Variability in Brain Structure and Function Reflects Lack of Peer Support

Abstract

Humans are a highly social species. Complex interactions for mutual support range from helping neighbors to building social welfare institutions. During times of distress or crisis, sharing life experiences within one's social circle is critical for well-being. By translating pattern-learning algorithms to the UK Biobank imaging-genetics cohort (n = ~40 000 participants), we have delineated manifestations of regular social support in multimodal whole-brain measurements. In structural brain variation, we identified characteristic volumetric signatures in the salience and limbic networks for high- versus low-social support individuals. In patterns derived from functional coupling, we also located interindividual differences in social support in action-perception circuits related to binding sensory cues and initiating behavioral responses. In line with our demographic profiling analysis, the uncovered neural substrates have potential implications for loneliness, substance misuse, and resilience to stress.

Keywords: Bayesian hierarchical modeling; machine learning; population neuroscience; salience network; social brain.

Figure 1

Gray matter variation across specific large-scale brain systems explains strong effects related to social support (SS). At the network level, our Bayesian hierarchical modeling framework directly estimated the varying effects of entire brain networks in explaining high versus low SS in the UK Biobank participants. The fully probabilistic modeling approach allowed volume variation effects to be estimated jointly in separate brain regions (see Fig. 2) and spatially distributed networks of constituent brain regions (shown in this figure). In rough analogy to ANOVA, the network definitions could be viewed as factors and the region definitions could be viewed as continuous factor levels. This analysis tactic enabled quantifying the extent to which spatially dispersed regional variation in gray matter volume can be coherently explained by differences among major brain networks (Bzdok et al. 2020; Kiesow et al. 2020). Histograms show marginal posterior distributions of the overall explanatory variance (sigma parameter) for each brain network (volume measures in standard units). Horizontal black bars indicate the highest-posterior density (HPD) interval of the model’s network variance parameters, ranging from 10 to 90% probability. Population-level volume variation in the salience and limbic networks emerged as preferentially linked to interindividual differences in SS. Note that the subcortical network is not shown in the cortical surface view.

Figure 2

Top brain regions that explain gray matter differences related to SS. At the region level, our Bayesian hierarchical modeling framework identified for which brain regions variability in gray matter volume explains the level of SS reported by the participants. Strongest associations to day-to-day SS (cf. Fig. 1) were determined based on effect sizes (mean parameters) of the marginal posterior parameter distributions (volume measures in standard units). Key region associations were located in parts of the salience network, including the anterior insula and anterior/mid cingulate cortex. Additional neural substrates of SS were located in regions of the limbic network, including the OFC. Red (blue) color indicates positive (negative volume) effects related to regular SS, for the top 10 regions found in this analysis. Abbreviations: ACC = anterior cingulate cortex; dlPFC = dorso-lateral prefrontal cortex; MCC = midcingulate cortex; OFC = orbitofrontal cortex; SMA = supplementary motor area; SMG = supramarginal gyrus; Temp. Pole = temporal pole; TPJ ant = anterior portion of the TPJ. The subcortical system is not shown in this view.

Figure 3

Leading functional coupling signature of SS suggests up-regulated action-perception systems and down-regulated interplay with internal-cognition systems. Functional connectivity shifts are shown for the dominant population mode related to everyday SS (connectivity relevancies in standard units). Statistical significance of this population mode of coherent functional coupling differences in high versus low SS participants was determined by nonparametric permutation testing (P < 0.05). Red connectivity links indicate compounded functional coupling for individuals with high amounts of regular SS, which suggests an up-regulation of “here-and-now” related networks, including the salience, dorsal attention, somatomotor, visual networks. Blue connectivity links indicate reduced coupling between regions in high SS.

Figure 4

Demographic profiling analysis identifies lifestyle factors related to brain substrates of SS. Multivariate pattern-learning (cf. Methods) was used to explore how the top brain regions (see Fig. 2) are linked to a variety of behavioral indicators in high-SS versus low-SS individuals. Behavioral markers covered domains of mental and physical well-being, lifestyle choices, and social embeddedness. In 1000 bootstrap resampling iterations, our entire pattern-learning pipeline in gray matter volume was repeated separately in the two participant groups: UK Biobank participants who regularly share life experience with close others and those with little such exchange of personal events. The computed differences in brain-behavior associations between both groups (i.e., diverging canonical vector entries) were gathered across the 1000 perturbed redraws of our original dataset to obtain faithful bootstrap intervals at the population level. Note that in the context of the used quantitative modeling framework, age and sex can show relevant effects in conjunction with other behavioral indicators, even if age/sex-related brain variation has been removed in a preceding deconfounding step. The derived estimates of uncertainty directly quantified how group-related deviations vary in the wider population. Asterisks indicate statistical relevance based on excluding zero between the 5/95% quantiles of the bootstrap distribution (cf. Methods, Supplementary Table S3). The boxplot whiskers show the interquartile range (i.e., 25/75% interquartile distance). The highlighted divergences between individuals with weak versus strong SS reveal characteristics of these population strata. Among them were multiple indicators of social embeddedness, health and substance use, as well as factors related to emotional tenseness and distress. An analogous analysis based on functional connectivity did not yield any statistically relevant brain-SS associations. This configuration of brain-behavior differences in high versus low SS speaks to multifaceted manifestations of stress-buffer capacities.