Distance and grid-like codes support the navigation of abstract social space in the human brain

The study investigates whether the human brain’s spatial navigation codes (distance and grid-like representations) generalize to navigation within an abstract social space. Grounded in the stereotype content model, social perception is structured by two core dimensions—warmth/trustworthiness and competence—which can be organized as a two-dimensional social cognitive map. Prior work on grid cells in entorhinal cortex and associated fMRI evidence shows hexadirectional modulation supporting spatial and nonspatial cognitive maps. The authors hypothesize that similar neural coding mechanisms (distance and grid-like codes) support representation and navigation in a continuous social value map and relate to social decision-making and traits.

Foundational work in spatial navigation shows entorhinal grid cells forming periodic hexagonal lattices with shared orientations among neighboring cells and directionally modulated firing patterns (Hafting et al., 2005; Sargolini et al., 2006). Human fMRI work detects grid-like codes via hexadirectional modulation aligned to putative grid orientation during virtual and imagined navigation (Doeller et al., 2010; Horner et al., 2016; Bellmund et al., 2016). Grid-like codes extend beyond physical space to conceptual, semantic, perceptual, and social hierarchy domains (Constantinescu et al., 2016; Viganò & Piazza, 2020; Park et al., 2021), suggesting a domain-general role in organizing cognitive maps. The precuneus and retrosplenial/posterior cingulate regions have been implicated in encoding path or goal distance and changes in social relationships (Patai et al., 2019; Tavares et al., 2015). These bodies of work motivate testing whether distance and grid-like codes underpin navigation of a continuous, orthogonal 2D social value space (competence and trustworthiness).

Participants: Forty-four participants (18 males, mean age 21.59 ± 2.56) completed behavioral training and fMRI scanning; exclusions for excessive motion (n=3) and low accuracy in scanner (<60%; n=3) left 38 participants for analyses (15 males, mean age 21.47 ± 2.64). Ethics approval was obtained; informed consent collected; participants compensated. Social value map: A continuous 2D space (0–1 on each axis) with competence (profit rate) and trustworthiness (return rate) operationalized via an investment game. Six avatars (faces) served as landmarks, with coordinates sampled around predefined centers (radius 1/30 units). Visual analog: a square box containing two adjacent bars whose heights encode competence and trustworthiness. Behavioral tasks: - Match: learn nonspatial controller mapping to morph bars to a target configuration. - Explore: free search to find avatars; avatars appear when within 0.01-unit radius of their location. - Collect: navigate from random starts to match the target avatar’s bar configuration; measures include first transition deviation from ideal trajectory and Euclidean distance error. - Recall: view 1 s bar morph at a predefined direction (ratio), then imagine continuation for equal time and speed; select the correct outcome from three options; 80 trials per block; trajectories uniformly sampled in direction; completed outside (2 blocks/session) and inside the scanner (4 blocks total, 320 trials). - Compare: pairwise comparisons on competence, trustworthiness, and willingness to cooperate (equal weighting of both dimensions); test distance effects on accuracy and response time. - Rank and Map tasks: rank avatars by dimensions; at end, place avatars on an explicit empty map by mouse click. MRI acquisition: 3T SIEMENS Prisma, 64-channel head coil; T2*-weighted functional images (TR=2000 ms, TE=30 ms, 33 slices, 3 mm, FoV=224 mm, voxel 3.5×3.5×3.5 mm); field map; T1-weighted structural (TR=2530 ms, TE=2.98 ms, voxel 0.5×0.5×1 mm). Preprocessing (SPM12): realignment, slice-timing correction, field map distortion correction, coregistration, segmentation and normalization to MNI, smoothing with 6 mm FWHM; explicit intracranial mask used to avoid signal loss in frontal and entorhinal regions. fMRI analyses: - GLM1 (quadrature filter functional localizer): morph and choice regressors; parametric modulators sin(6θ) and cos(6θ); F-tests for hexagonal modulation; Z-statistics used as modulation index. - Grid orientation estimation and cross-validated hexagonal consistency (GLM2): derive ROI grid orientation from averaged beta weights (φ = arctan(βsin/βcos)/6) using three-run estimating sets; in the held-out run, bin trials by alignment offset (12 bins; aligned vs misaligned); contrast aligned > misaligned tested at whole-brain level; sixfold specificity tested against 4-, 5-, 7-, 8-fold controls. ROIs: 5 mm spheres around GLM1 peaks (including right frontal pole); anatomical entorhinal subdivisions (pmEC, alEC; bilateral) per Maass et al., 2015. - GLM3 (distance code): parametric modulation of morph-stage regressor by traveled Euclidean distance; whole-brain tests for positive/negative correlations. EC ROI analyses: both univariate (GLM1 Z, GLM2 align>misalign) and multivariate RSA approaches. RSA used LSS per-trial estimates, orientation-independent similarity model (hexadirectional angular difference), and orientation-dependent aligned vs misaligned pattern similarity; FDR corrections applied. Behavioral analyses: Linear mixed-effects models for session effects; time-at-edges and time-at-avatars; collect and recall performance; compare task distance effects with random slopes per participant; pre- and post-experiment face ratings modeled with time, avatar characteristic, and their interaction; rank and map task accuracy. Behavioral relevance of neural codes: correlations between consistency effects and recall performance (scanner), and whole-brain covariate analyses with compare-task distance effects (cooperation) and social anxiety/avoidance scores.

