Cerebrospinal fluid metabolome‐wide association study of Alzheimer’s disease provides insight into molecular pathways

Abstract

AbstractBackgroundMany genetic risk variants have been identified for Alzheimer’s disease (AD), but their functional consequence on molecular pathways are unclear. Quantitative trait loci (QTL) studies help elucidate the functional effect of genetic variance on (neuro)biological processes. The understanding of AD mechanisms has been limited due to the relative inaccessibility of in vivo brain tissue and its surroundings. In this study, we integrated genetic and metabolomics cerebrospinal fluid (CSF) data, to understand how AD risk variants influence neurobiological processes.MethodWe performed a genome‐wide mQTL study on 5,543 CSF metabolite levels using data from 977 subjects (age 52.7±16.6 years, 63.8%female) from the Amsterdam Dementia Cohort (n = 487; 220 controls, 87 mild cognitive impairment, 180 AD‐type dementia) and the Utrecht cohort (n = 490 controls). We measured CSF metabolites with three platforms: GC‐TOF MS (primary metabolism, 393 metabolites), CSH‐QTOF MS/MS (complex lipids, 3,532 metabolites), and HILIC‐QTOF MS/MS (biogenic amines, 1,618 metabolites). Association signals between genetic variants and CSF metabolite levels were tested using linear regression, adjusted for principal components, age, sex and site (Amsterdam/Utrecht) using a Bonferroni‐corrected genome‐wide significance threshold (5.0e−8 /754 independent CSF signals = 6.0e−11). To identify metabolites with predicted CSF levels associated with AD, we integrated metabolite summary statistics with AD summary statistics (Kunkle et al., 2019; Bellenguez et al., 2022), using a metabolome‐wide association study (MWAS) approach (FUSION software).ResultWe identified 89 genome‐wide CSF mQTLs, including 89 independent loci for 62 CSF metabolite levels. Most mQTLs were biogenic amines (81), followed by mQTLs for the primary (5) metabolism and complex lipids (3) (Figure 1). MWAS results included 24 metabolite‐AD associations with P<0.05. The strongest associations included variants proximal to the CANX gene with unannotated CSF metabolite 1.09_397.32 (PBellenguez = 2.4e−6; PKunkle = 3.4e−3), and CCDC124 variants with 8.76_225.06 (PBellenguez = 2.9e−3; PKunkle = 2.8e−5) (Figure 2).ConclusionWe identified novel metabolite‐AD risk associations, including CANX and CCDC124 with unknown CSF metabolites. In future work we aim to identify these CSF metabolites using raw spectra and intensity data. This study shows that the multi‐modal integration of genetics with CSF metabolomics helps in the functional interpretation of genetic variance, and could therefore serve as useful approach for other brain‐related disorders.