Abstract
Epigenetic studies face persistent challenges related to small sample sizes, particularly when using epigenome-wide array technologies. Presumably it is this limitation that has hindered the discovery of replicable and robust findings, much like the early struggles of genome-wide association studies. To address this gap, we conducted one of the largest epigenetic investigations of generalized anxiety disorder (GAD) using 43,504 participant's data from the Million Veteran Program. Our analysis assessed differential DNA methylation between GAD cases and controls across three major genetic ancestry groups: European (EUR), African (AFR), and Latin American (AMR). We identified 49 CpG sites reaching epigenome-wide significance across these ancestries. However, when controlling for smoking either by adjusting for smoking status or restricting analyses to non-smokers in the EUR group only 2 and 5 significant CpG sites remained significant, respectively. To explore the predictive utility of these findings, we constructed methylation risk scores using a clumping and correlation method. The scores showed significant association with GAD phenotype in the leave-one out cross-validation within the MVP cohort but failed to replicate the association in an independent sample from Scotland. This may reflect insufficient power in the follow-up cohort, the effects of unmeasured confounding variables, or other unmeasured heterogeneity. Our findings underscore both the promise and the ongoing limitations of large-scale epigenetic studies, particularly the need for replication efforts and improved control of environmental confounders. Continued expansion of sample sizes, stratified analyses by ancestry, and careful consideration of lifestyle-related and other covariates will be essential in advancing the reliability and interpretability of peripheral epigenetic markers in psychiatric phenotypes like GAD.