Dissecting the midlife crisis: disentangling social, personality and demographic determinants in social brain anatomy

The study investigates how social behaviors, personality dispositions, and demographic factors uniquely relate to the anatomy of the human social brain during midlife (approximately ages 40–70). Midlife is marked by changes in social networks, prioritization of close relationships, and occupational milestones. Prior neuroimaging work suggests structural alterations in frontal regions, especially the medial prefrontal cortex (mPFC), during healthy aging, and links the mPFC to domains of social cognition, personality, and socioeconomic status. The authors hypothesized that lifestyle traits across social, personality, and demographic domains would show prominent associations with gray matter volume, particularly in the mPFC and medial temporal limbic regions, and sought to disentangle these effects using a fully Bayesian, region-by-region framework in a large UK Biobank cohort.

The paper synthesizes evidence that: (1) social networks narrow in midlife with greater emphasis on close ties, influencing well-being; (2) age-related gray matter reductions are prominent in frontal and prefrontal regions, though not uniformly across the brain; (3) the mPFC supports multiple social cognitive processes, including self-referential processing and mentalizing, and shows age-related functional changes; (4) personality traits relate to social behavior and brain structure—e.g., neuroticism has been linked to mPFC gray matter loss while conscientiousness and extraversion have been associated with larger prefrontal volumes; (5) demographic factors (e.g., socioeconomic status, financial hardship) associate with structural differences in mPFC, hippocampus, and amygdala; and (6) prior studies are often limited by small samples and siloed measures. This motivates a comprehensive, comparative analysis of multiple lifestyle traits within the same model to isolate their unique brain associations.

Data: 9,939 UK Biobank participants (48% male, 52% female), ages 40–69 (mean 55, SD 7.5), all scanned at the same site (Cheadle) on a 3T Siemens Skyra using standardized protocols. T1-weighted 3D MPRAGE at 1 mm isotropic resolution. Preprocessing included defacing, automated QC, gradient distortion correction, brain extraction (BET), registration (FLIRT/FNIRT), tissue segmentation (FAST), and SIENAX for volumetric measures normalized for head size; additional deconfounding removed head size and BMI effects from regional volumes. Brain features: Gray matter volume was extracted using a 36-region social brain atlas (derived from meta-analysis of ~4000 fMRI studies) spanning four networks: visual-sensory, limbic, intermediate, and higher-associative. A 5 mm FWHM smoothing was applied; volumes were averaged within 5 mm spheres at each atlas coordinate, then z-scored across participants. Traits: 40 lifestyle indicators from UK Biobank, binarized for comparability, grouped into social (12), personality (15), and demographic (13) domains (e.g., living with others, friendship satisfaction, morning/evening chronotype, neuroticism, income, job type, health satisfaction). Modeling: For each of the 36 regions, a Bayesian multiple regression modeled regional volume as a function of the 40 traits plus age as covariate. Priors: Normal(0,1) for coefficients, HalfCauchy for dispersion terms; inference via MCMC in PyMC3 (5000 draws, burn-in of 4000, R-hat < 1.02). Posterior summaries yielded marginal (partial) effects for each trait, with 95% highest posterior density intervals (HPDI). Posterior predictive checks (500 replicated datasets per model) provided explained variance (R²) per region. Subanalysis: Partial correlation preprocessing orthogonalized each region’s volume relative to the other 35 regions, then repeated the Bayesian analyses to highlight unique region-specific variance contributions. Sex differentiation and age–trait interactions: Analyses were performed separately in men and women; sex differences were assessed by subtracting marginal posterior distributions (female–male). Joint age–trait posterior distributions characterized how associations varied with age for each sex. Replication: Workflow reapplied to three independent ~10,000-participant splits from a newer 40,000-participant UK Biobank release; posterior means correlated with original results to assess replicability.

