Panel-Gene Transcriptomic Associations in Multiple Sclerosis: a three-cohort white-matter analysis of a locked 468-gene detoxification panel

Abstract Multiple sclerosis (MS) was examined with a locked 468-gene redox, metal, xenobiotic-metabolism and barrier panel in three adult white-matter cohorts from the Gene Expression Omnibus (GSE138614, GSE32915, GSE108000). A DerSimonian–Laird random-effects meta-analysis yielded 442 analyzable genes, 78 at Benjamini–Hochberg FDR < 0.05, and 30 high-confidence genes (FDR < 0.05, measured in ≥ 3 cohorts, direction-concordant in ≥ 70% of contributing cohorts, and nominally significant in ≥ 2 cohorts); 12 were higher in MS, led by CP and PPARG, and 18 lower, led by OPLAH and GAMT. The meta-analysis was reproduced exactly from the per-cohort GEO2R result tables. Panel genes were not more often FDR-significant than the rest of the transcriptome (17.2% versus 15.5%; Fisher P = 0.35), and 18% of 20,000 random sets of 442 measured genes had at least as many FDR-significant genes as the panel. Each cohort contributes several samples per donor; re-fitted with the donor as the unit, 17 genes met the high-confidence rule, 12 of them from the original 30, and adjustment for an oligodendrocyte marker score left 9. Under a Hartung–Knapp–Sidik–Jonkman test no gene reached FDR < 0.05. In an independent cohort (GSE123496), 25 of the 30 high-confidence genes changed in the same direction (exact binomial p = 0.0003; Pearson r = 0.73). In supporting lesion and grey-matter tiers (not pooled), higher HMOX1 and lower GPX4 (the latter not FDR-significant in the meta-analysis) recurred; oligodendrocyte loss with glial and vascular expansion may explain part of these and other signals. These are transcript-abundance associations in postmortem tissue; enzyme activity, iron content and lipid peroxidation were not measured, and the findings do not establish ferroptosis or a treatment target.