Main

Alzheimer’s disease (AD) is a chronic neurodegenerative disease characterized by several disease-escalating factors1,2, including local brain inflammation driven by innate immune cells, which significantly contributes to cognitive impairment3. Genome-wide association studies have identified immune-related genes as risk factors in disease onset and severity, some of which are expressed by microglia4,5,6, the resident innate immune cells of the brain7,8. Among the identified genes is triggering receptor expressed on myeloid cells 2 (TREM2), a polymorphism that is considered a risk factor in late-onset AD5,6. Microglia undergo profound changes in response to pathological conditions9. For example, in animal models of amyloidosis, the expression of TREM2 is intricately linked to the state of microglial activation9,10,11. Moreover, TREM2-dependent activated microglia, disease-associated microglia (DAM), were found to cluster around amyloid plaques9.

Aging is recognized as a major risk factor for sporadic AD12,13. A known manifestation of aging is senescent cell accumulation14. Senescent cells are defined by stable cell cycle arrest, persistent DNA damage response and a senescence-associated secretory phenotype (SASP)14. Through continuous secretion of pro-inflammatory cytokines, senescent cells contribute to the chronic tissue inflammation associated with aging14. Over the last two decades, accumulated evidence has supported the contention that the immune system plays a pivotal role in supporting lifelong brain maintenance and functional plasticity15,16,17 and has linked AD to the aging of the immune system18,19. With respect to microglia, it remains unclear whether the activated microglia that fail to support the brain in aging or in AD are themselves becoming senescent or whether they are outcomes of distinct cellular pathways.

In the present study, we identified a new state of microglia, termed senescent, which unexpectedly expressed a high level of TREM2 but exhibited a distinct protein signature than that of the TREM2-dependent activated microglia, known as DAM. A similar signature of senescent microglia was seen in aging as well as in mouse models of AD and tauopathy. Senolytic treatment reduced the level of TREM2-expressing senescent microglia, with no effect on the levels of the DAM population, highlighting the heterogeneity of TREM2-expressing microglia and their potential contrasting impact on disease pathology.

Results

Protein signature of senescent microglia

To identify and characterize microglia with a senescent signature, we assessed the expression of myeloid cell markers and markers characteristic of senescence (Extended Data Table 1). Using mass cytometry (CyTOF), we observed a distinct population of microglia with a characteristic senescent signature in the brains of aged mice (24 months; Fig. 1a–c), mouse models of familial AD (5×FAD20; Fig. 1d–f) and tauopathy (DMhTAU21; Fig. 1g,h). Microglia were defined as myeloid cells that do not express the monocyte markers Ly-6g/c and CCR2 (refs. 22,23). Resting microglia were defined as cells that do not express the activation marker CD11c (refs. 9,24,25) but expressed markers characteristic of homeostatic microglia (CD11b (refs. 9,26), P2RY12 (refs. 27,28) and CX3CR1 (refs. 9,29)). Senescent cells were defined as cells expressing senescence markers, including p16 (refs. 30,31,32), p19 (refs. 32,33), pp38 (refs. 32,34), γH2AX (refs. 32,35), p21 (refs. 32,36) and p53 (refs. 32,37). Unexpectedly, we found that the microglia that expressed senescence markers also expressed some of the characteristic features of homeostatic microglia9, including TMEM119 (refs. 9,38), P2RY12 (refs. 27,28) and CX3CR1 (refs. 9,29), as well as markers associated with activated microglia, including TREM2 (refs. 9,39,40), ApoE (refs. 9,41,42), C5aR (ref. 43), Cd115 (ref. 44), SiglecH (ref. 45) and CD39 (ref. 46) (Fig. 1a,d,g). Of note, aged wild-type (WT) mice had a higher number of senescent microglia with a higher variability among individuals within the group relative to the young mice (Fig. 1c). In addition, we found an accumulation of senescent microglia in 5×FAD mice when comparing with age-matched WT mice (Fig. 1f). Overall, the signature of the senescent microglia included multiple proteins characteristic of senescent cells as well as high levels of TREM2 and ApoE (Fig. 1g). The high correlation between the signature of the senescent microglia in aging and AD (Fig. 1h) could explain the strong association between these conditions.

Fig. 1: Senescent microglia with a distinct signature accumulate in aged mice and in 5×FAD mice.
figure 1

CyTOF analyses of brain microglia isolated from: a–c, 4-month-old C57BL/6 male and female mice (adult; n = 3) and 24-month-old C57BL/6 male and female mice (n = 5); d–f, 11-month-old female mice (WT; n = 4) and 11-month-old female 5×FAD mice (5×FAD; n = 5). a, Median marker expression values for each population. The presented matrix is the average profile across n = 8 mice. b, Representative t-SNE plot of each experimental group. t-SNE map displaying 155,504 randomly sampled CD45+ cells, 19,438 CD45+ cells per sample. Colors correspond to FlowSOM-guided clustering of cell populations. c, Quantitative analysis of the percentage of senescent microglia (as identified in a and b) in microglia (two-tailed Student’s t-test, P = 0.0112). d, Median marker expression values for each population. The presented matrix is the average heatmap across n = 9 mice. e, Representative t-SNE plot of each experimental group. t-SNE map displaying 210,906 randomly sampled CD45+ cells, 23,434 CD45+ cells per sample. Colors correspond to FlowSOM-guided clustering of cell populations. f, Quantitative analysis of the senescent microglial percentage in CNS microglia (two-tailed Student’s t-test, P = 0.0264). g, Senescent microglia average median marker expression values for each population, each line representing a different experimental group (top to bottom: the 5×FAD mouse model of amyloidosis, the DMhTAU mouse model of tauopathy and the third from aged mice). h, A correlation matrix of four cell types’ expression profiles across the three conditions. Cluster 4 represents the senescent microglia. Data are presented as mean ± s.e.m.; *P < 0.05. CNS, central nervous system.

