Introduction

We encounter many people in our lives but only a few of them will have a lasting impact on us. A large body of studies (e.g., refs. 1,2,3,4) suggest that social networks can be divided into five layers of personal closeness (social layers), successively identified as (1) support clique, (2) sympathy group, (3) affinity group, (4) active network, and (5) acquaintances. These social layers run outward from the self (so-called network owner), gradually increasing in size and decreasing in relationship quality and contact frequency. This structure has been consistently identified in a variety of social contexts, including hunter–gatherer societies2,5, modern armies6, communication databases7, social media3,8, virtual worlds9, and even non-human mammalian networks10. Though social layers have been investigated ethologically, sociologically and psychologically, not much is known about the brain processing of social layers. Are individuals from different social layers processed in different brain regions? If so, does this processing exhibit a gradual change based on social distance, akin to patterns observed in other domains11,12, or is it discontinuous? Finally, does the processing of different social layers involve distinct cognitive mechanisms?

Previous neuroimaging investigations of the effect of personal closeness on the processing of people compared individuals from one’s social network (i.e., all social layers together)13,14,15 or a single social layer16 to strangers. These studies identified differences in activity in specific brain regions, mainly the medial prefrontal and posterior cingulate cortices. More recent studies compared brain activity in response to individuals from different social layers (rather than strangers)17,18,19,20. Specifically, a notable study18 used a trait-judgment task to differentiate between romantic partners (layer 1) and other network members including parents (usually layer 1), close friends (layer 2) and acquaintances (layer 5). While processing of romantic partners involved a wide network of cortical regions, including the medial prefrontal cortex, the precuneus, and the temporoparietal junction, all other individuals engaged only the parahippocampal cortex and the temporal pole. The authors suggested that the cognitive representations of romantic partners are unique in content and use, chronically accessible, and serve emotion-regulatory functions. In another study, differences between closer individuals (layers 1–3) and less close individuals (layers 4–5) influenced activity patterns in the medial prefrontal, inferior frontal and medial parietal cortices, and the temporoparietal junction20. Two other studies21,22 investigated real-world social networks (layers 1–5) as a whole and found that activity in the temporoparietal junction, the posterior lateral temporal lobe and the retrosplenial cortex was correlated with the personal distance to individuals within these networks. So far, neuroimaging studies have examined a more coarsely defined scale of personal closeness, though evidence from other disciplines demonstrates the potential value of investigating a finer scale.

In an initial attempt to capture the entire range of social layers, Wlodarski & Dunbar used functional magnetic resonance imaging (fMRI) to examine social networks of participants23. Although the focus of this study was the differences between kin and non-kin, the authors also compared brain activity during processing of different social layers. In their analysis, the authors assumed a monotonic relationship between brain activity and social layer distance. As a result, this approach identified brain regions in which activity monotonically increased (prefrontal, lingual, and somatosensory cortices) or decreased (posterior cingulate and retrosplenial cortices) during the processing of more distant social layers. However, previous findings have implicated brain regions that displayed preference for one or more social layers, which do not follow this assumption of monotonicity18. Furthermore, in other domains such as numerosity, space, and time, neural activity exhibits a Gaussian-shaped rather than monotonic tuning curve12,24,25,26. Therefore, a systematic attempt to map the brain processing of different social layers is yet to be carried out.

Most of the brain regions mentioned thus far, including regions in the medial parietal, temporal, and prefrontal cortices constitute the default mode network (DMN), a cortical network that is active during internal mentation and self-referential processing and is implicated in various aspects of social cognition17,21,23,27,28,29,30,31,32,33,34. Previous work has decomposed the DMN into three distinct subsystems35, rigorously defined their boundaries through clustering of resting-state functional connectivity data from a cohort of 1000 participants36 and associated each subsystem with specific cognitive processes37. The first of these subsystems is the core subsystem (DMN-Core), which includes the ventromedial prefrontal cortex, the posterior cingulate cortex, and parts of the temporoparietal junction. This subsystem is supposed to facilitate self-related, emotional, and social processes, enabling the extraction of personal meaning from salient information. The second is the medial temporal subsystem (DMN-MT), including the hippocampus, the parahippocampal cortex, and the retrosplenial cortex. It supports mnemonic and spatiotemporal processes and is considered to provide episodic context and aid in constructing mental simulations. The third is the dorsal medial subsystem (DMN-DM), comprised of the dorsomedial prefrontal cortex, the lateral temporal cortex, parts of the temporoparietal junction, and the temporal pole. This subsystem presumably supports the retrieval of semantic and conceptual knowledge. Previous studies have recognized the major role of the DMN, particularly its DMN-Core subsystem, in personal closeness processing11,17,20,38,39; in contrast, the DMN-MT subsystem has been found to be associated with spatial processes20,38. Since processing spaces at different scales has been shown to involve distinct regions of the DMN26, we hypothesized that processing individuals from different social layers will differentially involve DMN subsystems.

