AbstractThe Alzheimer's disease (AD) literature treats the sulphur amino acid axis — methionine,homocysteine, cysteine, taurine, and their downstream effectors S-adenosylmethionine (SAM),glutathione (GSH), and hydrogen sulphide (H₂S) — as a loosely-coupled set of biomarkers andintervention candidates. This treatment has produced an apparent paradox. Plasmahomocysteine elevation is one of the most robustly replicated AD risk factors in the field 1–9,11,yet large randomised trials of homocysteine-lowering B-vitamin therapy have produced splitoutcomes — effective in subgroups stratified by baseline homocysteine and disease stage24–27,29–34, null in established disease and unstratified populations 35–40, and null inprimary-prevention populations even when baseline homocysteine is elevated. A 2026multicentre Phase 2 randomised controlled trial of S-adenosylmethionine supplementation inMCI and AD reported no significant effect on p-tau217 or cognition 41, apparently extending thedisappointment from substrate-supply intervention to downstream methylation substrate. Weargue that this paradox is an artefact of treating the sulphur amino acid axis as a singleintervention class. Drawing on the General Theory of Regulated Stability (GTRS) architectureintroduced in SIP-NEURO-LONG-01 * and the heterogeneous-class taxonomy fromSIP-NEURO-03 *, we propose a three-layer regulatory architecture: Layer 1 — substrate node(homocysteine as system σ-load marker), Layer 2 — flux partitioning switch (transsulfuration viacystathionine β-synthase, modulated by a structured Layer 2 input set with genetic andhormonal modulators as primary, APOE as candidate dominant modifier under furtherinvestigation, and microbiome as emerging input, with oxidative-metabolic state as cross-cuttinginput), Layer 3 — downstream regulatory effectors (GSH antioxidant reserve; H₂Spost-translational kinase regulation via dual sulfhydration mechanisms 47,51; SAM/SAHtranscriptional methylation control via PP2A activity 30,32,52; taurine neurotransmissionstabilisation). The architecture explains every major finding and apparent contradiction in thecorpus through a single regulatory logic. The split B-vitamin trial outcomes resolve as a Layer 1effect requiring elevated Layer 1 load AND adequate Layer 2 partitioning capacity to translateLayer 1 reduction into outcome change — a two-condition requirement that theprimary-prevention null trials (where Layer 1 load can be elevated but Layer 2 state isunselected) corroborate rather than contradict. The SAMe Phase 2 null resolves as Layer 3substrate provision without Layer 2 switch correction (with under-dosing as competingexplanation honestly engaged). The U-shaped homocysteine risk curve 10 resolves as aregulatory rather than monotonic phenomenon. The estrogen-modulated homocysteineresponse across the menopausal transition 50 resolves as a Layer 2 switch repositioning. Thedual sulfhydration cascade (GSK3β directly 47; AKT → GSK3β indirectly 51) resolves as aLayer 3 mechanism by which sulphur status directly governs the tau hyperphosphorylationcascade. The architecture's strongest empirical anchor is a documented field gap: across 7,097published papers at the AD × menopause intersection in PubMed (2010–2026), 0.13% examinethe sulphur amino acid axis, and 0.06% examine homocysteine specifically. We acknowledgethat the integrated regulatory reading is convergent with independent emerging consensus inthe field 53,54; this paper's distinct contribution is the operationalisation of that consensus intospecific testable predictions for trials currently in progress † and into intervention-designimplications the convergent reviews do not articulate. We acknowledge the APOE × MTHFR ×Hcy attenuation finding 55 as evidence that APOE may be a dominant rather than co-equalLayer 2 modulator, with the Layer 2 input set therefore organised as aprimary-plus-candidate-dominant-plus-emerging structure rather than four co-equals; and thefield's critical sex-stratification gap as a structural rather than methodological limitation.
Smith et al. (Tue,) studied this question.