Parallel systems for social and spatial cognition reaching the cortical apex

Significance

A long-standing debate in cognitive science and neuroscience concerns whether mechanisms for information processing in the human brain are domain-specific—comprising multiple systems specialized to process certain types of information, such as faces or information about other people—or domain-general, operating on diverse inputs. We address this question using fMRI during a range of behavioral tasks—viewing images, rating traits, or imagining events—involving either people or places. We find that processing social and spatial information elicits neural responses in distinct brain areas, regardless of task. Areas responsive to people and places formed two separate systems with a parallel organization across distributed brain areas. These two systems may each be involved in building internal relational models within distinct content domains.

Abstract

How are systems supporting high-level cognition organized in the human brain? We hypothesize that cognitive processes involved in understanding people and places are implemented by distinct neural systems with parallel anatomical organization. We test this hypothesis using precision neuroimaging of individual human brains on diverse tasks involving perception and cognition in the domains of familiar people, places, and objects. We find that thinking about people and places elicits responses in distinct areas of high-level association cortex within the default mode network, spanning the frontal, parietal, and temporal lobes. Person- and place-preferring brain regions are systematically spatially adjacent across cortical zones. These areas have strongly domain-specific response profiles across visual, semantic, and episodic tasks and are specifically functionally connected to other parts of association cortex with like domain preference. Social and spatial networks remain anatomically separated at the apex of a unimodal-to-transmodal gradient across cortex and include regions with anatomical connections to the hippocampal formation. These results demonstrate the existence of parallel, domain-specific networks reaching the cortical apex.

How are systems for high-level cognition structured in the human mind and brain? Cognitive scientists have long debated whether these systems are domain-general, acting on different types of information, or domain-specific, comprising multiple components specialized for understanding particular classes of input. Proponents of domain-general theories have argued that generic mechanisms of associative learning can parsimoniously account for a wide range of high-level cognitive abilities (1, 2). By contrast, proponents of domain-specific accounts argue that the human mind comprises a collection of separate systems that evolved to process specific classes of ecologically relevant input, such as people, objects, and places (3–5).

In the cognitive neuroscience literature, a related debate surrounds the function of the default mode network (DMN). The DMN constitutes a broad, distributed zone of cortex including regions in medial prefrontal cortex (MPFC), medial parietal cortex (MPC), temporo–parietal junction (TPJ), superior temporal sulcus (STS), and superior frontal gyrus (SFG). The DMN is anatomically positioned at the apex of an organizing gradient in cortex separating unimodal sensorimotor areas from transmodal association cortex (6, 7). While initially described as deactivating to cognitive tasks involving focused attention to external stimuli (8, 9), the DMN has since been implicated in a range of cognitive processes involving long-term memory and social cognition (10–13). These anatomical and functional properties make the DMN well-positioned to support high-level cognition in humans.

Despite widespread recognition of its importance, views on the structure and function of the DMN vary widely and remain debated. Early work identified a diverse set of tasks that engaged regions within the DMN, including theory of mind (ToM; reasoning about others’ mental states), episodic projection (recalling memories of past events, or imagining future events), and spatial navigation (12, 14). This evidence was used to argue for a common cognitive process underlying these distinct tasks, such as the mental construction of events or scenes (12, 15). Studies directly comparing responses to multiple tasks in the same participants came to a more nuanced conclusion, arguing that the DMN comprises separate but interacting subsystems: one specialized for social cognition, one for episodic projection, and a “core network” involved in both (16–19).

However, this line of research had a key methodological limitation: Evidence for common neural responses to different tasks and process came from group analysis of fMRI data, in which data from multiple individuals are combined in a standardized coordinate space. Because the precise organization of functional responses varies across individuals, group analysis may introduce spurious overlap that does not exist in the individual (20, 21). Studies measuring responses of brain areas defined functionally within individuals argued instead for regions that were selective for specific cognitive processes such as ToM (11, 22), but did not include mnemonic or episodic tasks.

More recently, substantial progress has come from the study of functional network organization within individual brains. This work has argued that the DMN does not in fact constitute a single, coherent functional system, but instead two separate networks comprising spatially interdigitated regions within a parallel large-scale organization (23–26), termed default network A (DN-A) and default network B (DN-B). These networks have also been found to be distinct from a third, spatially adjacent system sensitive to linguistic content (27, 28).

The presence of two separate networks within the DMN invites a reevaluation of prior work on its functional role. How should we understand the contribution of DN-A and DN-B to high-level cognition? Recent work has found that ToM tasks specifically engage DN-B, while episodic projection tasks specifically engage DN-A (29), although conflicting evidence has also been reported (30). This result has been used to argue for a cognitive division of labor between DN-A and DN-B similar to prior theoretical perspectives on DMN subsystems (16): that each network supports distinct cognitive processes, with DN-A specialized for mnemonic and episodic functions, and DN-B for social cognition (29, 31). This argument is supported by the finding that unlike DN-B, DN-A includes a component in parahippocampal cortex (PHC), and has functional connectivity with the hippocampal formation, indicating a close relationship with the medial temporal lobe long-term memory system (24, 32, 33).

An alternative hypothesis is that rather than supporting distinct cognitive processes, the dissociation between DN-A and DN-B can be explained in terms of the type of information they operate on, with DN-A supporting spatial cognition and DN-B supporting social cognition. Consistent with this view, an item-wise analysis of responses in an episodic projection task found that DN-A responses are predominantly driven by the demand to mentally construct a scene (34), leading subsequent work to describe the function of DN-A as “spatial/episodic” (35, 36). Furthermore, distinct areas within MPC respond to tasks involving faces or people and scenes or places, with an organization that appears similar to the interdigitated structure of DN-A and B (37–39). This content-based, rather than process-based, account of the functional dissociation between DN-A and B opens the possibility that DN-B may also be involved in mnemonic and episodic processes, but with a specialization for social content. While DN-B does not contain a component in PHC, preliminary evidence indicates that it does contain a component in the temporal pole (TP; 24, 40, 41), which also has a direct anatomical connection with the hippocampal formation in primates (42, 43).

In the current work, we rigorously evaluate the content-based account of DMN function. While prior work has studied DMN responses using a small number of task conditions, we take the approach of using a large number of tasks and conditions in the same set of participants to provide strong constraints on function while avoiding the group analytic approach required of meta-analyses (44). We have previously reported on functional responses within TP in this dataset (41) but here apply analyses targeting the DMN. We hypothesize the presence of distinct but parallel systems involved in processing information about people (social content) and places (spatial content). In contrast to process-based accounts, we propose that each system contributes to multiple cognitive processes—including understanding events in the present, retrieving information stored in long-term memory, and simulating future events—applied to information within its preferred content domain. Our results suggest that domain-specificity is a widespread organizing principle of cortical function reaching the cortical apex, and support a content-based account of DMN function.