This hypothesis concerns the relationship between social and spatial cognition. In recent years, evidence has accumulated to suggest that social and spatial cognitive mechanisms often operate in similar ways11,17,21,30,40,41,42,43,44,45,46. Does this similarity manifest also with respect to the segmentation of the spatial and social environments to different proximity levels? One organizing principle that may play a role is that of cortical gradient47,48. Cortical gradients are transitions in brain activity along continuous neuroanatomical axes that correspond to gradual changes in various aspects of information processing26,47,48,49,50. Previous studies have revealed cortical gradients according to proximity level in the spatial and temporal domains12,26. In the spatial domain, processing of close-to-distant spaces was found to involve posterior-to-anterior gradients in the retrosplenial cortex and the hippocampus26. Do similar organizing principles govern the processing of social layers?

Here, we aimed to characterize the relations between brain processing of individuals from all social layers. We hypothesized that the five distinct social layers would evoke different patterns of brain activity, reflecting the cortical organization into networks and paralleling the processing of different spatial distances26. To investigate this hypothesis, we analyzed participants’ brain activity while they processed people in the different layers of their own social network. We further compared the processing of individuals in different social layers to the processing of spaces in various scales. Finally, in light of the important role of the DMN in social cognition17,21,23,28,29,30,31,32, we compared the cortical regions that we found to be sensitive to social layers to the DMN and its subsystems.

Results

Twenty-one participants provided detailed descriptions of their own social networks, which enabled us to create individually tailored stimuli to engage each of their social layers. During fMRI scanning, participants were asked to compare their personal distance to individuals from their social network (Fig. 1). Each experimental block contained names of individuals from a single social layer. This setup allowed us to examine differences in behavior and brain response across distinct social layers by comparing between blocks.

Fig. 1: Experimental procedure.
figure 1

A Prior to the experiment, participants provided names of individuals from their social network and categorized them into five layers based on personal closeness, ranging from their closest friends to those with whom they have no contact. B fMRI task. In each trial, participants were presented with two names of individuals from their social network and asked to indicate which person they feel closer to. Participants were instructed beforehand to respond based on personal (rather than physical) distance. Each experimental block contained four trials and eight names, all from the same social layer.

Full size image

Behavioral results

Analysis of participants’ response times revealed a significant effect of social layers (F(4, 80) = 25.55, p < 0.001). A subsequent post hoc test showed a significant difference between layer 1 and the other social layers (p < 0.001), with no significant differences observed between any other consecutive social layer pairs (Supplementary Fig. 1, Supplementary Note 1).

Bipartition of the social layer sensitive cortex

To investigate brain activity sensitive to social layers, we applied a general linear model (GLM) analysis to each voxel. Five predictors were created for the five social layers, each modeling all experimental blocks corresponding to a single social layer. The model was fitted to the activity of each voxel, and the beta values corresponding to the five social layers were extracted. These beta values were then compared using repeated-measures ANOVA to identify clusters of significant sensitivity to social layers (Table 1). Subsequently, in order to determine the preferred social layers, we assigned to each voxel within these clusters the social layer to which it showed the maximal activity (i.e., the social layer with the largest averaged beta value). Strikingly, the vast majority of voxels showed preference for the first social layer (Layer 1: 79.83%; Layer 2: 0.09%; Layer 3: 3.01%; Layer 4: 4.14%; Layer 5: 12.92%). In addition, social layer sensitive clusters could be classified into two separate divisions: while preference for layer 1 (support clique) was observed in the precuneus, temporoparietal junction, dorsomedial prefrontal cortex, middle temporal gyrus, supplementary motor area and precentral sulcus, separate regions in the ventromedial prefrontal, retrosplenial and parahippocampal cortices and in the hippocampus showed preference for layers 2–5 (sympathy group, affinity group, active network, acquaintances) (Fig. 2A). Preference for layer 2 was minimal, observed only in the medial prefrontal and retrosplenial cortices. We subsequently aimed to further investigate this apparent dichotomous grouping of brain activity for social layers. The averaged brain activity for each social layer in each cluster was calculated and a hierarchical clustering analysis was performed (Fig. 2B). The analysis validated the separation between layer 1 and layers 2–5 as reflected in the association of each group with distinct brain regions. Similar results were observed when determining the preferred social layer at each voxel by fitting a Gaussian curve to the beta values and locating its peak (Supplementary Fig. 2. see Methods section for details).

