Genome-wide association studies (GWASes) find that a large number of genetic variants jointly influence the risk of depression, which is summarized by polygenic indices (PGSs) of depressive symptoms and major depression. But PGSs by design remain agnostic about the causal mechanisms linking genes to depression. Meanwhile, the role of adverse life experiences in shaping depression risk is well-documented, including via gene-environment correlation.
Building on theoretical work on dynamic and contingent genetic selection, we suggest that genetic influences may lead to differential selection into negative life experiences, forging gene-environment correlations that manifest in various permutations of depressive behaviors and environmental adversities.
Using data from two large surveys of middle-aged and older US adults, we investigate to what extent a PGS of depression predicts the risk of 27 different adversities. Further, to glean insights about the kinds of processes that might lead to gene-environment correlation, we augment these analyses with data from an original preregistered survey to measure cultural understandings of the behavioral dependence of various adversities.
We find that the PGS predicts the risk of the majority of adversities, net of class background and prior depression, and that the selection risk is greater for adversities typically perceived as being dependent on people’s own behaviors.
Taken together, our findings suggest that the PGS of depression largely picks up the risk of behaviorally-influenced adversities, but to a lesser degree also captures other environmental influences. The results invite further exploration into the behavioral and interactional processes that lie along the pathways intervening between genetic differences and wellbeing.
…Most PGS coefficients are statistically-significant (15⁄22 in Add Health; 12⁄22 in WLS). Null results are largely consistent: in both datasets, the PGS had no effect on physical unattractiveness, sibling death, cancer, and heavy alcohol use. Inconsistencies between samples (namely, unemployment, childhood sexual abuse, and parent death) have viable explanations: Unemployment and childhood sexual abuse are relatively rare, with correspondingly wider confidence intervals; the magnitude of the childhood sexual abuse estimate is similar across samples, as is the relative rank of the PGS coefficient for unemployment. In case of parent death, the inconsistency potentially stems from different variable definitions: Add Health includes non-parental guardians whereas WLS only includes parents. A restricted definition of parent death in Add Health yields results consistent with WLS (Appendix G1).
Many of the statistically-significant results concur with expectations based on prior research. To give some examples: Prior twin research finds convincing evidence of heritability of marital dissolution ( et al 2010). For unemployment, previous research has pointed to the aspects of “discipline and temperament” that predict getting laid-off (not to mention fired) as also being risk factors for divorce (2004). The same can be said for incarceration and depression risk ( et al 2021). For experiences of being insulted/disrespected, research has found a higher tendency among depressive individuals to negatively interpret neutral or ambiguous stimuli ( et al 2018; 2012).
However, previous scholarship is not as helpful in understanding the statistically-significant effect of the depression PGS on physical and sexual abuse and partner violence, which may be reasonably regarded as outside a person’s control. At least for childhood events, one explanation could be ‘indirect’ genetic effects, as discussed above. Such indirect effects could operate by, for instance, shaping the risk of selection into unhealthy relationships or risk-taking behavior. In other cases, the PGS could simply be capturing parental genetic influences on the family environment: for instance, supplementary analyses find that growing up in a single-parent household is also predicted by the depression PGS in Add Health (Append1ix G2).
In ancillary analyses, we estimated sibling FE models to net out indirect genetic effects (Appendix H). As noted, these results have diminished statistical power, especially for Add Health. Nevertheless, overall, we found statistically-significant within-sibling effects for 10⁄12 adversities in WLS that were also statistically-significantly predicted by the PGS in the main analyses, and for 5⁄15 adversities in Add Health. Yet, consistent with the interpretation of indirect genetic effects being captured by the PGS, we find that FE estimates were not statistically-significant for child or adult sexual abuse in either dataset and for partner abuse in WLS. But, perhaps surprisingly, FE models continued to show a statistically-significant effect of the PGS on childhood physical abuse (both datasets) and for partner abuse in Add Health.
The analyses in Figure 2 control for parental education and income. For most adversities at least one of these indicators has statistically-significant conditional effects. Indeed, the depression PGS has no statistically-significant association with sibling death, childhood disability, and being perceived as physically unattractive, but class background does. In other cases, estimates for class are comparatively small: in particular, child’s illness is statistically-significantly predicted by the PGS but not by indicators of class background. Unsurprisingly, then, the relative importance of genetic and environmental influences varies across types of adversities. The effect of sex on adversity risk also varies in direction and magnitude, as one would expect. Taken together, these results suggest that the depression PGS predicts selection into adversities through mechanisms that are not subsumed by some key observable dimensions of disadvantage, including class background and sex.
3.3. Does gene-environment correlation correspond with perceptions of behavioral dependence?
We now turn to the ratings of behavioral dependence of adversities. Overall, there was high inter-rater reliability across adversities (see Appendix Figure I3). Attribution ratings were generally lower for adversities that happen to family members or friends, result from the violence of another person or external factors, or may be perceived as inborn physiological characteristics; whereas events occurring later in life and pertaining to one’s own health, career, or relationships were viewed as more behaviorally dependent.
Figure 3 plots the weighted average logit coefficients of the depression PGS against the average ratings of behavioral attribution. The linear fit line indicates a positive correlation between the behavioral attribution ratings and the coefficients of the depression PGS (r = 0.33, p = 0.09).
This statistic does not meet the conventional statistical-significance threshold of 0.05, but given the small n = 27, this remains suggestive evidence that the genetic burden of depression is more strongly associated with adversities regarded as being influenced by a person’s behavior. The adversities in the top-right quadrant of the graph may be especially likely to reflect events/experiences that are notable pathways of genetic selection into depression, since these are regarded as resulting from greater behavioral input. In contrast, adversities in the top-left quadrant may be reflecting indirect genetic effects or evocative selection.