Results

We evaluate DMN function with a human fMRI experiment, using precision imaging of individual human brains across a range of tasks and conditions. To maximize the strength of memory-related responses, we use tasks involving closely familiar people and places for each participant. To determine whether responses are modulated by content domain consistently across specific task demands, we use tasks eliciting visual, semantic, and episodic processing. Tasks included visual perception (viewing images of familiar and unfamiliar faces and scenes, and generic objects); semantic judgment (answering questions about personality traits of familiar people, spatial properties of familiar places, and physical properties of generic objects); and episodic simulation (imagining familiar people talking about common conversation topics, navigating through familiar places, and physical interactions of generic objects; Fig. 1A and SI Appendix, Fig. S1). We also performed localizer scans for ToM, language comprehension, and dynamic visual perception. Optimized data acquisition and preprocessing methods yielded data with high temporal signal to noise ratio (tSNR, mean 158) despite minimal spatial smoothing, including in regions of signal loss such as the anterior medial temporal lobes (SI Appendix, Fig. S2).

Fig. 1.

Parallel, distributed responses to people and places across high-level association cortex. (A) Schematic of conditions from visual, semantic, and episodic tasks. (B) Whole-brain general linear model-based responses to people versus places, from one representative participant (semantic task, thresholded at a false discovery rate of q < 0.01 to correct for multiple comparisons across coordinates). (C) Responses across each task within medial prefrontal cortex, (D) medial parietal cortex, (E) superior frontal gyrus, (F) temporo–parietal junction, (G) superior temporal sulcus, (H) parahippocampal cortex, and (I) temporal pole. (J) Cingulate sulcus paths used to compute spatial autocorrelation. (K) Spatial autocorrelation of the contrast between people and places along the cingulate sulcus within medial prefrontal cortex from a representative individual participant. (L) Spatial autocorrelation within medial parietal cortex.

We first asked whether social and spatial cognition engage distinct regions within zones of cortex associated with the DMN. Comparing responses to people and places, we observed preferential responses for both domains among multiple zones distributed across association cortex, including MPFC, MPC, TPJ, and SFG, bilaterally (whole-brain general linear model-based analysis, corrected for temporal autocorrelation using prewhitening with an ARMA(1,1) model, and corrected for multiple comparisons across coordinates using a false discovery rate of q < 0.01; Fig. 1 B–F and SI Appendix, Figs. S6–19). Responses to people were additionally observed within middle and anterior regions of the STS. Importantly, a common pattern of preferential responses to people and places was observed across visual, semantic, and episodic tasks, despite substantial differences in the physical nature of the stimuli (images and words) and cognitive demands of the tasks. These preferences thus cannot be explained by a confound specific to any one task and are more parsimoniously explained as effects of content domain.

We next asked whether preferential responses to places and people are not simply segregated, but organized in parallel. Across multiple zones of association cortex, we found that regions responsive to people and places were systematically yoked, with adjacent or alternating parts of cortex showing opposing stimulus preferences. In the TPJ, a pair of regions was typically observed, with a person-preferring area just anterior to a place-preferring area (Fig. 1F). In MPFC and MPC, a series of interdigitated responses to people and places was observed along a curved axis following the cingulate sulcus, with 3 to 5 (MPFC) or 2 to 3 (MPC) pairs of regions in individual participants (Fig. 1 C and D). To formally analyze this pattern of spatially adjacent responses, we computed the spatial autocorrelation in the response to people versus places along paths on the cortical surface traversing the cingulate sulcus within MPFC and MPC (Fig. 1J and SI Appendix, Figs. S20–23). Observing a negative spatial autocorrelation would indicate that neighboring regions of cortex tend to have distinct domain preferences. For both MPFC and MPC bilaterally, we observed a negative spatial autocorrelation that peaked between 15 and 25 mm (Fig. 1 K and L). In every participant, hemisphere, and region (MPFC and MPC), this negative autocorrelation surpassed a statistical threshold established by a permutation test (P < 0.05; SI Appendix, Figs. S20–S23). These results demonstrate that responses to people and places in association cortex are distinct but spatially yoked.

The DMN is anatomically positioned at the apex of a unimodal-to-transmodal gradient, near in connection distance to the hippocampus and limbic system (7). We thus wondered whether person and place preferences exist in zones of cortex known to have anatomical connections with the hippocampal formation. We first tested for the presence of place responses in PHC and retrosplenial cortex (RSC), areas with bidirectional connections to the hippocampus via entorhinal cortex and the subicular complex (42, 45, 46), implicated in spatial cognition in rodents and primates (47, 48). PHC was hand-drawn on individual anatomical images based on anatomical landmarks related to cytoarchitectonic boundaries (49, 50), while RSC was defined using the Human Connectome Project’s multimodal cortical surface atlas (51), given the lack of established gross anatomical landmarks for this region in humans. Whole-brain analyses showed place responses in PHC and RSC were consistent across tasks and participants (left hemisphere, 60/60 comparisons; right hemisphere, 60/60 comparisons; Fig. 1H and SI Appendix, Figs. S8, S9, S16, and S17). We then tested for the presence of person responses in the TP. While our prior work identified reliable TP responses to visually presented faces in whole-brain analyses (41), the current analysis extends this result to semantic and episodic tasks. We observed preferential responses within TP to people over places consistently across tasks and participants (left hemisphere, 28/30 comparisons; right hemisphere, 28/30 comparisons; Fig. 1I and SI Appendix, Figs. S18 and S19). These results demonstrate the presence of person and place preferences within parts of cortex with anatomical connections to the hippocampal formation in primates, identifying potential relay areas between domain-preferring regions of association cortex and the hippocampus.

Our results thus far have demonstrated that regions with preferential responses to people and places have a parallel, distributed organization across association cortex. To what extent do these regions specifically process information from their preferred content domain? To address this question, we assessed the magnitude of responses across content domains and tasks. Functional regions-of-interest (ROIs) were defined as the top 5% of maximally person- or place-preferring coordinates (semantic task) within anatomical search spaces covering zones of the DMN: MPC, MPFC, TPJ, STS, SFG, TP, and PHC (Fig. 2 A and B). We then extracted response magnitudes from independent data, across all task conditions.

Fig. 2.

