Alzheimer's disease genetic risk variants show brain cell type-specific associations with protein levels in cerebrospinal fluid

BACKGROUND: Neuronal dysfunction is central to the clinical manifestation of Alzheimer's disease (AD). However, genome-wide studies also suggest important roles for non-neuronal brain cell-types such as microglia and astrocytes. Our objective was to study whether brain cell type-specific polygenic risk scores (PGRS) for AD, including single nucleotide polymorphisms (SNPs) of genes expressed in one brain cell type, showed relationships with levels of cerebrospinal fluid (CSF) AD markers in individuals across the clinical spectrum of AD. METHOD: We selected 1,535 subjects (552 controls/709 mild cognitive impairment/274 AD-dementia, age 71±8 years, 48%female) from the ADNI (N=617) and EMIF-AD MBD (N=918) study, who had genetic data available. We labelled AD risk genes as specific for neurons, astrocytes, microglia, oligodendrocytes or endothelial cells when more than 50% of the gene expression was produced by one cell type (otherwise as 'non-specific') according to the BRAIN RNASeq database (Zhang et al., 2014). We calculated cell type-specific-PGRS with cell type-corresponding SNPs (P<5e-8 ) using weights from De Rojas et al., (2020) AD case-control summary statistics. Associations between cell type-specific-PGRS and CSF biomarkers (amyloid-beta, total tau (t-tau), phosphorylated tau (p-tau), neurofilament light, neurogranin and YKL-40) were examined using linear regressions, adjusted for population structure, study, age and sex. RESULT: Of 40 AD risk genes, 26 had detectable RNA levels in at least one brain cell type. Of these, sixteen were cell type-specific: 11 were microglia-specific, 4 astrocyte-specific (including APOE) and 1 neuron-specific (Figure 1). Astrocyte-PGRS (P=1.4e-6 , Pwithout_APOE =.1) and microglia-PGRS (P=6.3e-3 ) were increased in AD-type dementia compared to controls, whereas other PGRS were not (Figure 2). Astrocyte-PGRS (including APOE) were most strongly associated with CSF amyloid-beta (P=1.4e-31 ), t-tau (P=6.3e-10 ), p-tau (P=4.9e-9 ), and neurogranin (P=0.01). Astrocyte-specific (excluding APOE) and microglia-specific PGRS-AD also associated with CSF amyloid-beta (P<0.05), and with p-tau at trend-level (Figure 2). Apart from the tentative association between neuron-PGRS and CSF YKL-40, no other associations were observed. CONCLUSION: Findings indicate that AD risk variants have cell type-specific associations with amyloid and tau pathology, which seems mostly expressed by astrocytes (also without APOE) and microglia, suggesting that these cell types play a role in amyloid pathogenesis.