Test Your Hypothesis

Battle-testing the mediation hypothesis

Verdict up front: plausible, mechanistically coherent, but empirically unproven. No supplied study measures longitudinal methylation change in glaucoma, so the mediation arm of the hypothesis rests on inference, not observation. What the supplied evidence does establish is (a) a strong, pathway-specific genetic substrate, (b) environmentally responsive cellular stress pathways in trabecular meshwork (TM) cells where methylation changes would have to act, and (c) vascular/systemic predictors that any combined model must integrate.

1. The genetic substrate is strong — methylation must mediate through known pathways

The clearest genetic signal comes from exfoliation syndrome: a rare LOXL1 Tyr407Phe allele shows ~25-fold protection (2/3,909 cases vs 68/5,338 controls; OR ≈ 0.04), and five new loci (POMP, TMEM136, AGPAT1, RBMS3, SEMA6A) implicate extracellular matrix and proteostasis pathways across 13,838 cases and 110,275 controls XFS GWAS meta-analysis loci. For the mediation hypothesis to work, methylation changes would need to plausibly modulate these ECM/proteostasis programs — a reasonable but untested inference, since the XFS study itself notes that RNA-level mechanisms for Tyr407Phe were not explored and functional assays were in vitro only.

2. TM stress pathways are environmentally responsive — the right biological arena

Two supplied studies define candidate mediation targets. MYOC S341P knock-in mice develop elevated IOP, TM cell loss, and RGC degeneration through ER stress and deregulated autophagy (increased p62/LC3B, impaired flux); rapamycin restored flux and 4-PBA partially rescued viability in human TM cells MYOC S341P autophagy model. Autophagy and ER-stress gene networks are classic methylation-sensitive and stress-responsive programs, making them credible intermediaries between environmental load and genetic risk. Independently, KALRN depletion in human TM cells disrupts ER calcium handling, mitochondrial ATP production, and autophagic flux, and TMCO1 knockdown phenocopies it; in vivo, both knockdowns elevate IOP in C57BL/6J mice KALRN-TMCO1 TM homeostasis axis. Because TMCO1 loss-of-function also causes systemic disease in humans TMCO1 recessive disorder, dosage-dependent epigenetic regulation of such genes is mechanistically plausible — but again, inferred, not measured.

3. What is missing — and where the hypothesis is weakest

• No longitudinal methylation data in the supplied evidence. The critical gap is repeated-measures DNA methylation linked to incident glaucoma, environmental exposure, and genotype in the same individuals. Cross-sectional EWAS signals elsewhere in the literature are preliminary; treating them as established mediation would be begging the question.
• Mediation vs confounding. Even if methylation tracks environment and predicts glaucoma, demonstrating that it statistically mediates GxE interaction requires formal causal modeling (e.g., longitudinal mediation with genetically informed instruments), which no supplied study performs.
• Reverse causation risk. Methylation changes in blood could reflect disease consequence (inflammation, ageing) rather than antecedent mediator. TM tissue is inaccessible serially in living humans, forcing reliance on surrogate tissues — a known blind spot.
• Ancestry generalizability. LOXL1 common haplotypes show effect reversal across populations; a methylation layer layered on ancestry-confounded PRS could inherit and even amplify that instability LOXL1 haplotype reversal.

4. Evidence map for the three-component model

Reported odds ratios from LOXL1 resequencing (5,570 cases / 6,279 controls) LOXL1 protective variant ORs. No equivalent longitudinal methylation effect sizes exist in the supplied evidence — that absence is itself the decisive finding.

5. Improved hypothesis and falsification

Improved version: In prospective, ancestrally diverse cohorts, serial methylation changes at ECM/proteostasis and stress-response loci (LOXL1, POMP, TMEM136, KALRN/TMCO1-linked programs) will statistically mediate a measurable fraction of GxE interaction on incident glaucoma, and a combined PRS + methylation-velocity + vascular marker (e.g., arterial stiffness) model will outperform PRS alone for early prediction, with incremental AUC and calibration improvement as the falsifiable endpoint. What would falsify it: longitudinal methylation velocity that fails to predict incident disease after PRS and vascular adjustment; mediation indirect effects near zero with tight confidence intervals; or methylation changes that track disease progression rather than preceding it.

Confidence note: moderate confidence in the mechanistic plausibility judgment, low confidence in the empirical prediction claim, because the decisive longitudinal methylation evidence does not yet exist in the supplied record.