Region-of-interest (ROI)-based analysis reveals domain-specific responses across tasks. (A) ROIs were defined as the top 5% of person- or place-preferring coordinates within anatomical search spaces. (B) Search spaces and example functional ROIs from one representative participant. (C) Responses (% signal change) extracted from functionally defined ROIs, across all task conditions. Error bars show SE across runs. *P < 0.05/7 = 0.0071, **P < 10⁻³, ***P < 10⁻⁴ (linear mixed model across runs, with participant included as random effect). Abbreviations: MPC, medial parietal cortex; MPFC, medial prefrontal cortex; TPJ, temporo–parietal junction; STS, superior temporal sulcus; SFG, superior frontal gyrus; TP, temporal pole; PHC, parahippocampal cortex.

This ROI analysis found that responses in functionally defined regions were strongly selective for their preferred stimulus domain (Fig. 2C). Across ROIs, responses to nonpreferred stimulus categories were typically at or below baseline. Person-preferring areas in MPC, MPFC, TPJ, STS, SFG, and TP responded significantly more strongly to seeing images of familiar people versus objects and familiar scenes; making judgments about familiar people versus objects and familiar places; and imagining events involving familiar people versus objects and familiar places (all 36 P’s < 0.0071 = 0.05/7, applying Bonferroni correction across the full set of 7 ROIs; individual values in SI Appendix, Table S3; linear mixed effects model across runs, with participant included as a random effect). Similarly, place-preferring areas in MPC, MPFC, TPJ, STS, SFG, and PHC responded significantly more strongly to seeing images of familiar scenes versus objects and familiar people; making judgments about familiar places versus objects and familiar people; and imagining events involved familiar places versus objects and familiar people (all 36 P’s < 0.0071; individual values in SI Appendix, Table S4). Strong domain preferences not only observed for the focal, maximally responsive areas used in this analysis, but across a range of ROI sizes, from the top 5 to 40% of person- or place-preferring coordinates (SI Appendix, Fig. S24). Other regions of association cortex responded selectively to object or language conditions (SI Appendix, Figs. S25 and S26 and Tables S5 and S6), arguing against an explanation of person or place responses in terms of generic factors like attention or task engagement. These results demonstrate domain-specificity as an organizing principle of association cortex, arguing for distinct systems for social and spatial cognition.

Given its anatomical positioning near the medial temporal lobe, the DMN is well positioned to contribute to long-term memory. How do person- and place-preferring regions respond when processing familiar and unfamiliar entities? Person-preferring areas in MPC, MPFC, TPJ, STS, SFG, and TP responded significantly more strongly to familiar versus unfamiliar face images (all 6 P’s < 0.0071; SI Appendix, Table S3). Similarly, place-preferring areas in MPC, MPFC, TPJ, STS, SFG, and PHC responded significantly more strongly to familiar versus unfamiliar scene images (all 6 P’s < 0.0071). These complementary familiarity effects suggest distinct neural systems supporting long-term memory for social and spatial information.

How do the familiar person responses described here relate to well-established patterns of response to ToM tasks? Some theoretical perspectives have argued that others’ mental states are processed by specialized cognitive and neural mechanisms, distinct from other aspects of person understanding such as personality traits, group membership, relationships, and other semantic information about people (3, 22, 52). Other perspectives have suggested that other types of information about people are also processed and represented within the DMN (53–55). However, prior studies have not directly compared familiar person and ToM responses within individual brains.

To address this question, we asked whether brain regions responsive to familiar person tasks also respond when participants reason about the mental states of unfamiliar characters in a story. Person-preferring areas in MPC, MPFC, TPJ, STS, and SFG responded significantly more strongly during reasoning about false beliefs relative to false “photographs” or physical representations (all 6 P’s < 0.0004 = 0.05/12, applying Bonferroni correction across the full set of person- and place-preferring ROIs; Fig. 2C and SI Appendix, Table S3). Person-preferring TP also showed a preferential response to the false belief condition, although this effect did not survive multiple comparison correction (P = 0.0013). Conversely, ROIs in MPC, MPFC, TPJ, STS, SFG, and TP defined by the ToM localizer responded significantly more strongly to familiar people over objects and familiar places, across visual, semantic, and episodic tasks (all 36 P’s < 0.0071; SI Appendix, Fig. S27 and Table S7). In contrast, place-preferring areas in MPC, MPFC, TPJ, STS, SFG, and PHC did not respond more strongly to false beliefs over photographs (all 6 P’s > 0.9; SI Appendix, Table S4). These results argue for a common system for ToM and social memory. They support our hypothesis of a set of regions with domain-specific responses to social content across a range of tasks.

How are person- and place-preferring areas of association cortex situated along a unimodal-to-transmodal cortical gradient? We hypothesize that person- and place-preferring brain regions extend to cortical apex, or the transmodal end of this gradient. To identify cortical gradients, we computed low-dimensional embeddings of graphs defined by resting-state correlation distance in individual participants, using the diffusion maps algorithm to effectively capture both local and global distance structure (7, 56). Consistent with prior results from group-level data, the space spanned by the first two principal dimensions included a “principal gradient” separating unimodal sensorimotor cortex from a transmodal cortical apex, along with a “visuomotor gradient” separating visual from somatomotor and auditory cortices (Fig. 3A). For visualization purposes, a boundary was placed at an arbitrary position along the principal gradient, separating areas roughly considered within versus outside of the cortical apex.

Fig. 3.

Person- and place-preferring areas across cortex are functionally coupled. (A) Diffusion embedding of resting-state functional connectivity data reveals large-scale cortical gradients, separate the cortical apex (pink) from visual (cyan) and somatomotor (yellow) cortices. (B) Person-, place-, and object-preferring regions (coordinate-wise Z > 2.3 across visual, semantic, and episodic tasks), shown on the cortical surface and in the diffusion embedding. All results from one representative participant. (C) Matrix of resting-state correlations among person- and place-preferring areas, and hierarchical clustering results. (D) Mean correlations within person- and place-preferring areas, and between the two. Error bars show SE across correlation values. **P < 10⁻³, ***P < 10⁻⁴ (permutation test). Abbreviations: MPC, medial parietal cortex; MPFC, medial prefrontal cortex; TPJ, temporo–parietal junction; STS, superior temporal sulcus; SFG, superior frontal gyrus; TP, temporal pole; PHC, parahippocampal cortex.

