from the present data (n = 10), and panel B shows the magnified inset from the same data. Panel C shows the averaged map from an independent subject pool (standardized Free Surfer surface; n=40), and D is the magnified inset. Gyri are abbreviated in green (PH = parahippocampal; L = lingual; TO = temporal occipital; IT = inferior temporal; MF = medial fusiform; LF = lateral fusiform). Sulci are abbreviated in red (C = collateral; TO = temporal occipital; IT = inferior temporal). The right hemisphere is illustrated; data were similar in the left hemisphere. In both hemispheres, and in both cortical surfaces, the fusiform gyrus is subdivided into two parallel branches by the middle fusiform sulcus (MFS, white arrowhead).

Confirmation of a middle fusiform sulcus in individual ex vivo human brains. Panels A C show ventral views of the temporal lobe in three right hemispheres (anterior = rightmost; lateral = uppermost). Sulci and gyri are indicated as in Figure 2. In all three examples, a clear middle fusiform sulcus (‘MFS’, labeled in white) divides the fusiform gyrus into two branches.

Scene-selective activity from the present experiment was consistently centered on the lip of the medial fusiform gyrus and collateral sulcus. The right hemisphere is illustrated; data was similar in the left hemisphere. Panel A: group-averaged activity in response to the scene-vs.-face (i.e. PPA vs. FFA) localizing stimuli (scenes = red/yellow; faces = blue/cyan), rendered on the group-averaged cortical surface (minimum = p < 10 ^ -10; maximum = p < 10 ^ -50). Panel B shows the location of the inset on the medial view of the entire hemisphere. The white circle in panel C shows the centroid (the vertex showing the highest scene bias) in the group-averaged map of ‘PPA’. Sulci (C = collateral; MF = middle fusiform) and gyri (PH = parahippocampal; MF = medial fusiform; LF = lateral fusiform) are abbreviated. Panel D shows the location of the centroid in each individual case; all are located on the lip of the collateral sulcus, on the fusiform side.

The topographical shape and center of PPA remains essentially constant, even when produced by different stimuli. All panels show group-averaged scene-selective activity (red/yellow) in medial views of the right hemispheres, equivalent to the views in Figure 4. Panels A–C show activity in one set of subjects (n=4 from group 1; Table 1), and panels D–F show activity in a different set of subjects (n=7 from group 2; Table 1); relevant comparisons are between panels from a common row (A–C or D–F). The color scaling was adjusted to topographically comparable levels, to offset differences in statistical power in each comparison, as a result of differences in subject number and variations in stimulus effectiveness (panels A–B: minimum = p < 10 ^ -10; maximum = p < 10 ^ -50; panel C: p < 10 ^ -3; maximum = p < 10 ^ -16; panels D and E: minimum = p < 10 ^ -5; maximum = p < 10 ^ -30; panel F: minimum = p < 10 ^ -2; maximum = p < 10 ^ -10). Panel A shows activation produced by localizing stimuli analogous to those used by Epstein and Kanwisher (1998): a wide range of scenes (scene image set #1: indoor rooms, empty rooms, outdoor city scenes, and landscapes) was contrasted with individual faces (image face set #1). Panel B shows activity in response to a more limited set of scenes (scene set #2, indoors, from the scanning laboratory), compared to images of groups of faces (face set #2). The scene and face stimuli in panel C were equivalent to those used in panel B, except they were summed from three retinotopically-limited ring apertures, at complementary ring diameters (retinotopic set #1). Panel D shows the activity produced by a different set of scenes (set #3; indoor and outdoor), compared with computer-generated faces (face set #3). Panel E shows the activity produced by a different set of scenes (scene set #4, indoor scenes from an unfamiliar site), compared with arbitrary computer-generated objects (‘blobs’; Yue et al., 2010). Panel F shows activity produced by scene set #3, compared with scrambled versions of the same scene stimuli, based on perturbing a random noise filed in different scales to match the original image statistics (Portilla and Simoncelli, 2000).

A meta-analysis of the previous literature localizes ‘PPA’ to the lip of the collateral sulcus/medial fusiform gyrus, consistent with the current data (Figures 4 and 5). Data are rendered on a standardized cortical surface (FreeSurfer). Panels B and D show the whole normal and inflated cortical surfaces, respectively. Panels A and C show the magnified insets. The centers of PPA activity in previous publications have been translated to this common surface based on Talairach coordinates (indicated with white letters in panels A and C). When distinguished in the individual reports, coordinates from left and right hemispheres were averaged together, and displayed here on the right hemisphere. The references corresponding to those letters are given in Table 1. For comparison, an asterisk marks the center of PPA in the current dataset (from Figure 4C). None of these centers was located on the crown of the parahippocampal gyrus, but five were located on the crown of the medial fusiform gyrus. Remaining centers were located on the lips (but not the fundus) of the collateral sulcus.

