Replicative senescence dictates the emergence of disease-associated microglia and contributes to Aβ pathology

Abstract

The sustained proliferation of microglia is a key hallmark of Alzheimer's disease (AD), accelerating its progression. Here, we aim to understand the long-term impact of the early and prolonged microglial proliferation observed in AD, hypothesizing that extensive and repeated cycling would engender a distinct transcriptional and phenotypic trajectory. We show that the early and sustained microglial proliferation seen in an AD-like model promotes replicative senescence, characterized by increased βgal activity, a senescence-associated transcriptional signature, and telomere shortening, correlating with the appearance of disease-associated microglia (DAM) and senescent microglial profiles in human post-mortem AD cases. The prevention of early microglial proliferation hinders the development of senescence and DAM, impairing the accumulation of Aβ, as well as associated neuritic and synaptic damage. Overall, our results indicate that excessive microglial proliferation leads to the generation of senescent DAM, which contributes to early Aβ pathology in AD.

Keywords: APP/PS1; Alzheimer's disease; CSF1R; disease-associated microglia (DAM).

Graphical abstract

Figure 1

Dynamics and phenotypic specification of DAM in APP/PS1 mice (A) Time course of the microglial (IBA1⁺) density in APP/PS1 mice and age-matched controls, analyzed by IHC. (B) Representation of the cumulative number of microglial division cycles from early colonization of the brain primordium to aging (20 months of age) in WT and APP/PS1 mice, based on retrospective analysis of published data (Askew et al., 2017; Olmos-Alonso et al., 2016; Alliot et al., 1999; Nikodemova et al., 2015). (C–E) Time course of the density and plaque association (%) of DAM, identified as CLEC7A⁺ (C), CD11C⁺ (D), or MHCII⁺ (E) cells, in APP/PS1 mice and WT littermate controls, analyzed by immunohistochemistry (IHC) (black). Aβ plaques labeled with Congo Red (C)–(E). (F) Confocal imaging of the co-localization of MHCII, CD11C, and CLECL7A in plaque-associated microglia (IBA1⁺) in APP/PS1 mice. Nuclei stained with DAPI, shown in grayscale. (G–I) Flow cytometry analysis of the expression of markers of DAM (CLEC7A, CD11C) in microglia (CD11B⁺CSF1R⁺) in APP/PS1 and WT controls. Immunonegative versus immunopositive gates for CD11C represented as cell count in a histogram plot. Co-expression pattern of CLEC7A and CD11C in microglia (H), analyzed as in (G). Quantification of the frequency of CLEC7A⁺ or CD11C⁺ microglia shown in (I). IHC data collected from average of parietal, auditory, and entorhinal cortex. Flow performed in samples from cerebral cortex. Scale bars in (C)–(F), 20 μm, shown in (C) and (F). Data shown in (A), (C)–(E), and (I) represented as means SEMs. N = 4–5 (A–F), N = 6–8 (G–I). Statistical differences: ^∗p < 0.05, ^∗∗p < 0.01, ^∗∗∗p < 0.001, and ^∗∗∗∗p < 0.0001 versus age-matched controls. Data were analyzed with a 2-way ANOVA and post hoc Tukey tests.

Figure 2

Microglial senescence in APP/PS1 mice (A) Time course of the density of βgal⁺ (blue) microglia (IBA1⁺; brown) in APP/PS1 mice and age-matched controls, analyzed by IHC. Representative βgal⁺IBA1⁺ cells identified with an arrowhead. Time course of the plaque association (%) of βgal⁺IBA1⁺ microglia, in APP/PS1 mice. (B) Correlation of the density of senescent microglia (βgal⁺IBA1⁺) with the total density of microglia (IBA1⁺) or the density of DAM (CLEC7A⁺). R² of linear regression analysis shown in plots. (C) Expression of βgal in CLEC7A⁺ cells in APP/PS1 mice and age-matched controls, analyzed by IHC and shown as percentage of total CLEC7A⁺ cells. Representative βgal⁺CLEC7A⁺ cells identified with an arrowhead. (D–F) Flow-FISH analysis of the telomere length (Cy-5 probe) in microglia (CD11B⁺CD45^low), identifying DAM by CD11C⁺ expression in APP/PS1 mice and WT littermates. Immunonegative versus immunopositive gates for the telomere probe in WT and APP/PS1 microglia, as well as negative controls, represented as cell count in a histogram plot. Relative telomere length (percentage corrected to internal T1301 signal and relative to WT) in total microglia (E) and in homeostatic (Hom; CD11C⁻) versus DAM (CD11C⁺) in APP/PS1 mice (F). (G–I) Analysis of the relative telomere length in specific subpopulations of microglia (CD11B⁺CD45^low), identified by CD11C expression as negative (CD11C^neg), low (CD11C^low), intermediate (CD11C^int), and high (CD11C^high). (G) Representative gates and their cell count for the telomere probe shown in a histogram plot. (H) Quantification of the relative telomere length (percentage relative to individual average CD11C MFI) in 4 CD11C gates in APP/PS1 mice. (I) Correlation of CD11C intensity versus telomere probe intensity in a representative APP/PS1 mouse. R² of the correlation analyses shown in (H and I). IHC data collected from average of parietal, auditory, and entorhinal cortex. Scale bars in (A and C), 20 μm. Flow cytometry performed in samples from cerebral cortex. Data represented as means ± SEMs. N = 4–5 (A–C), N = 8–14 (D–I). Statistical differences: ^∗∗p < 0.01 versus age-matched control (A); ^∗p < 0.05, ^∗∗p < 0.01, and ^∗∗∗p < 0.001 correlation analysis, linear regression (B, H, and I); ^∗∗p < 0.01 versus homeostatic microglia (Hom; F). Data were analyzed with a 2-way ANOVA and post hoc Tukey tests (A) and unpaired t test (E and F).

