https://orcid.org/0000-0001-8585-9858
Abstract
The hippocampus is largely recognized for its integral contributions to memory processing. By contrast, its role in perceptual processing remains less clear. Hippocampal properties vary along the anterior–posterior (AP) axis. Based on past research suggesting a gradient in the scale of features processed along the AP extent of the hippocampus, the representations have been proposed to vary as a function of granularity along this axis. One way to quantify such granularity is with population receptive field (pRF) size measured during visual processing, which has so far received little attention. In this study, we compare the pRF sizes within the hippocampus to its activation for images of scenes versus faces. We also measure these functional properties in surrounding medial temporal lobe (MTL) structures. Consistent with past research, we find pRFs to be larger in the anterior than in the posterior hippocampus. Critically, our analysis of surrounding MTL regions, the perirhinal cortex, entorhinal cortex, and parahippocampal cortex shows a similar correlation between scene sensitivity and larger pRF size. These findings provide conclusive evidence for a tight relationship between the pRF size and the sensitivity to image content in the hippocampus and adjacent medial temporal cortex.
Introduction
It has recently become clear that visual processing triggers hippocampal activation, wherein visual representations are sourced from either an external visual input or from the visual information constructed from memory. As a terminus for feature-based processing along the visual hierarchy, the structure of visual receptive fields within the hippocampus and its surrounding medial temporal cortical regions is of key importance to understanding their role in visual perception and memory. In this study, we investigated the organization of population receptive fields (pRFs) in the hippocampus and adjacent medial temporal lobe (MTL) regions by using a large set of high resolution retinotopy data acquired at a field strength of 7T (Benson et al. 2018). We focused on the organization of receptive field size along the anterior–posterior (AP) axis of the hippocampus and the surrounding regions that are known to project to the hippocampus: the perirhinal cortex, entorhinal cortex, and the parahippocampal cortex. As the hippocampus is stimulated by visual input and has a strong link specifically to navigation (Hirshhorn et al. 2012), we also explored the organization of pRF parameters with the activation for images of scenes versus faces. As there is evidence that scene versus face stimuli are preferentially processed within areas along the hippocampal long axis, as well as in the perirhinal cortex and in the surrounding MTL structures (Robin et al. 2019), we investigated the activation for images of scenes versus faces in these areas in relation to the organization of pRF size within these areas.
Retinotopy is a well-established characteristic of visual processing, including a contralateral representation in early visual cortical areas (Sereno et al. 1995). Recent work has shown a similar contralateral organization within the hippocampus, specifically in the middle hippocampus (Silson et al. 2021). This result gives more credence to the classification of the hippocampus as a visual processing area. Among the retinotopic mapping parameters of polar angle, eccentricity of the pRF, and size of receptive field, it has been suggested that the receptive field size is related to the higher-order processing capacities in the hippocampus such that larger pRFs are expected to relate to global and coarse-grained representations and smaller pRFs relate to fine-grained details. However, these predictions have not yet been tested using retinotopic mapping in humans (Evensmoen et al. 2013; Robin et al. 2016). Interestingly, this purported organization of the human hippocampus maps onto observed gradations in the receptive field sizes of place cells (place fields) along the dorsoventral axis of the rodent hippocampus. These place cells generally display larger place fields in the ventral hippocampus (homologous to the human anterior hippocampus) than in the dorsal hippocampus (homologous to the human posterior hippocampus). Place cells are organized along the AP axis, analogous to humans, where fine-grained details of higher-order stimuli are processed within posterior hippocampus (Kjelstrup et al. 2008; Keinath et al. 2014; Nyberg et al. 2022), and global representations are found in the anterior hippocampus (Genon et al. 2021; Poppenk et al. 2013; Strange et al. 2014.). The representation of visual space within the hippocampus is also aligned with spatial view cells that map allocentric coordinates, which are found within the hippocampus (Rolls 1999).