Table 1 Clusters of significant sensitivity to social layer
Full size table
Fig. 2: Cortical social layer sensitivity shows a distinction between layer 1 (support clique) and more distant social layers.
figure 2

A Social layer sensitive clusters were identified using voxel-wise ANOVA on beta values, with thresholds of FDR-corrected p < 0.01 at the voxel level and cluster volume >37 voxels of 3 × 3 × 3 mm3. Colors indicate the social layer of maximal activity, determined as the social layer with the highest beta value after averaging across participants. Note the dominance of layer 1 in the lateral cortical wall as well as in the dorsal aspect of the medial wall. B Hierarchical clustering analysis reveals a bipartition between social layer sensitive brain regions that are tuned to layer 1 and layers 3–5, with layer 2 clustered closer to layers 3–5. (l left, r right, dmPFC dorsomedial prefrontal cortex, HC hippocampus, MTG middle temporal gyrus, PCS precentral sulcus, PHC parahippocampal cortex, Prc precuneus, RSC retrosplenial cortex, SMA supplementary motor area, TPJ temporoparietal junction, vmPFC ventromedial prefrontal cortex).

Full size image

Overlap between social layers and spatial scales

Previous work showed that spatiotemporal and social functions rely on similar brain networks11,17,42,43,44,45,51 and that spatiotemporal processing is scale-sensitive, that is, differs according to the size of the environment or timeframe being processed12,26. In the spatial domain, brain activity associated with different spatial scales (room, building, neighborhood, city, country, continent) exhibits a systematic organization along cortical gradients26. To better understand the relations between social layers and spatial scales, we performed a voxel-wise comparison between social layer preference, as identified here, and spatial scale preference, as identified previously in a different cohort26. Small spatial scales (room, building) showed the largest overlap with the closest social layer (layer 1, or the support clique); medium spatial scales (neighborhood, city) showed the largest overlap with the most distant social layers (layers 3–5); large spatial scales (country, continent) showed the largest overlap with the most distant social layer (layer 5) and, to a lesser degree, with the intermediate layer 3 (Fig. 3). Interestingly, when comparing the distribution of social layers with two partially overlapping components of the brain’s orientation system, closer social layers overlapped with the social component, while more distant social layers overlapped with the spatial component (Supplementary Fig. 3)11,17,38.

Fig. 3: Overlap between social layers and spatial scales.
figure 3

A–C Social layer sensitive activity in regions that also showed preference for small (room, building), medium (neighborhood, city) or large (country, continent) spatial scales. Spatial scale sensitive ROIs (taken from ref. 26) are outlined in black. Clusters showing significant sensitivity to social layer (identified by voxel-wise ANOVA with thresholds of FDR-corrected p < 0.01 at the voxel level and cluster volume >37 voxels of 3 × 3 × 3 mm3) are colored according to the social layer with maximal activity. D Jaccard indices (intersection over union) represent the overlap between each of the six spatial scales and each of the five social layers. While social layer 1 overlapped with small spatial scales, social layers 3–5 overlapped with medium (and, to a lesser degree, large) spatial scales. *p < 0.05.

Full size image

Social layer preference in subsystems of the default mode network

Processing of social closeness regardless of social layer, as estimated by comparing the social orientation task to a lexical control task, implicated the precuneus, ventromedial prefrontal cortex, bilateral temporoparietal junction and left superior temporal sulcus, all of them components of the DMN (Fig. 4A). To understand the relations between the social layers and the DMN, we compared the social layer sensitive clusters with a parcellation of the human brain into seven large-scale cortical networks including the DMN, and with three subsystems of the DMN (DMN-Core, DMN-DM, DMN-MT), as identified in data from 1000 participants36. Among the seven cortical networks, the DMN exhibited the most significant overlap with brain activity related to social closeness processing (Jaccard index = 0.16, FDR-corrected p < 0.001) and with social layer sensitive clusters (Jaccard index = 0.12, FDR-corrected p < 0.001) (Fig. 4B). Examination of the relationship between the social layer sensitive clusters and the DMN subsystems revealed that DMN-Core exhibited a significant preference for layer 1 (with 13% of DMN-Core overlapping with layer 1), while DMN-MT showed a significant preference for the more distant layers 4 (7%) and 5 (8.4%) (all FDR-corrected p < 0.001) (Fig. 4C, D). Preference for layer 3 was evident in both subsystems (1.1% of DMN-Core; 0.8% of DMN-MT; both FDR-corrected p < 0.001), while preference for layer 2 was not observed in any of the subsystems.