Figure 3

DAM display a senescent transcriptional signature (A) Heatmap representation of the log2 fold expression of genes from the DAM signature (Keren-Shaul et al., 2017) in WT CD11C⁻ microglia (blue), APP/PS1 CD11C⁻ microglia (green), and APP/PS1 CD11C⁺ microglia (red), using the pheatmap package. (B) Correlation analysis of the top 100 genes with highest and lowest fold change from Keren-Shaul et al. (2017) alongside the log2 fold change comparison of CD11C⁺ versus CD11C⁻ microglia from APP/PS1 mice, using the ggplot2 package. (C) Correlation analysis of the fold change of genes from the core senescence signature (Hernandez-Segura et al., 2017), with low read genes filtered out, alongside the log2 fold change comparison of APP/PS1 CD11C⁺ microglia versus WT CD11C⁻ microglia, using pheatmap and corrplot packages. (D) Correlation analysis of the genes from the senescence-associated signature of melanocytes, keratinocytes, astrocytes, fibroblasts, and core senescence signature (boxed in green) (Hernandez-Segura et al., 2017), with low read genes filtered out, with microglia from APP/PS1 and WT mice, using the corrplot package. (E) Gene set enrichment analysis (Mootha et al., 2003; Subramanian et al., 2005) of signatures upregulated or downregulated in senescence cells (Hernandez-Segura et al., 2017; Fridman and Tainsky, 2008; Casella et al., 2019; Kamminga et al., 2006), as well as a custom signature of genes highly associated with senescent cells (see Results section). Normalized enrichment score (NES) shown for the comparison of DAM (CD11C⁺) versus homeostatic microglia (CD11C⁻) from APP/PS1 mice. NES reaching a p < 0.05 and FDR < 0.25 highlighted by a squared NES. (F–J) Analysis of the single-cell dataset from Van Hove et al. (2019). (F) Uniform manifold approximation and projection (UMAP) plot of the microglial clusters identified from the original dataset after subsetting based on enriched expression of Sall1, Gpr34, Tmem119, Hexb, P2ry12, and Cx3cr1 from the whole brains of 16-month-old APP/PS1 and WT mice. (G) Feature plot of the DAM signature (Cst7, Csf1, Lpl, Apoe, Spp1, Cd74, Itgax), identifying cluster 2 as DAM. Further annotation of the clusters (see Figure S5) identified cluster 1 as homeostatic microglia in C57BL/6 mice, cluster 0 as homeostatic microglia in APP/PS1 mice, cluster 3 as white matter microglia (highly expressing Usp18; Goldmann et al., 2015), and cluster 2 as DAM. (H) Feature plot of the custom senescence signature (Cdkn2a, Cdkn1a, Cdkn2d, Casp8, Il1b, Glb1, Serpine1) identifying the DAM cluster 2 as enriched in senescent genes. (I) Bar plots of the percentage of cells per genotype (left) or per cluster (right), with a low, medium, or high enrichment in the custom senescence signature (from all cells with senescence scores > 0). (J) Dot plot representing the expression of the individual genes of the custom senescence signature in the 4 clusters identified in (F). N = 3–4 (A–E).

