Expression of DAM and senescence genes, including p16
ink4a and Lgals3, increase in aged brain white matter
We previously demonstrated that the aged female hippocampal region harbors increased DAM and senescence markers20. We investigated senescence and microglial gene expression patterns in microdissected fimbria-fornix and corpus callosum white matter relative to hippocampal gray matter of aged female and male mice (Fig. 1). We distinguish between detection of Cdkn2a (referring to the broader p19arf and p16ink4a locus) and p16ink4a alone. p16ink4a (Fig. 1a), Cdkn2a (Fig. 1b), Lgals3 (Fig. 1c), Cdkn1a/p21 (Fig. 1d) and Ccl2 (Fig. 1e) expression were significantly higher in fimbria-fornix and corpus callosum compared with hippocampus. In aged female mice, Ccl5 (Fig. 1f), Apoe (Fig. 1g) and Gpr34 (Fig. 1h) expression were significantly higher in fimbria-fornix and corpus callosum compared with hippocampus. Relative to young counterparts, p16ink4a, Cdkn2a, Csf1r, Gpr34 and Trem2 expression increased in both gray and white matter microdissections of old male and female mice (Fig. 1i,j). Apoe, Bcl2a1a, Cd11b, Klk8, Lgals3, Cdkn1a/p21 and Tmem173 expression increased in old female and male fimbria-fornix and corpus callosum. Ccl2, Ccl5, Il1b and Tyrobp increased in old male fimbria-fornix and corpus callosum (Fig. 1j). In aged female and male mice, senescence and DAM gene expression was more pronounced in white matter relative to hippocampal gray matter. At the regional level, we observed age-dependent increases in genes such as p16ink4a, Cdkn2a, Lgals3, Apoe, Bcl2a1a, Cd11b, Csf1r, Klk8, Tmem173, Trem2 and Tyrobp more consistently in old fimbria-fornix and corpus callosum (Fig. 1k) than in the hippocampus or cerebellum, emphasizing that white matter is susceptible to late-life changes in senescence and DAM transcriptional signatures.
Fig. 1: Senescent and DAM gene expression increases in aged brain white matter tracts.
a–h, Summarized are RT-PCR relative expression (RE) values from old female (pink) and old male (blue) hippocampus (HIP, circles) or fimbria-fornix and corpus callosum (FFCC, triangles). Values are normalized relative to HIP expression per sex. Comparisons are shown for p16ink4a (a), Cdkn2a (b), Lgals3 (c), Cdkn1a/p21 (d), Ccl2 (e), Ccl5 (f), Apoe (g) and Gpr34 (h). P values denote significance by two-tailed unpaired t-tests with Welch’s correction. n = 7 female, n = 6 male samples per region. i, Female mouse RT-PCR RE values from old HIP (red circles) and old FFCC (pink diamonds) normalized to expression of the same tissue from young female mice. n = 6 young, n = 9 old female samples per region. j, The male mouse RT-PCR RE values from old HIP (gray circles) and old FFCC (blue diamonds) normalized to expression of the same tissue from young male mice. n = 5 young, n = 7 old male samples per region. i,j, Two-tailed unpaired t-tests with FDR correction, *q < 0.05 versus young tissue. k, log2 fold changes in RT-PCR gene expression in old brain regions of HIP, FFCC and cerebellum (CB) in female (F) and male (M) mice as compared with young, sex-matched brain regions, as in normalizations from i and j. Bars represent mean ± s.e.m.
GAL3+ myeloid cells accumulate in the hippocampal-adjacent white matter of old mice
Based on the observation that senescent DAM cells express Lgals3/GAL320,35, we investigated the abundance of GAL3+ immunoreactive cells across the hippocampus and adjacent white matter in young and old brain sections (Extended Data Fig. 1). GAL3+ cells were rare in young mouse brains. In aged brains, we observed GAL3+ cells in multiple limbic regions (Extended Data Fig. 1a) with dense localization in white matter tracts, especially the fimbria (Extended Data Fig. 1a). In aged fimbria, GAL3 signal colocalized with the myeloid/microglial marker IBA1 but not with the astrocyte marker GFAP (Extended Data Fig. 1b,c). GAL3 immunoreactivity comprised a larger area of old versus young female fimbria; however, GAL3 area was not significantly different in fimbria between old and young male mice (Extended Data Fig. 1d). In the old female fimbria, IBA1+GAL3+ cells were larger relative to IBA1+GAL3− cells, (Extended Data Fig. 1e,f). PCA of IBA1+GAL3+ and IBA1+GAL3− populations from old mouse fimbria revealed size-specific principal components based on cellular morphology and distinguished by GAL3+ immunoreactivity (Extended Data Fig. 1g). Similar to the fimbria, anterior commissure and corpus callosum white matter tracts exhibited strong GAL3+ immunoreactivity (Extended Data Fig. 1h,i). IBA1 and GAL3 also colocalized in deep cerebellar nuclei (Extended Data Fig. 1j). Thus, immunofluorescent imaging and gene expression demonstrate increased abundance of the senescent and DAM marker GAL3/Lgals3 in aged white matter, including the fimbria, which was more pronounced in aged female mice.
IMC demonstrates microglia positive for senescent and DAM markers are abundant in the aged fimbria
To investigate the molecular phenotypes of aged microglia across microenvironments, we implemented IMC, which combines immunolabeling and time-of-flight mass spectrometry to map and quantify ten protein markers with 1-µm spatial resolution in the aged female mouse brain (Fig. 2). We measured the abundance and colocalization of senescent and DAM markers (GAL3, TMEM173, MHCII, uPAR, dPP4/CD26, phospho-p38MAPK and CD38) with IBA1 and CX3CR1 as myeloid markers and CD45 as a pan-immune marker (Fig. 2a). Fimbria myeloid cells exhibited greater intensity of IBA1, GAL3, CD38, uPAR, dPP4 and CD45, relative to hippocampal myeloid cells, which exhibited greater CX3CR1 protein levels (Fig. 2b). Stratification of GAL3+ and GAL3− cells in the fimbria showed that GAL3+ cells harbored trending or significantly higher levels of senescence-related markers uPAR and CD38 (Fig. 2c). These protein-based results validate and extend our observations from regional gene expression and immunofluorescent analyses, demonstrating that the aged fimbria accumulates senescent and DAM markers, including GAL3 and uPAR.