Representational differences within the hippocampus are also suggested by differences in the connectivity between the anterior and posterior sections of the hippocampus with MTL cortex and other cortical regions. Coarse granularity within the anterior hippocampus is supported by the functional connectivity of the subfields located in the anterior hippocampus to the perirhinal cortex (Dalton et al. 2019). The anterior hippocampus is responsible for processing abstract representations and is involved in tasks that require scene construction (Libby et al. 2012; Moscovitch et al. 2016; Dalton et al. 2019, 2022). The posterior hippocampus is primarily connected to areas involved in visual processing, such as the retrosplenial cortex and parahippocampal cortex (Zeidman and Maguire 2016; Barnett et al. 2021; Dalton et al. 2022). These areas are all related to scene processing broadly, which can be broken into functional objectives, such as: for the purpose of navigation or for perception (Epstein and Kanwisher 1998; Hassabis et al. 2009).
The level of spatial detail related to the visual specialization found in the hippocampus activations extends to higher-order representations that support coarse-grained visual processing functions such as scene image processing (Silson et al. 2015). Fine-grained feature discrimination in medial temporal regions such as the perirhinal cortex (Bussey and Saksida 2007) and the posterior hippocampus (Woollett et al. 2012) lead us to expect smaller pRFs in these areas. The parahippocampal cortex has also been shown to exhibit larger pRFs than other high-level visual areas; thus, our predictions of larger pRFs in anterior hippocampus provides further evidence for the relationship between perceptual scene processing and relatively larger pRFs. Given the expected relationship between spatial scale of feature preference, or category of specialization, and pRF size in the human visual system, we are interested in whether memory-related areas also show a similar relationship. Particularly, we are interested in the specialization for processing scenes versus faces, that is, larger activation by scenes than faces, as these show the greatest difference in the spatial field coverage in pRF maps of functionally localized areas (Silson et al. 2016).
The current study addresses the relationship between the high-level image category preferences and the size of pRFs between the anterior and posterior hippocampus and the adjacent medial temporal cortex. Specifically, we ask whether the organization of pRF size along the geodesically segmented AP axis of the hippocampus and in adjacent cortical regions is congruent with the specialization for processing image categories, given these regions’ involvement in a variety of various behavioral tasks. As the activation within surrounding medial temporal regions have also been shown to exhibit visual specialization for stimuli such as scenes, objects, and faces across a variety of tasks (Barense et al. 2005; Bussey and Saksida 2007; Murray et al. 2007; O’Neil et al. 2009; O’Neil et al. 2013), we compare the hippocampus, the perirhinal cortex, the parahippocampal cortex, and the entorhinal cortex with respect to pRF size and scene versus face activation.
Methods
Participants
We used data of 181 participants (109 females, 72 males, age = 22–35) from the Human Connectome Project dataset (HCP) (Van Essen et al. 2013), which were scanned at both 7 Tesla and 3 Tesla. Of the 181, we included 157 participants in our subsequent analyses (mean age = 29.5 years, 63 males, 94 females). We excluded the data of 24 participants from the analysis either due to alignment failure or because we were unable to ascertain a robust scenes versus faces contrast.
Hippocampal and MTL segmentation
The T1-weighted anatomical images from the 3 Tesla session were processed with the HippUnfold algorithm (DeKraker et al. 2022). As a result, a Laplace gradient was calculated between the anterior (hippocampal-amygdalar transition area) and posterior (indusium griseum) endpoints of the hippocampus as described in detail in DeKraker et al. (2018). This gradient is geodesically constrained to only hippocampal gray matter and ranges from 0 (most anterior) to 1 (most posterior). This gradient is used, separately in each participant, to define coordinates in a 2D “unfolded” index of points along the hippocampus (Fig. 1A). Subfields are defined according to an unfolded atlas from 3D BigBrain (Amunts et al. 2013; DeKraker et al. 2020). Laplace coordinates that determined the AP axis were binned into 3 discrete equal segments running along the geodesic AP extent of each hippocampus in native space (Fig. 1B). All gradient analyses along the continuous geodesic AP axis within the hippocampus used the native space Laplace coordinates to maintain the subjects-specific curved shape of the hippocampus.