Figure 4

Microglial senescence in human AD (A and B) Analysis of the expression of cell-cycle inhibitors associated with senescence P16 (A) and P21 (B) in microglia (IBA1⁺) in the gray matter of the temporal cortex of human AD cases and non-demented controls (NDCs). Density of IBA1⁺P16⁺ or IBA1⁺P21⁺ cells represented as means ± SEMs. Representative immunopositive cells identified with an arrowhead. (C) Morphological analysis of P16⁺ microglia (Iba1⁺) in the gray matter of the temporal cortex of human AD cases and NDCs. (D) mRNA expression of selected markers of senescence (PAI1, P19, P16, P21, CASP8) or senescence-associated secretory patters (SASP; IL-1β, IL-6), in the temporal cortex of human AD cases and NDCs. Data represented as means ± SEMs and indicated as relative expression to the normalization factor (geometric mean of 4 housekeeping genes; HPRT and GUSB) using the 2-ΔΔCT method. N = 10. Scale bars in (A) and (B), 40 μm. N = 6–7 (A–C). Statistical differences: ^∗p < 0.05, ^∗∗p < 0.01, ^∗∗∗p < 0.001, and ^∗∗∗∗p < 0.0001 versus NDC. Data were analyzed with an unpaired t test (A and B) or with a 2-tailed Fisher t test with correction for multiple comparisons (C).

Figure 5

Prevention of microglial proliferation impairs the development of DAM (A) Microglial (IBA1⁺) density and plaque association in APP/PS1 mice and WT littermates after treatment with a diet containing GW2580 (CSF1R inhibitor) or a control diet (RM1) for 4 months from the pre-plaque stage (3.5 months of age), analyzed by IHC (brown). Aβ plaques labeled with Congo Red. (B) Density and frequency (percentage of total microglial population) of senescent microglia (βgal⁺, blue; IBA1⁺, brown) in APP/PS1 mice and age-matched controls, after treatment as in (A), analyzed by IHC. (C) Density of DAM, identified as CLEC7A⁺, CD11C⁺, or MHCII⁺ cells, in APP/PS1 and WT littermate controls, after treatment as in (A), analyzed by IHC (black). Aβ plaques labeled with Congo Red. (D) Flow cytometry analysis of the frequency of DAM (CD11B⁺CD45^lowCD11C⁺) and homeostatic microglia (CD11B⁺CD45^lowCD11C⁻) in APP/PS1 and WT controls, after treatment as in (A). (E) Confocal imaging of the co-localization of cleaved caspase-3 (red) and CLECL7A (green) in APP/PS1 mice treated with GW2580 as in (A). Nuclei stained with DAPI, shown in blue. Representative example of a caspase-3⁺CLEC7A⁻ cell identified with an arrowhead (right). IHC data collected from average of parietal, auditory, and entorhinal cortex. Scale bars in (A)–(C) and (E), 40 μm. Flow cytometry performed in samples from the cerebral cortex. Data shown in (A)–(D) represented as means ± SEMs. N = 5–7. Statistical differences: ^∗p < 0.05, ^∗∗p < 0.01, ^∗∗∗p < 0.001, and ^∗∗∗∗p < 0.0001 versus age-matched control; ^#p < 0.05, ^##p < 0.01, ^###p < 0.001, and ^####p < 0.0001 versus RM1 group. Data were analyzed with a 2-way ANOVA and post hoc Tukey tests.

Figure 6

Prevention of microglial senescence ameliorates amyloid-related pathology (A–C) Analysis of the amyloid pathology, quantified as density (B) and area covered (C) of Aβ plaques (6E10⁺, brown) in APP/PS1 mice and age-matched controls, after treatment with a diet containing GW2580 (CSF1R inhibitor) or a control diet (RM1) for 4 months from the pre-plaque stage (3.5 months of age), analyzed by IHC. Representative overview and detailed images shown in (A). (D–F) Analysis of the axonal dystrophy pathology, quantified as density (E) and area covered (F) of LAMP1⁺ plaques (green) in APP/PS1 mice and age-matched controls, after treatment as in (A)–(C), analyzed by IHC. Representative overview and detail images shown in (D). (G–J) Analysis of the synaptic pathology, quantified as optical density of PSD95 (post-synaptic marker; G and I) or synaptophysin (post-synaptic marker; H and J) in APP/PS1 mice and age-matched controls, after treatment as in (A)–(C), analyzed by IHC. IHC data collected from an average of parietal, auditory, and entorhinal cortex (Aβ and LAMP1) or hippocampus (pyramidal layer and stratum radiatum). Scale bars in (A), (D, overview), 500 μm, shown in (D); (A) and (D, detail), 50 μm, shown in (D); (G) and (H), 50 μm. Data shown in (B)—(F) represented as means ± SEMs. N = 5–7. Statistical differences: ^∗p < 0.05, ^∗∗p < 0.01, and ^∗∗∗∗p < 0.0001 versus age-matched controls; ^#p < 0.05, ^##p < 0.01, and ^####p < 0.0001 versus RM1 group. Data were analyzed with a 2-way ANOVA and post hoc Tukey tests.