Fig. 2: IMC reveals senescent and DAM protein signatures increase in aged fimbria.
a, Representative reconstructed IMC image in old female hippocampus (left) and fimbria (right) with protein labels for IBA1 (green), CX3CR1 (yellow), GAL3 (magenta), CD38 (cyan) and uPAR (red). Scale bars, 200 μm; inset scale bars, 50 μm. b, Background-subtracted protein values for IBA1/CX3CR1+ myeloid cells from the hippocampus (black circles) and fimbria (red squares) of old brains. n = 33 HIP-myeloid cells, n = 56 FIM-myeloid cells. c, Summarized are background-subtracted protein values for fimbria-specific IBA1/CX3CR1+ cells stratified into GAL3− (green squares) versus GAL3+ (purple triangles). n = 35 FIM-myeloid GAL3− cells, n = 21 FIM-myeloid GAL3+ cells. Cells were assayed from IMC of four old female mice. Statistical q-values denote significance in two-tailed unpaired t-tests with FDR correction. Bars represent mean ± s.e.m.
GeoMx DSP confirms enrichment of conserved DAM and expanded gene signatures in aged fimbria white matter
We next examined microglial transcriptional signatures specific to the hippocampus and fimbria of old female mice using GeoMx DSP. We segmented the hippocampus and fimbria regions with positive selection based on GAL3, IBA1 and SYTO83 labels and negative selection by GFAP (Fig. 3a). First, we compared IBA1+ populations in the hippocampus versus fimbria (Fig. 3b). Established senescence and DAM genes Fth1, Ctss and Apoe and candidate senescence-related genes, Ccno and Sesn3, were enriched in IBA1+ cells of the fimbria. Adam3, Cdc20b, Ciao3, Efcab8, Enkd1 and Ttc21b were enriched in hippocampal IBA1+ cells. We next compared transcriptional profiles of IBA1+GAL3+ and IBA1+GAL3− microglia specifically in aged fimbria (Fig. 3c, d). Apoe was the most abundantly expressed gene in IBA1+GAL3+ cells, relative to IBA1+GAL3− cells.
Fig. 3: GeoMx DSP demonstrates canonical DAM and emergent genes increase in aged fimbria microglia.
a, Representative staining of sections from (i) fimbria and (ii) hippocampus with labels for GAL3 (red), IBA1 (green), SYTO83 (blue) and GFAP (pink) in old mouse brain sections. Right: IBA1 staining alone in green. Scale bars, 200 µm. b, Volcano plot comparing IBA1+ gene expression in the hippocampus versus fimbria. Blue points indicate higher enrichment in hippocampus; red points indicate higher enrichment in fimbria (DESeq2, P < 0.05, n = 8 HIP, 15 FIM sections sampled from 5 mice). c, Representative images of staining and segmentation of fimbria for distinguishing IBA1+GAL3− and IBA1+GAL3+ cells. Scale bars, 200 µm. d, Gene expression volcano plot comparing IBA1+GAL3− versus IBA1+GAL3+ AOI in old fimbria (DESeq2, P < 0.05, n = 15 sections sampled from 5 mice). For b and d, differential expression was analyzed with two-sided DESeq2 Wald test.
CosMx SMI reveals diverse aged microglial identities, including white matter-concentrated DAM- and senescence-linked states
We used CosMx SMI to define microglial transcriptional signatures at single-cell resolution in the hippocampus, corpus callosum, fimbria, cortex, dorsal thalamus and choroid plexus of aged female mice (Fig. 4). We developed a 50-gene custom panel of aging- and senescence-associated genes based on a two-step process of (1) selection of genes from our and others’ published senescence profiles20,36,37,38 and (2) confirmation of expression in aged brain single-cell RNA sequencing and spatial transcriptome datasets20,33,39,40. We used the custom panel combined with the 950-plex CosMx mouse neuroscience panel, enabling investigation of 102 senescence-related genes that we term the ‘SenBrain’ panel, which we analyzed alongside brain function, cell-typing and control probes (see the Methods for the SenBrain description; Supplementary Tables 1 and 2).
Fig. 4: CosMx SMI resolves DAM- and senescence-linked microglial identities in aged brain.
a, CosMx images of microglial centroid spatial localizations across hippocampus, cortex, thalamus and white matter regions in 24-month-old female mouse brains. Scale bar, 500 μm. b, UMAP Leiden clustering of 2620 microglia as represented spatially in a, separated into 9 clusters. The shapes correspond to mouse ID#231-234. Leiden algorithm clusters were created from 1,000-plex genes with no spatial information as input. c, UMAP of microglia, color coded by spatial localization of either white matter (magenta) or gray matter (green). d, Stacked bar plot of percentage of cells in each cluster spatially localized to either white matter (magenta) or gray matter (green). e, Volcano plot of differentially expressed genes present in white matter-localized versus gray matter-localized microglia. Positive fold change indicates genes relatively increased in white matter (magenta dots), and negative fold change indicates genes relatively increased in gray matter microglia (green dots). Differential expression was analyzed by two-sided Wilcoxon rank-sum test. f, A heat map of the top five differentially expressed genes in clusters 1–9 as shown in b. Color values per heat map cell represent normalized z-scores; white-light orange denotes lower and red denotes higher relative expression. g, Dot plot of cluster z-scaled average (color) and cell percentage (size) expression of DAM and a subset of SenBrain genes from clusters 1–9. Percent of cell quantification is inclusive of cells in each cluster that contain two or more transcript localizations. h, Volcano plot from pooled microglia from four 24-month-old mouse brains of genes upregulated in Lgals3-enriched microglia (with two or greater Lgals3 transcript counts per cell), compared with microglia with one or no counts of Lgals3 per cell. Differential expression was analyzed by two-sided Wilcoxon rank-sum test with Bonferroni correction for multiple comparisons. i, UMAP of pooled microglia as in b, colored by Lgals3 expression (red). j, UMAP representation of AddModuleScore for the SenBrain gene panel (blue, low expression; red, high expression). k, Dot plot of SenBrain, DAM and cell-type contract-tracing gene set modules for clusters 1–9. l, Violin plot of SenBrain panel AddModuleScore for each cluster (1–9). Box plots denote quartile 1 (25th percentile) lower bounds, quartile 2 (50th percentile median) center and quartile 3 (75th percentile) upper bounds and whiskers show minimum and maximum values for each cluster. m, Dot plot of senescence domain gene sets within the SenBrain panel, showing average gene set expression for clusters 1–9. FC, fold change.
