Abstract Body: Background A comprehensive blood metabolome fingerprint of incident stroke and the temporal variation of metabolite levels across pre-diagnostic trajectories remain poorly defined.
Methods We included 38,594 stroke-free adults from seven multi-ethnic TOPMed cohorts with 1,245 named circulating metabolites (Fig1.a). Incident stroke (n = 1,628) was ascertained over 7.5–19.3 years. Results from cohort-specific Cox models [adjusted for sociodemographic characteristics, behavioral factors and medications use] were pooled by random-effects meta-analysis. We examined race/ethnicity-specific associations, pre-diagnostic trajectories of metabolites, and risk prediction with metabolite panels.
Results We identified 141 metabolites (FDR < 0.05) associated with incident stroke by pooling results from 7 cohorts after multivariate adjustment, with 97 metabolites independent of major cardiometabolic traits (e.g., glucose, lipids, blood pressures; Fig1.b). Over 80% of identified metabolites showed positive associations, predominantly belonging to phosphatidyl lipids, steroids, glutamate/glutamyl amino acids, aromatic amino acids, and ceramides (Fig1.b,c). Race/ethnicity stratified analyses revealed 21 out of 141 identified metabolites (FDR<0.05) exhibiting significant heterogeneities in associations with stroke across ancestry groups. Of note, 9 metabolites (e.g., MTA, HPLA, oxalate) showed stronger or even opposite associations in Hispanics compared to other ancestry groups (Fig1.d). In Study of Latinos (SOL, n=13,453), temporal analysis identified 3 clusters of metabolites exhibiting nonlinear variational patterns in levels throughout 12 years before stroke diagnosis. Furthermore, higher weighted scores for metabolites in clusters 1&2 were linked with elevated risk of stroke, whereas a reduced risk was found for cluster 3, in which metabolites maintained comparatively low levels throughout the pre-diagnostic process of stroke (Fig.1e). Adding metabolites to conventional risk factors significantly improved risk prediction of incident stroke (AUC improved from 0.78 to 0.84; p<0.001) (Fig1.f).
Conclusion Our study characterized most comprehensive metabolomic signatures of incident stroke to date and revealed potentially ancestry specific signals. Our results further characterized complex temporal dynamics of identified metabolites across pre-diagnostic process and reinforced the value of metabolites in stroke risk prediction.