We next identified person-, place-, and object-preferring coordinates in this space, only including coordinates that responded significantly across visual, semantic, and episodic tasks (Fig. 3B and SI Appendix, Fig. S28; Z > 2.3 for each task). The three sets of regions, while distributed and interdigitated on the cortical surface (as shown before, Fig. 1), were each clustered in connectivity distance space. Person-preferring regions were positioned at or near the cortical apex. Place-preferring areas spanned nearly the whole length of the principal gradient, stretching from visual cortex up to the cortical apex. On the cortical surface, place responses often straddled the boundary between visual and transmodal cortex, in medial and lateral parietal cortex as well as ventral temporal cortex. By contrast, object-preferring areas were typically clustered toward the middle or low end of the principal gradient, adjacent to somatomotor cortex. These results demonstrate that person- and place-preferring areas reach the apex of a unimodal-to-transmodal gradient in cortex. The results also show that they are not equal: place-preferring regions extend “lower” into visual cortex than person-preferring ones.

To what extent are person- and place-preferring regions across multiple zones of cortex functionally coupled? While the results shown in Fig. 3B indicate that areas with common stimulus preferences also share patterns of functional connectivity, we next tested this hypothesis explicitly using the ROIs defined above (Fig. 2B). We found that resting-state correlations were substantially stronger within person- or place-preferring areas than between the two (Fig. 3 C and D; permutation test; P < 10⁻⁴, person vs between; P < 10⁻³, place vs between). Hierarchical clustering of regions based on correlation distance revealed a dominant two-cluster solution, separating person- and place-preferring regions (Fig. 3C). Consistent results were obtained whether or not global signal removal was included as a preprocessing step (SI Appendix, Fig. S29). Whole-brain functional connectivity analyses using person- and place-preferring regions as seeds confirmed this dissociation, showing resting-state correlations to other parts of the frontal, temporal, and parietal lobes with similar stimulus preferences (SI Appendix, Figs. S30–S33). These results demonstrated that person- and place- preferring areas are functionally coupled with other regions across cortex with similar content preferences.

To what extent do the person- and place-preferring areas described here correspond with DMN subsystems DN-A and DN-B? To define these networks and others, we used the multisession Bayesian hierarchical modeling (MSHBM) algorithm (57) to cluster coordinates into networks in a manner that is specific to individual participants, but constrained by a group-level prior (Fig. 4A and SI Appendix, Fig. S34). We used a 15-network prior including DN-A and B, the language network (LANG), frontoparietal network (FPN)-A and B, salience/parietal memory network (SAL/PMN), cingulo-opercular network (CG-OP), dorsal attention network (dATN)-A and B, premotor/posterior parietal network (PM-PPr), somatomotor network (SMOT)-A and B, auditory network (AUD), and central (VIS-C) and peripheral (VIS-P) visual networks (36).

Fig. 4.

Default network (DN)-B and A respond to people and places, respectively. (A) Individual-specific parcellation of cerebral cortex into fifteen functional networks. (B) Proportion of person- and place-preferring coordinates falling within each functional network. Error bars show SE across participants. (C) Task responses averaged across DN-B and DN-A. Error bars show SE across runs. *P < 0.01, **P < 10⁻³, ***P < 10⁻⁴ (linear mixed model across runs, with participant included as random effect).

We next determined the proportion of person- and place-preferring coordinates contained within each network. Person- and place-preferring coordinates were defined as above, as those that responded significantly across visual, semantic, and episodic tasks. The majority (63.8%) of person-preferring coordinates fell within DN-B, with smaller proportions within DN-A (13.5%) and LANG (12.9%, Fig. 4B). A plurality of place-preferring coordinates (32.9%) fell within DN-A. The remainder of place-preferring coordinates were separated across multiple networks: VIS-P (24.3%), dATN-A (15.7%), dATN-B (10.6%), and FPN-A (8.0%). When only considering coordinates with the search spaces used for the ROI analysis, we observed an increased proportion of person-preferring coordinates within DN-B (70.1%) and place-preferring coordinates with DN-A (47.2%), suggesting that the functional ROIs characterized above fall largely within these networks (SI Appendix, Fig. S35). These results demonstrate that individually defined networks DN-B and DN-A contain domain-preferring responses. However, like the functional connectivity gradient analysis, they reveal a difference between people and places: While person responses were largely confined to one network, place responses extended across multiple networks.

Finally, we asked how DN-B and DN-A respond across task conditions, by conducting an ROI analysis using the functional networks as ROIs (Fig. 4C). DN-B responded significantly more strongly to person versus object and place conditions across visual, semantic, and episodic tasks; responded more to familiar versus unfamiliar people; and responded more strongly to the false belief versus false photograph condition in the ToM localizer (all P’s < 0.05, SI Appendix, Table S8). DN-A responded significantly more strongly to the place versus object and person conditions across visual, semantic, and episodic tasks, and responded more to familiar versus unfamiliar places (all P’s < 0.05, SI Appendix, Table S9). DN-A also responded to the false belief versus false photograph comparison (P < 0.05), which was not predicted, although its response to the false belief condition was at baseline. These results complement the functional ROI analysis, demonstrating that domain-specific responses emerge from networks defined using resting-state data, without selecting for areas with particular task responses. They demonstrate that DN-B and A are functionally dissociated by their response to people and places across a range of specific tasks.

Discussion

Our results suggest that social and spatial cognition involve separate neural systems with a similar functional organization. Areas preferring social or spatial information were observed in multiple zones of cortex associated with the DMN, including MPFC, MPC, TPJ, and SFG. Person- and place-preferring areas were observed in spatially adjacent locations. These two features—similar cortex-wide structure and yoking of local spatial position—suggest a parallel functional organization. Areas responded to information from their preferred domain across a range of specific task contexts—visual perception, semantic judgment, and episodic simulation—suggesting a functional dissociation based on information content rather than cognitive process. Person-preferring areas corresponded closely to DN-B. Place responses were observed both within DN-A and other functional systems, including peripheral visual cortex and dATN, extending along the length of a unimodal-to-transmodal gradient. These results suggest that domain-specificity is a pervasive feature of cortical organization, extending beyond sensory regions to the apex of transmodal association cortex.

Our results provide converging evidence that the DMN comprises two anatomically and functionally distinct networks (24, 29, 36). While this line of work has studied DN-B function using a ToM localizer task, we find a response to numerous additional tasks involving familiar people, adding to evidence that this network specifically processes information from the social domain. In contrast, we find that DN-A responds to task conditions involving places or spatial information. This converges with the finding that across individual trials in an episodic projection task, DN-A activity specifically tracks the extent to which items elicit mental construction of a scene (34). These results also build upon evidence for domain-specific responses within medial parietal cortex, which has shown a similar pattern of alternating regions with MPC preferring faces and scenes (37–39). We demonstrate that these preferences exist beyond MPC, across parts of cortex associated with the DMN.