We initially assayed CosMx on five brain sections from four aged female mice (for one aged mouse, two technical replicates were included) and identified 2,620 total microglial cells, which clustered into nine populations (Fig. 4a,b and Extended Data Fig. 2a,b). Microglia were identified based on expression of canonical markers, including Csf1r, Ctss, Cx3cr1, Hexb, Selplg, Itgam, P2ry12, Tmem119 and Trem241,42. We excluded nonmicroglial immune cells (185 cells), which separated clearly from microglia based on uniform manifold approximation and projection (UMAP) dimensionality reduction, gene expression profiles and spatial location (Supplementary Fig. 1a–d). We detected immune cells in the choroid plexus and brain parenchyma with shared molecular signatures (Supplementary Fig. 1b). Although rare in tissue sections compared to microglia, immune cells expressed Bsg, Crip1, Cd74, Cd63, H2-Aa and H2-Ab1, with low or no expression of microglia-specific markers (Supplementary Fig. 1c–e).
All microglial populations comprised cells from both white and gray matter anatomical regions; however, clusters 1, 2, 4, 7 and 9 were proportionally enriched in microglia localized to fimbria and corpus callosum white matter regions (Fig. 4c,d and Extended Data Fig. 2c,d). An inherent artifact of current single-cell spatial platforms and segmentation algorithms is low-level detection of transcripts from adjacent or interacting cells. By exploiting this signal as a variant of molecular ‘contact-tracing’43, we observed that white matter-localized microglia were enriched for oligodendrocyte transcripts Mbp, Mobp and Plp1, whereas gray matter-localized microglia were distinctly enriched for neuronal transcripts Camk2a, Adcy1, Sncb and Snap25, indicating close proximity to, or potentially prior engulfment of these cells (Fig. 4e). Distinct astrocytic markers were enriched in gray (for example, Aldoc) and white (for example, Gfap) matter-localized microglia (Fig. 4e).
Analysis of top differentially expressed genes revealed distinct and shared signatures across the nine microglia clusters (Fig.4f). Clusters 3, 5, 6 and 8 were characterized by low expression of Apoe and Fth1 and moderate-to-high expression of Cx3cr1, Hexb, P2ry12, Selplg and Tmem119, suggestive of homeostatic populations (Fig. 4f and Extended Data Fig. 2e). Clusters 2 and 6 demonstrated elevated immune regulatory genes (for example, B2m, C1qa, Cst3 and Tmem173), while retaining high expression of homeostatic genes. However, DAM-related transcripts Apoe, B2m, Cd63, Clec7a, Csf1, Cst7, Ctsb, Ctsd, Fth1, Ftl1, Lyz1/2, Spp1, Trem2 and Tyrobp were higher in clusters 1, 2 and 7 (Fig. 4g and Extended Data Fig. 2e).
Across pooled Lgals3+ microglia and among Lgals3+ microglia from each individual brain, DAM- and senescence-associated genes Apoe, B2m, Clec7a, Ctsb, Ctsd, Cd9, Cd63, Csf1, Fth1, Ftl1, Gpnmb, Itgax, Lilrb4a/b, Lyz1/2, Lpl, Mmp12 and Spp1 were co-expressed (Fig. 4h and Extended Data Fig. 2f–i). Microglial cluster 7 harbored the greatest frequency of Lgals3+ cells (Fig. 4i and Extended Data Fig. 2j) and demonstrated greater frequency and/or higher intensity expression of Apoe, Axl, Ccl3–4, Cd63, Clec7a, Csf1, Ctsb, Ctsd, Fth1, Ftl1, Gpnmb, Itgax, Lgals3, Lilrb4a/b, Lpl, Mmp12, Plaur, Spp1 and Tyrobp (Extended Data Fig. 2e,k). Lgals3+ cells were also present in clusters that expressed Mbp and Mobp (Extended Data Fig. 2l). Among all aged microglia, we observed that the expression profiles across distinct and total Cdkn+ microglia were heterogenous (Supplementary Fig. 2). Furthermore, we detected low-level expression of multiple Cdkns across most individual microglial populations; cluster 8 harbored the highest expression (Extended Data Fig. 2j).
To profile transcripts from neighboring and/or interacting cells, we devised canonical gene set modules of distinct brain cell types (Supplementary Table 2) using previously published data39,42 and interrogated the ‘contact-tracing modules’ and enrichment for SenBrain and DAM signatures in clusters (Fig. 4j–k). We detected enrichment for immune cell and choroid plexus related gene modules in cluster 1, 7 and 9, oligodendrocyte module in clusters 1 and 4, neuronal module in cluster 3 and astrocyte and vascular modules in cluster 9 (Fig. 4k). We identified the highest enrichment of the SenBrain gene module in aged microglial cluster 7 and lowest enrichment of the SenBrain module in cluster 3 (Fig. 4j–l), suggesting that senescence genes were least prevalent in neuron-adjacent microglia.