Full size image

TREM2 is associated with a state switch of microglia from resting to activated in animal models of amyloidosis (for example, 5×FAD), where microglia acquire the DAM state9,10,11,47. This state represents TREM2-expressing activated cells. To evaluate the potential connection between TREM2 expression and the senescent microglial phenotype, we compared, within the same individual 5×FAD mice, the levels of TREM2 in microglia bearing the DAM signature versus microglia with a senescent signature. We found that, within each tested mouse, levels of TREM2 expressed by senescent microglia were significantly higher than TREM2 levels expressed by the DAM (Fig. 2a). Moreover, within the senescent microglial population, levels of TREM2 significantly correlated with the expression levels of the senescence markers (Fig. 2b). To further verify the potential functional link between TREM2 expression and senescent microglia accumulation, we examined whether fewer senescent microglia would accumulate in 5×FAD mice that are missing TREM2 (TREM2-null 5×FAD). Thus, we assessed the levels of senescent microglia in TREM2-null 5×FAD mice. The number of senescent microglia in these mice was significantly lower than in age-matched TREM2-intact 5×FAD mice, suggesting that TREM2 is involved in senescent microglia accumulation (Fig. 2c–e).

Fig. 2: High levels of TREM2 are associated with expression of senescence markers in microglia.
figure 2

a–e, Seven-month-old female mice (WT; n = 4) and 7-month-old female mice TREM2−/− (WT) (n = 3) compared to 7-month-old female TREM+/+5×FAD mice (n = 5) and 7-month-old female TREM2−/−5×FAD (n = 6). a, Quantitative analysis of the median expression across the cell population of each mouse of TREM2 in DAM and in senescent microglia; each dot represents single mouse (two-tailed Student’s t-test, P < 0.0001), marked with identical color in the two cell populations. Box plots show the data range (whiskers), 25th and 75th percentiles (box) and median (gray line). b, Pearson’s correlation of the expression of TREM2 with senescence markers p16, p19 and p21. Each color represents a different experiment from Fig. 1; the median expression was z-scored within each experimental group to account for batch identity. c, Median marker expression values for each population, one for each condition of TREM2+/+5×FAD and TREM2−/−5×FAD. The presented matrix is the average profile across samples. d, Quantitative analysis of the senescent microglia percentage within total CNS microglia (one-way ANOVA, Tukey’s multiple comparisons test, \(P_{({\mathrm{WT}}\; {\mathrm{vs}}.\; {\mathrm{5xFAD}}\; {\mathrm{Trem}}2^{+/+})}< 0.0001\), \(P_{({\mathrm{WT}}\; {\mathrm{vs}}.\; {\mathrm{5xFAD}}\; {\mathrm{Trem}}2^{-/-})}=0.0465\) and \(P_{({\mathrm{5xFAD}}\; {\mathrm{Trem}}2^{+/+} {\mathrm{vs}}.\; {\mathrm{5xFAD}}\; {\mathrm{Trem}}2^{-/-})}\left.< 0.0001\right)\). e, Representative t-SNE plot of each experimental group. The t-SNE map was created using 343,520 randomly sampled CD45+ cells, 17,176 CD45+ cells per sample, analyzed by mass cytometry. Colors correspond to FlowSOM-guided clustering of cell populations. Data are presented as mean ± s.e.m. One-way ANOVA was used for the analyses; *P < 0.05 and ***P < 0.001.

Full size image

TREM2-activated microglia were previously seen in close proximity to amyloid plaque accumulation48; therefore, we performed immunostaining to examine whether senesecent microglia have a different spatial distribution. We immunostained cortical sections of mice and postmortem human AD, using antibodies directed to IBA-1, γH2AX and amyloid-beta (Aβ) (Fig. 3). In both postmortem human brain samples and mouse brains, we detected γH2AX immunostained microglia in brain area where amyloid plaques were seen, although these microglia were not necessarily present in close proximity to the plaques (Fig. 3a,b).

Fig. 3: Senescent microglia are found in postmortem brain of patients with AD and in 5×FAD mice.
figure 3

a, Confocal images showing Aβ (magenta), IBA-1 (green) and γH2AX (red) immunoreactivities in sections of cortex from an 11-month-old 5×FAD mouse. There were three biological repeats. b, Confocal images showing Aβ (magenta), IBA-1 (green) and γH2AX (red) immunoreactivities in sections of temporal cortex from a patient with AD. There were two biological repeats. Scale bar, 10 µm.