Fig. 4: Personal closeness processing and social layer preference in different cortical networks.
figure 4

A Personal closeness processing with respect to the DMN subsystems. Voxels which were significantly more active during the social task (regardless of social layer) compared to a lexical control task (FDR-corrected p < 0.01 at the voxel level) are colored. DMN subsystems are shown in greyscale outlines. B Left: Overlap between the seven cortical networks and clusters implicated in personal closeness processing (depicted in A). Only the DMN was significantly implicated in the social task. Right: Overlap between cortical networks and social layer sensitive clusters (depicted in D). The DMN showed the largest overlap with social layer sensitive clusters. C Distribution of social layers within subsystems of the DMN (Core, dorsal medial (DM) and medial temporal (MT)36,37). Percentages indicate the proportion of each subsystem showing preference for each social layer. D Cortical distribution of social layers with respect to DMN subsystems. Social layer sensitive clusters are identified and colored as in Fig. 2. DMN subsystems are shown in greyscale outlines. Layer 1 was dominant in DMN-Core and DMN-DM, while layers 4–5 were dominant in DMN-MT. *p < 0.05.

Full size image

Discussion

Examining how humans process members of their social network in different levels of personal closeness during fMRI scanning revealed several novel findings. Firstly, the vast majority (80%) of the cortical area that responded differentially to various layers of social networks showed preference for individuals in the closest social layer; this area was anatomically segregated from regions that showed preference for all other social layers. Secondly, brain activity during processing of the closest social layer and distant social layers overlapped with brain activity during processing of small and medium spatial scales, respectively. Thirdly, out of all the cortical networks, the DMN exhibited the greatest social layer preference; specifically, DMN-Core showed preference for the closest social layer while DMN-MT showed preference for distant social layers.

The large cortical volume dedicated to the closest social layer may reflect the significance of this layer to one’s social world. The closest social layer has been termed the support clique1, since it encompasses the few closest people who are most likely to provide emotional and instrumental support in time of need1,2,4,52,53. The vast majority of one’s social cognitive efforts are dedicated to individuals in the support clique54. Furthermore, social reward provided by individuals in this social layer significantly surpasses reward by individuals in all other social layers and has been shown to be crucial to emotional well-being, stress reduction and healthy social interaction4,55,56. The disproportional cognitive effort and social reward associated with the support clique may account for its extensive representation in the brain despite its small size (Supplementary Fig. 1). The distinction is similar to findings in the most social non-human primates (cercopithecine monkeys and apes)57,58,59, involving essentially the same brain regions60,61.

Examination of the social layer preference distribution along the cortex revealed a pronounced separation between regions that are tuned to the support clique and regions that are tuned to more distant social layers. This separation may reflect the engagement of distinct cognitive mechanisms in assessing personal distance within different social layers. Indeed, participants’ self-reports support this possibility. When asked whether their approach to the question differed between close and distant individuals, 57% of the participants indicated that they relied more on frequency or recency of meetings, or on memories of mutual events, when considering distant individuals, while for closer individuals, they relied more on the spontaneous sentiments evoked when thinking about them (Supplementary Note 1). Differences in the involvement of mechanisms related to episodic memory or temporal cognition are also suggested by the differences in response times, which were significantly shorter for the support clique compared to all other social layers, even among participants who reported using the same strategy for all social layers. Similar findings were reported by Wlodarski and Dunbar23, who found that when participants made decisions about traits of their social network members, reaction times were shorter for those in the innermost social layer (equivalent to our ‘close’ layer) compared to those in more distant social layers. The involvement of mnemonic mechanisms in processing distant social layers is also supported by hippocampal activity in response to these layers, along with the activity in the ventromedial prefrontal cortex, retrosplenial cortex, and parahippocampal cortex, all of which are also implicated in memory processing62,63,64,65,66,67,68,69,70. The variance in the utilization of episodic memory mechanisms across social layers may be linked to the time elapsed since last encountering their members. When asked about the various individuals used in the study, participants reported that, on average, they met individuals from distant social layers less frequently (t = −17.03, p < 0.001) and less recently (t = 3.6, p < 0.001) compared to those from close social layers (Supplementary Note 2). Comparing individuals in more distant social layers may require more effort in recall, as more extensive memory search may be necessary. In contrast, regions exhibiting a preference for processing support clique individuals, including the temporoparietal junction, the precuneus, the dorsomedial prefrontal cortex, and the middle temporal gyrus, are consistently involved in processing the self and the mental states of others15,18,71,72,73,74,75. Hence, the cue-related processing of close individuals may rely more on such self-related processes. The aforementioned reports and reaction times seem to support this hypothesis, but further studies are needed to validate it. It should be noted that self-referential processes are potentially an integration of various common cognitive processes, including some forms of memory; a precise dissection of these processes is beyond the scope of the current work and merits further exploration76. Interestingly, though the activity patterns of layer 2 (sympathy group) were more similar to layers 3–5 than to the support clique, only a few voxels in the medial prefrontal and retrosplenial cortices showed preference for this social layer. Further research should examine the minimal distribution of layer 2 and its relations to the support clique and layers 3–5.