We parsed the SenBrain panel into senescence domains (Supplementary Table 2) and investigated the scores and percentage of cells enriched for individual modules44. The white matter-enriched cluster 7 demonstrated pronounced frequency and expression enrichment across all investigated senescence domains, including cell cycle regulation, nuclear stress, SASP, senescent cell anti-apoptotic pathway (SCAP), organelle stress and cell surface markers (Fig. 4m). This corresponded to comparatively elevated expression of specific senescence-related genes, including Cdkn2d/p19ink4d (cell cycle regulation); Cenpb, Hmgb1, Pak2 and Smad2 (nuclear stress); Ccl3–4, Csf1, Mmp12 and Spp1 (SASP); Akt1 and Bcl2l1 (SCAP); Cyba, Lgals3, Lgals3bp and Lpl (organelle stress); and Axl, Cd9, Gpnmb and Plaur (cell surface markers) (Fig. 4g). Cluster 2, also proportionally concentrated in white matter, was characterized by the second highest SenBrain score and demonstrated enrichment for cell cycle, nuclear stress, SASP, organelle stress and cell surface domains (Fig. 4m). Collectively, CosMx delineated diverse microglial populations defined by molecular and anatomical features, including a distinct white matter-enriched population characterized by DAM and senescence gene expression.
Young versus old CosMx spatial maps reveal age-altered, white matter-enriched microglia
We next leveraged CosMx SMI to explore age-dependent microglial states in an independent dataset of corticolimbic sections from four old 24-month-old versus four young 4-month-old female mice (Figs. 5 and 6). We first separated microglia from nonmicroglial immune cells (Supplementary Fig. 1f). We generated a UMAP of 316 nonmicroglial immune cells and observed that young and old samples heterogeneously contributed to three distinct clusters (Supplementary Fig. 3a). Gradient expression of Ttr, a choroid plexus-enriched marker, broadly defined cells more likely to be choroid plexus localized (clusters 1 and 2) versus infiltrated parenchymal cells (cluster 3) (Supplementary Fig. 3b). Parenchymal immune cells expressed Aldoc, Apod, Cpe, C1qa, C1qb, C1qc, Csf1r, Fth1, Gfap, Hmgb1, Malat1, Mbp, Mobp, Ptn, Slc1a2, Sparcl1 and Tmsb4x, whereas Ttr-high immune cells were enriched for markers expressed in choroid plexus epithelium, including Atp1b1, Bsg, Cab39l, Chchd10, Cox4i1, Htr2c, Ndufa4 and Psap45,46 (Supplementary Fig. 3c). Relative to young immune cells, higher levels of astrocyte (Gfap), oligodendrocyte (Plp1, Mbp and Mobp), and microglia (C1qa, C1qb and Ctss) markers were detected in old immune cells, suggesting proximity to brain-resident glia (Supplementary Fig. 3d).
Fig. 5: CosMx SMI reveals high-resolution molecular signatures of age-altered and region-defined microglia.
a, Volcano plot of differentially detected transcripts assayed by CosMx SMI in 4-month young (1,116 cells; blue dot genes are relatively higher in young) and 24-month old (1,811 cells; orange dot genes are relatively higher in old) microglia from 4 young and 4 old female sagittal corticolimbic brain samples. Differential expression was analyzed by two-sided Wilcoxon rank-sum test with Bonferroni correction for multiple comparisons. b, Representative CosMx spatial images of a young (left) and old (right) mouse corticolimbic brain section, depicting cell segments with color-coded microglial clusters corresponding to c. c, UMAP of 15 microglial clusters from 4 young (open shapes) and 4 old (closed shapes) mouse brains with each cell color coded according to Leiden cluster algorithm. Point shape is specific to mouse of origin (see Supplementary Fig. 4 for spatial maps for each mouse). d, UMAP as in c, color coded by age (young, YNG, blue; old, OLD, orange). e, (i) Bar chart of percentage of cells per age from YNG (blue) and OLD (orange) brains from combined microglia clusters 1–15. Blue bars total 100% and orange bars total 100%. (ii) Bar chart of percentage of cells contributing to each microglia cluster from YNG (blue) and OLD (orange) brains. The blue and orange bars from each cluster column total 100%. f, Dot plot of cluster z-scaled average (color) and cell percentage (size) expression of age-increased genes identified in a. Percent of cell quantification is inclusive of cells in each cluster that contain two or more transcripts. g, UMAP as in c and d, colored according to microglial spatial localization in gray matter (green) or white matter (pink) brain regions. h, Bar charts of percentage of gray matter (green) and white matter (pink) cells in each cluster for YNG and OLD mice. i, UMAP as in c colored according to anatomical localization. j, Dot plot showing enrichment for age, identity and contact-tracing gene set module scores within microglial populations. Age is a composite score of binary categorization (YNG = 0, OLD = 1) with a higher score reflecting relatively higher percentage of cells from old samples. VLMC, vascular leptomeningeal cells.
Fig. 6: Analysis of the CosMx SenBrain gene panel and senescence domains reveals a microglial senotype in the aged brain.
a, UMAP of scaled Lgals3 expression across microglial clusters 1–15 in YNG (triangles) and OLD (circles) mice, defined in Fig. 5. b, UMAP of 15 microglia clusters as in Fig. 5c, with heat map color according to SenBrain AddModuleScore, with blue indicating relatively low score and red indicating relatively high score. c, Violin plot of the SenBrain AddModuleScore for microglial clusters with YNG (blue) and OLD (orange) cells indicated. Values denote P value for nonparametric two-sided pairwise Mann–Whitney U test. Cluster 10 was untested due to absence of YNG cells. d, Dot plot showing gene set module scores for SenBrain senescence domains of cell cycle regulation, nuclear stress, SASP, SCAP, organelle stress and cell surface domains for each of the 15 microglial clusters. e, Dot plot of cluster z-scaled average (color) and cell percentage (size) expression of DAM genes and a subset of SenBrain senescence genes from clusters 1–15. Percent of cell quantification is inclusive of cells in each cluster that contain two or more transcripts. f, Monocle3 trajectory inference analyses of microglia cluster gene expression based on UMAP reduction of clusters within the reachable core partition. Both young and old microglia are included. g, Pseudotime quantification of microglia within UMAP along trajectories using either (i) cluster 1 or (ii) cluster 10 as the selected root.
