Epigenetic prediction of major depressive disorder

Abstract

Variation in DNA methylation (DNAm) is associated with lifestyle factors such as smoking and body mass index (BMI) but there has been little research exploring its ability to identify individuals with major depressive disorder (MDD). Using penalised regression on genome-wide CpG methylation, we tested whether DNAm risk scores (MRS), trained on 1223 MDD cases and 1824 controls, could discriminate between cases (n = 363) and controls (n = 1417) in an independent sample, comparing their predictive accuracy to polygenic risk scores (PRS). The MRS explained 1.75% of the variance in MDD (β = 0.338, p = 1.17 × 10^-7) and remained associated after adjustment for lifestyle factors (β = 0.219, p = 0.001, R² = 0.68%). When modelled alongside PRS (β = 0.384, p = 4.69 × 10^-9) the MRS remained associated with MDD (β = 0.327, p = 5.66 × 10^-7). The MRS was also associated with incident cases of MDD who were well at recruitment but went on to develop MDD at a later assessment (β = 0.193, p = 0.016, R² = 0.52%). Heritability analyses found additive genetic effects explained 22% of variance in the MRS, with a further 19% explained by pedigree-associated genetic effects and 16% by the shared couple environment. Smoking status was also strongly associated with MRS (β = 0.440, p ≤ 2 × 10^-16). After removing smokers from the training set, the MRS strongly associated with BMI (β = 0.053, p = 0.021). We tested the association of MRS with 61 behavioural phenotypes and found that whilst PRS were associated with psychosocial and mental health phenotypes, MRS were more strongly associated with lifestyle and sociodemographic factors. DNAm-based risk scores of MDD significantly discriminated MDD cases from controls in an independent dataset and may represent an archive of exposures to lifestyle factors that are relevant to the prediction of MDD.

Fig. 1. Prediction of MDD case-control status.

a Receiver Operating Characteristic (ROC) curve indicating the sensitivity (y-axis) and specificity (x-axis) of methylation risk score (MRS) and methylation risk score trained on non-smokers (MRS-ns) for both prevalent and incident MDD. The AUC estimates are indicated for each predictor in the legend. b Variance in prevalent MDD (indicated by R² (%) on the y-axis) explained by MRS and PRS alone when fitting MDD as the outcome variable and fitting age, sex and ten genetics principal components as covariates. MRS and PRS are then fit in the same model (PRS + MRS) to show their additive contribution to variance explained in MDD.

Fig. 2

a Variance in MDD (indicated by R² (%) on the y-axis) explained by four lifestyle factors and MRS. b Variance in MDD (indicated by R² (%) on the y-axis) explained by four lifestyle factors and MRS-ns. Lifestyle factors = BMI, alcohol consumption, smoking status and pack years. Light and dark pink bars indicate the additive variance explained by all lifestyle factors combined in incident (I) and prevalent (P) MDD; the light and dark green bars indicate the additive variance explained by all lifestyle factors with the addition of the MRS to the model.

Fig. 3. Phenotypic associations with MRS and PRS.

Associations between mental health, sociodemographic, lifestyle, physical and cognitive measures and methylation risk score (MRS) in red and polygenic risk score (PRS) in blue; the x-axis represents the standardised effect size for each outcome variable listed on the y-axis. Error bars represent standard errors of the effect size.