Full size image

In our initial analysis using CyTOF, we found that senescent microglia exhibit high levels of both TREM2 and ApoE (Fig. 1g). It is worth noting that the ApoE/TREM2/ERK signaling pathway was previously linked to the induction of senescent neutrophils49. Therefore, we further investigated whether ApoE expression and TREM2 expression in our dataset are interconnected. Our analysis revealed a significant positive correlation in expression of these two markers (Extended Data Fig. 2). Furthermore, ApoE expression demonstrated a positive correlation with markers of senescence, similar to the observed association with TREM2 (Extended Data Fig. 2). These findings suggest that senescent microglia may accumulate through a mechanism involving the ApoE/TREM2/ERK signaling pathway, similar to senescent neutrophils.

The transcriptional profile of senescent microglia

The comparison between the protein expression signature of DAM and senescent microglia, described here, revealed that senescent microglia have a signature distinct from the DAM. In contrast to the DAM, senescent microglia express high levels of homeostatic microglial markers (Extended Data Fig. 1). We further investigated the transcriptomic signature of the senescent microglia by conducting a meta-analysis of previously published single-nucleus RNA sequencing (snRNA-seq) data of WT, TREM2−/− WT, TREM2+/+5×FAD and TREM2−/−5×FAD mice50 (Fig. 4a–c). We detected a cluster of senescent microglia, cluster 2, that appeared in TREM2+/+5×FAD mice but was almost entirely absent in the other groups (Fig. 4a,d–f). This cluster was identified as senescent based on the SenMayo gene set51 (Fig. 4g) and showed a significant upregulation of TREM2, ApoE, CD9 and CD11c (Fig. 4h), similar to the senescent microglial cluster identified at the protein level by CyTOF (Fig. 1). The transcriptional signature of cluster 2 was similar to a previously described subtype of microglia, highly activated microglia, which appear only in aged mice41 (Fig. 4h) and, thus, shared with these microglia many differentially expressed genes (DEGs), including ApoE, Lpl, Lgals3, Cst7, Cd74, Cd63, Lilrb4a, Axl, Itgax, Cd83, Cd9, Lgals3bp, B2m, Csf2ra, Tyrobp, Crlf2, Cd34, Ccl4, Lyz2, Hif1a, Csf1 and Cd68 (ref. 41) (the list of all DEGs of each cluster is available in Supplementary Table 1). Notably, although the two microglia populations highly expressing TREM2, namely the DAM and senescent microglia, displayed distinct protein profiles (Extended Data Fig. 1), discerning the differences at the transcriptional level proved challenging. This observation is consistent with the notion that transcriptomic analysis does not always reflect protein expression due to sensitivity limitations, stability concerns and the influence of post-translational modifications. Previous research demonstrated that TREM2 is expressed by white matter-associated microglia (WAM) in aged mice52, which partially share the signature characteristics of DAM. WAM formation is TREM2 dependent but ApoE independent and occurs as part of the normal aging process52. We, therefore, hypothesized that analysis of the WAM population might allow us to identify distinct signatures of TREM2-activated microglia and senescent microglia at the transcriptomic level. We conducted a comparative analysis of the transcriptomic signatures of cluster 2 senescent cells and WAM52. Our analysis revealed 102 DEGs by senescent microglia cluster; only 15 of them shared with the 241 DEGs of WAM, implying that these two populations represent distinct subtypes of TREM2-expressing microglia. The fact that, at the transcriptomic level, the WAM, unlike DAM, showed a different transcriptomic signature compared to senescent microglia suggests possible regional variations among activated microglial phenotypes within the brain.

Fig. 4: Transcriptional signature of senescent microglia.
figure 4

a–h, Analysis of snRNA-seq50 of 7-month-old female mice (WT; n = 3) and 7-month-old female TREM2−/− mice (WT; n = 3) compared to 7-month-old female TREM+/+5×FAD mice (n = 3) and 7-month-old female TREM2−/−5×FAD mice (n = 3)50. a, Uniform manifold approximation and projection (UMAP) plots of re-clustered microglia and macrophages. b, UMAP plots of re-clustered microglia and macrophages colored based on cell type. c, Pathways that were found significant in enrichment analysis. d, UMAP plots of re-clustered microglia and macrophages colored based on genotype. e, The percentage of each cluster. f, The percentage of cluster 2 across all groups (one-way ANOVA, Tukey’s multiple comparisons test, \(P_{({\mathrm{WT}}\; {\mathrm{vs}}.\; {\mathrm{5xFAD}}\; {\mathrm{Trem}}2^{+/+})}< 0.0001\), \(P_{({\mathrm{WT}}\; {\mathrm{vs}}.\; {\mathrm{5xFAD}}\; {\mathrm{Trem}}2^{-/-})}< 0.0001\) and \(P_{({\mathrm{5xFAD}}\; {\mathrm{Trem}}2^{+/+} {\mathrm{vs}}.\; {\mathrm{5xFAD}}\; {\mathrm{Trem}}2^{-/-})}\left.< 0.0001\right)\). g, Module score of senescence signature based on SenMayo gene list for mice51; only 67 of 120 genes were recognized in the data, identifying cluster 2 as enriched in senescent genes. h, Volcano plot describing the results of differential analysis of genes by Wilcoxon rank-sum test for the differential gene expression between cluster 2 and the rest of the clusters. Each dot represents one transcript. The horizontal dashed line marks the significance threshold (adjusted P value based on Bonferroni correction using all genes in the dataset). The vertical dashed lines represent two-fold differences in expression. Data are presented as mean ± s.e.m.; *P < 0.05, **P < 0.01 and ***P < 0.001. NS, not significant.

