A specific brain network for a social map in the human brain

Abstract

Individuals use social information to guide social interactions and to update relationships along multiple social dimensions. However, it is unclear what neural basis underlies this process of abstract “social navigation”. In the current study, we recruited twenty-nine participants who performed a choose-your-own-adventure game in which they interacted with fictional characters during fMRI scanning. Using a whole-brain GLM approach, we found that vectors encoding two-dimensional information about the relationships predicted BOLD responses in the hippocampus and the precuneus, replicating previous work. We also explored whether these geometric representations were related to key brain regions previously identified in physical and abstract spatial navigation studies, but we did not find involvement of the entorhinal cortex, parahippocampal gyrus or the retrosplenial cortex. Finally, we used psychophysiological interaction analysis and identified a network of regions that correlated during participants’ decisions, including the left posterior hippocampus, precuneus, dorsolateral prefrontal cortex (dlPFC), and the insula. Our findings suggest a brain network for social navigation in multiple abstract, social dimensions that includes the hippocampus, precuneus, dlPFC, and insula.

Hippocampal Place-like Signal in Latent Space

“…The posterior hippocampus in particular seems to encode directional vectors from the animal's current location to goal locations (Sarel et al, 2017). Similarly, as participants make their decisions in our social navigation task, the left mid/posterior hippocampus parametrically tracks the directional angle from the participant's theoretical point-of-view to the updated social places of the characters (Tavares et al, 2015;Zhang et al, 2022)-a finding we replicated here in our validation sample. Previous fMRI work in physical space mirrors this long-axis distinction: the anterior hippocampus represents locations and their global properties, whereas the posterior hippocampus represents directions and detailed trajectories (Howard et al, 2014;Javadi et al, 2017;Kim et al, 2017).…”

“…The angle analyses' hippocampal clusters, as well a cluster from a recent replication (Zhang et al, 2022 , were more posterior than the place searchlight's cluster (see fig. 4).…”

“…Two previous studies found that left hippocampal BOLD activity correlated with the angle of the vector from the participant's theoretical point-of-view to the current character's social place (Tavares et al, 2015;Zhang et al, 2022). This analysis assumes that while making decisions, participants represent the characters' locations in an "egocentric" fashion: from their own first-person, subjective point-ofview.…”

“…To replicate the effect in the validation sample, we created an ROI from the results of the previous two studies. We re-ran the analysis in the initial sample and combined the resulting peak left hippocampal coordinate with the peak left hippocampal coordinate reported in Zhang et al (2022) to generate a sample size weighted coordinate around which we defined a spherical ROI (radius = 5mm). This ROI was used to test the validation sample's parametric modulator contrasts for significant positive effects against 0 with a 1-sample t-test.…”

“…In the validation sample, we replicated the left hippocampal effect of the trial-wise social angle to the character from the theoretical point-of-view of the participant. To first generate an ROI to test for this effect in the validation sample, we combined the peak left hippocampus voxel from a re-analysis of the initial sample (peak: -24/-22/-22; t(17)=5.38, p FWER =0.011, with a p<0.05 cluster extent=18 3mm 3 voxels) and the peak left hippocampus voxel from a recently published replication (Zhang et al, 2022) to calculate a sample size weighted average peak coordinate (-23/-26/-11), around which we defined a spherical ROI. Then, using this ROI, we replicated the angle parametric modulation effect in the validation sample (peak: -25/-24/-7; t(31)=3.02, p FWER =0.026, p<0.05 cluster extent=18 2.1mm 3 voxels).…”