A) The gradient from anterior to posterior of the unfurled hippocampus is depicted in the sagittal image on the left. B) Geodesic parcellation of the hippocampus as well as relative locations of the entorhinal, perirhinal, and parahippocampal cortexes are shown in the middle. C) The right image is a close-up of the hippocampus, showing its division into 3 equal-sized parts as well as the aforementioned cortical areas. D) Average contrast coefficient for scenes versus faces depicted within the anatomical bounds of the ROIs. E) Average size of receptive field depicted within anatomical bounds of the ROIs. F) Quantile plots for distribution of scene versus face coefficient for each of the 7 ROIs. G) Quantile plots for distribution of pRF size across the 7 ROIs. The positive relationship between scene selectivity (F) and size of receptive field (G) along the hippocampal main axis (from posterior to anterior) is indicated by the dotted regression lines. The pattern of smaller receptive fields and less scene specialization is evident in the perirhinal cortex and posterior hippocampus.
Medial temporal neocortical structures, including perirhinal, entorhinal, and parahippocampal cortex, were delineated using automated segmentation (Wisse et al. 2016). We manually segmented each subject’s entorhinal cortex into the anterolateral entorhinal cortex and postero-medial entorhinal cortex using the functional connectivity-derived protocol developed by Yeung et al. (2017), which is based on Maass et al.’s (2015) connectivity findings (Fig. 1B and C).
Category-specific responses in MTL regions
The 3T session, that is part of the Human Connectome Project, includes a working memory task, which is comprised of images of places, faces, tools, and body parts, which are taken from an assortment of studies, which focused on the differences in the size of stimuli per category and range of category types (O’Craven and Kanwisher 2000; Kanwisher 2001; Downing et al. 2006; Peelen and Downing 2007; Park and Chun 2009; Pinsk et al. 2009; Bracci et al. 2010). Whole-brain data were collected with a 2-mm isotropic resolution and a repetition time (TR) of 0.7 s. The 3T scan session also contained a T1-weighted anatomical volume with 0.7 mm isotropic resolution. We aligned the 3T functional data to the 7T data of the same subject by computing the alignment of the 3T T1 scan to the motion-corrected 3T functional data and to the T1w scan from the 7T session. We compiled regressors for faces, places, and objects from the task stimuli files available from the HCP dataset resources. We regressed the activation for the contrast of scenes versus faces as a measure of scene versus face preference (3dDeconvolve in Afni). The maps were subsequently aligned to the 7T scans using an affine transform (3dAllineate in Afni) with the alignment parameters computed from aligning the anatomical scans. The quality of alignment was visually checked.
To assess the quality of the viability of the face, place, and object regressors, we performed localization of the parahippocampal place area and the occipital place area using a (places > faces, objects) linear contrast and the fusiform face area (FFA) using a (faces > places, objects) contrast. ROIs were localized as contiguous clusters of voxels with statistically significant (P > 10−4) contrasts. These functionally localized ROIs are not used in this study, but failure to localize the ROIs (24 of 181 participants) was used as a criterion for excluding participants from the analysis.
Population receptive fields
We used the retinotopy experiment performed at 7 Tesla to measure pRFs (Benson et al. 2018). The experiment consisted of 6 retinotopic runs during which participants were shown moving bars, rotating wedges, and expanding and contracting rings exposing a mélange of object, scene, and face images that are known to stimulate high-level visual cortex (Benson et al. 2018). Whole-brain functional data were collected at a 1.6 mm isotropic resolution with a TR of 1 s. The 7T data set also included a T1-weighted anatomical scan with an isotropic resolution of 0.7 mm, which we used for data alignment with the 3 Tesla scanning session.
We converted the retinotopy data from the surface-based coordinates (for cortical ROIs) and common-space volumetric coordinates (for subcortical ROIs), also known as grayordinate space, into individual subjects’ volumetric voxel space using the Neuropythy toolbox (Benson et al. 2018). We then masked the pRF data with the segmented ROI voxels (see below for ROI localization) within each participant. The pRF of each voxel was computed using the nonlinear regression-based compressive spatial summation (CSS) algorithm (Kay et al. 2013), which computes the activation of a field of the viewing plane that activates a population of neurons that are measured within each voxel of a given ROI. The benefit of using the CSS algorithm is that it fits a gain function to each voxel in order to account for the increasing pRF size on average from the early visual cortex to high-level visual areas. We only included voxels with positive R2 values, that is, for which reasonable pRF fits could be computed.
Anatomical organization of pRFs
Differences in the average size of receptive field and scene versus face contrasts across all ROIs were analyzed with a repeated measures 2-by-7 ANOVA of hemisphere (2 levels) by segment (7 levels).