Full size image

Senolysis of senescent microglia in AD

To examine the potential impact of TREM2-expressing senescent microglia on AD manifestation, we treated 10–11-month-old 5×FAD mice with either 25 mg kg−1 day−1 of the senolytic Bcl-2 family inhibitor ABT-737 (refs. 53,54) or vehicle control. Two cohorts of mice were used in this study. The mice from the two cohorts were evaluated by behavioral tests. After testing, the brains from the first cohort were excised to determine the protein expression profile of microglia, and the brains from the second cohort were excised and tested for their level of inflammatory cytokines. Analysis of the microglial protein expression profile after the treatment showed a significant reduction in the accumulation of senescent microglia in the mice treated with the senolytic drug relative to control mice (Fig. 5a–d). Senolytic treatment did not affect DAM percentages (Extended Data Fig. 3), but TREM2-independent activated microglia percentage increased (Extended Data Fig. 3). Of note, 2 weeks after ABT-737 injections, splenocytes were collected, and the peripheral immune cell profiling revealed no changes caused by the senolytic drug (Extended Data Fig. 4). The behavioral results from both cohorts were pooled together for analyses. Specifically, we tested the animals for anxiety, mobility and recognition memory by using the open field and novel object recognition (NOR) tests, respectively55 (Fig. 5e). We found no effect on anxiety as assessed by time spent in the center of the arena (Extended Data Fig. 5a) and no effect in open field behavior, as measured by the speed of movement and distance (Extended Data Fig. 5b,c). However, we observed a significant improvement in object recognition memory, measured by NOR, after the senolytic treatment (Fig. 5e). Analysis by quantitative real-time polymerase chain reaction (PCR) of the levels of inflammatory cytokines and chemokines in the excised brains revealed a significant reduction in their levels in the 5×FAD mice treated with the senolytic drug relative to those treated with vehicle control (Fig. 5f). These results support our contention that TREM2-expressing microglia represent heterogeneous populations of microglia with distinct signatures and differential activity and indicate that senescent microglia might play a role in neuroinflammation and cognitive impairment.

Fig. 5: Senolytic therapy using ABT-737 reduced the accumulation of senescent microglia, cognitive decline and neuroinflammation in 5×FAD mice.
figure 5

a–f, Two cohorts of mice were combined for the analysis. First cohort of mice: 10–11-month-old female mice, vehicle control (WT, vehicle; n = 4), compared to 10–11-month-old 5×FAD female mice, vehicle control (5×FAD, vehicle; n = 5), and 10–11-month-old 5×FAD female mice, treated with ABT-737 (5×FAD, ABT; n = 4). Second cohort of mice: 10–11-month-old male mice, vehicle control (WT, vehicle; n = 5), compared to 10–11-month-old 5×FAD male and female, vehicle control (5×FAD, vehicle; n = 6), and 10–11-month-old 5×FAD male and female mice, treated with ABT-737 (5×FAD, ABT; n = 7). a, Schematic representation of the experimental approach. b, Median marker expression values for each population. The presented matrix is the average profile across n = 13 mice. c, Representative t-SNE plot of each experimental group. t-SNE map displaying 224,146 randomly sampled CD45+ cells, 17,242 CD45+ cells per sample, analyzed by mass cytometry. Colors correspond to FlowSOM-guided clustering of cell populations. d, Quantitative analysis of the senescent microglia percentage in CNS microglia (one-way ANOVA, P(WT vs. AD vehicle) = 0.0003, P(WT vs. AD ABT) = 0.0208 and P(AD vehicle vs. AD ABT) = 0.0283). e, Quantification of object recognition memory, presented as percent time the mouse interacted with the novel object divided by the total time spent with both objects. One-way ANOVA, Tukey’s multiple comparisons test, P(WT vs. AD vehicle) < 0.0001, P(WT vs. AD ABT) = 0.4851 and P(AD vehicle vs. AD ABT) < 0.0001. f, rt–qPCR results; one-way ANOVA was used for the analyses. Cxcl10: P(WT vs. AD vehicle) = 0.0004, P(WT vs. AD ABT) = 0.1504 and P(AD vehicle vs. AD ABT) = 0.0213. Cxcl9: P(WT vs. AD vehicle) = 0.0013, P(WT vs. AD ABT) = 0.8065 and P(AD vehicle vs. AD ABT) = 0.0113. Ccl5: P(WT vs. AD vehicle) = 0.0054, P(WT vs. AD ABT) = 0.4785 and P(AD vehicle vs. AD ABT) = 0.0440. IL-6: P(WT vs. AD vehicle) = 0.0160, P(WT vs. AD ABT) = 0.8050 and P(AD vehicle vs. AD ABT) = 0.0457. IL-1a: P(WT vs. AD vehicle) < 0.0001, P(WT vs. AD ABT) = 0.1262 and P(AD vehicle vs. AD ABT) = 0.0184. IFN-γ: P(WT vs. AD vehicle) < 0.0001, P(WT vs. AD ABT) = 0.0035 and P(AD vehicle vs. AD ABT) = 0.0160. Data are presented as mean ± s.e.m. One-way ANOVA was used for the analyses; *P < 0.05, **P < 0.01 and ***P < 0.001. CNS, central nervous system.

