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

Abstract

In any stage of life, humans crave connection with other people. In midlife, transitions in social networks can relate to new leadership roles at work or becoming a caregiver for aging parents. Previous neuroimaging studies have pinpointed the medial prefrontal cortex (mPFC) to undergo structural remodelling during midlife. Social behavior, personality predisposition, and demographic profile all have intimate links to the mPFC according in largely disconnected literatures. Here, we explicitly estimated their unique associations with brain structure using a fully Bayesian framework. We weighed against each other a rich collection of 40 UK Biobank traits with their interindividual variation in social brain morphology in ~10,000 middle-aged participants. Household size and daily routines showed several of the largest effects in explaining variation in social brain regions. We also revealed male-biased effects in the dorsal mPFC and amygdala for job income, and a female-biased effect in the ventral mPFC for health satisfaction.

Fig. 1. Social brain variation is preferentially explained by social traits in women, and by personality and demographic traits in men.

Extending previous social neuroscience studies, the richness of the UK Biobank resource allowed uniquely isolating marginal correlations, which prevail over a wide variety of competing explanatory factors. A generative probabilistic model was estimated for each of the 36 social brain regions in our population sample of middle-aged adults. These region-by-region analyses revealed numerous dominant specific (i.e., partial) trait associations in social brain structure. Colors indicate which individual traits have the largest magnitude (i.e., strongest positive or negative association) in explaining its regional gray matter volume, relative to the other 39 out of 40 total examined traits (cf. Supplementary Data 1). Red indicates markers in the social trait category, purple indicates the personality trait category, and blue indicates the demographic trait category. Sharing one’s home with other individuals was the single most frequent trait association to show the largest magnitude in explaining social brain volume for women, in atlas regions including the AI, AM, HC, IFG TPJ and pSTS (left). The personality trait of being a morning person was the most common trait to show the strongest trait association in social brain structure for men, in atlas regions including the AI, HC, FP, IFG, PCC and pMCC (right). Dominant trait associations from the partial correlation analysis are shown in Supplementary Fig. 1 (cf. Table 3 and Supplementary Data 2 for a description of the social brain region abbreviations).

Fig. 2. vmPFC anatomy is linked to social lifestyle markers of social, personality, and demographic traits with sex-differentiated population effects.

During midlife, regions of the mPFC have been found to show an accelerated decline in gray matter structure (cf. introduction). For the sake of illustration, the marginal posterior population distributions from our vmPFC analysis are depicted using raincloud plots (1 of the 36 region-by-region analyses that we have conducted on the UK Biobank). We reveal to what extent vmPFC region volume is specifically explained by the 40 examined lifestyle traits at the population level (cf. Supplementary Data 1). The half violin plots show the posterior parameter distributions of the specific contributions of each single trait to vmPFC volume, not explained by the other traits, in middle-aged men and women. The boxplots and scatterplots underneath depict the probabilistic parameter guesses that together form the marginal posterior distributions. Each brain-behavior HPDI communicates three kinds of information: direction (e.g., a positive parameter value indicates a tendency for a higher brain volume in the presence of that particular trait), magnitude (e.g., a parameter mean further above or below zero indicates a bigger dependence of region volume on that particular trait), and certainty (e.g., a narrower interval indicates that the model is more sure about the estimated direction and magnitude of a particular brain-behavior effect). For middle-aged women, satisfaction with health contributed most to explaining vmPFC region volume. In contrast, for middle-aged men, earning a higher job income explained most vmPFC gray matter volume. Error bars/dispersion shows uncertainty of Bayesian posterior parameter distributions. Source data are provided in Supplementary Data 3.

Fig. 3. Participant age is a driving factor in how dominant lifestyle traits are linked to midline social brain regions.

The co-relationship between age and a trait association in explaining region variation is quantified by the joint posterior parameter distribution for one particular social brain region (black arrow) for men (blue) and women (pink). This summary visualization exposes the traits with top effects in the full analysis (cf. Fig. 1; cf. Supplementary Data 2 for a description of the social brain region abbreviations). The left column shows these brain-trait associations for women and the right column shows the top trait contributions for men. Middle-aged men and women showed diverging age-trait associations in dmPFC volume in the context of social interaction quality and social status. However, in the context of two socioeconomic status measures, men and women were more similar in age-trait associations with vmPFC volume. The joint posteriors of trait associations in the FP revealed socioeconomic status as measured by job type and personality to show incongruent population parameter distributions. The limbic AM and HC regions additionally showed non-overlapping posterior distributions between middle-aged individuals for social network size and social lifestyle. Supplementary Fig. 3 showcases the age-trait associations in the midline and limbic regions from the partial correlation analysis. Error bars/dispersion shows uncertainty of Bayesian posterior parameter distributions.

Fig. 4. Explained variance of brain-trait associations differs across social brain networks.

In each of the four networks of our social brain atlas (cf. Methods; cf. Supplementary Data 2 for a description of the social brain region abbreviations), we computed posterior predictive checks for every analysis of a given target region. These model-based simulations of replicated data were then compared to the actually observed data to compute the overall explained variance (coefficient of determination, R²) for A the whole population cohort and B men and women, separately. Posterior predictive checks thus safeguarded against several important issues related to model fit by evaluating model-simulated empirical expectations of target region volumes. Intuitively, we asked the Bayesian model: “Based on drawing examples from the previously inferred model posterior, what should the region volume in each particular participant be given his or her 40 trait indicators?”. We thus evaluated model-predicted data that could have been observed or will potentially be observed in the future. This practical check of model-based predictions of observations is a well-recognized approximation to external validation given the actual data at hand. The collective population-level results suggest that in each of the four subnetworks of the social brain atlas, at least one region showed an explained variance of >10% in our middle-aged participant cohort. Source data are provided in Supplementary Data 3.

Fig. 5. Degree of sex bias in brain-trait associations in the social brain midline.

Left/Right: In each of the 40 examined traits (cf. Supplementary Data 1), boxplots show the difference contrasts between the marginal posterior population distributions of each sex (female–male). Means of posterior parameter distribution above zero indicate a relatively female-biased effect for a specific trait association (pink). For means below zero, there is a relatively male-biased effect for that specific trait (blue). Middle: To provide a summary visualization, we counted across the 40 trait associations, for each brain region, to see how many traits were biased predominantly towards males (blue) or females (pink). Purple shows an equal number of male- and female-biased trait associations. Transparency indicates the strength of the sex divergence. Overall, a male bias in volume effects becomes apparent in almost all examined medial prefrontal and limbic regions. In the dmPFC and AM_R, yearly job income showed a stronger effect in men compared to women. However, women drive the trait associations in the FP of the higher-associative social brain, especially with regards to several demographic traits, such as the age of full-time education completion and working a manual job (cf. Supplementary Fig. 5 for sex differentiation in lifestyle trait associations from the partial correlation analysis; cf. Supplementary Data 2 for a description of the social brain region abbreviations). Error bars/dispersion shows uncertainty of Bayesian posterior parameter distributions. Source data are provided in Supplementary Data 3.