We assessed the organization of receptive field size and scene versus face contrast coefficients along the AP axis gradient within the hippocampus using a linear mixed-effects model, with participants as the random effect and location within the coordinate space along the unfolded AP axis as a fixed effect to predict either size of the pRFs within the hippocampus or the contrast coefficient for scenes versus faces.
Results
Category selectivity
Our aim was to first compare visual specialization by analyzing the coefficients from the regression analysis of neural activation for images of scenes contrasted with activation elicited by face images. To investigate the category preferences for scene versus face images within the hippocampus and in adjacent medial temporal cortex, we used a linear contrast between these types of stimuli to account for differences in the absolute signal strength between and within regions of interest at the level of individual voxels across all participants.
The contrast showed a greater activation for scenes than faces in the hippocampus as a whole within the right hemisphere, but not the left hemisphere (left hemisphere m = −0.026, right hemisphere m = 0.047). We analyzed the contrast coefficient for scenes versus faces with a 7 × 2 rmANOVA for each of the 7 ROIs and found a significant main effect for ROI segments, F(4.13, 644.71) = 299.1, P < 0.001, and not for hemisphere, F(1,156) = 2.62, P = 0.11 (Fig. 1D and F). There was a significant interaction between the segment and hemisphere F(3.03, 472.6) = 12.37, P < 0.001, within the hippocampus and surrounding regions: anterolateral entorhinal, and postero-medial entorhinal, parahippocampal, and perirhinal cortexes. Pairwise tests with Bonferroni correction for multiple comparisons showed that the contrast coefficient values were significantly greater in the postero-medial entorhinal cortex than the posterior hippocampus (P < 0.001, t(156) = 3.65 and 3.81 in the left and right hemispheres, respectively), and the contrast is greater for the anterior hippocampus than the anterolateral entorhinal cortex in the right hemisphere (t(156) = 5.95, P < 0.001), but we found no significant difference in the left hemisphere (t(156) = 1.34, P = 0.18). The perirhinal cortex showed less scene activation than the postero-medial entorhinal cortex (P < 0.001, t(156) = −5.95 and −6.90 in the left and right hemispheres, respectively) and less than the middle hippocampus (P < 0.001, t(156) = −4.66 and t(156) = −4.31, in the left and right hemispheres, respectively). Not surprisingly, the parahippocampal cortex showed greater scene activation than all the other ROIs (P < 0.001) (Fig. 1D and F). The selectivity of the parahippocampal cortex for scenes is well documented (Epstein and Kanwisher 1998), and it is significantly stronger than for the anterolateral entorhinal cortex, postero-medial entorhinal cortex, perirhinal cortex, and even the hippocampus.
Since the magnitude of the contrast appeared to vary systematically over the length of the hippocampus, we measured the gradient of the scene versus face contrast coefficient activation along the AP axis of the unfurled hippocampus. We found that the scene advantage is greater in anterior than the posterior hippocampus (Fig. 1D). In fact, a mixed-effects linear regression of the face–scene contrast within each hippocampus voxel as a function of the geodesically constrained AP gradient (fixed effect) and participant (random effect) showed a significant negative slope within the left hemisphere (beta = −0.035, sem = 0.001, and t(1191231) = −34.86, P < 0.001) as well as the right hemisphere (beta = −0.022, sem = 0.001, t(1192409) = −21.16, P < 0.001). The contrast coefficient of scenes versus faces is significant within each hemisphere, that is, the specialization for scene processing in the hippocampus is organized in the form of a gradient along the AP axis, with greater scene specialization in anterior areas (Fig. 2A).
A) The scenes versus faces contrast clearly decreases from anterior to posterior hippocampus in both the left and the right hemispheres. This slope is significant (P < 0.001). B) The pRF size also varies linearly from the anterior to posterior hippocampus in both hemispheres (P < 0.001), with larger receptive field size in the anterior hippocampus. C) The size of pRF is significantly related to the contrast coefficient for scenes versus faces (P < 0.001 bilaterally).