• Overall pattern: Social and demographic traits more frequently showed dominant associations with social brain gray matter than personality traits. Contrary to the primary hypothesis, midline mPFC regions were not uniformly the strongest contributors across traits, though notable effects were observed in mPFC and limbic regions.
• Social traits—living with others and close-relationship markers: • Living with others (vs living alone) was the most frequent dominant trait across >50% of the 36 regions. Examples (women): AM_L: posterior mean 0.035, 95% HPDI 0.005–0.069; HC_L: 0.040, 0.012–0.072; HC_R: 0.056, 0.019–0.093; TPJ_L: 0.070, 0.015–0.127; TPJ_R: 0.059, 0.010–0.109; AI_R: 0.058, 0.024–0.100; IFG_R: 0.065, 0.014–0.118; pSTS_L: 0.058, 0.010–0.110; MT/V5_L: 0.050, 0.004–0.097. • Friendship satisfaction was dominant in dmPFC (women: 0.039, 0.000–0.078). • Lifetime number of romantic partners was dominant in AM_R (women: 0.031, 0.009–0.056) and also in FG_L and rACC in both sexes (e.g., women FG_L: 0.032, 0.005–0.060; rACC: 0.057, 0.018–0.099; men FG_L: 0.048, 0.009–0.086). • Sports club membership showed dominance in FG_R (men: 0.045, 0.002–0.086). • Social support dominated NAC (women: NAC_L: 0.076, 0.018–0.133; NAC_R: 0.040, −0.010–0.091). Loneliness showed negative associations (men: NAC_R: −0.051, −0.128–0.020; aMCC: −0.051, −0.114–0.011).
• Personality traits—daily routine and well-being: • Morning vs evening person dominated 11 regions, especially in men: FP (0.052, 0.002–0.104), HC_L (0.051, 0.020–0.084), HC_R (0.054, 0.022–0.087), AI_R (0.050, 0.015–0.089), CB_L (0.069, 0.016–0.123), CB_R (0.083, 0.028–0.140), SMA_L (0.056, 0.014–0.101), IFG_R (0.058, 0.015–0.106), MTG_L (0.043, 0.006–0.081), PCC (0.051, −0.013–0.088), pMCC (0.044, 0.012–0.075). • Happy mood dominated rACC (men: 0.059, 0.017–0.100). Neuroticism showed negative dominance in TPJ_R and SMG_L (men: TPJ_R: −0.044, −0.106–0.015; SMG_L: −0.050, −0.113–0.009).
• Demographic traits—income and occupation: • High income dominated vmPFC and bilateral amygdala in men: vmPFC 0.053, 0.011–0.100; AM_L 0.060, 0.022–0.104; AM_R 0.048, 0.017–0.084; also MTG_R (men: 0.069, 0.027–0.111; women: 0.047, 0.006–0.090) and pSTS_L (men: 0.065, 0.016–0.113). • Health satisfaction dominated vmPFC in women: 0.067, 0.030–0.105. • Manual job dominated FP in women: 0.059, 0.007–0.115; also observed in MTV5_R and SMG_R in men, PCC and SMA_R in women. • Walking/standing job dominated TP_L in men: 0.045, 0.009–0.078. • Job satisfaction contributed to MTV5_L, SMA_R, pSTS_R in men (e.g., pSTS_R: 0.059, 0.018–0.102).
• Age–trait interactions and sex differences: • Joint age–trait posteriors often diverged by sex in dmPFC, FP, vmPFC, AM, and HC. For example, age×friendship satisfaction (women) vs age×income (men) showed divergent dmPFC patterns; age×health satisfaction (women) vs age×income (men) in vmPFC showed differing profiles; age×morning chronotype and age×living with others showed opposite or non-overlapping trends in hippocampus by sex. • Sex bias contrasts showed a male bias for income effects in dmPFC and amygdala; female bias for health satisfaction in vmPFC; and stronger female effects of living with others in hippocampus, whereas weekly contact had stronger male effects in hippocampus.
• Explained variance (posterior predictive R²): Highest in pSTS_R (~16%) and pSTS_L (~15%) within visual-sensory network; aMCC (~16%) and AI_L (~12%) in intermediate network; TPJ_L (~16%) and TPJ_R (~12%) in higher-associative network; HC_R and NAC_L (~11%) in limbic network. At least one region per network achieved >10% explained variance.
• Partial correlation subanalysis: Revealed additional dominant demographic and personality effects when shared inter-regional variance was removed (e.g., education completion affecting limbic and FP regions; working >40 hours/week affecting vmPFC). Replicated core patterns while highlighting region-unique contributions.

The study set out to determine whether social, personality, and demographic traits uniquely map onto variation in social brain gray matter during midlife, anticipating prominent mPFC and limbic contributions. Findings partly supported this: while mPFC and limbic regions showed notable associations (e.g., vmPFC with health satisfaction and income, amygdala with income and romantic partner history), the strongest explanatory power was not uniformly concentrated in midline regions. Instead, robust effects appeared across the social brain hierarchy, prominently in pSTS, TPJ, aMCC, and insula, indicating distributed network sensitivity to lifestyle factors. Interpersonal closeness and embeddedness (living with others, friendship satisfaction, romantic partners, social support) were consistently linked to structural variation in limbic (amygdala, hippocampus, NAC, rACC) and association regions (dmPFC, TPJ), aligning with literature on social reward (NAC) and social pain (aMCC) and with midlife tendencies to prioritize close ties. Demographic status—especially income and occupational characteristics—related to vmPFC, amygdala, MTG, and pSTS, consistent with neural processing of social hierarchy and status signaling. Personality markers tied to daily routines and well-being (morning chronotype, happy mood, low neuroticism) emerged as key correlates, particularly in FP, hippocampus, and intermediate network nodes, echoing associations with conscientiousness and prosocial orientations. Sex differences and age–trait interactions suggest that the relationship between lifestyle factors and brain structure evolves differently for men and women across midlife: income exerted stronger effects in men in dmPFC and amygdala, whereas health satisfaction showed stronger effects in women in vmPFC; living with others differentially related to hippocampal volumes by sex. Collectively, results indicate that midlife social support and social status are major determinants of interindividual variability in social brain anatomy, distributed across networks rather than localized solely to mPFC.

By deploying a Bayesian, region-by-region framework in ~10,000 UK Biobank participants, the study disentangled unique associations of 40 social, personality, and demographic traits with social brain gray matter. It identifies social embeddedness (e.g., living with others, friendship satisfaction) and demographic status (e.g., income, occupational features) as leading contributors to structural variation, with notable but not exclusive involvement of mPFC and limbic regions. Personality traits associated with daily routines and well-being (e.g., morning chronotype) also showed widespread effects. The work underscores sex-differentiated and age-dependent patterns in these associations. Future directions include: incorporating richer, standardized psychological constructs; longitudinal and interventional approaches (e.g., TMS) to probe causality; multimodal integration to bridge structural and functional networks; enhanced QC at scale; and exploration of mechanisms linking socioeconomic context, social engagement, and brain structure across the lifespan.

• Correlational design precludes causal inference or directionality of effects.
• Social brain atlas derived from fMRI applied to structural MRI may introduce modality misalignment; functional–structural correspondence is imperfect.
• Automated QC pipelines at large scale may leave residual inaccuracies; manual QC infeasible for ~10,000 scans.
• UK Biobank trait measures do not fully align with classical psychological constructs; demographic coverage is necessarily partial, potentially limiting construct validity.
• Binary encoding of traits simplifies complex dimensions and may reduce sensitivity.
• Potential unmeasured confounders; despite deconfounding for head size and BMI, other factors could influence volumes.
• Cross-sectional midlife sample limits inference on developmental trajectories.