Prediction of alzheimer's disease biomark-ers using genomics and cerebrospinal fluid metabolomics

Metabolite measurements from the cerebrospinal fluid (CSF) have been employed as diagnostic tools for neurodegenerative disorders, such as Alzheimer's disease (AD). AD's pathogenesis is identified by lowered measured levels of CSF amyloid Beta 1-42 (AB), and increased levels of phosphorylated tau (pTau) and total tau (tTau). Due to their direct contact with the brain, metabolites from the CSF are relatively close to these phenotypes and could be used to study biological pathways and signatures associated with neurological disorders. Furthermore, CSF metabolites could help elucidate the effect of the genetic architecture of these disorders, as genome wide association studies (GWAS) have shown that metabolic features can have a heritable component. While the APOE e4 allele has been implicated as the main AD risk locus, other sites of the genome have been found to be associated with AD, indicating a complex, polygenic architecture. Therefore, CSF metabolites are suitable to identify biological pathways impacted by both AD genetic factors and its hallmark biomarkers. For this purpose, we collected more than 5,000 CSF metabolite measurements from AD and mild cognitive impairment patients, and healthy individuals from the Amsterdam Dementia cohort (ADC), as well as imputed genotype data. Using the most recent AD GWAS to calculate a polygenic risk score (PRS) for AD that adds the effect of the APOE alleles as a weighted sum, we found a modest prediction of pTau and tTau by the PRS in linear models after correcting for age and sex, however this effect was mainly driven by APOE e4 allele counts, and is smaller in comparison to the large effect size that is seen for AB. No CSF metabolites were found to be significantly predicted by the PRS or APOE e4 allele counts. We then used an elastic net model to predict AD biomarkers levels using the measured CSF metabolites. This approach identified 18 and 20 metabolites that predicted CSF pTau and CSF tTau levels respectively, among them Cholesterol, Isothreonic acid, and Nicotinoylglycine. Within these CSF metabolites, we did not find any that predicted CSF AB levels. These results indicate that these biomarkers show independent, but biologically correlated aspects of AD pathogenesis. AB reflects the influence of the APOE e4 allele while pTau and tTau are associated with ongoing cellular processes. This is consistent with prior reports that sporadic AD pathogenesis could be driven by a dual-cascade. Further analysis on a separate cohort of healthy individuals, in which the same CSF metabolites were measured, is currently ongoing to replicate these results.