Size of pRFs
We compared the size of pRFs in the medial temporal cortex regions surrounding hippocampus, which are known to have direct inputs to the hippocampus: the entorhinal cortex, the parahippocampal cortex, and the perirhinal cortex. Our analysis of pRF size (sigma of the fitted 2D Gaussian) via a 7 × 2 rmANOVA yielded a significant main effect for segments, F(4.28, 667.82) = 26.29, P < 0.001, but not for hemisphere, F(1, 156) = 0.48, P = 0.49 (after Greenhouse–Geisser correction for sphericity). There was a significant interaction between the segment and hemisphere F(4.96, 774.22) = 6.32 P < 0.001 in the hippocampus and surrounding ROIs. Post hoc analyses, which were conducted using the Bonferroni method to correct for multiple comparisons, showed that the main effect for pRF size was different between the parahippocampal cortex and all other segments bilaterally, P < 0.001, except for the left anterior and middle hippocampus, where the pRF size did not show a significant difference with that of the parahippocampal cortex (Fig. 1E and G). Within the hippocampus, voxels in the anterior and middle segments had significantly larger receptive fields than those found in the posterior hippocampus regions (P < 0.001), bilaterally. The middle hippocampus and anterior hippocampus each had larger receptive fields than the perirhinal cortex, within the left hemisphere (t(156) = 3.38, and t(156) = 3.37, P < 0.001 for both segments, respectively), but not in the right hemisphere (t(156) = 2.15, P = 0.0033, t(156) = 2.91, P = 0.0042, respectively) after Bonferroni correction. The anterior hippocampus also showed larger pRFs than the postero-medial entorhinal cortex and anterolateral entorhinal cortex averages, within the right hemisphere (P < 0.001, for both segments, t(156) = 3.87, t(156) = 3.46, respectively). Within the left hemisphere, only the postero-medial segment of the entorhinal cortex showed significantly smaller pRFs than the anterior hippocampus: (t(156) = −4.16, P < 0.001). The pRF sizes within the left anterolateral entorhinal cortex were not significantly different from the pRFs in the anterior hippocampus after Bonferroni correction (t(156) = −3.1, P = 0.0023).
We measured the average size of pRFs for each voxel within the hippocampus as well as adjacent cortical regions (Fig. 1E). Within the hippocampus, we find a clear gradient along the AP axis with significantly larger pRFs in the anterior regions than in the posterior hippocampal regions (Fig. 2B): Within the left hemisphere, the mixed-effects regression with AP axis location as a fixed effect and participants as a random effect showed a significant negative relationship beta = −0.768, sem = 0.0092, t(1191242) = −83.13, P < 0.001. The right hemisphere showed a similar significant negative relationship, beta = −0.81, sem = 0.0094, t(1192424) = −85.58, P < 0.001.
Relationship between category selectivity and pRF size
The larger pRFs in the anterior compared to the posterior hippocampus align well with the stronger preferences for scenes in the anterior compared to the posterior hippocampus. We tested this relationship formally with a mixed-effects regression model with contrast coefficient as the predicted variable, pRF size of hippocampus voxels as a fixed effect, and participants as a random effect. We found a significant relationship with pRF size in the left hippocampus, beta = 0.025, sem = 0.0001, t(1191340) = 25.22, P < 0.001, as well as the right hippocampus, beta = 0.001, sem = 0.0001, t(1192491) = 11.01, P < 0.001. That is, scenes versus faces selectivity is strongly related to the pRF size in both hemispheres (Fig. 2C).
The contrast coefficient differences for scenes versus faces across the medial temporal ROIs show a similar increase of the average size of receptive fields across hippocampus segments, where generally larger pRFs showed a greater activation for scenes. To test this relationship explicitly, we performed the same mixed-effects regression analysis as above, separately for the voxels within each of the cortical ROIs, with participants as the random factor and pRF size as the independent variable. Results showed that the scene versus face contrast coefficients are significantly related to size of receptive field, bilaterally, within parahippocampal cortex (left: beta = 0.52, sem = 0.0098, t(435479.9) = 52.42, P < 0.001; right: beta = 0.753, sem = 0.0093, t(432548) = 81.45), P < 0.001. Within the right perirhinal cortex, we found a significant relationship of larger receptive field size with greater scene activation (beta = 0.13, sem = 0.0051, t(1171611) = 24.86, P < 0.001), and within the left perirhinal cortex, a negative relationship between pRF size and scene versus face contrast coefficient (beta = −0.079, sem = 0.0052, t(1201247) = −15.05, P < 0.001). There was no consistent pattern of relationship between the pRF size and scene selectivity within the postero-medial entorhinal cortex (left hemisphere was nonsignificant, beta = −0.021, sem = 0.0161, t(111459.5) = −1.25, P = 0.21; right hemisphere: beta = 0.16, sem = 0.017, t(95035.31) = 9.3, P < 0.0015). The relationship within anterolateral entorhinal cortex was also inconsistent across the hemispheres (left:, beta = 0.17, sem = 0.013, t(121161.2) = 12.50, P < 0.001; right: beta = −0.087, sem = 0.012, t(128125.3) = −7.07, P < 0.001).