We identified 1,116 young and 1,811 aged microglial cells from the eight mouse brains. Apod, Apoe, Axl, B2m, C4b, Cd9, Cst7, Ctsb, Ctsd, Fth1, Ftl1, Gfap, Gpnmb, Lgals3, Lgals3bp, Lyz1/2, Malat1, Serpina3n, Serinc3, Spp1, Tmsb4x and Vim were among the most abundant transcripts that distinguished old compared with young microglia (Fig. 5a). Calm1, Camk2a, Ckb, Csf1r, Dlg4, Glul, Mbp, Mobp, P2ry12 and Sparc were more abundant in young microglia (Fig. 5a). Leiden clustering revealed 15 microglial populations (Fig. 5b,c, Supplementary Fig. 4 and Extended Data Fig. 3a). We assessed if old versus young samples distinctly contributed to microglial populations and found three populations that were rare or absent in young brains: cluster 3, 10 and 15 (Fig. 5d,e). Cluster 10 was characterized by increased expression of canonical DAM- and senescence-related genes, including Apoe, Axl, B2m, Cd9, Cd63, Clec7a, Csf1, Cst7, Ctsb, Ctsd, Cyba, Fth1, Ftl1, Gpnmb, Itgax, Lgals3, Lgals3bp, Lilrb4a/b, Lpl, Lyz1/2, Mmp12, Spp1 and Tyrobp (Fig. 5f and Extended Data Fig. 4e). Cluster 10 also demonstrated a molecular profile overlap with the previous senescent DAM population identified in aged microglia from Fig. 4 (cluster 7). Cluster 1 was comprising young and old microglia and defined by higher expression of homeostatic microglial markers, including P2ry12, Tmem119 and Csf1r (Extended Data Fig. 3a). GSEA-based comparison of cluster 10 versus 1 revealed enrichment of regulatory pathways associated with immune cell activation, chemokines/chemotaxis and lymphocyte apoptotic processes in cluster 10 (Extended Data Fig. 3b). Investigation of DAM and microglia homeostasis gene set modules revealed a gradient increase in DAM and gradient decrease in microglia homeostasis from cluster 1 toward cluster 10 (Extended Data Fig. 3c,d). Although cluster 1 displayed homeostatic characteristics, interestingly, aged microglia from cluster 1 displayed increased expression of canonical aging- and DAM-linked genes, including Apoe, Fth1 and Malat1 compared with cluster 1 young microglia (Extended Data Fig. 4a,b). Cluster 2 was enriched for oligodendrocyte- and myelin-associated genes, including Cryab, Mag, Mog and Plp1 (Extended Data Fig. 4c). Age-dominant clusters 3, 10 and 15 harbored high proportion and expression of age-dependent, differentially expressed markers detected in Fig. 5a (Fig. 5f). Furthermore, differentially expressed genes in clusters 3, 10 and 15 exhibited conserved senescence, DAM and white matter-specific genes (Fig. 5f and Extended Data Fig. 4d–f).
We next interrogated anatomical composition by annotating microglia as either gray matter (cortex, hippocampus or thalamus) or white matter (corpus callosum or fimbria) localized (Fig. 5g,h). In the aged brains, clusters 2, 3, 9, 10 and 15 were white matter-enriched. Clusters 1, 4, 7 and 11–14 had 70% or greater gray matter resident microglia in both young and old mice. Consistent with these anatomical characterizations, Mbp and Mobp oligodendrocyte markers were detected in white matter-majority clusters 2, 3, 9, 10 and 15, and Snap25 and Syt1 were detected in clusters 4, 7 and 14 (Extended Data Fig. 4g,h). We further annotated microglial anatomical locations within the caudate putamen, cingulate cortex, corpus callosum, cortex, dentate gyrus, fimbria, hippocampus and dorsal thalamus (Fig. 5i and Extended Data Fig. 4i). Cluster 7 and cluster 12 were primarily localized in the dorsal thalamus. Cluster 2, 3, 10 and 15 were identified in similar proportions in the corpus callosum and fimbria. Clusters 5, 6 and 8 had a greater proportion of corpus callosum-localized microglia, whereas cluster 9 had a greater proportion of fimbria-localized microglia (Extended Data Fig. 4i). Regional siloing confirmed an expansion of age-related microglial clusters 3, 10 and 15 in white matter, with aged brain-exclusive cluster 10 displaying the most pronounced, reproducible DAM gene expression profile.
To further investigate microenvironment characteristics, we implemented the contact-tracing gene set modules initially described in Fig. 4 (Fig. 5j and Supplementary Fig. 5). Microglial clusters 2, 3, 9, 10 and 15 exhibited higher oligodendrocyte module scores, whereas clusters 4, 7 and 14 exhibited higher neuron module scores, corroborating the white and gray matter-majority anatomical localizations (Fig. 5j, Extended Data Fig. 4c–i and Supplementary Fig. 5a). Clusters 1 and 13 were enriched for the microglia cell-type module, suggestive of homeostatic populations, whereas clusters 3, 10 and 13 were enriched for the DAM module (Fig. 5j). Cluster 10 was characterized by the highest score and percentage of cells for the DAM module and was positive for the immune cell module, which likely reflects an overlapping inflammatory profile and/or proximity to peripheral immune cells (Fig. 5j). Based on spatial proximity, aged microglia were closer to the nearest-neighbor immune cell than young microglia (Supplementary Fig. 3e,f), potentially indicating changes in microglial and immune cell abundance in the aged brain and increased interactions, as suggested by GSEA analysis of aged-increased cluster 10 (Extended Data Fig. 3b). Cluster 13 was proportionally greater in the aged brain and demonstrated both DAM and homeostatic characteristics. Cluster 3 was more abundant in the aged brain and demonstrated DAM and immune enrichment. Cluster 15 was enriched for an ependymocyte module, suggesting proximity to ventricular borders (Fig. 5j and Supplementary Fig. 5d).
Aged senescence- and DAM-linked microglia are transcriptionally distinct from young and old homeostatic microglia
We next investigated region- and age-defined molecular senescent characteristics in old versus young CosMx-defined microglia. Lgals3 expression was most prominent in the age-selective, white matter-concentrated and DAM-expressing cluster 10 (Fig. 6a and Extended Data Fig. 4e). By differentiating Lgals3-enriched microglia from both young and old mice, we identified a conserved profile with many of the same highly expressed genes as in Fig. 4, including Clec7a, Ctsb, Ftl1, Spp1, Trem2 and Tyrobp (Extended Data Fig. 5). GSEA of Lgals3+ microglia revealed enrichment for genes encoding cell surface signaling and immune response regulatory functions that suggest interactions with infiltrating immune cells (Extended Data Fig. 5b).
