Processing of visual and non-visual naturalistic spatial information in the "parahippocampal place area"

The “parahippocampal place area” (PPA) in the human ventral visual stream exhibits increased hemodynamic activity correlated with the perception of landscape photos compared to faces or objects. Here, we investigate the perception of scene-related, spatial information embedded in two naturalistic stimuli. The same 14 participants were watching a Hollywood movie and listening to its audio-description as part of the open-data resource studyforrest.org. We model hemodynamic activity based on annotations of selected stimulus features, and compare results to a block-design visual localizer. On a group level, increased activation correlating with visual spatial information occurring in the movie is overlapping with a traditionally localized PPA. Activation correlating with semantic spatial information occurring in the audio-description is more restricted to the anterior PPA. On an individual level, we find significant bilateral activity in the PPA of nine individuals and unilateral activity in one individual. Results suggest that activation in the PPA generalizes to spatial information embedded in a movie and an auditory narrative, and may call for considering a functional subdivision of the PPA.


Surface plots of individual results
The results of the xed-e ects second-level analysis of the movie's primary contrast (vse_new > vpe_old) and audio-description's primary contrast (geo, groom > non-spatial noun categories) mapped onto subject-speci c cortical surfaces can be seen in Figure S1. FreeSurfer v5.3.0 1 was used to reconstruct surfaces from T1-weighted images and an additional high-resolution T2-weighted image (s. github.com/ psychoinformatics-de/studyforrest-data-freesurfer). Thresholded -maps ( >3.4; <.05, cluster-corrected) were projected onto the cortical surfaces using FreeSurfer's 1 'mri_vol2surf' command (unthresholded -maps are provided at neurovault.org/collections/KADGMGVZ). After projection, individual PPA ROIs were spatially smoothed by applying a Gaussian kernel with full width at half maximum (FWHM) of 2.0 mm. Figure S1: Fixed-e ects individual-level GLM results ( >3.4; <.05 cluster-corrected) projected onto reconstructed subject-speci c brain surfaces. The results of the audio-description's primary -contrast (blue) that compares geometry related nouns to non-geometry related nouns spoken by the narrator (geo, groom > all non-geo) are overlaid over the movie's primary -contrast (red) that compares cuts to a setting depicted for the rst time with cuts within a recurring setting (vse_new > vpe_old). Black: outline of participant-speci c PPA ROIs reported by Sengupta et al. 2 that was spatially smoothed by applying a Gaussian kernel with full width at half maximum (FWHM) of 2.0 mm.
In order to test the robustness of our approach, we created overall ve -contrasts for the movie stimulus, and overall eight -contrast for the audio-description (see Table 5). Contrasts di er in contrasted categories based on "how well" averaged events within categories were considered to represent spatial and non-spatial information, and the number of events in the stimulus. Unthresholded -maps of all contrasts on a group level co-registered to the group-template (MNI152 space) can be found at neurovault.org/collections/KADGMGVZ.
In results from the analysis of the audio-description, all contrasts except contrast 7 and 8 yielded signi cant bilateral clusters in anterior regions of the group PPA overlap (see Figure S2). Contrasts of the audio-description also yielded bilaterally signi cant clusters in the ventral precuneus (all contrasts except contrasts 8) and lateral occipital cortex (bilateral in contrasts 1 to 4; unilateral in contrasts 6 and 7). Results across contrasts of both naturalistic stimuli indicate that our ndings regarding the PPA but also RSC and LOC are robust and do not depend on the design of one speci c contrasts. Nevertheless, results are sensitive to the contrasted categories and the amount of available data. For example, contrast 7 and 8 (se_new, se_old > non-spatial categories, and se_new > non-spatial categories) that used the most heterogeneous categories (nouns indicating switches to other settings) and a low amount of events yielded neither a signi cant cluster in the right-hemispheric nor left-hemispheric PPA. Hence, investigators that use model-driven analyses have to consider how many events a naturalistic stimulus may provide and how homogeneous the events to be averaged might be.  Table 5) . a) results as brain slices on top of the MNI152 T1-weighted head template, with the acquisition eld-of-view for the audio-description study highlighted. For comparison depicted as a black outline, the union of the individual PPA localizations reported by Sengupta et al. 2 that was spatially smoothed by applying a Gaussian kernel with full width at half maximum (FWHM) of 2.0 mm. b) results projected onto the reconstructed surface of the MNI152 T1-weighted brain template. After projection, the union of individual PPA localizations was spatially smoothed by a Gaussian kernel with FWHM of 2.0 mm Several contrasts yielded signi cant clusters in superior temporal cortices (bilateral in contrasts 3, 6, 7; unilateral right in contrast 2). Further, contrasts that compare nouns indicating a switch to another setting (categories se_new or se_old) with nouns from non-spatial categories yielded signi cant clusters in primary and secondary auditory cortices (bilateral: contrasts 3, 6, 7; unilateral right: contrast 2). This suggests the presence of a low-level auditory processing confound that was not completely captured by the employed nuisance regressors. The bias of the GLM contrasts towards low-level auditory perceptual processes might be attributed to changes in the soundscape or the start of a new song that accompany the narrator when he is indicating a switch to another setting.
Lastly, contrasts of the audio-description stimulus yielded clusters in regions that were classi ed as statistically signi cant in two or less contrasts. In detail, contrast 2 yielded clusters in the anterior cingulate gyrus (bilateral), right frontal pole, right frontal medial cortex, and right putamen. Contrast 3 shows clusters in the anterior cingulate gyrus (bilateral), left frontal pole, and left insular. Contrast 6 shows one additional cluster in the right medial frontal cortex. Contrast 7 shows clusters in the left frontal pole and left frontal operculum, left insular. Here again, signi cantly increased activation beyond areas known to be involved in processing of spatial information could be attributed to stimulus features of the naturalistic stimulus that were not su ciently controlled leading to a bias in the contrasts.