Eccentricity
When we compared the eccentricity across all ROIs via a repeated measures ANOVA, we found a main effect for segment, F(4.98, 776.52) = 52.77, P < 0.001, and for hemisphere, F(1,156) = 6.23, P = 0.014. There was no significant interaction between the segment and hemisphere, F(4.96, 773.48) = 1.37, P = 0.23 (see Supplementary Figs. 1 and 2). Post hoc tests conducted within segments of the hippocampus showed that the average eccentricity within the posterior and middle segments was larger than the anterior segments, P = 0.0099, and P < 0.001, respectively, averaged across the hemispheres per segment. These findings suggest that the relation between the pRF size and eccentricity is not consistent across the AP axis of the hippocampus.
Variance explained
The variance explained by the pRF analysis differs along the AP axis. It increases from posterior to anterior in both hemispheres (left hemisphere: beta = −0.0014, sem = 0.000017, t(1191234) = −85.10, P < 0.001; right hemisphere: beta = −0.019, sem = 0.000016, t(1192417) = −120.41, P < 0.001). However, poor quality fits are related to large estimates of pRF size. For this reason, we repeated the analyses with R2 as a fixed-effects covariate. Even with R2 regressed out, we still find a significant effect of pRF size along the AP axis in both hemispheres (left hemisphere: AP axis beta = −0.62, sem = 0.0091, t(1191265) = −68.56, P < 0.001, right hemisphere: beta: −0.58, sem = 0.0093, t(1192471) = −83.13, P < 0.001). The relationship between the scene/face contrast versus pRF size when regressing out R2 as a covariate in the right hemisphere is, beta = 0.00197, sem = 0.00010, t(1192510) = 19.346, P < 0.001, and in the left hemisphere: beta = 0 0.00051, sem = 0.00052, t(1191360) = 5.092, P < 0.001.
Discussion
Our findings generally confirmed our prediction that pRF size would vary along the longitudinal axis of the hippocampus, decreasing from anterior to posterior. We also found the pRF size to be related to the specialization for processing higher-order stimuli, with lower scene preference in the posterior hippocampus with its relatively small pRFs and higher scene preference in the anterior hippocampus with its larger pRFs (Fig. 2). Larger pRFs in the anterior hippocampus complement research, showing that the scale of feature-based processing varies along the AP axis, with more fine-grained, feature-based processing occurring in the posterior and with more schematic, or coarser, processing occurring in the anterior hippocampus (Poppenk et al. 2013; Brunec et al. 2018).
By and large, we found similar relations between the sizes of pRFs and specialization of stimulus processing in the MTL regions surrounding the hippocampus. The scene-selective parahippocampal cortex showed significantly larger pRFs than the less scene-selective areas, namely: the perirhinal cortex and the entorhinal cortex (Arcaro et al. 2009).
Specialization for perceptual processing of scene versus face images in the high-level visual cortex is a well-documented phenomenon. The hippocampus is integral to memory creation and retention and is not often characterized as a visual perception region. Here, we observed a similar relationship for large receptive fields and scene processing within the anterior hippocampus as we did in the parahippocampal cortex. Our results showed differential activation to be greater for scenes in the anterior hippocampus than it was in the posterior hippocampus (Fig. 1D), which supports the theory that the hippocampus is involved in perceptual processing of high-level visual features. Our findings also support the hypothesis that larger, more global features, and schematic representations activate the anterior hippocampus during visual processing (Audrain and McAndrews 2022; Farzanfar et al. 2023), which is consistent with the view that the anterior hippocampus specializes in scene reconstruction (Dalton and Maguire 2017).