Regarding the relationship between social layers and spatial scales, the support clique was found to activate posterior regions, previously associated with processing of small-scale environments; in contrast, layers 3–5 activated more anterior regions, associated with medium-scale environments. One explanation for the posterior-to-anterior organization of the spatial scales is that small environments are represented in a more precise manner, heavily influenced by visual perception, therefore processed posteriorly, closer to visual areas; larger environments evoke more flexible representations that are processed in higher-level cortical areas26,51,77,78. This view is in accordance with construal-level theory, which posits that distances across different cognitive domains, including the spatial and social domains, share a common underlying meaning: psychological distance, representing the degree of separation from the current egocentric experience79. The cortical processing of the support clique and larger social layers along the small-to-large spatial scale gradient supports these ideas. The association of distant social layers (layers 3–5) with medium spatial scales (neighborhood, city) may signify the use of more flexible representations for processing more distant social layers, though this hypothesis was not explicitly tested. Interestingly, a much weaker association was found between these social layers and large spatial scales (country, continent). This suggests that the representation and processing of such large spaces might involve different cognitive mechanisms. Small and medium spatial scales might be captured through visual perception from a first-person perspective, as vista spaces78,80,81, in contrast to large spatial scales that require map-like representations from a third-person perspective26,51. Since all social layers in our study involve familiar individuals, their processing may relate more closely to the first kind of spaces. It would be intriguing to investigate the processing of less familiar individuals, famous people, or even complete strangers, which might be associated with large spatial scales30,46,82,83. Alternatively, it is possible that the differences between spatial and social processing become more pronounced with larger distances, such that no social parallel can be found for large spatial scales. This hypothesis is supported by the observation that, with the processing of larger spatial scales, cortical activity shifts anteriorly, moving away from the retrosplenial cortex and the parietooccipital sulcus, regions involved in processing both spatial and social distances11,26. Further research is needed to decide between these hypotheses.

The parcellation of the cortex into networks of functional connectivity may provide further explanation for the observed bipartition. Personal closeness processing significantly involved all main hubs of the DMN, a cortical network known for its activity during self-referential and internal mentation27,33,34. While the distribution of support clique preference primarily aligned with the DMN-Core subsystem, the distribution of distant social layers predominantly engaged the DMN-MT subsystem. Though the functional difference between these subsystems is not fully understood, tasks involving theory of mind, or mentalizing - that is, the attribution of mental states such as thoughts, beliefs, and desires to others84 - consistently involve DMN-Core, while DMN-MT more commonly supports mnemonic and spatiotemporal processes37. Likewise, when cortical parcellation is defined at the individual subject level, one subsystem of the DMN (similar to DMN-Core) is consistently implicated in theory of mind, while another subsystem (resembling DMN-MT) is implicated in episodic memory85,86. Our results corroborate and extend this by indicating a differential recruitment of the two subsystems according to social layer. The association of distant social layers with the DMN-MT subsystem may be attributed to episodes memory; nonetheless, its presence across all participant groups, irrespective of judgment strategy (Supplementary Note 1), in tandem with the association of distant social layers with larger spatial scales, reinforces the view that DMN-MT supports the organization and manipulation of relational knowledge, beyond the representation of spatial or temporal scales45,87,88,89. Concurrently, the DMN-Core subsystem, associated with mentalizing37, exhibited preference for the support clique. This finding aligns with previous studies indicating that DMN-Core responds more strongly to the mental states of close individuals compared to those of distant ones90,91. It should be noted, however, that broader definitions of mentalizing have been suggested92, and several meta-analyses have found that this function involves wide cortical areas, including virtually all social layer sensitive regions93,94. Interestingly, regions that were here implicated by distant social layers (such as the retrosplenial and ventromedial prefrontal cortex), were previously more strongly activated by tasks that require ‘cognitive’ mentalizing (e.g., ‘false belief’ tasks) while regions associated with the support clique (such as the temporoparietal junction) were related to ‘affective’ mentalizing (e.g., ‘reading the mind in the eyes’), supporting this distinction.

Finally, the cortical distribution of social layers in relation to the general activity during personal closeness processing (Fig. 4A) may also suggest an organizing principle for the cognitive mechanisms involved. The ventral portion of the precuneus was significantly engaged during personal closeness processing regardless of social layer. While the adjacent dorsal precuneus showed preference for the closest social layer, adjacent ventral regions including the retrosplenial cortex showed preference for remote social layers. Interestingly, an analogous similar division may be found in the spatial and temporal domains, where the dorsal and ventral parts of the medial parietal lobe process small and large scales, respectively12,26. This pattern, also evident in the medial prefrontal cortex, implies a dorsal-ventral gradient along the medial wall, where dorsal regions are engaged for processing close layers and ventral regions are recruited for processing distant layers, according to task demands. A similar gradient was discovered in an analysis of global connectivity based on a large dataset of brain activity at rest95,96. While the first gradient (i.e., the gradient that explained the most variance) positioned the regions implicated in personal closeness processing at one end of the spectrum and surrounding dorsal and ventral regions at the other end, the third gradient was distributed along the dorsal-ventral axis of the medial cortical surface. This pattern is also associated with gradual changes in hippocampal connectivity, with greater concentration of fornix fibers in more ventral regions, consistent with the implication of the hippocampus in processing distant social layers97.