Existing theoretical perspectives on the functional dissociation between DN-A and B have emphasized their role in long-term memory as a point of distinction, arguing for a role for DN-A but not DN-B for remembering past events (29, 31). In addition to task responses, this argument has been supported by the distinct anatomical organization of DN-A and B: DN-A contains a component in PHC, which has well established anatomical connections with the hippocampal formation. The current results, however, suggest that DN-B may contribute to long-term memory as well. The DN-B responded substantially more strongly to images of familiar relative to unfamiliar people. Our prior work did not find such an effect in face-preferring regions of visual cortex such as the fusiform face area and argued that such a preference could not be explained as a multiplicative attentional modulation (41). Furthermore, responses to semantic and episodic tasks involving personally familiar entities were observed in both DN-B and A, depending on whether the event involved social versus spatial content. A role for DN-B in social aspects of long-term memory could account for long-standing results in the literature on DMN responses during familiar face perception (58) and trait judgment (59), which have not previously been specifically localized to DN-B.

The view that DN-B supports social aspects of long-term memory may also help resolve a debate in the literature about the role of the DMN in episodic recall. Some studies have argued for distinct systems for ToM and episodic memory (31), while others have argued for a common system (19, 30). Representations of events from memory, or imagined future events, typically incorporate a diverse array of information, including both social and spatial content (60). Social information may include specific people, their relationships, their actions and interactions, and their inferred mental states. Spatial information may include specific locations, scene structure, movement through space, and a temporal sequence of multiple locations. Our results indicate that DN-B may processing social information from events, while DN-A processes spatial information (see 18 for a related perspective). On this view, episodic recall or future simulation in naturalistic contexts would be expected to engage both networks. The current study focused on future events, because it was easier to dissociate social and spatial information with a simulation task—past events usually include both. However, future work should investigate whether and how DN-A and DN-B interact when constructing events with integrated social and spatial information—e.g., remembering who was where and when.

If DN-B does contribute to long-term memory, how might it interact with the hippocampal formation? The current results, along with our prior work (41), suggest the temporal pole as a potential intermediary. While summaries of cortical inputs to the hippocampus focus on PHC and perirhinal cortex, tracer studies in the macaque demonstrate that TP has a direct, bidirectional connection with entorhinal cortex of similar magnitude (42, 43). Neuropsychological evidence supports a role of TP in long-term memory: Patients with TP dysfunction show impaired recognition of familiar people and objects, in the context of frontotemporal dementia and brain damage (61–63). Prior studies have found preliminary evidence that TP constitutes a node of DN-B (24, 40). Here, we demonstrate that TP is similar in response profile and functional connectivity with other person-preferring regions associated with DN-B. Taken together, these results suggest that TP is well positioned to relay information between person-preferring regions of association cortex and the hippocampal formation.

On this interpretation, DN-B may be understood as a component of the brain’s corticohippocampal long-term memory system that selectively processes social information. This view builds upon existing theoretical accounts of a functional division of labor among cortical inputs to the hippocampus. Prominent accounts have argued for two separate systems, converging at the hippocampus: one providing information about “items” (e.g., objects, people, or concepts) via perirhinal cortex and another providing information about “context” or relations between items via PHC (64–66). Extending these ideas, DN-B could be understood as a third system providing social information to the hippocampus via the TP (see also 67). While prior accounts have argued that PHC processes contextual and relational information regardless of content domain (64, 68)—including spatial relations, social relationships, and causal links between mental states and outcomes—we hypothesize that relational information is processed by distinct systems based on content domain. Specifically, we hypothesize that DN-B process social relational information, while DN-A processes spatial relational information, consistent with long-standing evidence for a role of parahippocampal cortex in spatial cognition (47, 69). Consistent with this view, narrative context in stories with social content modulates responses in areas identified with a ToM localizer, likely corresponding to DN-B (70).

Why might the brain employ systems with parallel organization for understanding people and places? While these two problems may appear different on their face, they share a similar structure. In navigating the physical world, we decide how to move around in an uncertain, constantly changing environment, in order to execute plans and optimize rewards. In navigating the social world, we decide what to say and do around people with uncertain, constantly changing internal states, in order to achieve personal and collective goals. In both cases, researchers have argued that the mind solves these problems using abstract, internal models—theories of mind (71–73), and cognitive maps or graphs (74, 75)—combined with concrete information about specific familiar people and places derived from experience (72, 76, 77). One potential explanation for the parallel organization observed here is that DN-A and B support the common cognitive operation of building internal relational models, but within distinct content domains. This perspective fits with theories arguing for a common neural substrate for memory, cognition, and planning (78), and theories of the corticohippocampal long-term memory system emphasizing its role in generating flexible relational models (64, 79, 80).

While this work focuses on parallels between systems for social and spatial cognition, we also observed differences between their organization. Person responses were restricted to association cortex and largely found within DN-B. By contrast, place responses extended throughout the unimodal-to-transmodal gradient and were found within multiple networks, including DN-A, VIS-P, and dATN-A and B. The existence of place preferences across both visual and association cortex is consistent with prior results (81–83). This line of research has identified an anterior–posterior dissociation in place responses across the parahippocampal cortex, MPC, and lateral occipital and parietal cortex, with posterior regions driven more by perceptual than mnemonic tasks and shower stronger functional connectivity to visual cortex.

While areas involved in scene perception and memory appear to be spatially adjacent in cortex (81), this is not the case in the social domain: Face-preferring areas such as the fusiform and occipital face area are not anatomically positioned in the vicinity of social cognitive responses within DN-B, and have strongly dissociable response profiles (41). While the current results do not speak directly to the basis of this divergence, we speculate that this could relate to a difference in the degree of separation between systems for cognition and perception in the social and spatial domains. Much information relevant for social cognition (e.g., social relationships, mental states) is not immediately accessible from visual input. By contrast, information relevant for spatial cognition (scene layout, spatial relationships) often is, which may have led to closer functional and anatomical relationship between systems for spatial cognition and scene perception.

Several limitations of the current work should be noted. First, the task conditions used in the study are complex and likely rely on numerous component processes. While we refer to tasks as “visual,” “semantic,” and “episodic,” we do not mean to imply that these tasks only involve the corresponding cognitive process. For instance, visual perception of familiar faces or scenes could evoke semantic or episodic information stored in memory. Episodic simulation could engage semantic knowledge about specific people and places. Self-reports of the extent to which each task elicited episodic recall suggest that this occurred a minority of the time and consistently across tasks, but there was substantial variability across participants, highlighting the possibility that individual participants engaged different component processes (SI Appendix, Fig. S4). Given these considerations, we do not argue that responses to the visual task strictly reflect visual processing, and likewise for semantic and episodic tasks. Instead, we leverage our precision fMRI approach to draw conclusions from the consistency of effects across multiple tasks (44, 84). In particular, we argue that the consistency of effects of content domain (people versus places) across varying task contexts provides strong evidence that these effects are in fact driven by content domain: Any confounding factor would have to be present across all tasks to fully explain the observed response profiles.