Category preference in perirhinal cortex aligns with lesion studies specifically pointing toward its importance in object recognition (Barense et al. 2005). Previous fMRI research on perirhinal cortex showed a greater activation for faces than for scenes, where visual representations are products of complex visual features (Bussey and Saksida 2007; Murray et al. 2007; O’Neil et al. 2009; O’Neil et al. 2013). Our results suggest lateralization within the perirhinal cortex, where the relationship between the pRF size and scene activation was found within the right but not the left hemisphere. This finding is interesting, considering the role of the right perirhinal cortex of perceptual discrimination (Inhoff et al. 2019). The preferential activation of the perirhinal cortex for faces is consistent with the connectivity between perirhinal cortex and FFA (O’Neil et al. 2014; Hodgetts et al. 2015). These areas function as part of a larger face processing network that is separate from the posterior network that includes the parahippocampal cortex (Ranganath and Ritchie 2012). We also note that the cognitive network in the temporal lobe supports detail sharpening as a function of the perirhinal cortex. The network is hypothesized to support the spatial context-based details that are processed within the parahippocampal cortex (Ranganath and Ritchie 2012).
The anterolateral entorhinal cortex showed greater face activation than the postero-medial entorhinal cortex. This finding is consistent with literature showing a greater connectivity of perirhinal cortex to anterolateral than postero-medial entorhinal cortex, especially in the right hemisphere (Maass et al. 2015; Ferko et al. 2022). The fine-grained processing requirements for face recognition align with the smaller average pRF size in perirhinal cortex as well as in the anterolateral entorhinal cortex, relative to the larger pRFs found in the scene-specialized anterior hippocampus.
In contrast to the hippocampus, the postero-medial entorhinal cortex did not show a particularly strong preference for faces or scenes, nor was the size of receptive fields in this region related to perceptual specialization for scenes or faces. This deviation from the intuitive relationship of the pRF size and feature specialization could be explained by the spatial integration that is thought to occur in postero-medial entorhinal cortex (Avidan and Behrmann 2021). It is a region that is less involved in processing single features, but it is involved in combining high-level feature maps within the MTL regions for organization of location information in physical space.
We expected to find larger receptive fields in parahippocampal cortex, given past literature showing that large pRFs are a characteristic of perceptual scene-processing areas (Arcaro et al. 2009). The large pRF size supports the theory that scene images tend to be processed more globally (Oliva and Torralba 2006). We also expected to find larger pRFs in the anterior hippocampus, given that it is specialized for processing spatial schemas. This finding provides an insight into information encoding within the anterior hippocampus, given that it is typically not considered to be a perceptual processing area due to its role in processing internal recoveries from memory rather than online processing of stimuli in the environment (Poppenk et al. 2013; Steel et al. 2021).
Our findings revealed a close relationship between pRFs with visual activation, suggesting that this relationship is linked more strongly to perceptual processing. Although contralateral hemifield retinotopy, which is a characteristic of perceptual processing areas, has been found in the hippocampus (Silson et al. 2021), those results were inconsistent across all 3 segments of the hippocampus. This inconsistency in finding a hemifield bias could reflect a weaker involvement of the hippocampus in the processing of visual features. Specialization within the hippocampus for the visual perception of scenes versus faces was evident in a systematic review that showed an inconsistent activation between the anterior and posterior parts of the hippocampus (Robin et al. 2019). Our results of greater scene image activation within anterior and middle hippocampus align with most studies that used the scenes versus faces contrast, as mentioned in Robin et al. (2019). The discrepancy across studies may be specific to the one-back task that required greater memory recall instead of the detailed spatial navigation that has been shown to recruit the posterior hippocampus (Woollett and Maguire 2011; Hirshhorn et al. 2012).
The contrast of scenes versus faces does not answer questions about other types of features. The utility of the scenes versus faces contrast is its composition of perceptual features that are also used for functionally localizing high-level visual areas within visual cortex. As such, the activation patterns that we found within the medial temporal ROIs and the hippocampus provide a comparison between visual specialization and the pRF size. This relationship can be employed to investigate further questions about visual representations in memory-related areas. Though we can look at the activation for category-level specialization within the hippocampus and other memory-related ROIs, we cannot make claims from our results about the low-level features, such as specialization for spatial frequencies.
