Senescent glia link mitochondrial dysfunction and lipid accumulation

Abstract

Senescence is a cellular state linked to ageing and age-onset disease across many mammalian species^1,2. Acutely, senescent cells promote wound healing^3,4 and prevent tumour formation⁵; but they are also pro-inflammatory, thus chronically exacerbate tissue decline. Whereas senescent cells are active targets for anti-ageing therapy^6-11, why these cells form in vivo, how they affect tissue ageing and the effect of their elimination remain unclear^12,13. Here we identify naturally occurring senescent glia in ageing Drosophila brains and decipher their origin and influence. Using Activator protein 1 (AP1) activity to screen for senescence^14,15, we determine that senescent glia can appear in response to neuronal mitochondrial dysfunction. In turn, senescent glia promote lipid accumulation in non-senescent glia; similar effects are seen in senescent human fibroblasts in culture. Targeting AP1 activity in senescent glia mitigates senescence biomarkers, extends fly lifespan and health span, and prevents lipid accumulation. However, these benefits come at the cost of increased oxidative damage in the brain, and neuronal mitochondrial function remains poor. Altogether, our results map the trajectory of naturally occurring senescent glia in vivo and indicate that these cells link key ageing phenomena: mitochondrial dysfunction and lipid accumulation.

Fig. 1. AP1⁺ glia appear with age and have a senescent phenotype.

a, Lifespan of genomic AP1 reporter line (TRE-dsRed; n = 200 flies). b, Representative images of fly brains showing age-onset AP1 activity (dsRed; top) is mostly in glial cells (repo; bottom). c, Quantification of dsRed intensity shows that AP1 activity is higher the central brain (left) versus optic lobes (right) (TRE-dsRed; n = 93 brains). See Extended Data Fig. 1a–c for high-magnification images and quantification of dsRed colocalization with glial versus neuronal markers. a.u., arbitrary units. d,e, Representative images showing SA-β-Gal activity increases with age (d) with quantification (e) (TRE-dsRed; n = 119 brains). f, Brain γH2Av levels increase with age (TRE-dsRed; n = 8 brains per replicate). For gel source data, see Supplementary Fig. 1a. g, Bulk RNA-seq of brains shows that senescence-associated genes increase with age (w¹¹¹⁸; n = 20 brains per replicate). h, Cells were FACS-isolated from 40-day-old brains for bulk RNA-seq (repo-GAL4 > TRE-dsRed;UAS-GFP; n = 500 cells per replicate). i,j, Expression of AP1 subunits (dFos, dJun), AP1-target genes (i) and senescence-associated genes (j) is highest in AP1⁺ glia. See Supplementary Data 1 for differential expression genes. k, Most AP1⁺ glia are non-dividing by EdU labelling at 40 days (TRE-dsRed; n = 39 brains). l, Analysis of live FACS-isolated cells from 40-day-old brains shows that AP1⁺ glia are larger (left) with normal DNA content (right) (n = 4,748 neurons, n = 14,326 AP1^neg glia, n = 490 AP1⁺ glia). m, Distribution of γH2Av staining in fixed FACS-isolated cells from 40-day-old brains (n = 8,609 neurons, n = 845 AP1^neg glia, n = 302 AP1⁺ glia). For all bar graphs, data shown are means. Each point in a microscopy experiment represents one brain; in immunoblot or bulk RNA-seq experiments it represents one biological replicate. All data were collected from two or three independent experiments. Pearson’s correlation (c,e,f). Precise n and P values are in the Source Data. *P-adjusted < 0.05 for sequencing data; ***P < 0.001; **P < 0.01, *P < 0.05 for all other data. All scale bars, 100 μm. Source Data

Fig. 2. Neuronal mitochondrial dysfunction triggers senescent AP1⁺ glia.