Inevitably, our study has limitations. Since people are different in their social attributes, the definitions of each social layer may differ between participants. For example, one participant may consider the frequency of contact to be a major factor in identifying close friends, while for another, this factor may hold little significance. However, this general measure of psychological distance has proven itself reliable in many studies4,11,20,90. To obtain valid representations of participants’ social networks, we did not provide exact definitions for each social layer but instead allowed them to choose the definitions that best fit their network. Inconsistencies in the social layer definitions between participants may thus introduce error variance. It is possible that some of the intertwining between the brain distribution of social layer preference stem from this source of variance. Nevertheless, the clear distinction between the support clique and the distant social layers is robust and in line with previous findings as described above. Other consequences of the flexibility in social network structure are the integration of kin and non-kin, and the potential variance between social layers in terms of age, group affiliation or other factors. This may be important due to the difference in social information processing between friends and family23, and should be investigated in future studies. Finally, the comparison between social layers and spatial scales was performed using two different cohorts. While these cohorts were similar in terms of age and occupation, identification of the social layers and spatial scales in the same cohort is required to validate our findings.

In conclusion, we have presented behavioral and neuroimaging evidence supporting the division of the brain system underlying social orientation into two separate networks - one for processing very close individuals and another for the broader social network. This division corresponds to distances in the spatial and presumably other cognitive domains, illuminating potential global organizational principles across the human cortex. Our findings emphasize the unique role close relationships play in shaping our social lives and highlight the diverse array of cognitive processes that contribute to our complex and multifaceted social world.

Methods

Participants

Twenty-one healthy participants (10 women, mean age = 26.1 ± 4.3 years) participated in the study. Five additional participants were excluded from the study; four due to an inability to complete the paradigm, and one due to missing more than 5% of all trials. All participants provided written informed consent, and the study was approved by the ethical committee of the Hadassah Hebrew University Medical Center in conformity with the Declaration of Helsinki (2013). All ethical regulations relevant to human research participants were followed.

Acquisition of social network data

A week before the experiment, participants were asked to provide names of people with whom they have an active relationship, and to classify them into four categories representing the first four social layers (“closest people”, “close friends”, “friends”, “people I am in contact with”). Participants were exposed only to the social layer descriptions (e.g., ”closest people”) and not to the numbers (“layer 1”) or titles (“support clique”) of the social layers. Participants were encouraged to use their phone contact list or social media to avoid forgetting anyone within these social layers. Participants who failed to provide at least four names in each social layer were excluded from the experiment. For the fifth social layer, participants were asked to provide the names of 30 acquaintances with whom they do not have an active relationship (Fig. 1A).

Experimental design

Upon arrival at the neuroimaging unit, participants were informed that they are about to perform a task in which they will choose between different names of individuals from their own social network. They were instructed to respond according to their personal closeness to each individual. To ensure participants’ comprehension of the task, a training trial was administered prior to the scanning session, using stimuli identical to the experimental ones (as described below) but with fictional names. Following the completion of the training, all participants confirmed their comprehension of the task.

During the fMRI scan, participants performed a social orientation task11,20,39, in which they were presented with textual stimuli containing pairs of names of individuals from their social network, both belonging to the same social layer, and were asked to choose the name of the individual who was closer to them (Fig. 1B). Participants indicated their response by pressing either the left or the right button of an MRI-compatible button box. Stimuli were presented in a randomized block design, each block containing names of individuals from a single social layer. Before each block, the question “Who is closer to you?” appeared on the screen for 5 s. Next, four pairs of names were sequentially presented, each for 2.5 s, with no interstimulus interval. Each block was followed by a fixation cross for 5 s. Participants were instructed to respond accurately but as fast as possible. The experiment consisted of five experimental runs for each participant, each run containing 20 blocks in randomized order (four blocks for each of the five social layers). In addition, after the third experimental run, participants completed a lexical control task. In this task, the same stimuli were presented, but participants were asked to indicate which of the two names is closer to a target name in alphabetical lexical order. All stimuli were presented using the Presentation software (Version 18.3, Neurobehavioral Systems, Inc, Berkeley, CA, www.neurobs.com, RRID: SCR_002521). In total, participants performed 400 comparisons of personal closeness during the experiment. All participants, except one, responded to over 95% of the stimuli. One participant did not meet this criterion and was excluded from subsequent analyses.