Full size image

Discussion

Here we show accumulation of senescent microglia with a conserved signature across brain conditions, including aging and amyloidosis. We further show that high levels of TREM2 expression characterize this shared senescent microglial signature. The TREM2-dependent senescent microglia and the TREM2-dependent activated microglia, DAM, display distinct protein signatures. Senolytic treatment reduced the level of TREM2-expressing senescent microglia, with no effect on the levels of the DAM population. This indicates that TREM2-expressing microglia are not a homogenous population of cells. Moreover, depletion of senescent cells, including, but not limited to, microglia, led to improved cognitive performance and reduced expression levels of inflammatory cytokines and chemokines in the brain.

Our present results demonstrate a distinction between activated and senescent microglia. These findings are in line with reported results, showing that degenerative neuronal structures positive for tau (neuropil threads, neurofibrillary tangles and neurotic plaques) are invariably co-localized with severely dystrophic, rather than activated, microglial cells56,57. These studies further suggested that progressive, aging-related dystrophic microglia contribute more to AD than activated ones. Other studies identified senescent microglia and astrocytes in mouse models of tauopathy58 and aging59. Accordingly, consistent with our findings, several studies highlighted the key role played by senescent cells in the brain during AD.

Polymorphism in the TREM2 gene has been linked to the disease severity in patients with AD50. Multiple studies documented the presence of microglia expressing TREM2 in close proximity to the amyloid plaques9. Furthermore, TREM2 appears to promote the proliferation of microglia and their adoption of the DAM profile9,60. Nevertheless, it is important to note that the TREM2-expressing microglia may not constitute a homogenous population, a possibility that was overlooked in those studies. Moreover, the TREM2-activated microglial contribution to the disease progresson could differ in different disease stages. Our data show that TREM2-dependent senescent microglia and homeostatic microglia share markers, such as TMEM119, P2YR12 and CX3CR1. These results, together with the analysis of published transcriptomic data, suggest that TREM2-expressing activated microglia, DAM, and the senescent microglial population are not identical cells and do not represent a continuum of activation states but, rather, are the outcome of activation of a distinct cellular pathway. These results could also explain why attempts to directly activate TREM2 in mouse models of amyloidosis as a therapeutic strategy have shown only limited beneficial effects in preclinical studies61. In support of our findings, a recent study identified a microglial cluster that is upregulated in aged mice compared to juvenile mice and expresses elevated levels of TREM2 (ref. 62). In another study, a subtype of microglia with high expression of ApoE emerged after injection of apoptotic neurons into WT mouse brains42. This microglial subtype was detected in aged and APP-PS1 mice but was diminished in APP-PS1 TREM2−/− animals42. Similarly, the cluster of senescent microglia in our dataset showed high expression of ApoE in both 5×FAD mice and aged mice, and the percentage of this cluster was lower in TREM2−/−5×FAD mice. Our results highlight a strong association among TREM2, senescent microglia and AD. Recently, ApoE/TREM2/ERK signaling was linked to the induction of senescent neutrophils in prostate cancer. TREM2 is activated by tumor ApoE secretion and promotes senescence in immune-suppressive neutrophils through ERK (ref. 49). As senescent microglia upregulate both ApoE and TREM2, it is possible that they contribute to the senescence induction in microglia through a mechanism similar to their involvement in neutrophils. In future studies, it will be intriguing to investigate this mechanism and the accumulation of senescent microglia in TREM2-null mice in different models, such as aging mice. Such investigations will contribute to further understanding of whether this mechanism is specific to AD or it is a more general phenomenon.

The treatment with the senolytic Bcl2 family inhibitor drug ABT-737 (refs. 53,54) led to the elimination of TREM2-dependent senescent microglia, but not of DAM, and was accompanied by improved cognitive performance in an AD mouse model. However, we cannot rule out the possibility that the favorable outcome on cognitive performance was also due to the elimination of senescent cells of other cell lineages beyond microglia, as other studies identified additional brain senescent cells, such as oligodendrocyte progenitor cells, in an AD mouse model63. In addition, no effect on levels of peripheral immune cells was observed at the time that the mice were assessed for behavior (Extended Data Fig. 4), which may suggest that the observed major effect of the senolytic drug was due to removal of senescent cells within the brain. However, we cannot rule out the possibility that the senolytic drugs affected peripheral immune cells at earlier stages or affected skull bone marrow–derived cells, which were not included in the present study. After the elimination of senescent cells, the improved cognitive performance in 5×FAD mice was associated with a reduction in pro-inflammatory cytokines. This supports the suggestion that local brain inflammation is a major pathological factor contributing to cognitive deterioration, as found in other studies in mice18,64 and in human disease3.

Overall, this study suggests that TREM2 displays a dual activity in microglia, which should be carefully considered when contemplating TREM2 as a therapeutic target65. In addition, this study suggests that targeting senescent cells may open new potential therapeutic strategies for AD, which might be effective even at late disease stages. The senolytic drugs available today have limited applicability to humans due to their side effects and lack of specificity53,66. In the future, efforts should be made to develop new senolytic therapies that exclusively target senescent cells or, specifically, senescent microglia. Recent findings have shown a beneficial effect of targeting the inhibitory immune checkpoint pathway PD-1/PD-L1 in animal models of AD and tauopathy, through a mechanism that involves reduction of inflammation67,68,69,70; additionally, it was shown that the same treatment reduces the abundance of senescent cells in all tested tissues71,72. It is, therefore, possible that the beneficial effect of blocking PD-L1 in animal models of AD could also involve elimination of senescent microglia.

