Testing a polygenic score for microglial activation against Alzheimer's disease pathology and cognition.

Abstract

Neuroinflammation and activation of the brain's immune cells, microglia, are among the earliest changes in Alzheimer's diseases (AD). However, lack of direct access to living brain tissue has impeded the development of useful, non-invasive biomarkers of microglial states. Here, we tested multiple novel polygenic risk scores (PRS) of microglial activation for their ability to predict clinical diagnoses of AD, differences in late-life cognition, and levels of specific AD-related neuropathologies measured in vivo. We recently performed a genome-wide association study (GWAS) of morphologically activated microglia measured postmortem in two regions of human cortical tissue (midfrontal and inferior temporal) using immunohistochemistry. Using the GWAS summary statistics, we used the clumping and thresholding method (PRSice-2) to calculate microglial activation PRSs per brain region in imputed genotype data from n=939 participants from the Alzheimer's Disease Neuroimaging Initiative (ADNI). Linear models were used to test associations between each PRS and AD diagnosis at baseline, ADAS-cog scores at baseline, and levels of beta-amyloid (CSF, AV-45 PET), total tau (CSF, AV-1451 PET), and p-tau (CSF), with appropriate covariates (including genomic PCs). Our inferior temporal-PRSs showed significant positive associations with AD diagnosis at low SNP-inclusion p-value thresholds, strongest when comparing the top and remaining PRS deciles (p=1.3x10-3 ). This pattern was also observed for ADAS-cog scores (p=3.2x10-5 ), PET Tau (p=9.7x10-4 ), CSF Tau (p=7.2x10-3 ), and, to a lesser extent, PET beta-amyloid (p=1.8x10-4 ). Conversely, our midfrontal-PRSs showed significant positive associations with amyloid measured in the brain (p=8.3x10-4 ) and CSF (p=0.02) at higher p-value thresholds. Notably, associations of both PRSs showed strong p-value threshold-dependence, where scores at higher thresholds yielded opposite directions of effect on AD-related traits. In this initial experiment of novel microglial activation PRSs, we found several strong associations with AD-related phenotypes resembling the direct phenotypic associations identified in earlier work of postmortem neuropathology. However, we also found strong variability in PRS performance across p-value threshold iterations, possibly due to selection bias in ADNI or the small sample size of the source GWAS. Further replication is required to establish the reliability of a polygenic score for morphological microglial activation in models of neuroinflammatory diseases like AD.