Mendelian Randomization Screen of Genetically Predicted Plasma Metabolite Ratios Against Cerebrospinal Fluid Amyloid-β and Tau Biomarkers

Background: Glutamatergic and metabolic dysregulation are implicated in Alzheimer’s disease (AD), making circulating metabolites attractive targets. Prior Mendelian randomization (MR) studies mostly tested binary clinical diagnoses—a heterogeneous outcome—leaving causal links between peripheral metabolism and core AD pathology unresolved. Objective: We aimed to investigate whether genetically predicted plasma metabolite ratios causally influence core cerebrospinal fluid (CSF) AD biomarkers, amyloid-β 42 (Aβ42) and phosphorylated tau 181 (p-tau181). Methods: We conducted a pathology-aware two-sample MR using instrumental variables located in the regions of annotated effector genes for 143 of 309 available plasma metabolite ratios (n = 8299) against genome-wide association study (GWAS) summary statistics for CSF Aβ42 (n = 12,583) and p-tau181 (n = 12,017). Cross-biomarker pathological coherence (Aβ42 falling and p-tau181 rising with worsening pathology) was examined as an exploratory, descriptive criterion whose chance rate was estimated empirically. Instruments in the extended APOE region (chr19:44.4–45.4 Mb) were flagged, and analyses were repeated with and without the corresponding ratios. Results: Across 134 testable ratios, no ratio demonstrated a robust causal effect surviving false discovery rate (FDR) correction on either biomarker. One FDR-significant signal arose at the APOE region (rs59007384, TOMM40) and reflects horizontal pleiotropy; eleven ratios carried APOE-region instruments, and after their exclusion (123 ratios) no ratio reached FDR < 0.05. Pathology coherence was 62/123 (50%) across ratios and 79/138 (57%) across unique instrument variants, not exceeding the 58% expected by chance given the negative correlation of the two outcomes. Three non-APOE ratios were nominally significant for both biomarkers in a coherent direction. The strongest signal, the proline/glutamate ratio, showed a pathology-protective direction (higher Aβ42, lower p-tau181; p-tau181 β = −0.37, p = 1.0 × 10−3, FDR = 0.13), though not surviving multiple testing correction. Conclusions: This MR screen detected no robust causal effect of plasma metabolite ratios on core CSF AD biomarkers at the available statistical power; this does not exclude small effects or tissue-specific mechanisms. The proline/glutamate ratio (PRODH locus) is a hypothesis-generating signal only and requires colocalization and independent-instrument support.