Taken together, the present study highlights an unprecedented dual role for TREM2 in microglial activation and senescence. Although, in the past, TREM2 was uniformly considered to be protective in late-onset AD, here we show that microglia expressing high levels of TREM2 could be detrimental. This complexity should be carefully considered when targeting TREM2 as a therapeutic approach.

Methods

Mice

Three mouse models were used in this study: (1) heterozygous 5×FAD transgenic mice (on a C57/BL6-SJL background), which express familial AD mutant forms of human APP (the Swedish mutation, K670N/M671L; the Florida mutation, I716V; and the London mutation, V717I), and PS1 (M146L/L286V) transgenes under transcriptional control of the neuron-specific mouse Ty-1 promoter20 (5×FAD line Tg6799, The Jackson Laboratory); (2) Trem2−/−5×FAD and Trem2+/+5×FAD mice on a C57/BL6 background, obtained from the laboratory of Marco Colonna (Washington University), where they were generated as previously described73; and (3) heterozygous DMhTAU transgenic mice, bearing two mutations (K257T/P301S) in the human-tau (hTAU) gene (double mutant (DM); on a BALBc-C57/BL6 background), associated with severe disease manifestations of frontotemporal dementia in humans21, expressed under the natural tau promoter. Genotyping was performed by PCR analysis of tail DNA. Throughout the study, WT controls in each experiment were non-transgenic littermates from the relevant mouse colonies. Mice were bred and maintained by the animal breeding center of the Weizmann Institute of Science. All experiments detailed here were approved by the Institutional Animal Care and Use Committee (IACUC) of the Weizmann Institute of Science.

Elimination of senescent cells in vivo

For selective elimination of senescent cells, mice were subjected to three consecutive daily intraperitoneal injections of ABT-737 (25 mg kg−1 body weight; Selleck Chemicals) or vehicle as a control, 2 weeks before the behavior assessment, administered as described54. ABT-737 and DMSO-based vehicle were prepared in a working solution (30% propylene glycol, 5% Tween 80 and 3.3% dextrose in water, pH 4–5).

Brain dissociation to a single-cell suspension

Mice were euthanized using ketamine–xylazine, followed by transcardial perfusion with cold PBS and whole-brain excision. Next, the brains were diced and incubated with RPMI supplemented with 0.4 mg ml−1 Collagenase IV (Worthington Biochemical), 2 mM HEPES (Biological Industries), 10 μg ml−1 DNase (Sigma-Aldrich) and 2% FCS. Then, the first brain program of gentleMACS (Miltenyi Biotec) was used for tissue dissociation, followed by incubation for 20 min in a 37 °C shaker. After 10-min incubation, cells were homogenized again, using the gentleMACS brain program 2. After the incubation, the samples were processed by brain program 3 in the gentleMACS for three repeats. The enzymatic reaction was stopped using cold RPMI supplemented with 0.5 M EDTA (0.2%). Then, the cells were filtered (100-μm nylon mesh) and isolated using 30% Percoll (GE Healthcare (17–0891–01)) density gradient centrifugation (23,500g, 30 min, 4 °C) and pelleted, filtered again (70-μm nylon mesh) and rewashed.

Conjugation of metals to antibodies

Antibodies were conjugated to metals using the MIBItag Conjugation Kit (IONpath), according to the manufacturer’s protocols. Metals for conjugation were chosen to minimize noise and spillover between channels, according to guidelines in Han et al.74. Extended Data Table 1 specifies the antibodies and the metal used to label each antibody in Fig. 1, and Extended Data Table 2 specifies the antibodies and the metal used to label each antibody in Figs. 2 and 5.

CyTOF sample preparation

The single-cell suspensions of brain cells were stained by 1.25 μM Cell-ID cisplatin in Maxpar Cell Staining Buffer (Fluidigm). Then, the samples were washed twice with Maxpar Cell Staining Buffer. Next, they were incubated with Fc‐block CD16/32 (BD Biosciences; 10 min, room temperature), followed by incubation with the antibodies for extracellular markers (60 min, 4 °C). The samples were fixed and resuspended with 4% formaldehyde (Pierce) for 10 min, and washed and kept on ice for 10 min. The cells were permeabilized using 90% methanol for 15 min on ice, blocked with 1% donkey serum and stained with antibodies for intracellular markers in buffer with 1% phosphatase inhibitor for 60 min at room temperature. Next, the samples were washed twice and kept in 4% formaldehyde (Pierce) with iridium (125 pM) at 4 °C overnight. On the day of analysis, the samples were washed twice with Maxpar Cell Staining Buffer and then washed twice with Maxpar water (Fluidigm) before acquisition, using a CyTOF 2 upgraded to a Helios system (Fluidigm); before the reading, the cells were filtered (40-μm nylon mesh). A detailed list of antibodies used is provided in the supplementary materials (Extended Data Table 1). No barcoding was used.

CyTOF data processing and analysis

CyTOF data underwent the following pre-processing before analyses. The CyTOF software by Fluidigm was used to normalize and concatenate the acquired data. Then, gates were applied using the Cytobank platform (Beckman Coulter). First, the CD11b stable signal across time was gated and then the event length and the Gaussian residual parameters. Then, the beads were gated out using the 140Ce beads channel. Live single cells were gated using the cisplatin 195Pt and the iridium DNA label in 193Ir, and single cells were gated using the two channels for iridium. Lastly, we gated for CD45+ cells. The gating strategy is illustrated in Extended Data Fig. 6.

