Abstract Body: Introduction: Soy intake improves dementia risk factors in clinical trials. However, population studies reported inconsistent associations with cognitive outcomes, likely due to self-report bias. We aimed to develop an objective poly-metabolite score (PMS) for soy intake and test its association with cognition.
Methods: The PMS was derived for self-reported soy intake in 9,992 participants from the Canadian Longitudinal Study on Aging (CLSA) and externally validated in 80 participants from the Protein and Blood Pressure (ProBP) Trial, where participants received soy protein, milk protein, and carbohydrates supplement in a random order for 8 weeks each, separated by 3-week washouts. Plasma metabolomics were profiled using the untargeted Metabolon platform. CLSA participants were randomly split into training (80%) and testing (20%) sets. Candidate metabolites were identified by comparing very frequent consumers (>7 times/week) with never consumers in the training set. An elastic net regression with 10-fold cross-validation was used to construct the PMS, optimizing for AUC and model parsimony. PMS sensitivity and specificity were tested in ProBP by comparing post–soy vs. baseline and soy vs. non-soy interventions. Cognition in CLSA was assessed at baseline and 3-year follow-up using a neuropsychological test battery for memory, executive function, and global cognition. Associations between PMS and cognition were examined using linear regression adjusted for age, sex, race, education, smoking, drinking, physical activity, BMI, SBP, depression, diabetes, cardiovascular disease, hypertension medication, and baseline cognition (for longitudinal outcomes).
Results: The final PMS contains 153 metabolites, including soy biomarkers (daidzein sulfate), gut microbiota products (4-ethylphenylsulfate), and internal metabolites. The PMS showed robust discrimination (AUC_Training=0.84, AUC_Testing=0.80, AUC_ProBP=0.78). In ProBP, PMS increased substantially after soy protein diet compared with baseline (Cohen's d=0.67, p=2.9×10^-8) and non-soy diet (Cohen's d =1.11, p=8.6×10^-16). In CLSA, higher PMS was associated with better memory at baseline (β=0.46, p<0.0001) and 3-year follow-up (β=0.35, p=0.0003), and nominally with global cognition (β=0.14, p=0.02 at baseline; β=0.11, p=0.03 at follow-up).
Conclusion: We developed a PMS of habitual soy intake that generalizes across populations and interventions. This PMS was positively associated with memory and global cognition.