To rule out the possibility that participants’ responses were influenced by their physical proximity to individuals, participants completed a post-scan questionnaire in which were asked, among other questions, to describe their understanding of the question “who is closer to you”. Mentioned factors included emotion (76.19%), frequency of contact or time since the most recent encounter (52.38%), familiarity of the individual with the participant’s life (33.33%), empathy (9.52%), significance to the participant’s life (9.52%) and content of recent encounters (4.76%). Notably, no participant mentioned that physical distance influenced their decisions.

Behavioral analysis

For each participant, the average response time in each social layer was calculated. To examine whether there are differences in response times across social layers, repeated-measures ANOVA was conducted with average response time as the dependent variable, using social layer as a fixed factor and participant as a random factor. Subsequently, Tukey–Kramer post hoc test was performed to identify the specific social layers that contributed to the observed effect. To evaluate the effect of strategy used by participants on response times (Supplementary Note 1), we conducted a repeated-measures ANOVA with (A) judgment strategy and (B) the interaction between social layer and judgment strategy as additional fixed factors. Subsequently, we conducted a simple effects analysis, examining the effect of social layer on response times for each strategy, followed by Tukey–Kramer post hoc tests.

MRI acquisition

Participants were scanned in a 3-T Siemens Skyra MRI at the Edmond and Lily Safra Center neuroimaging unit. BOLD contrast was obtained with an EPI sequence (repetition time [TR] = 2500 ms; echo time [TE] = 30 ms; flip angle = 72°; field of view = 192 mm; matrix size = 96 × 96; voxel size = 2 × 2 × 2 mm; 60 slices, multi-band acceleration factor = 2, interleaved acquisition order; 164 volumes per run). In addition, T1-weighted high-resolution (1 × 1 × 1 mm, 160 slices) anatomical images were acquired for each participant using the magnetization prepared rapid gradient echo (MPRAGE) sequence (TR = 2300 m, TE = 2.98 ms, flip angle = 9°, field of view = 256 mm).

MRI preprocessing

fMRI data were processed and analyzed using the BrainVoyager 20.6 software package (R. Goebel, Brain Innovation, Maastricht, The Netherlands, RRID:SCR_013057), Neuroelf v1.1 (www.neuroelf.net, RRID:SCR_014147), and in-house Matlab (Mathworks, version 2020a, RRID:SCR_001622) scripts98. Pre-processing of functional scans included slice timing correction (cubic spline interpolation), 3D motion correction by realignment to the first run image (trilinear detection and sinc interpolation), high-pass filtering (up to two cycles), smoothing (full width at half maximum (FWHM) = 4 mm), exclusion of voxels below intensity values of 100, and co-registration to the anatomical T1 images. Anatomical brain images were corrected for signal inhomogeneity and skull-stripped. All images were subsequently normalized to the Montreal Neurological Institute (MNI) space (trilinear interpolation) and downsampled to a resolution of 3 × 3 × 3 mm3.

fMRI analysis

Estimation of cortical responses to each social layer

A GLM analysis99 was applied at each voxel, where predictors corresponded to the five social layers. Each predictor included all experimental blocks of a single social layer, with each block being modeled using a boxcar function. Predictors were convolved with a canonical hemodynamic response function, and the model was fitted to the BOLD time-course at each voxel. To account for noise in the signal, motion parameters as well as average white matter and CSF signals were included as nuisance regressors in the GLM. White matter and CSF signals were obtained by first creating masks of white matter and CSF for each participant using BrainVoyager (intensity >150 for the white-matter mask; intensity <10 and a bounding box around the lateral ventricles for the CSF mask), then averaging the EPI time series data within these masks. Data were corrected for serial correlations using the AR(2) model and transformed to units of percent signal change. Subsequently, a random-effects analysis was performed across all participants to obtain group-level beta values for each predictor.

Identification of brain regions with social layer sensitive activity

To identify brain regions exhibiting variations in brain activity in response to various social layers, we first extracted the beta values of social layer specific predictors from each voxel across all participants. For each voxel, we conducted a single-factor repeated-measures ANOVA on these beta values, with social layer as the single within-subjects factor. Results were corrected for multiple comparisons at the voxel level using FDR for a corrected p < 0.01. In addition, a cluster extent threshold of volume > 37 voxels of 3 × 3 × 3 mm3 was applied, to yield a whole-brain alpha of p < 0.01, determined using a Monte-Carlo simulation with 10,000 iterations. Results were then clustered using BrainVoyager, resulting in twelve clusters of significant sensitivity to social layer. The ANOVA was later repeated, with judgment strategy and its interaction with social layer included as additional between-subjects factors (Supplementary Note 1); results remained unchanged, with the additional factors found to be non-significant.

