Blood transcriptomic overlap between major depressive disorder and experimental sleep restriction in a reanalysis of public microarray data

Abstract Background Insomnia and depression are closely related, but evidence for a shared blood transcriptional signature depends on how the clinical groups and sleep responses are defined. We examined the consistency of transcriptomic overlap across sample inclusion rules and statistical models. Methods We reanalysed public microarray data from major depressive disorder (MDD; GSE98793) and experimental sleep restriction (GSE39445) using standard limma. Sleep responses were estimated from participant means and from repeated measurements with adjustment for sampling time and circadian phase. Signed Stouffer tests assessed combined associations; partial-conjunction tests required evidence in both phenotypes. A clinical-insomnia dataset was examined for confounding and excluded from validation. Results The sleep analysis included 21 participants with 185 paired measurements. In the 192-array MDD dataset, the contrast for MDD without generalized anxiety disorder identified 93 genes at a false discovery rate below 0.05, compared with 13 for pooled MDD. No gene met this threshold in the participant-mean sleep model, whereas 15 did in the time-and-phase model. Among 16180 common genes, the corresponding Stouffer analyses identified 134 and 617 associations with concordant directions. Neither analysis identified a significant partial-conjunction result; this remained the case across all 25 model combinations. Mean participant-level alignment was 0.0611 (95% bootstrap interval − 0.0174 to 0.1401; one-sided sign-flip P = 0.0737). The original 22-array MDD quality-control exclusion list could not be authenticated. Conclusions Apparent transcriptomic overlap was sensitive to the MDD contrast, sample inclusion and temporal modelling. These data did not establish a shared signature with support in both phenotypes after false discovery rate control. Limited power and unresolved historical quality control prevent interpreting this finding as evidence that shared biology is absent.