Second, while we argue that TP may provide a link between DN-B and the hippocampal formation, the current study does not directly investigate interactions between the hippocampus and cortical systems. Recent studies on functional connectivity between the hippocampus and individually defined functional networks have focused on connections to DN-A and SAL/PMN, but not DN-B (32, 33, but see 85). However, modest resting-state correlations between DN-B and the anterior hippocampus have been reported, using an anatomical approach that did not specifically search for areas with such a correlation (32). Based on the current results, future work should explicitly test the hypothesis that DN-B interacts with the hippocampal formation via TP.

In summary, the current work argues for parallel but dissociable systems for spatial and social cognition. The results offer a unique perspective on the functional role of the DMN, arguing that subsystems DN-A and DN-B are distinguished by the type of information they act upon, rather than the cognitive process they support. These results may guide future work addressing the mystery of how humans understand other people. Just as interactions between the hippocampus and functionally specific cortical areas have been argued to underlie the ability to learn cognitive maps of the spatial environment (48, 75), we speculate that interactions between the hippocampus and a person-preferring network of association cortex support learning internal models of specific familiar people. These results inspire several directions for future research, into how the brain builds representations of familiar people during learning, and how networks for social and spatial cognition emerged in evolution and differentiate in development.

Methods

Participants.

Ten human participants (5 male, 5 female; age 28 to 40) were scanned across multiple sessions using fMRI. Participants were healthy with normal or corrected vision, right-handed, and native English speakers. The experimental protocol was approved by the Rockefeller University Institutional Review Board, and participants provided written, informed consent. Separate analyses of this dataset have been reported elsewhere (41).

Tasks.

We used a range of perceptual and cognitive tasks involving familiar people and places. Participants were asked to choose six of their top ten most familiar people and places, and tasks involved processing these six people and places. Main tasks included visual perception of familiar and unfamiliar faces and scenes, and generic objects; semantic judgment about familiar people and places, and generic objects; and episodic simulation of events involving familiar people and places, and generic objects (Fig. 1A and SI Appendix, Fig. S1). Additionally, localizer tasks from the existing literature were run, including tasks eliciting theory of mind (ToM, reading and answering questions about stories involving false beliefs or false physical representations, 86); language comprehension (reading sentences or nonword lists, 28); and dynamic visual perception (watching videos of moving faces, objects, and scenes, 87). In each of three scans, one main task and one localizer were run (order: visual and ToM, semantic and dynamic perception, episodic and language). Stimulus presentation scripts can be found at https://osf.io/5yjgh/.

In the visual perception task, participants viewed serially presented naturalistic images of faces, objects, and scenes. For each participant, we obtained 20 images each of six familiar faces and scenes. Familiar face images were obtained directly from friends or family members, without the participant seeing them. Control images were defined as six matched unfamiliar faces and scenes. Familiar and unfamiliar faces were matched on age group (young adult, middle-aged, old), race, and gender; familiar and unfamiliar scenes were matched on rough semantic category (e.g., outdoor street view; building interior). Object images were of six generic objects with varying physical properties - a banana, a baseball, a feather, a rock, a sponge, and a wrench. Face and object images contained no clear spatial structure (e.g., corners), and had minimal contextual cues beyond the background. Scene images contained no people. All five image categories were matched, for each participant, on the mean and variance across images of luminance, Rms contrast, and saturation (in CIE-Lch space, with a D65 illuminant): all P’s > 0.05, one-way ANOVA and Bartlett test. Images were presented at 768 × 768 resolution, 12.8 × 12.8 degrees of visual angle, for 1.85 s each with a 150 ms interstimulus interval, in 18 s blocks of images of one identity. Participants performed a one-back task, pressing a button when an image was repeated. A postscan questionnaire verified that participants could recognize the person or place in the familiar images (mean: 100% for faces, 84% for scenes), but not the unfamiliar images (mean: 0% for faces, 2% for scenes, SI Appendix, Fig. S4).

In the semantic task, participants rated traits of familiar people and places, and generic objects, on a 0 to 4 scale. This included personality traits of people (e.g., confident, angry, intelligent), spatial or navigational properties of places (e.g., cramped, large, has walls), and physical properties of objects (e.g., soft, heavy, rough; SI Appendix, Table S1). Participants rated by moving an icon left or right, over 18 s blocks of four questions for a given identity, for a total of 20 questions per condition and identity.

Prior to the episodic task, participants listed five common conversation topics for each familiar person, and five familiar subregions of each familiar place. In the scan, they were asked to imagine familiar people talking about common topics, navigating specific subregions of familiar places, and objects engaged in physical interactions (e.g., rolling or sliding down a hill, falling into water). Participants were specifically asked to generate a novel event, rather than remember a past event. Imagination blocks lasted 18 s, with a 3 s verbal prompt, and a 1 s hand icon at the end of the block, which participants responded to with a button press. After the scan, participants were asked (for 2/5 of trials) questions about difficulty, detail, visual imagery, emotional response, and memory recall (SI Appendix, Fig. S5). Additionally, a free response description was collected, to ensure that participants could describe what they imagined. 9/10 participants were able to describe all events; the remaining one recalled 75% of events.

Across the three main tasks, blocks were separated by 4 s of resting fixation, and presented in five 8 to 13 min runs per task, with palindromic block orders, counterbalanced across runs and participants. Fixation blocks were included in the beginning, middle, and end of the experiment to estimate a resting baseline. Localizer tasks were split into 4 to 5 min runs, with four runs for theory of mind and language tasks, and six for dynamic perception. Scanner task performance was high, indicating sustained attention during scan sessions (SI Appendix, Fig. S3). After each scan, participants were asked to rate on a 0 to 10 scale to what extent they recalled memories of events during the person and place conditions (SI Appendix, Fig. S4).

Behavioral Data Analysis.

To evaluate participants’ self-reported experience in the episodic task, we compared behavioral responses across each pair of conditions (person, object, place), for each of the five questions. Statistics were computed using a linear mixed effects model (MATLAB’s fitlme, unpaired, two-tailed comparisons) across items or event ratings, with random intercepts for participant.

MRI Acquisition.