The results from the pRF analyses also suggest that the middle and anterior segments of the hippocampus have different organization, given that the middle segment contains pRFs with larger “eccentricity” than the anterior segment, but the “size” of pRFs is statistically similar. These findings suggest that differences between the anterior and middle parts of the hippocampus show different pRF organizations than the anterior or the posterior areas when taking the relationship between pRF size and eccentricity into account. Thus, the relationship between size and eccentricity that is found within the early and high-level visual areas (Kay et al. 2013) does not necessarily generalize to the visual tuning properties across the 3 equal parts of the hippocampus.
The model fits we utilized were derived from the CSS algorithm (Kay et al. 2013), which produces robust eccentricity and angle parameter fits, but it is known to produce large variance with respect to pRF size. These effects can be mitigated through use of suprathresholding of the data (Lage-Castellanos et al. 2020). Our dataset was large, allowing us to include many data points across the 157 subjects even though we restricted all model fits to R2 > 0. The analysis by Lage-Castellanos et al. (2020) demonstrated that small pRF sizes <0.91°, such as those within early visual cortex (Benson and Winawer 2018), were most susceptible to the effects of skewness. Our data show that the median value of pRF size was above the threshold for all areas, which would mitigate any significant biasing of pRF size in our results.
Surprisingly, our results of larger pRFs in the anterior hippocampus seem to conflict with the retinotopic connectivity of V1 with the anterior hippocampus during movie watching (Knapen 2021), given the prevalence of smaller pRFs in V1. Stronger activation of the anterior hippocampus and, as a consequence, stronger functional connectivity during movie viewing than for resting state could be explained by the movies’ story narrative, which requires the representation of longer timeframes. Given that the anterior hippocampus supports longer spans of time as shown by Brunec et al. (2018) and Bouffard et al. (2023), the temporal organization could be another explanation for finding larger pRFs in the anterior hippocampus. This interpretation would lead to important questions for future research about how the temporal scale interacts with spatial scale in the hippocampus.
Conclusions
Visual specialization in the hippocampus is known to vary along the AP axis, where anterior areas have a global bias, and posterior areas are more involved in fine-grained processing. Here, we found that visual specialization of parts of the hippocampus tends to mirror the size of pRFs, where greater specialization of the anterior hippocampus for scenes was associated with larger pRFs and less specialization of the posterior hippocampus for scenes was associated with smaller pRFs.
This organization of pRF size along the AP axis combined with the specialization for particular image content within areas of the hippocampus overlap with mnemonic processing of visual information. Our findings suggest that pRF representations in the hippocampus and surrounding medial temporal regions are indicative of high-level complex representations that contribute to memory and perception.
Acknowledgments
We would like to thank to Kayla Ferko and Alexander Minos for the manual segmentation work on the entorhinal cortex and to Claudia Damiano for help with processing the 3T data for the scene-face contrast.
CRediT author statement
Charlotte A. Leferink (Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Software, Validation, Visualization, Writing—original draft, Writing—review & editing), Jordan DeKraker (Data curation, Investigation, Methodology, Software, Visualization, Writing—review & editing), Iva Brunec (Conceptualization, Methodology, Writing—review & editing), Stefan Köhler (Conceptualization, Methodology, Supervision, Writing—review & editing), Morris Moscovitch (Conceptualization, Methodology, Supervision, Writing—review & editing), and Dirk B. Walther (Conceptualization, Funding acquisition, Investigation, Methodology, Project administration, Supervision, Validation, Visualization, Writing—original draft, Writing—review & editing)
Funding
This research was supported by Natural Sciences and Engineering Research Council (NSERC) Discovery Grants RGPIN-2020-04097 to DBW, RGPIN 2018-05770 to SK. CAL was funded by an NSERC Postgraduate Scholarship-Doctoral (PGS D). JD is funded by an NSERC Post-Doctoral Fellowship (NSERC-PDF), and MM by NSERC Grant A8347.
Conflict of interest statement: None declared.
Data availability
The code and data supporting the findings of this study are available from OSF (https://osf.io/j98kc/).
