Human longevity and Alzheimer's disease variants act via microglia and oligodendrocyte gene networks

Abstract

Ageing underlies functional decline of the brain and is the primary risk factor for several neurodegenerative conditions, including Alzheimer's disease (AD). However, the molecular mechanisms that cause functional decline of the brain during ageing, and how these contribute to AD pathogenesis, are not well understood. The objective of this study was to identify biological processes that are altered during ageing in the hippocampus and that modify Ad risk and lifespan, and then to identify putative gene drivers of these programmes. We integrated common human genetic variation associated with human lifespan or Ad from genome-wide association studies with co-expression transcriptome networks altered with age in the mouse and human hippocampus. Our work confirmed that genetic variation associated with Ad was enriched in gene networks expressed by microglia responding to ageing and revealed that they were also enriched in an oligodendrocytic gene network. Compellingly, longevity-associated genetic variation was enriched in a gene network expressed by homeostatic microglia whose expression declined with age. The genes driving this enrichment include CASP8 and STAT3, highlighting a potential role for these longevity-associated genes in the homeostatic functions of innate immune cells, and these genes might drive 'inflammageing'. Thus, we observed that gene variants contributing to ageing and AD balance different aspects of microglial and oligodendrocytic function. Furthermore, we also highlight putative Ad risk genes, such as LAPTM5, ITGAM and LILRB4, whose association with Ad falls below genome-wide significance but show strong co-expression with known Ad risk genes in these networks. Indeed, five of the putative risk genes highlighted by our analysis, ANKH, GRN, PLEKHA1, SNX1 and UNC5CL, have subsequently been identified as genome-wide significant risk genes in a subsequent genome-wide association study with larger sample size, validating our analysis. This work identifies new genes that influence ageing and AD pathogenesis, and highlights the importance of microglia and oligodendrocytes in the resilience of the brain against ageing and AD pathogenesis. Our findings have implications for developing markers indicating the physiological age of the brain and new targets for therapeutic intervention.

Keywords: Alzheimer’s disease; ageing; homeostasis; microglia; myelin; oligodendrocytes.

Figure 1

Experimental workflow to integrate human gene variants associated with AD and longevity from GWAS with gene expression networks that show ageing-dependent changes. Gene-based SNP analysis from GWAS for human longevity^, and AD (left). Gene co-expression networks that were ageing dependent were generated from RNA sequencing of bulk human hippocampus, bulk C57BL/6J mouse hippocampus and single-cell C57BL/6J mouse hippocampal microglia (middle). The gene-based GWAS were then used to identify enrichment of genes with variants in ageing gene co-expression networks or in TWAS from various expression datasets related to brain ageing (right). Details of the approach are in the ‘Materials and methods’ section and the Supplementary material. AD = Alzheimer’s disease; FDR = false discovery rate; GTEx = Genotype-Tissue Expression Project; GWAS = genome-wide association studies; MAGMA = multi-marker analysis of genomic annotation; SNP = single nucleotide polymorphism; TWAS = transcriptome-wide association study.

Figure 2

Co-expression analysis of the human hippocampus reveals aspects of ageing and features preserved in mice or unique to humans. (A) Co-expression analysis produced 17 gene modules whose expression is significantly correlated with age in the human hippocampus (correlation > 0.4), from the GTEx data.^, ***P < 0.001. (B) Network diagram of the most central 152 genes from the human microglial network. Red nodes represent hub genes, and blue nodes represent known and putative AD risk genes. (C) Network diagram of the most central 150 genes from the human oligodendrocytic network. (D) Mouse co-expression modules are, in general, moderately preserved in humans. Therefore, it appears that only subsections of the mouse gene networks are preserved in humans. Human GTEx co-expression module preservation in mouse data (left), and mouse co-expression module preservation in human GTEx RNA-sequencing data (right). z.summary is an amalgamation of other preservation statistics (z.density, mean connection strength per gene; and z.connectivity, sum of all connections) found to depict module preservation better than these statistics alone or simple gene overlap measures. Human data were compared with our Mouseac data. Full human networks given in Supplementary Table 2. Mouse networks given in Supplementary Figs 1 and 2, and Supplementary Table 3. GTEx = Genotype-Tissue Expression Project.