We integrated our SenBrain panel into the microglia cluster definitions from both young and old mice and observed that for several clusters, SenBrain enrichment was inversely related to the homeostatic-associated microglia module (Figs. 5j and 6b,c and Extended Data Fig. 3c,d). Cluster 10, which was only present in the aged brain, showed the highest overall SenBrain score (Fig. 6c). Interestingly, multiple old microglia clusters exhibited significantly higher SenBrain scores relative to young, including cluster 2, 3, 6, 7, 8, 12, 14 and 15 (Fig. 6c). However, cluster 15 only contained five young cells and cluster 10 had none (Fig. 6c).
Age-exclusive cluster 10 was enriched for senescence domains including cell surface, organelle stress and SASP, implicating this population as a dominant proinflammatory microglial phenotype (Fig. 6d and Supplementary Table 2). Specific senescence-related genes enriched in cluster 10 included Cdkn2c/p18ink4c (cell cycle regulation); Ccl3, Ccl5, Csf1, Mmp12 and Spp1 (SASP); Cyba, Lgals3 and Lpl (organelle stress); and Axl and Gpnmb (cell surface markers). Age-dominant cluster 3 was enriched for cell surface, organelle stress and nuclear stress domains (Fig. 6d). Examination of the cell cycle regulation domain revealed that young and old microglia broadly and lowly expressed cell cycle regulatory markers (Extended Data Fig. 6). In examination of microglia expressing any Cdkn, the genes Cdkn1b, Cdkn2a, Cdkn2c, Apoe, Csf3r, Hmgb1, Tgfbr1, Cyfip1, P2ry12 and Sparc were increased relative to microglia without Cdkn expression (Extended Data Fig. 6a). As a top senescence biomarker44, we investigated Cdkn2a+ microglia, which exhibited a heterogenous profile (Extended Data Fig. 6b–d). Across all microglial clusters, 1–5% of cells were positive for Cdkn2a. Among total aged versus young Cdkn2a+ microglia, we observed age-increased expression of Cdkn2a, Apoe, C1qa and Clu, and age-decreased expression of Trem2 and Serpine1, among other factors, suggesting distinct roles for Cdkn2a+ microglia in the aged versus young brain (Extended Data Fig. 6b). The expanded identity and impacts of cells expressing this heterogeneous, low-level expression biomarker warrant further investigation. Overall, the young-versus-old CosMx analysis identified a dominant age-increased microglial population 10 that exhibited multiple domain and biomarker senescence features, which robustly overlapped with a canonical DAM profile. Population 3 was also enriched for the SenBrain signature to a lesser degree. Organelle stress and cell surface remodeling emerged as shared senescence domains, whereas SASP and nuclear stress domains were distinctly enriched among the two age-increased populations.
To further confirm the conserved senescent and DAM signatures, we performed cross-validation of the two CosMx microglia datasets (Fig. 4, dataset 1, and Figs. 5 and 6, dataset 2). Using Harmony batch correction47, the two datasets from combined 9 old and 4 young brain samples were merged and re-clustered (Extended Data Fig. 7a–d). Harmonized cluster 6 exhibited a gene expression profile highly concordant with cluster 7 from Fig. 4 and cluster 10 from Figs. 5 and 6, characterized by expression of Apoe, Ctsd, Ftl1, Fth1, Gpnmb, Lilrb4a/b, Lyz1/2, Cd9, Spp1 and Tyrobp (Extended Data Fig. 7d,i). Cluster 6 displayed the highest SenBrain score (Extended Data Fig. 7e,f) and together with the combined Lgals3-enriched microglial gene set, demonstrated reproducible senescence and DAM signatures across both datasets (Extended Data Fig. 7g–i). Cluster 6 contained enrichment for the SASP, organelle stress and cell surface senescence domains (Extended Data Fig. 7j). Therefore, the senescence- and DAM-linked microglial profile is a conserved feature of aged corticolimbic brain regions and CosMx SMI can reproducibly detect conserved microglial states across independent datasets comprising different sample sizes and distinct age groups.
We next modeled fate trajectories that may give rise to age-increased microglial populations within the aged (n = 4) versus young (n = 4) dataset (Figs. 5 and 6). We examined SenBrain gene expression gradients that may indicate a trajectory from homeostatic microglia (cluster 1) toward the age-exclusive cluster 10 (Fig. 6e). Across clusters 1, 13, 3 and 10, we observed gradient increases in the percentage of cells and average expression for several senescence and DAM-associated markers, including Apoe, Axl, C4bi, Cd9, Clec7a, Ctsb, Lgals3 and Tyrobp. We applied Monocle3 trajectory inference48, excluding peripheral clusters (2, 6, 11 and 12, as defined by the UMAP) that were unreachable from the main core partition (Fig. 6f). Our initial model included homeostatic cluster 1 as the pseudotime root and proceeded to cluster 13, following to 3, which then branched into 9, 10 or 15 → 8 terminal populations (Fig. 6g(i)). This 1 → 13 → 3 → 10 trajectory tracked with the age-increased transcript gradients described above (Fig. 6e) and was characterized by proportional decreases in young cell abundance (Fig. 5e(ii)). To explore alternative trajectories resulting in the age-increased populations, we interrogated a retrograde model with terminal cluster 10 as the pseudotime root (Fig. 6g(ii)). Inferred retrograde trajectories branched from cluster 10 to 3, bearing both DAM and SenBrain-enriched genes, then branched toward three trajectories: (1) ependymocyte-, astrocyte- and oligodendrocyte-associated cluster 15 and astrocyte-associated cluster 8, (2) choroid plexus and oligodendrocyte-associated cluster 9 and (3) DAM- and homeostatic-microglia-associated cluster 13 with continued branching toward neuron-associated populations, including homeostatic cluster 1 (Fig. 6g(ii)). Oriented from cluster 10 as the root, pseudotime values were inversely correlated with SenBrain and DAM module scores, and the greatest pseudotime distance from cluster 10 was represented in terminal gray matter populations (Supplementary Fig. 6). Based on trajectory branching of subpopulations, we determined that the retrograde-inferred astrocyte-, ependymocyte- and oligodendrocyte-associated microglial trajectory involving population 15 (8 → 15 → 3 → 10) was the most direct route to cluster 10 and involved the fewest branch points, whereas the gray matter-enriched homeostatic microglial trajectory (1 → 13 → 3 → 10) contained more branch points (Fig. 6g(ii)). Critically, gene expression gradients and the pseudotime trajectory modeling suggest that microglia harboring senescent and DAM molecular characteristics (including Lgals3 expression) are not a single static population. Age-increased populations 3 and 15 are possible precursors to the age-exclusive senescent DAM population 10 (Fig. 6e–g). The models suggest that senescent DAM cells may emerge along multiple possible molecular trajectories, probably resultant of anatomically defined microenvironment priming that ultimately confers common age-selective transcriptional signatures.
