MMS-32: Development and Evaluation of a 32-CpG Salivary DNA Methylation Score Associated with Subjective Memory Ratings in Males

Abstract

Background: DNA methylation (DNAm) signatures may capture molecular variation associated with cognitive phenotypes, but saliva-based DNAm scores for subjective memory remain poorly characterised. We developed MMS-32, a 32-CpG salivary DNAm score designed to distinguish individuals reporting extreme differences in subjective memory. Methods: MMS-32 was derived from a discovery comparison of individuals reporting Very bad memory (n = 1117) versus Very good memory (n = 2400). Approximately 32,000 predefined CpGs were screened using Mann–Whitney U tests, with signed discovery mean differences used as score weights. Reduced panel sizes were explored in a separate, non-overlapping male dataset initially comprising 1033 participants. MMS-32 v1.0 was subsequently frozen using the discovery-statistic criterion p < 1 × 10−4; accordingly, performance in this male dataset represents development/evaluation rather than untouched external validation. Saliva cell composition was estimated using EpiDISH robust partial correlations based on 648 of 716 EpiFibIC epithelial/fibroblast/immune reference CpGs (90.5% coverage). Results: Among participants providing single-category memory responses, MMS-32 discriminated Very bad (n = 16) from Very good (n = 63) ratings with an area under the receiver-operating-characteristic curve (AUC) of 0.871 (95% stratified-bootstrap CI 0.750–0.963; p = 5.21 × 10−6). Average precision was 0.670 compared with a class-prevalence baseline of 0.203. Performance was substantially weaker for the broader Bad/Very bad versus Good/Very good contrast (AUC = 0.605, 95% CI 0.532–0.677), and the association across all five memory-rating levels was weak (Spearman ρ = −0.117). Chronological age explained only 0.33% of MMS-32 variance in a linear model. Adjustment for estimated epithelial and fibroblast fractions did not materially attenuate the observed MMS-32 associations. Conclusions: MMS-32 is a research-stage salivary methylation signature associated primarily with an extreme subjective memory rating. It should not currently be interpreted as a validated general memory biomarker or biological-age score. Untouched external validation incorporating objective cognitive measures and longitudinal follow-up is required.