The CyTOF data were analyzed using Cytobank and MATLAB. In Cytobank, CD45+ cells were processed with equal sampling. The two-dimensional reduction was applied using the Cytobank ‘visne’ (Vi-distributed stochastic neighbor embedding (Vi-SNE)) method, and FlowSOM-based clustering was performed. Figure 1: the aged mice experiment included 19,438 CD45+ cells per sample; the 5×FAD mice experiment included 23,434 CD45+ cells per sample. Figure 2: the experiment included 17,176 CD45+ cells per sample. Figure 4: the experiment included 17,242 CD45+ cells per sample. The median expression of each cluster was averaged to create a single heatmap for each experiment using MATLAB. The t-tests and correlation analysis were done in MATLAB. See raw data for the complete MATLAB code.

Cognitive assessment

Each mouse was subjected to a daily 3-min handling session for five successive days before the first behavioral test. The investigators performing behavioral testing were blinded to the treatment group of the mice throughout the experiments. Testing sessions were recorded and analyzed using the EthoVision tracking system XT 11 (Noldus Information Technology). The behavioral results shown are combined from two independent experiments.

NOR

The NOR test, which provides an index of recognition memory55, was performed using a protocol modified from Bevins and Besheer75. A square gray box (45 × 45 × 50 cm) with visual cues on the walls was used. The task spanned 2 days and three trials: a habituation trial, a 20-min session in the empty apparatus (day 1); a familiarization trial, a 10-min session allowing the mice to interact with two identical objects (day 2); and a test trial: after a 1-h inter-trial interval, each mouse was returned to the apparatus for a 6-min session, in which one of the objects was replaced by a novel one. Novel object preference was calculated as follows: percent novel object exploration = ((novel object exploration time) / (novel object exploration time + familiar object exploration time)) × 100.

Open field and anxiety

A square gray box (45 × 45 × 50 cm) was used to perform the test. Videos of a habituation trial consisting of a 20-min session in the empty apparatus (day 1 of NOR) were analyzed for cumulative duration in the center zone(s), distance moved (cm), mean velocity, frequency entering the center zone and latency to the first entrance to the center zone.

RNA purification, cDNA synthesis and quantitative real-time PCR analysis

Mice were transcardially perfused with PBS before tissue excision. Cortex tissues were isolated from the brain under a dissecting microscope (Stemi DV4, Zeiss) and snap frozen in liquid nitrogen. Total RNA was extracted using the NucleoSpin RNA Mini Kit (Macherey-Nagel, 740955.50), and mRNA (2 μg) was converted into cDNA using a High Capacity cDNA Reverse Transcription Kit (Applied Biosystems). The expression of specific mRNAs was measured using fluorescence-based quantitative real-time PCR (rt–qPCR; Fast-SYBR PCR Master Mix, Applied Biosystems). Quantification reactions were performed in duplicates for each sample using the ‘ΔΔCt’ method. Hypoxanthine phosphoribosyltransferase 1 (HPRT1) was chosen as a reference (housekeeping) gene.

The primers used are shown in Table 1.

Table 1 Primers
Full size table

rt–qPCR reactions were performed and analyzed using StepOne software version 2.2.2 (Applied Biosystems) and QuantStudio 3 software.

Flow cytometry

Mice were transcardially perfused with PBS before tissue extraction. Spleens were mashed with the plunger of a syringe and treated with ammonium chloride potassium (ACK) lysing buffer to remove erythrocytes. Before immunostaining, all samples were filtered through a 70-μm nylon mesh and blocked with anti-Fc CD16/32 (1:40; BD Biosciences) (Table 2). The samples were fixed, permeabilized and subsequently stained using FOXP3/Transcription Factor Staining Buffer Set (eBioscience, 00–5523-00), according to the manufacturer’s instructions.

Table 2 Immunostaining panel
Full size table

Analysis of snRNA-seq data

We obtained data from Zhou et al. (GSE140511)50. Details about sample processing, sequencing and initial quality control can be found in the original publication50. We accessed the gene–cell count matrices and cell barcode data from 7-month-old female WT mice (n = 3); 7-month-old female TREM2−/− mice (n = 3); 7-month-old female TREM+/+5×FAD mice (n = 3); and 7-month-old female TREM2−/−5×FAD mice (n = 3). Analyses were performed using RStudio 2022.12.0. Following Seurat’s (version 3.2.2) guidelines for quality control76 and Zhou et al.’s fine-tuning parameters, we created a Seurat object and performed statistical analyses of the data. In brief, cells having fewer than 200 genes and genes that appear in fewer than five cells were filtered out. Similarly, cells having more than 5% mitochondrial genes were also filtered out. Then, NormalizeData, ScaleData, RunPCA, FindNeighbors, FindClusters and RunUMAP functions were activated according to object-tailored dimensions for principal components (PCs) (PCs were estimated based on ‘JackStraw’ analysis and ElbowPlot). Cell type was determined using unsupervised mapping of the ChenBrainData dataset from the scRNA-seq package and the MouseRNAseqData dataset from the celldex package. Afterwards, microglia and macrophages were subsetted, and Seurat objects were integrated to create a single integrated object of microglia and macrophages, on which we recalculated data dimensionality and performed scaling using SCTransform, RunPCA, FindNeighbors, FindClusters and RunUMAP functions. Clustering was performed by selecting eight PCs after ‘JackStraw’ and ElbowPlot analyses, after which ‘clustree’ guided our selection for a resolution of 0.4 in FindClusters. Accordingly, RunUMAP was established.