Beta values from identified voxels were used to determine ‘preference’ for social layers (i.e., the social layer for which a voxel is most active) using two methods. Firstly, beta values for each social layer were averaged across participants and the social layer showing the maximal activity was selected. Social layer preference was verified by comparing brain activity for the chosen social layer to the mean activity for all other social layers across all subjects (paired-sample t-test, voxel-level FDR-corrected p < 0.05); 97% of the voxels that were found significant in the ANOVA also met this threshold and were used in further analyses. Secondly, to account for neural responses, which can be characterized by Gaussian-shaped tuning curves, the preferred social layer was estimated using a Gaussian distribution26. For each voxel of each participant, separately, five beta values corresponding to the five social layers were extracted. After subtracting the minimum beta value, five datapoints were created with social layer numbers (1–5) as x-coordinates and adjusted beta values as y-coordinates. A Gaussian curve was fitted to the datapoints using MATLAB. The Gaussian model was defined as \(y=a{e}^{-{\left(\frac{x-b}{c}\right)}^{2}}\), where \(a\) is the amplitude, \(b\) is the mean, and \(c\) is the standard deviation. The fitting constraints were \(0\le a\le 100\); \(-100\le b\le 100\); \(0\le c\le 100\). Subsequently, voxels were selected to include only those where the Gaussian curve fit to the data points surpassed a threshold of r2 > 0.8. Within these voxels, the x-coordinates of the Gaussian curve peaks, representing the numbers of the preferred social layers, were averaged across participants after removing outliers (i.e., curves whose peaks fell outside three standard deviations from the mean. Note that they constituted only <0.1% of the data). The resulting means were rounded to the nearest integer to determine the preferred social layer. All analyses were performed using a volume-based approach, with the results projected onto the cortical surface solely for visualization purposes.

Hierarchical clustering analysis

To explore the relationship between social layer sensitive brain regions and the social layers, hierarchical clustering was conducted using the MATLAB clustergram function. In each region of interest (ROI, i.e., cluster of social layer sensitivity, as detailed in Table 1), beta values were averaged across voxels to yield a single value for each participant and social layer. Averaged beta values were subsequently standardized across all ROIs. Clustering was carried out initially based on social layers and subsequently on ROIs, using Euclidean distances and the average linkage method.

Comparison between social and spatial scale sensitive regions

Using data from a previous fMRI study on spatial scales26, we calculated the overlap between the brain ROIs that activate differentially according to social layers and those that do so according to spatial scales. In that study, nineteen healthy participants (7 women, mean age = 27.7 ± 4.4 years) were presented with textual stimuli containing pairs of names of places within a single scale of proximity (room, building, neighborhood, city, country and continent), and were subsequently asked to choose the one closest to a target. Brain activity during processing of different spatial scales was analyzed using voxel-wise GLM and single-factor repeated-measures ANOVA with spatial scale as the single within-subjects factor. In voxels that exhibited variations in activity across various spatial scales (spatial scale sensitive voxels), beta values for each scale were averaged across participants, and the scale showing the maximal activity was subsequently selected. Thus, we obtained six brain masks representing sensitivity to six spatial scales, which could be compared with the five brain masks representing sensitivity to five social layers. Overlap between spatial scale sensitive and social layer sensitive voxels was calculated for each of the six spatial scales and each of the five social layers using the Jaccard index (intersection over union). To assess the statistical significance of the overlap, the scale sensitive voxels were shuffled 100,000 times (randomizing their location within the spatial scale sensitive regions), each time calculating the Jaccard index for each social-spatial pair. Subsequently, the original Jaccard index was compared to the results of the permutations to obtain a p value. All p values were corrected for multiple comparisons using the FDR method. Additionally, we calculated the overlap between social layer sensitive voxels and two brain masks, representing regions that were previously implicated in orientation in the spatial and social domains, regardless of scale or layer11.

Comparison between the social task and the lexical control task

Regressors for the lexical control task were added to the social layer predictors in the GLM analysis, and a new design matrix was computed for each participant. A group analysis (corrected for serial correlations, AR(2)) was performed for each voxel, and activity during the social task in all five social layers was contrasted with the activity during the lexical control task.

Assessment of the distribution of social layer preference across large-scale resting-state networks

A whole-brain parcellation of seven large-scale brain networks was used as a template to identify resting-state networks36. We calculated the overlap between the brain volume involved in personal closeness processing (derived from comparing the social task with the lexical control task) and each of the seven cortical networks using the Jaccard index. Similarly, we used the Jaccard index to estimate the overlap between each cortical network and the aggregate of all social layer sensitive clusters. To examine social layer preference in the DMN subsystems, we used three resting-state networks derived from a cortical parcellation of 17 brain networks36 and identified as subsystems of the DMN37. Statistical significance was evaluated using a permutation test as described above for the spatial and social layer sensitive regions (see Comparison between social and spatial scale sensitive regions), and an FDR-correction was applied.

Statistics and Reproducibility

Statistical analyses were conducted as described above.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.