A Metabolome-Wide Study of Dry Eye Disease Reveals Serum Androgens as Biomarkers

Purpose To test the association between serum metabolites and dry eye disease (DED) using a hypothesis-free metabolomics approach. Design Cross-sectional association study. Participants A total of 2819 subjects from the population-representative TwinsUK cohort in the United Kingdom, with a mean age of 57 years (range, 17–82 years). Methods We tested associations between 222 known serum metabolites and DED. All subjects underwent nontargeted metabolomic analysis of plasma samples using gas and liquid chromatography in combination with mass spectrometry (Metabolon Inc., Durham, NC). Dry eye disease was defined from the validated Short Questionnaire for Dry Eye Syndrome (SQDES) as a previous diagnosis of DED by a clinician or “often” or “constant” symptoms of dryness and irritation. Analyses were performed with linear mixed effect models that included age, BMI, and sex as covariates, corrected for multiple testing. Main Outcome Measures Primary outcome was DED as defined by the SQDES, and secondary outcomes were symptom score of DED and a clinical diagnosis of DED. Results Prevalence of DED as defined by the SQDES was 15.5% (n = 436). A strong and metabolome-wide significant association with DED was found with decreased levels of the metabolites androsterone sulfate ( P = 0.00030) and epiandrosterone sulfate ( P = 0.00036). Three other metabolites involved in androgen metabolism, 4-androsten-3beta,17beta-diol disulfate 1 and 2, and dehydroepiandrosterone sulfate, were the next most strongly associated of the 222 metabolites, but did not reach metabolome-wide significance. Dryness and irritation symptoms, as opposed to a clinical diagnosis, were particularly strongly associated with decreased androgen steroid metabolites, with all reaching metabolome-wide significance (androsterone sulfate, P = 0.000000029; epiandrosterone sulfate, P = 0.0000040; 4-androsten-3beta,17beta-diol disulfate 1, P = 0.000016; 4-androsten-3beta,17beta-diol disulfate 2, P = 0.000064; and dehydroepiandrosterone sulfate, P = 0.00011). Of these 5 androgens, epiandrosterone sulfate ( P = 0.0076) was most associated with 2-year incidence of clinician-diagnosed DED. In addition, we found decreased glycerophosphocholines to be associated with DED, although not at metabolome-wide significance. Conclusions This hypothesis-free metabolomic approach found decreased serum androgens to be highly associated with DED and adds important evidence to the growing body of research that links andro