Figure 4a,b,d,e,g,h was created using the DimPlot, FeaturePlot, ggplot2 and EnhancedVolcano packages. Figure 4c is based on DEGs found between cells of cluster 2 and the rest of the cells in the object. Pathways were deducted from the KEGG database using Cytoscape version 3.9.1 and the STRING application. Significant pathways having q values less than 0.05 are described.

To compare between our cluster 2 and WAM DEGs, we used Venn diagrams to represent mutual and unique genes found in each group. Cluster 2’s DEG list was produced by employing FindAllMarkers (from the Seurat R package) and extracting cluster 2’s DEGs (117 genes, using adjusted P < 0.05 and log2fold change (FC) > 0.25). The WAM DEG list was adopted from Supplementary Table 3 from Safaiyan et al.52. Venn diagrams were produced with the online tool https://bioinformatics.psb.ugent.be/webtools/Venn/. Pathway analysis was done using Cytoscape after importing the mutual and unique DEGs in each group.

Immunofluorescence and imaging

For immunofluorescence studies, mice were euthanized and intracardially perfused with PBS, and brains were extracted and fixed in 4% paraformaldehyde overnight and then moved to 1% paraformaldehyde for overnight or longer. Tissue processing and immunohistochemistry were performed on paraffin-embedded sectioned mouse brain (6 μm thick). After de-paraffinization, antigen retrieval was performed in citric acid (pH 6). Sudan black (0.1%) treatment (room temperature, 10 min) and acetone (−20 °C, 7 min) were performed before staining to reduce autofluorescence. Slides were blocked in Cas (Invitrogen, 008120) with 0.5% Triton and then stained using the following primary antibodies: mouse anti-Aβ (1:100, BioLegend, 803001); rabbit anti-γH2AX (1:100, Cell Signaling Technology, CST-2577); and rat anti-IBA-1 (1:50, Abcam, ab283346). For γH2AX, additional amplification was used with anti-rabbit biotin (Jackson ImmunoResearch, 711-065-152, 1:100) in PBS with 1% Cas and 0.5% Triton. Secondary antibodies used included donkey streptavidin-cy2 (Jackson ImmunoResearch, 016-160-084, 1:100), donkey anti-rat cy3 (Jackson ImmunoResearch, 705-165-147, 1:150) and donkey anti-mouse cy5 (Jackson ImmunoResearch, 715-175-151, 1:150). The slides were exposed to Hoechst nuclear staining (1:4,000; Invitrogen probes) for 30 s.

For immunofluorescence on human samples, paraffin-embedded human brain temporal cortex sections of postmortem AD patients were obtained from the Oxford Brain Bank (formerly known as the Thomas Willis Oxford Brain Collection) with appropriate consent and ethics committee approval by the Weizmann Institutional Review Board. After de-paraffinization, antigen retrieval was performed in Tris-EDTA (pH 9). Sudan black (0.1%) treatment (room temperature, 10 min) and acetone (−20 °C, 7 min) were performed before staining to reduce autofluorescence. The following primary antibodies were used: mouse anti-Aβ (1:100, BioLegend, 803001); rabbit anti-γH2AX (1:100, Cell Signaling Technology, CST-2577); and rat anti-IBA-1 (1:50, Abcam, ab283346). For signal amplification for γH2AX detection, anti-rabbit biotin (Jackson ImmunoResearch, 711-065-152, 1:100) was used in PBS with 1% Cas and 0.5% Triton. Secondary antibodies used included donkey streptavidin-cy2 (Jackson ImmunoResearch, 016-160-084, 1:100), donkey anti-rat cy3 (Jackson ImmunoResearch, 705-165-147, 1:150) and donkey anti-mouse cy5 (Jackson ImmunoResearch, 715-175-151, 1:150). The slides were exposed to Hoechst nuclear staining (1:4,000; Invitrogen probes) for 30 s. Representative images were captured using a confocal microscope (Andor Dragonfly spinning disk confocal microscope, ×20 lens), and Fusion software was used for image capture. Representative images were merged and optimized using ImageJ software.

Statistical analysis

Data distribution was assumed to be normal, but this was not formally tested. The data were analyzed using a two-tailed Student’s t-test to compare between two groups; one-way ANOVA was used to compare several groups; and Fisher’s least significant difference test or Tukey or Bonferroni corrections were used for pairwise comparisons. Data from behavioral tests were analyzed using ANOVA and a Fisher’s least significant difference test for pairwise follow-up comparison. Sample sizes were chosen with adequate statistical power based on the literature and past experience, and mice were allocated to experimental groups according to age, sex and genotype. Investigators were blinded to the identity of the groups during experiments and outcome assessment. All inclusion and exclusion criteria were pre-established according to Institutional Animal Care and Use Committee guidelines. Results are presented as mean ± s.e.m. Statistical calculations were performed using GraphPad Prism software (GraphPad Software).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.