Background
Genetic studies have highlighted the shared architecture of susceptibility among inflammatory diseases [
1]; however, in unaffected individuals, the significance of this genetic predisposition to immune dysfunction for aging-related processes is unknown. Recent genetic studies of neurodegenerative diseases have also implicated the immune system – particularly the innate component – in susceptibility to aging-related conditions [
2], with Alzheimer’s disease (AD) susceptibility being most clearly enriched for variants that influence the expression and splicing of genes expressed in myeloid cells [
3,
4]. In the brain, microglia, the resident myeloid cells, have a prominent role, but infiltrating macrophages, B cells, and T cells also contribute to brain pathology in older age. Today, it is not clear whether a predisposition to pathologic inflammatory responses influences the function of organ systems in individuals that are not affected by an autoimmune disease. Here, we evaluate whether a propensity for excessive inflammation that has not manifested itself clinically in one’s life course can influence the likelihood of (1) late-life cognitive deficits, (2) the accumulation of common neuropathology, including amyloid-β (Aβ) and hyperphosphorylated tau, or (3) microglial recruitment and activation.
The genetic architecture of aging-related traits is not yet well understood; although their prevalence in the population and the heterogeneity observed among older individuals suggest that the allelic spectrum influencing susceptibility includes many common variants of modest effect [
5‐
7]. To assess this hypothesis, we used a common strategy in which the effects of validated susceptibility variants are aggregated into a single additive genetic risk score (GRS). While useful as a screening tool, a GRS obscures the intrinsic granularity of the genomic risk landscape; therefore, they need to be complemented with targeted investigations of individual variants, where possible. We evaluated two representative inflammatory diseases for which the genetic architecture of susceptibility has been well described: MS, which targets the central nervous system, and rheumatoid arthritis (RA), for which a recent study of medical claims data from over 8.5 million adults reported an increased relative risk of AD among patients with RA. Further, anti-TNFα therapy for RA may lower this risk [
8]. In MS, a neuropathological study reported no significant difference in the presence of AD pathologies compared to controls [
9]; however, the prevalence of late-life pathologies and cognitive deficits in genetically-defined populations remains to be characterized.
We also evaluated GRS for four other traits that could influence brain aging and have been characterized genetically: Parkinson’s disease (PD), telomere length, coronary artery disease (CAD), as well as schizophrenia, which may also have an inflammatory component [
10]. Specifically, we deployed our GRS in two deeply characterized cohorts of aging individuals, deconstructed polygenic associations to resolve whether an association is driven by selected variants or a broad distribution of variants, and accessed cortical RNA sequence data to further develop our mechanistic understanding of observed genetic effects.
Discussion
We show that polygenic risk burden for RA and CAD significantly impact microglial count density, in different regions and at different stages of activation, in postmortem brain of elderly individuals. However, these associations were driven by only one or two variants within each GRS, highlighting a key limitation in the use of polygenic risk models of complex traits. In joint analyses of GRS, microglial densities, and RNA sequencing from the frontal cortex, we found no significant direct associations between GRS and immune gene module expression. However, when evaluating pleiotropy among GRS variants, microglial density and immune gene expression, we noted significant associations of MS, CAD, and RA risk variants with brain-wide microglial density and expression of at least one immune module. In parallel, a high genetic MS burden was linked to a loss of modules that are enriched for neuronal or mitochondrial genes, suggesting that it may play a role in exacerbating the dysfunction or loss of neurons or their transcriptional programs.
Several GRS tested showed effects in the directions expected based on existing literature: both the AD
+APOE and AD
-APOE GRS were significantly associated with cognitive decline at corrected thresholds, but, interestingly, the AD
-APOE GRS was only significantly associated with tau-related neuropathology, complimenting existing evidence from in vivo PET imaging and CSF analyses showing an effect of
APOE ε4 on amyloid- but not tau-related biomarkers in healthy elderly [
35]. While our observed association of the CAD GRS with microglial density in the inferior temporal cortex was not expected, the variant driving this association, rs4977574, is in high linkage disequilibrium (r
2 = 0.89) with another variant, rs1333049, that has been associated at genome-wide significance with risk for ischemic stroke [
36]. It is possible that the overlapping susceptibility at this locus for CAD and stroke drives cerebrovascular changes that lead to the recruitment and activation of microglia. Analyses of regional interactions between microglia density and other types of brain pathology, such as silent infarction, is beyond the scope of our current study and is a topic of future interest. For our observed association of the RA GRS with microglia across multiple activation stages, there appears to be little regional specificity other than that the association is much less pronounced is the inferior temporal gyrus. Interestingly, this region is the earliest affected in AD [
37] and appears to behave differently from the other three brain regions in our analyses: it harbors additional, significant (CAD) and suggestive (MS and AD) associations in secondary analyses, particularly in relation to the activated, stage 3 microglia. Finally, we note that, while we elected to use an FDR-based correction in our original analysis plan, our main results also meet more conservative thresholds of significance, such as Bonferroni correction.
Many tests revealed a lack of GRS effects on our outcomes: for example, the schizophrenia GRS was not associated with any measure of cognitive decline or neuropathology. This seemingly contradicts previous evidence of increased HLA-DR
+ microglia in brains of schizophrenia patients compared to age-matched controls [
38]. Our lack of association of schizophrenia GRS with microglial density in any area and at any stage of activation suggests that the mechanism behind aberrant microglial recruitment in the schizophrenia brain is less likely to be due to schizophrenia-specific genetic risk factors; rather, it may be driven by environmental factors or be a consequence of processes downstream of the onset of schizophrenia. In addition, for telomere length variants, we find a lack of association with AD pathology or cognitive decline, in contrast to a recent meta-analysis finding that AD patients have shorter telomeres than controls [
39], as well as Mendelian randomization analyses suggesting causal links between telomere length and AD [
40]. Also, our top signals for association across all GRS (including RA) localized to the MHC region, which influences many immune traits, and resolving the mechanism of this association to the HLA class II region within the MHC will be difficult given the extensive linkage disequilibrium that exists in this unique genomic region [
41].
The main limitation of our GRS is that they were derived from lists of genome-wide significant loci only. We chose this approach due to (1) the lack of unrestricted availability of full summary statistics for all diseases tested, and (2) validity assumptions of individual variant analyses. If many variants well below genome-wide significance from each GWAS were included in our GRS calculations, then post-hoc associations of individual variants may not be relevant in the context of risk for the disease of interest. Moreover, our post-mortem measure of microglial activation is based on morphologic criteria and is subject to error associated with misclassification of individual cells to specific microglial stages of activation. However, by analyzing multiple binned groups of microglia by stage in our detailed analyses, the confounding of misclassification over the spectrum of groups has likely been mitigated. Also, the challenge of objectively classifying microglial activation states is not unique to our study; microglial staging is an active field of investigation [
42].
Conclusions
Together, our findings demonstrate limited links between microglial activation and liability for archetypal inflammatory diseases of both the central nervous system (MS) and periphery (RA). Thus, the immune dysfunction involved in AD susceptibility seems to be largely distinct from those genes and pathways that are involved in susceptibility to inflammatory disease in young and middle-aged adults. Nonetheless, we have uncovered a handful of variants that have strong effects on both inflammatory disease risk and microglial density, which informs our understanding of human microglial biology in aging which remains poorly understood today.