Control contrasts
For each naturalistic stimulus, we created several control contrasts. To test for the speci city of the main ndings within the movie fMRI data, we created four contrasts that compared hemodynamic activity during frames within movie shots (no_cut category) with movie cut related categories (see Table 5). None of these four contrasts yielded a signi cant cluster, suggesting that the observed di erential hemodynamic response in the PPA for events de ned by particular movie cuts is not observed for other arbitrarily selected time points in the movie.
The correlation of regressors across stimuli ( Figure 4) showed a moderate correlation among cuts to a setting depicted for the rst time (movie vse_new) and nouns indicating a switch to a setting occurring for the rst time (audio-description se_new), and cuts to a recurringly depicted setting (movie vse_old) and nouns indicating a switch a recurring setting (audio-description se_old). Hence, we computed an indentical contrast (se_new > se_old), comparing the audio-description's voice-over narrator nouns for the movie data and the audio-description. While for the movie data the contrast yielded a significant cluster in the right PPA, not in its anterior portion, and bilateral clusters in the superior lateral occipital cortices (the right hemispheric cluster extending into the superior parietal cortex), no signicant di erences where found for the audio-description (unthresholded contrast maps are available on NeuroVault). These results suggest that in the absence of the actual visual stimulation, the narrator's description of scene properties correlated with setting changes did not co-occur with increased hemodynamic activity in areas associated with the processing of spatial information.
Finally for the analysis of the audio-description, we computed control contrasts based on events of the annotated movie cuts (see Table 5) in order to evaluate the in uence of stimulus features correlated with these visual events that remain present in the audio-description, such as changes in the soundscape. Two of ve contrasts yielded signi cant clusters. Contrast 9 (vse_new > pe_old) revealed one signi cant cluster in the left inferior prefrontal cortex (pars triangularis) and right ventral precuneus/posterior cingulate gyrus. Contrast 12 (vse_new > vse_old, vpe_old) revealed one cluster in the right posterior hippocampus. The absence of observable hemodynamic response di erences in the PPA suggests that without the visual information in the movie stimulus, no, or substantially less, processing of stimulus content with respect to spatial information co-occurred with the timing of the movie cuts.