Participants were scanned on a Siemens 3 T Prisma across three 2.5-h sessions, which included task acquisitions as well as 40 min of high-resolution anatomical images, 60 min of resting-state acquisitions, and spin echo acquisitions for distortion correction. Three each of T1- and T2-weighted anatomical images were acquired at 8 mm isotropic resolution. Task and resting-state data were acquired using a multiband, multiecho EPI pulse sequence, optimized to boost temporal signal-to-noise ratio (tSNR) throughout the brain and specifically in the anterior medial temporal lobes (TR = 2 s, TE = 14.4, 33.9, 53.4, 72.9, and 92.4 ms, 2.4 × 2.4 × 2.5 mm resolution, 48 oblique axial slices with near whole brain coverage, multiband acceleration 3x, GRAPPA acceleration 2x, interleaved slice acquisition). 3 to 4 parameter-matched spin echo acquisitions were acquired per scan, between every four runs of task. Raw MRI data can be found at https://openneuro.org/datasets/ds003814 (88).

MRI Preprocessing.

Data were preprocessed and analyzed using a custom pipeline, integrating software elements from multiple software packages: FSL (6.0.3), Freesurfer (7.1.1), AFNI, Connectome Workbench 1.5, tedana 0.0.10, and Multimodal Surface Matching (MSM). The code is available at https://github.com/bmdeen/fmriPermPipe/releases/tag/v2.0.2, with dataset-specific wrapper scripts at https://github.com/bmdeen/identAnalysis.

Anatomical images were preprocessed using an approach based on the Human Connectome Project (HCP) pipeline (89). The three images for each modality were linearly registered using FLIRT (90) and averaged; registered from T2- to T1-weighted; aligned to ACPC orientation using a rigid-body registration to MNI152NLin6Asym space; and bias-corrected using the sqrt(T1*T2) image (91). Cortical surface reconstructions and subcortical parcellations were generated using Freesurfer’s recon-all (92). Surface-based registration (MSMSulc) was used to register individual surfaces to fsLR average space (93). This registration was used to project the HCP multimodal cortical parcellation (51) onto individual surfaces.

Functional data were preprocessed using a pipeline tailored to multiecho data, aiming to optimize tSNR while maintaining high spatial resolution. Motion parameters were first estimated using MCFLIRT (94). Intensity outliers were removed using AFNI’s 3dDespike, and slice timing correction was performed using FSL’s slicetimer. Motion correction was then applied, in combination with topup-based distortion correction (95), and rigid registration to a functional template image, in a single-shot transformation with one linear interpolation, to minimize spatial blurring. Multiecho ICA was performed using tedana, with manual adjustments, to remove non-blood-oxygen-dependent (non-BOLD) noise components (96). We used denoised data with ICA-derived non-BOLD components removed, with data combined across echoes with an optimal weighted average based on locally estimated T₂^* values. Data were intensity normalized and resampled to an individual-specific CIFTI space aligned with the anatomical template image, with 32 k density fsLR surface coordinates, and 2 mm volumetric subcortical coordinates. Registration between the functional and anatomical templates was computed using boundary-based registration (bbregister, 97). Surface data and subcortical volumetric data were both smoothed with a 2 mm-FWHM Gaussian kernel. For resting-state data, the global mean signal was removed via linear regression, to diminish global respiratory artifacts not removed by multiecho ICA (98). Data were also analyzed without global signal removal, and the results relevant to our arguments did not differ (SI Appendix, Fig. S29). Pairs of time points with excessive head movement between them (framewise displacement > 0.5 mm for task data, 0.25 mm for resting-state data) were excluded from subsequent analysis.

Whole-Brain Analysis.

Whole-brain statistical analyses were performed in individuals using AFNI’s 3dREMLfit, modeling temporal autocorrelation with a coordinate-wise ARMA(1,1) model (99). Results were thresholded using a false discovery rate of q < 0.01 (two-tailed), to correct for multiple comparisons across ~90 K CIFTI coordinates (100). We compared responses to people versus places, and objects versus places, across tasks (boxcar regressors convolved with canonical double gamma hemodynamic response function).

Definition of Anatomical Regions.

To accurately localize functional responses to anatomical areas, PHC and TP were hand-drawn on individual brains. Regions were drawn on coronal slices of MNI-aligned images at 2 mm resolution, using .8mm T1 images as a reference. They were subsequently resampled to 32 k-vertex surface hemispheres, and then modified in the volume to remove any discontinuities observed on the surface. Given the lack of established gross anatomical features delimited the cytoarchitectonic boundaries of human RSC, this area was defined using the HCP multimodal cortical parcellation.

PHC and TP were defined using previously established anatomical features corresponding to cytoarchitectonic boundaries (49, 50). TP extended anteriorly to the tip of temporal cortex. The posterior boundary of TP was placed at whichever of two landmarks was farther anterior: the anterior tip of the collateral sulcus, or one slice (2 mm) anterior to the fronto-temporal junction (FTP, the first slice containing white matter connecting the frontal and temporal lobes, also termed the limen insulae). On its dorsomedial surface, the TP extended laterally to the lateral bank of the lateral-most temporopolar sulcus. On anterior-most slices in which the temporopolar sulcus was not visible, there was no lateral border. On its ventrolateral surface, the TP extended laterally to the medial lip of the inferior temporal sulcus, or if that was not visible, the superior temporal sulcus. On the 3 to 4 anterior-most slices in which neither sulcus was visible, there was no lateral border. Anterior to the FTP, TP had no medial boundary. Posterior to the FTP, its medial boundary was the medial edge of the parahippocampal gyrus (PHG).

The anterior boundary of PHC was placed two slices (4 mm) posterior to the last slices containing the head of the hippocampus, determined by the presence of the gyrus intralimbicus. The posterior boundary of PHC was placed at the last slice containing the hippocampus. The lateral boundary of PHC was defined as the lateral edge of the collateral sulcus. The medial boundary of PHC was defined as the medial edge of the PHG.

Autocorrelation Analysis.

This analysis computed 1D spatial correlations in the person versus place contrast map along the cingulate sulcus within MPFC and MPC. We chose to analyze responses along the cingulate sulcus because it provides a clear anatomical landmark that can be used to define position along the dorsal-to-ventral axis in MPFC and posterior-to-anterior axis in MPC, in a manner that accounts for the curved shape of the cingulate and adjacent gyri. To define paths, we first identified a circular trajectory in spherical fsLR surface space that best corresponded, by visual inspection, to the location of the cingulate gyrus. We then defined two segments of this circle, confined to MPFC and MPC, based on the boundaries of search spaces used for region-of-interest analysis (described below). For each region, hemisphere, and coordinate on the path, we extract the person versus place contrast value, averaged aross visual, semantic, and episodic tasks. Values were interpolated to an evenly spaced coordinate frame based on the distance between coordinates on the cortical surface. Using contrast values in this coordinate frame, 1D spatial autocorrelation was computed for each hemisphere and region. To assess the significance of autocorrelation values, a permutation test was used, in which coordinates were randomly spatially permuted across 10,000 iterations.