Figure 3

Gene co-expression modules expressed by distinct microglial subpopulations are dysregulated with age in the mouse hippocampus. (A) Association, P(−log10), between the expression of co-expression modules formed by co-expression analysis of data generated by single-cell RNA sequencing of microglia (Mg) isolated from wild-type mice at 3, 6, 12 and 21 months of age, and female sex, sequencing plate and six microglial subpopulations identified by the original authors. Subpopulations: HM1 = homeostatic microglia 1; HM2 = homeostatic microglia 2; TRM = transiting microglia; ARM = activated response microglia; IRM = interferon responsive microglia. Negative P(−log10) indicates a negative association. (B) Pseudotime analysis with Monocle 2 identified a bifurcating trajectory, with HM1 as a root state and with ARM and IRM as two terminal states. (C, E and G) Expression of the differentially expressed genes in the ARM-associated (C), interferon (E) and HM2-associated (G) co-expression modules shown along this Pseudotime trajectory. (D, F and H) Comparison of expression of the 100 most central genes (as a proxy for module expression) from the ARM-associated (D), interferon (F) and HM2-associated (H) modules in microglia isolated from the mouse hippocampus at 6–8 and 16–18 months of age and profiled by RNA sequencing.  D, F and H:  n = 6 mice per age group, Student's t-test; *P < 0.05, **P < 0.01, ***P < 0.001. ARM = activated response microglia; HM2 = homeostatic microglial subcluster 2.

Figure 4

Gene co-expression networks from microglia analysed by single-cell RNA sequencing associated with ageing. The genetic networks from microglial cells isolated from wild-type mice at 3, 6, 12 and 21 months of age analysed by single-cell RNA sequencing. Network plot of the 45 most connected genes in the ARM-associated (A), interferon (B), HM2-associated (C) and phagolysosomal (D) modules. Green nodes represent genes, edge lines represent co-expression connections, and the central large red nodes are the hub genes. Full networks are given in Supplementary Table 6. ARM = activated response microglia; HM2 = homeostatic microglial subcluster 2.

Figure 5

Gene co-expression network changes in microglia isolated from mice treated with cuprizone and analysed using single-cell RNA sequencing. (A) Preservation analysis of age-related modules in a dataset generated by single-cell RNA sequencing of microglia isolated from the corpus callosum of 9- to 11-month-old WT mice fed a control or demyelinating cuprizone diet for 5 or 12 weeks. Control module represents results for randomly chosen genes of a similar module size. (B–G) Comparison of expression of the 100 most central genes (as a proxy of module expression) from the ARM-associated (B), interferon (C), ribosomal (D), HM2-associated (E), HM1-associated (F) and TGF-β (G) modules in microglia isolated from the corpus callosum fed a control diet or a demyelinating cuprizone diet for 5 or 12 weeks and profiled by RNA sequencing.  B–G:  n = 3–5 mice per diet group, module expression changes between cuprizone treatment groups was assessed by one-way ANOVA, if the ANOVA indicated a significant difference between treatment groups, a pairwise comparison of cuprizone diet timepoints to control diet was performed using Dunnett’s test; *P < 0.05, **P < 0.01, ***P < 0.001. ARM = activated response microglia; HM1 = homeostatic microglial subcluster 1; HM2 = homeostatic microglial subcluster 2.

Figure 6

Manhattan plots of TWAS results identifying genes whose expression is associated with longevity in whole blood, CD14⁺ myeloid cells, cortex and hippocampus. The y-axis is the Z-score of the association between gene expression and longevity in (A) whole blood from Young Finns Study; (B) naïve (CD14) monocytes from the Fairfax dataset; (C) brain cortex from the GTEx Project^,; and (D) hippocampus from the GTEx Project.^, Genes that showed significant association following Bonferroni correction for multiple testing are labelled (ageing risk genes, red).

Figure 7

Hypothesis for regulation of homeostatic microglia by genes associated with longevity, and on the other side regulation of microglial activity and oligodendrocyte function by genes associated with Alzheimer’s disease risk. Our data suggest a model whereby genes associated with longevity are involved in the homeostatic functions of microglia and perhaps other innate immune cells, and that if these immune cells escape this state of homeostasis by activating stimuli such as amyloid pathology and age-dependent myelin fragmentation, then genes associated with Alzheimer's disease (AD) determine how microglia are activated and how microglia interact with remyelinating oligodendrocytes. The age-associated genes we identified in this study that putatively drive the genetic networks associated with innate immune homeostatic processes and activation might underlie ‘inflammageing’.