Group-averaged cortical surface maps showing a macaque homologue of human PPA. The data show the averaged fMRI activity (n = 3) from fixating monkeys presented with localizing stimuli identical to those used in the human maps (scene set #2 vs. face set #2). Panels A and B show activity in the folded cortical surface, in the right and left hemispheres, respectively. The left hemisphere has been mirror-reversed for ease of comparison. Panels C and D show the same data in flattened cortical format. Scene-biased activity (red/yellow) is shown across the entire cortex (minimum = p < 10 ^ -12; maximum = p < 10 ^ -24). The peak of face-biased activity in the posterior (‘mFFA’) and anterior face patches (‘mATFP’) are indicated with a black and a white asterisk respectively. Two patches of scene-biased activity are evident in this group-averaged data: one dorsal and another ventral. A consistent ventral scene patch (mPPA) is located adjacent and ventral to the posterior face patch, analogous to the relationship between FFA and PPA in humans. The dorsal scene-biased patch (‘mTOS’) is the presumptive macaque homologue of human TOS.

Group-averaged map of mPPA and mFFA from an independent set of experiments. This group average (n = 4) was based on fMRI acquired at NIMH, based on an independent set of stimuli (monkey faces, and scenes and objects from the housing and training environment of the monkeys at NIMH). Panels A and C show activity in the right hemisphere; panels B and D show the left hemisphere, mirror-reversed for ease of comparison. Panels A and B show the contrast of scenes (red/yellow) vs. faces (blue/cyan) (minimum = p < 10 ^ -6; maximum = p < 10 ^ -12). As a control, panels C and D show the contrast of scenes vs. [faces + objects]/2. In this NIMH data, the slice prescription did not include regions posterior to the superior temporal sulcus, to the left of the dashed line (e.g. V1, V3A). Despite these experimental differences, the overall results shown in Figure 7 were confirmed: 1) two face patches were found in the expected locations (asterisk above the posterior patch (mFFA); anterior patch indicated with a black arrowhead), and 2) a scene-biased patch is evident immediately ventral to the posterior face patch (white arrowheads), consistent with that in Figure 7, and with the position of FFA and PPA in the human map. Anterior to that, additional scene-biased activity was also found. When objects were included as a stimulus contrast (panels C and D), an additional (‘third’) face patch could be seen, depending on the level of thresholding. However additional evidence suggests that this patch reflects retinotopic differences, rather than differences related to semantic category.

In human cortex, the scene-selective region ‘TOS’ is centered on the lateral occipital gyrus. All panels show the right hemisphere from a posterior-lateral viewpoint, as in Figures 1B and 1D. Panels A–C are cortical surface maps in the folded format, and panels D–F show the same region in inflated surface format. Panels A, B, D and E show the right hemisphere, and panels C and F show the left hemisphere after mirror reversal. In panels A and D, gyri are abbreviated in green (LO = lateral occipital; IPP = inferior posterior parietal; A = anectant; MT = middle temporal), and sulci are abbreviated in red (TO = transverse occipital; IP = inferior parietal; L = lateral; LO = lateral occipital; IO = inferior occipital; PTO = posterior temporal occipital; ST = superior temporal; MT = middle temporal). The cyan lines in the remaining panels indicate the threshold (p < 10 ^-10) of group-averaged (n = 10) scene-biased activity in this region (‘TOS’). The white circles indicate the vertex showing the highest activity in the group map; on average (combining results from both hemispheres) this point lies along the crown of the lateral occipital gyrus.

In humans, scene-selective ‘TOS’ is consistently located immediately anterior and ventral to retinotopically defined V3A. Panels A–D show data from the flattened right hemisphere of a single subject. Panel A is a summary map of retinotopic areas (solid black lines = vertical meridian; dotted black lines – horizontal meridian; dashed black lines = foveal representation), scene-selective regions (labeled in yellow) and face-responsive regions (labeled in cyan), in one hemisphere. Panels B–D show the original maps on which panel A is based (panel B = the upper vs. lower field retinotopy; panel C = the vertical vs. horizontal meridian retinotopy, and panel D = scenes vs. faces, with overlaid retinotopy. Panels E–I show the combined maps (as in panel D) from five additional hemispheres. In all six hemispheres, the peak of scene-selective TOS is located immediately anterior and ventral to retinotopically defined area V3A. For all panels, minimum = p < 10 ^ -12 and maximum = p < 10 ^ -24.