Region-Of-Interest (ROI) Analysis.

ROI-based analyses were conducted to assess how person- and place-preferring brain areas respond across a range of task conditions. ROIs were defined as the top 5% of person- or place-preferring coordinates (semantic task) within anatomical search spaces capturing zones within the cortical apex: medial frontal cortex, medial parietal cortex, temporo–parietal junction, superior frontal gyrus, superior temporal sulcus, parahippocampal cortex, and temporal pole. Search spaces were composed of regions from the HCP parcellation (SI Appendix, Table S2). To extract responses to the task used to define ROIs, a leave-one-run-out analysis was performed, in which ROIs were defined in all but one run of data, and responses extracted in the left out run. Statistics were performed on percent signal change values across runs, using a linear mixed effects model (MATLAB’s fitlme), with participant included as a random effect. Similar analyses were performed for object-sensitive regions, and for the language and theory of mind localizers. Analyses were also performed for a range of ROI sizes, from 5 to 40% (SI Appendix, Fig. S12).

For ROIs defined as person-preferring or by the language or ToM tasks, responses to people were compared with places and objects, across visual, semantic, episodic, and dynamic tasks. Additionally, responses to familiar/unfamiliar faces were compared. For place-preferring ROIs, responses to places were compared with people and objects across tasks, and responses to familiar/unfamiliar faces were compared. For object-preferring ROIs, responses to objects were compared with people and places, across tasks. For all ROIs, we compared responses to false belief versus false photo and sentence versus nonword list conditions. All comparisons were paired and one-tailed.

Connectivity Gradient Analysis.

To position functional responses along large-scale connectional gradients in cortex, we computed a low-dimensional embedding of resting-state functional connectivity similarity, using the diffusion maps algorithm (56). Following Margulies et al. (7), we computed a pairwise correlation matrix between resting-state time series from ~59 K cortical surface coordinates, kept only the top 10% of values per row, zeroed negative values, and computed cosine similarity between rows to generate a positive, symmetric similarity matrix. We then generated a diffusion map embedding, using α = 0.5 (Fokker–Planck normalization), and automated estimation of diffusion time via spectral regularization, by multiplying Laplacian eigenvalues ${\lambda }_i$ by $\frac{1}{1-{\lambda }_i}$. Across participants, the space spanned by the first two principal directions included previously described anatomical gradients (7): one between sensory cortex and transmodal association cortex (principal gradient) and one between sensorimotor and visual cortex (visuomotor gradient, Fig. 3A). In order to align the principal gradient with the y-axis, a manual rotation of 15° was applied to this space for visualization. Because the orientation of the apex and visuomotor gradients in diffusion embedding space varied across participants, a Procrustes transformation was used to map each participant’s embedding to the first participant’s, using the top three dimensions. To overlay functional responses on the diffusion embedding, we identified coordinates that were responsive to either 1) people versus objects and people, 2) places versus people and objects, or 3) objects versus people and places, across visual, semantic, and episodic tasks (coordinate-wise Z > 2.3 for each task). These coordinates were displayed in color in diffusion space. Only a small proportion (<0.15% for each participant) of coordinates were responsive to multiple contrasts. These were colored with the following precedence: object-preferring, place-preferring, and person-preferring. For the sake of visually comparing functional responses to the principal gradient, boundaries were drawn on the cortical surface at the 70th percentile of coordinates along the principal gradient, using a smoothed map of gradient values (surface-based 8 mm-FWHM Gaussian kernel).

ROI-Based Functional Connectivity Analysis.

Resting-state correlations were computed between person- and place-preferring areas defined similarly to the ROI analysis, but using separate search spaces for each hemisphere. Place-preferring STS subregions were not included, because such responses were not reliably observed across participants. The regions were hierarchically clustered based on correlation distance, and regions were ordered in a way that minimized distance between adjacent pairs, without separating clusters (Fig. 3C; MATLAB’s optimalleaforder). Within- versus between-network correlations were compared using a permutation test, permuting regions (10,000 iterations). These analyses were additionally performed on data preprocessed without global signal removal (SI Appendix, Fig. S29). Whole-brain correlation maps, with functional ROIs as seeds, were computed for person- and place-preferring regions of left MPC and SFG, within each individual participant. These two regions were chosen for presentation because the functional connectivity of other ROIs generally resembled one of these two.

Individualized Network Parcellation Analysis.

To parcellate cerebral cortex into multiple large-scale functional networks defined within individuals, we implemented the MSHBM algorithm (57). For each coordinate, a functional connectivity profile was computed based on resting-state correlations with each of 1,483 ROIs, consisting of single vertices evenly spaced across the fsLR surface mesh. The resulting correlation matrix was binarized by keep the top 10% of correlations. Coordinates were then clustered based on their functional connectivity profile, based on a hierarchical Bayesian model incorporating both group-level connectivity profiles and individual variability (57). Model parameters were estimated using the expectation–maximization algorithm. As a group-level prior, we used a 15-network parcellation derived from data from the Human Connectome Project and introduced in a previous study (36). This procedure produced a parcellation of cortex into 15 networks within each participant. In order to quantify how person- and place-responsive coordinates fall within this network organization, we computed the proportion of these coordinates within each network. Person- and place-preferring coordinates were defined as described above (Connectivity gradient analysis section), using a conjunction of effects across visual, semantic, and episodic tasks. To determine how DN-A and B respond across task conditions, we conducted an ROI analysis using individually defined networks as ROIs. Statistical tests for differences in response across conditions were conducted as described above (ROI analysis section). We hypothesized that DN-B would respond preferentially to people, and DN-A to places. Given these hypotheses, we tested the same set of comparisons in DN-B as for person-preferring ROIs, and DN-A as for place-preferring ROIs (see ROI analysis).

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

Raw data are available at https://openneuro.org/datasets/ds003814 (88). Stimulus materials are available at https://osf.io/5yjgh/. Analysis code is available at https://github.com/bmdeen/fmriPermPipe/releases/tag/v2.0.2 (generic analysis tools) and https://github.com/bmdeen/identAnalysis (dataset-specific wrapper scripts).