Senotherapeutics blunt age-related senescence and DAM gene expression in white matter
We next tested the hypothesis that systemic targeting of senescent cells would reduce the abundance of senescence and DAM markers in aged fimbria-fornix and corpus collosum. We treated old p16-InkAttac mice with either vehicle, AP20187, which induces cell death in p16-expressing cells49, or the senolytic BCL2-inhibitor venetoclax50,51. Young vehicle-treated mice were also assessed. AP20187 reduced p16ink4a expression in female fimbria-fornix and corpus collosum white matter (Fig. 7a) and hippocampal gray matter (Fig. 7b). AP20187 and venetoclax reduced age-increased Lgals3 in female fimbria-fornix and corpus collosum (Fig. 7c). AP20187 also reduced Trem2, Tmem173 and Tyrobp in aged female fimbria-fornix and corpus collosum (Fig. 7d–f). Transcriptional profiles were more heterogenous in male mice (Fig. 7).
Fig. 7: Senotherapeutics reduce senescence and DAM gene expression in aged white matter.
a–f, Summarized are RT-PCR RE values from white matter FFCC (a,c–f) and hippocampal (HIP) microdissections from female (pink) and male (blue) brains (b), in comparison with young vehicle-treated (YNG CON, open circles, n = 5 female, n = 5 male), old vehicle-treated (OLD CON, closed circles, n = 6 female, n = 8 male), old AP20187-treated (OLD AP, closed squares, n = 4 female for a, n = 5 female for c–f, n = 8 male) and old venetoclax-treated (OLD VEN, closed triangles, n = 5 female, n = 6 male) groups. Values are normalized RE of sex-matched YNG CON. y axis of panels denote the gene measured: FFCC-p16ink4a (a), HIP-p16ink4a (b), FFCC-Lgals3 (c), FFCC-Trem2 (d), FFCC-Tmem173 (e) and FFCC-Tyrobp (f) gene expressions were compared by one-way ANOVA, and exact P values are shown per group versus OLD CON. Bars represent mean ± s.e.m.
Senotherapeutics restore aged fimbria microglial morphology, identity and abundance to a more youthful state
We next explored how senotherapeutics modulate aged white matter microglial identity, frequency, morphology and distribution. Using the opposite hemisphere of brains applied to gene expression profiling (Fig. 7), we conducted IBA1+ immunofluorescent imaging of microglia in young and old vehicle-treated, old AP20187-treated and old venetoclax-treated p16-InkAttac female fimbria (Fig. 8). Fimbria IBA1+ cells frequently exhibited spindle- or ameboid-like morphologies52 (Fig. 8a and Extended Data Fig. 8a,b). We observed greater abundance of IBA1+ cells in old versus young fimbria (Fig. 8a,b), and old IBA1+ cells exhibited larger cell areas (Fig. 8c). AP20187 treatment in old mice reduced IBA1+ cell density in the fimbria to youthful levels (Fig. 8b). AP20187 or venetoclax treatment decreased IBA1+ cell size, compared with old controls (Fig. 8c). We explored microglia organization to fiber tract architecture by analyzing the long-axis angular offset of each IBA1+ cell to the nearest neuronal fiber tract (Fig. 8d(i)). Microglia from old mice displayed greater alignment to fiber tracts than young controls, and AP20187 treatment shifted angular offset of microglia toward youth-like organization (Fig. 8d(ii)).
Fig. 8: Senotherapeutics revert age-associated IBA1+, GAL3+ and APOE+ microglial changes in the fimbria.
a, Representative images of IBA1+ (green) and GAL3+ (red) cells costained with DAPI (blue) in the fimbria from YNG CON, OLD CON, OLD AP and OLD VEN female mice. Scale bars, 100 μm. b, Summarized quantification of IBA1+ cell density (per mm2) in YNG CON (open circles), OLD CON (gray circles), OLD AP (red squares) and OLD VEN (orange triangles) fimbria. c, Summarized quantification of IBA1+ cell size (µm2) in YNG CON, OLD CON, OLD AP and OLD VEN fimbria (for b and c, the P values denote significance in one-way ANOVA tests with multiple comparisons correction, n = 5 YNG CON, n = 6 OLD CON, n = 5 OLD AP, n = 6 OLD VEN mice). IBA1+ cell size included soma and any connected processes fluorescently labeled above local background. d, (i) Representative IBA1+ (green) and DAPI (blue) counterstain of fimbria and DAPI-estimated fiber tracts with computed angle of the fiber tract (magenta) and angle of longest Feret axis of each IBA1+ cell. Scale bar, 20 µm. (ii) Summarized histogram of measured angular offset of IBA1+ cells to fiber tracts from YNG CON, OLD CON and OLD AP fimbria (YNG CON versus OLD CON P < 0.001 and OLD CON versus OLD AP P < 0.001 by two-sided Kolmogorov–Smirnov test compared with OLD CON, KSD(YNG CON vesus OLD CON) = 0.1686, KSD(OLD AP versus OLD CON) = 0.1098). e, Principal component analysis of morphological features of fimbria IBA1+ cells showing group-specific clustering of YNG CON, OLD CON, OLD AP and OLD VEN. Each point represents one mouse. f,g, Summarized quantification of cellular area of GAL3 immunoreactivity in fimbria as percentage of total fimbria area (f) and percentage of IBA1+ cells colocalized with GAL3, in YNG CON, OLD CON, OLD AP and OLD VEN groups (g) (for f and g, the P values denote one-way ANOVA with multiple comparisons correction, n = 5 YNG CON, n = 6 OLD CON, n = 5 OLD AP, n = 6 OLD VEN mice). h, Cumulative distribution plot of IBA1+ cell populations according to cellular GAL3 immunofluorescence intensity from OLD CON (gray) and OLD AP (red) groups. Bars represent mean ± s.e.m. The inset shows histogram depiction of the same data. OLD CON versus OLD AP P < 0.001 denotes significance in two-sided Kolmogorov–Smirnov test relative to OLD CON, KSD = 18.2, n = 6 OLD CON/5 OLD AP mice. i, Representative immunofluorescence images for IBA1 (green), APOE (red) and GAL3 (blue) in fimbria per experimental group. Scale bars, 100 μm. j, Summarized quantification of APOE+ cell fluorescence intensity per group. k,l, Summarized quantification of percentage of APOE+ cells colocalized with (k) IBA1 or (l) GAL3. m, Summarized quantification of distance between APOE+ cells and the fimbria midline in two spatial dimensions. For j–m, the P values denote significance in one-way ANOVA tests with multiple comparisons correction. n = 4 YNG CON, n = 6 OLD CON, n = 5 OLD AP, n = 5 OLD VEN. All bars represent mean ± s.e.m.
