Complex trait methylation scores in the prediction of major depressive disorder

BACKGROUND: DNA methylation (DNAm) is associated with time-varying environmental factors that contribute to major depressive disorder (MDD) risk. We sought to test whether DNAm signatures of lifestyle and biochemical factors were associated with MDD to reveal dynamic biomarkers of MDD risk that may be amenable to lifestyle interventions.

METHODS: Here, we calculated methylation scores (MS) at multiple p-value thresholds for lifestyle (BMI, smoking, alcohol consumption, and educational attainment) and biochemical (high-density lipoprotein (HDL) and total cholesterol) factors in Generation Scotland (GS) (N=9,502) and in a replication cohort (ALSPAC adults, N=565), using CpG sites reported in previous well-powered methylome-wide association studies. We also compared their predictive accuracy for MDD to a MDD MS in an independent GS sub-sample (N=4,432).

FINDINGS: Each trait MS was significantly associated with its corresponding phenotype in GS (β range=0.089-1.457) and in ALSPAC (β range=0.078-2.533). Each MS was also significantly associated with MDD before and after adjustment for its corresponding phenotype in GS (β range=0.053-0.145). After accounting for relevant lifestyle factors, MS for educational attainment (β=0.094) and alcohol consumption (MS p-value<0.01-0.5; β range=-0.069-0.083) remained significantly associated with MDD in GS. Smoking (AUC=0.569) and educational attainment (AUC=0.585) MSs could discriminate MDD from controls better than the MDD MS (AUC=0.553) in the independent GS sub-sample. Analyses implicating MDD did not replicate across ALSPAC, although the direction of effect was consistent for all traits when adjusting for the MS corresponding phenotypes.

INTERPRETATION: We showed that lifestyle and biochemical MS were associated with MDD before and after adjustment for their corresponding phenotypes (p nominal<0.05), but not when smoking, alcohol consumption, and BMI were also included as covariates. MDD results did not replicate in the smaller, female-only independent ALSPAC cohort (N ALSPAC=565; N GS=9,502), potentially due to demographic differences or low statistical power, but effect sizes were consistent with the direction reported in GS. DNAm scores for modifiable MDD risk factors may contribute to disease vulnerability and, in some cases, explain additional variance to their observed phenotypes.

FUNDING: Wellcome Trust.