Within-hemisphere comparisons of retinotopy and scene-selective activity in macaque TOS. The format is similar to that in Figure 10. Panels A–C are taken from a single hemisphere, analogous to panels B – D in Figure 10. Panel A is a retinotopic map produced by vertical vs. horizontal meridians (blue/cyan vs. red/yellow, respectively). Panel B = upper vs. lower visual field retinotopy (red/yellow vs. blue/cyan respectively). In panels A–B, minimum = p < 10 ^ -5 and maximum = p < 10 ^ -10. Panel C = activity due to scenes vs. faces (red/yellow vs. blue/cyan). Panels D–F show the combined maps as in panel C, from three additional hemispheres. In panels C–F, minimum = p < 10 ^ -5 and maximum = p < 10 ^ -20. All maps are shown in right hemisphere format. All scene vs. face stimuli were identical to those used in Figure 10.

The human scene-selective region ‘RSC’ is located adjacent to peripheral V1. All panels show a medial view of the inflated cortical surface. Panels A and B show the right and left hemispheres (respectively), illustrating the group-averaged scene-biased activity, including the sulcal/gyral labels (minimum = p < 10 ^ -10; maximum = p < 10 ^ -30); both RSC (top) and PPA (bottom) are visible. Subsequent panels show magnified views of the inset region (yellow borders, in panel A). Panel C shows the borders (white) of group-averaged RSC (white lines), plus the full extent of V1 based on the high myelination in layer 4B, in group-averaged data (Hinds et al., 2008). The location of the horizontal meridian representation in V1 is indicated with a dotted black line. Panel D shows RSC and PPA, as revealed in one individual with the scene vs. face contrast (minimum = p < 10 ^ -20; maximum = p < 10 ^ -30). Panel E shows borders of the group-averaged myelination map, plus the more central retinotopy (up to 10° eccentricity, based on vertical vs. horizontal meridians, in blue/cyan and red/yellow, respectively (minimum = p < 10 ^ -3; maximum = p < 10 ^ -5) from that same subject. Panel F shows the activity produced by flickering checkerboards positioned outside vs. inside the visible limit (minimum = p < 10 ^ -10; maximum = p < 10 ^ -30). The dotted line indicates the limits of retinotopic activity, from the data shown in panel E.

Evidence for RSC in one macaque monkey. A patch of scene-biased activity was present bilaterally, in a location consistent with the location of RSC in humans (i.e. in POm) (minimum = p < 10 ^ -5; maximum = p < 10 ^ -10). Panels A and B show medial views of this activity, in the right and left hemispheres, respectively. Relevant sulci are labeled in red (C = calcarine; PO = parietal occipital).

Diagram of visual cortical areas, relative to regions of scene-biased fMRI activity, in flattened visual cortex in humans (panel A) and macaque monkeys (panel B). Scene-biased regions are indicated in gray. Both maps are based on fMRI maps of retinotopy, motion selectivity plus face/scene selectivity in a representative single subject. Less understood regions, and regions not mapped directly in the present study, and regions that do not have accepted homologues in both species are indicated by dotted lines, or labeled without borders.
The interspecies correspondence is relatively good. Retinotopic areas V1, V2, V3, V3A and V4v, and motion-selective MT/V5 are similar in both species. The correspondence of FFA with mFFA, and the anterior temporal face patch (ATFP) is also excellent, based on quantitative cortical transformations (Tsao et al., 2003; Rajimehr et al., 2009). The adjacent arrangement of mFFA with mPPA is essentially identical to the arrangement of human FFA and PPA. Macaque cortex also shows a dorsal patch of scene-responsive activity, likely homologous with human TOS. Scene-responsive RSC is well established in humans, but less certain in monkeys.
Despite this correspondence in local neighborhood relationships, the maps suggest overall map differences relative to sites located farther from PPA. For instance, the distance between mPPA relative to ATFP is ~ 6 mm in macaque (center-to-center), but much further apart (~ 35 mm) in humans. Part (but not all) of this difference can be accounted for by relative expansion of the temporal lobe in humans relative to macaques, since ATFP remains ~ 6 mm from the anterior tip of the temporal lobe in both species. A similar situation is evident between (m)PPA relative to other distinguishable areas in anterior temporal lobe, including the subiculum.
Comparison of the maps also raises the possibility of the converse expansion in macaques relative to human, after equating relative surface areas. In humans, PPA abuts ventral V2 (Figure 12). However in macaques, there is a large region between mPPA and this retinotopic area, in which visual areas are poorly defined. It is possible that scene-responsive activity extends farther ventrally compared to the consistent ‘mPPA’ region than shown here (e.g. Figure 7B, D, and Figure 14B). Moreover, V4v and TEO may extend far enough ventrally to help fill this gap in the map.