Given the high prevalence of ameboid- and spindle-shaped microglia in aged white matter, we applied a microglial phenotyping composite index that captures neuroinflammatory morphologies53 to quantify morphological properties of fimbria IBA1+ microglia across the experimental conditions (Fig. 8e and Extended Data Table 1). Old control microglia were distinct from young, old AP20187 and old venetoclax microglia, defined by larger cell size. Microglia from AP20187-treated mice shifted toward the principal component profiles of young mice, associated with cell density and tract-angle offset properties. Microglia from venetoclax-treated mice and two AP20187-treated mice clustered tightly together, mainly defined by increased cell circularity and decreased cell size. These findings suggest that morphological properties of aged microglial populations in fimbria are partially reverted by senotherapeutic interventions.
We also tested whether the senotherapeutics alter microglia density and morphology in the young brain. Vehicle, AP20187 or venetoclax was administered by the same regimen to 4-month female p16-InkAttac mice (Supplementary Fig. 8). Venetoclax decreased IBA1+ microglial abundance in the young fimbria (Supplementary Fig. 8a,b). Both AP20187 and venetoclax increased IBA1+ cell size (Supplementary Fig. 8c), decreased longest-axis cell length (Supplementary Fig. 8d) and decreased cellular density (Supplementary Fig. 8e) compared with young controls. These results indicate that p16 or BCL2 targeting impact microglia morphology in the young brain (Supplementary Fig. 8f), albeit in a different manner than in the old brain.
Senotherapeutics revert age-altered GAL3+ and APOE+ cell abundance and organization in the fimbria
We next assessed how senotherapeutics influence microglia positive for DAM and senescence markers in the aged fimbria. The average size of aged IBA1+GAL3− cells (46.8 ± 2.0 μm2) was similar to IBA1+ cells from young mice (51.1 ± 2.7 μm2), whereas aged IBA1+GAL3+ cell size was 109.6 ± 15.4 μm2 (Extended Data Fig. 1f), suggesting that the age-dependent increase in IBA1+ cell size in fimbria partially corresponds to IBA1+GAL3+ microglia. In old control female mice, GAL3+ cells occupied greater fimbria area compared with young (Fig. 8f). Accordingly, we tested whether senotherapeutics alter age-increased IBA1+GAL3+ signals (Fig. 8a,f,h). AP20187 or venetoclax reduced the area and frequency of IBA1+GAL3+ cells in aged female fimbria (Fig. 8f,g). AP20187 did not significantly alter GAL3 area in old male fimbria compared to control (Extended Data Fig. 8c). GAL3 fluorescence intensity of remaining IBA1+GAL3+ cells from AP20187-treated female mice was lower compared with old controls (Fig. 8h). Although venetoclax decreased Lgals3 transcript (Fig. 7c) and reduced the total number of IBA1+GAL3+ cells in white matter (Fig. 8g), we did not detect differences in GAL3 fluorescence intensity of remaining IBA1+GAL3+ cells (Extended Data Fig. 8d,e). This collectively suggests distinct effector populations in venetoclax-mediated BCL2 versus AP20187-mediated p16 targeting converge on GAL3+ microglia.
Based on the aged fimbria GeoMx and CosMx profiles demonstrating pronounced Apoe expression in GAL3/Lgals3+ microglia, we performed immunostaining for APOE, GAL3 and IBA1 (Fig. 8i). We observed high APOE+ cell density in the fimbria and corpus callosum (Extended Data Fig. 9a), as well as colocalization of APOE and GAL3 (Fig. 8i). APOE+ cells were seldom found in the hippocampus (Extended Data Fig. 9a). Old control mice had significantly greater APOE immunofluorescence intensity than young mice, and AP20187 and venetoclax both reduced APOE signal (Fig. 8j and Extended Data Fig. 9b). AP20187 also reduced the percentage of APOE cells colocalized with IBA1 (Fig. 8k). Both senotherapeutics reduced the percentage of APOE+ cells colocalized with GAL3 (Fig. 8l). Interestingly, we discovered a distinct architectural pattern in white matter, characterized by localization of APOE+ cells to the lateral ventricle in old mice (Extended Data Fig. 9a). We quantified the relative distance of APOE+ cells from the fimbria midline in two dimensions (Fig. 8m). AP20187- or venetoclax-treated mice exhibited more dispersed APOE+ cell organization characterized by trends of decreased distance from the midline, relative to old control mice (Fig. 8i,m and Extended Data Fig. 9c–f). Collectively, the senotherapeutics reverted microglial morphological and molecular properties in the aged fimbria to more youthful states.