- Participants constructed an internal 2D social value map and improved navigation/memory across sessions: faster explore times (βsession = −67.589, t = −37.604, p < 0.001); decreased time at edges (β = −0.104, t = −9.070, p < 0.001); increased time at avatars (βsession = 0.023, t = 7.367, p < 0.001). Collect task improved (first transition deviation <15°: βsession = 0.0504, t = 3.219, p = 0.002; distance <0.01 units: βsession = 0.047, t = 3.002, p = 0.003). Recall accuracy improved (βsession = 0.020, t = 4.213, p < 0.001). Compare task exhibited robust distance effects on accuracy and response time (all p < 0.001). - Distance code during social navigation (GLM3): precuneus activity positively correlated with traveled Euclidean distance (R: [12, −56, 26], t ≈ 4.53, PFWE ≈ 0.054; L: [−12, −54, 18], t ≈ 4.95, PFWE ≈ 0.082). Negative correlations with distance observed in bilateral fusiform gyrus (R: [36, −48, −20], PFWE < 0.001; L: [−38, −56, −10], PFWE = 0.008) and right middle occipital gyrus ([42, −80, 8], PFWE = 0.018). - Hexagonal modulation functional localizer (GLM1) identified regions sensitive to sixfold modulation, including right superior parietal gyrus (peak t = 5.021, cluster PFWE < 0.001), left precuneus (t = 5.707, PFWE < 0.001), right middle frontal gyrus (t = 5.221, PFWE < 0.001), right paracentral lobule (t = 4.835, PFWE < 0.001), right middle frontal gyrus (t = 4.849, PFWE = 0.001), bilateral frontal pole (L: t = 4.269, PFWE = 0.014; R: t = 4.494, PFWE = 0.017), and left angular gyrus (t = 4.718, PFWE = 0.041). - Cross-validated hexagonal consistency aligned to putative grid orientations (GLM2): alignment to right frontal pole (FP) orientation revealed sixfold-specific consistency in bilateral vmPFC (e.g., [−14, 56, −12], cluster PFWE = 0.033) and left entorhinal/parahippocampal region ([−16, −22, −26], PFWE = 0.024). Alignment to right posterior-medial EC (pmEC) orientation yielded sixfold-specific consistency in bilateral vmPFC (right rectus: [10, 34, −18], PFWE = 0.033; left ACC: [−8, 32, 6], PFWE = 0.009) and right superior temporal pole ([36, 20, −30], PFWE = 0.018). No robust evidence for EC grid-like code aligned to its own orientation in ROI-based analyses. - Specificity: sixfold effects exceeded 4-, 5-, 7-, 8-fold controls; controlled periodicities showed no significant modulation. - Behavioral relevance: greater compare-task distance effects (cooperation) correlated positively with hexagonal consistency in temporal lobe. For right FP orientation: left middle temporal gyrus (peak t = 4.528, cluster PFWE < 0.001) and right lingual gyrus (t = 4.369, PFWE < 0.001). For right pmEC orientation: bilateral lingual gyrus (L: t = 4.355, PFWE = 0.077; R: t = 3.934, PFWE = 0.032). Hexagonal consistency in left precuneus/PCC ([0, −64, 28], t = 4.965, PFWE < 0.001) was negatively correlated with social avoidance scores. No significant correlations between grid/distance indices and recall performance in-scanner.

Findings indicate that the human brain recruits spatial navigation-like codes to represent and traverse a continuous social value map defined by competence and trustworthiness. The precuneus exhibits positive distance coding, consistent with its role in tracking route lengths and social relationship changes, while occipitotemporal decreases with distance may reflect reduced engagement of landmark-related visual processing as abstract traversal increases. Grid-like codes appear in vmPFC and temporal regions and show orientation alignment to prefrontal and entorhinal-derived orientations, supporting a distributed, domain-general mechanism for organizing abstract social knowledge. Although EC did not show robust alignment to its own orientation under stringent tests, hexagonal consistency scaling with decision distance effects suggests that grid-like representations contribute to leveraging map structure for social decisions (e.g., cooperation preferences). The negative relationship between precuneus hexagonal consistency and social avoidance hints at links between map-like coding and social well-being, possibly mediated by default mode/prefrontal networks. Overall, results extend grid and distance coding beyond physical space to social cognition, aligning with conceptual and semantic mapping literature.

The study demonstrates that humans can construct and navigate a continuous two-dimensional social cognitive map, and that distance and grid-like neural codes (notably in precuneus, vmPFC, and temporal regions) support such navigation. Grid-like consistency relates to behavioral utilization of map distance in social decisions and to social avoidance traits, underscoring the relevance of spatial-like coding to social functioning. These findings bolster the domain-general view of cognitive maps and suggest that spatial coding principles underpin flexible inference and decision-making in abstract social spaces.

1) Visual analog-driven bar morphing may confound abstract grid-like coding with changes in sensory/visual processing; future designs should better control visual stimulus changes. 2) The scanner task involved navigation in a static social environment without explicit social interaction or decision-making; generalization to dynamic social contexts remains to be tested. 3) The social map was deliberately orthogonal and equally scaled (two dimensions, 0–1), which may lack ecological generality; real social spaces are likely nonuniform, high-dimensional, with non-orthonormal and possibly correlated dimensions, potentially distorting grid periodicity. 4) EC signals may suffer low tSNR, and despite checks, robust EC self-aligned grid-like effects were not observed, leaving origins of prefrontal alignment effects unresolved. 5) Some anatomical peak discrepancies and marginal significance for precuneus distance effects suggest caution in interpretation and the need for replication.