a, Reactome pathway enrichment shows neuronal mitochondrial function decreases with age. See Supplementary Data 3 and 4 for differential expression genes and gene ontology (GO) terms. b, Quantification of dsRed intensity shows knockdown of inner complex genes in neurons increases (red) or decreases AP1 activity (black) relative to control (white) (TRE-dsRed;elav-GS>UAS-RNAi as indicated on the x axis; n = 11–19 brains per genotype). c,d, Neuronal knockdown of inner complex genes elicits AP1⁺ glia (c) similar to natural ageing (d). See Extended Data Fig. 3b,c for quantification and high-magnification images. e, PCA of bulk RNA-sequenced brains at 10 days of age (n = 20 brains per replicate). f–i, Neuronal loss of ND42 reduces neuron-specific processes (f) and genes (g) with increased DNA damage pathways (h) and genes (i). See Supplementary Data 6 and 7 for enriched terms and differential expression genes (TRE-dsRed;elav-GS>UAS-ND42-RNAi versus >UAS-mCherry-RNAi). j, Representative immunoblot showing neuronal ND42 knockdown increases brain γH2Av levels, legend as in e; quantification in Extended Data Fig. 5c. For gel source data, see Supplementary Fig. 1b. k, Real-time qPCR of 10-day-old brains shows neuronal loss of ND75 and NP15.6 (two AP1-activating RNAi lines, b) increases Irbp while reducing Vglut and Dop1R (TRE-dsRed;elav-GS>UAS-RNAi as indicated on the x axis; n = 20 brains per replicate). l, FACS-isolated and bulk RNA-sequenced 10-day-old RNAi-induced dsRed⁺ cells express senescence-associated genes (n = 500 cells per replicate). See Supplementary Data 8 for differential expression genes. m, PCA shows 10-day-old RNAi-induced dsRed⁺ cells cluster with naturally occurring AP1⁺ glia. For bar graphs (g,i,l), data shown are means. Each point in a microscopy experiment represents one brain, and in immunoblot, real-time PCR and bulk RNA-seq experiments it represents one biological replicate. All data were collected from two or three independent experiments. A two sample t-test was used (b,k). Precise n and P values are in the Source Data. *P-adjusted <0.05 for sequencing data; ***P < 0.001; **P < 0.01, *P < 0.05 for all other data. All scale bars, 100 μm. Source Data

Fig. 3. Dampening glial AP1 activity extends animal lifespan and health span.

a, Lifespan of flies with glial AP1 activity blocked by expressing puckered (UAS-puc) or a dominant negative dFos (UAS-dFos^DN) for 7 (top), 3 (middle) or 1 day(s) per week (bottom) (n = 100 flies per genotype and condition). Grey lines indicate days animals were fed the geneSwitch activating drug, RU-486. See Extended Data Fig. 6a for comparison to UAS-GFP controls. b,c, Blocking glial AP1 activity for 1 day per week improves climbing ability at 35 days of age, measured by vial height reached after 30 s (n = 30) (b) and heat shock (HS) survival measured at 24 h after mild heat exposure (n = 15 per cohort) (c). Legend as in a. d, Blocking glial AP1 activity for 1 day per week reduces SA-β-Gal activity (n = 13–29 brains). e, Representative images of d. See Extended Data Fig. 6b–e for additional images and experimental conditions. Each point in a microscopy experiment represents one brain, and in climbing experiments it depicts one fly (averaged from three independent measurements), and in heat shock it depicts one cohort of flies. All data were collected from two or three independent experiments. Kaplan–Meier survival with pairwise comparison by log-rank test was used in a. Two-way analysis of variance (ANOVA) with Tukey’s comparison was used in b,c. One-way ANOVA with Tukey’s comparison was used in d. Precise n and P values are provided in the Source Data. *P-adjusted <0.05 for sequencing data; ***P < 0.001; **P < 0.01, *P < 0.05 for all other data. NS, not significant. Source Data

Fig. 4. AP1 contributes to FFAs in AP1⁺ glia and TAGs in AP1^neg glia.

a, Bulk RNA-seq of brains shows that blocking glial AP1 activity for 1 day per week reduces lipid metabolism genes at 42 days of age (analysis on shared differential expression (DE) genes between repo-GS>UAS-dFos^DN versus >UAS-GFP and repo-GS > UAS-puc versus >UAS-GFP). See Supplementary Data 9 and 10 for terms and differentially expressed genes. b,c, BODIPY⁺ LDs increase with age (TRE-dsRed; n = 13–15 brains) (b) with representative images (c). d, Representative images showing most BODIPY⁺ LD (green) are in glia (red; repo-GAL4>UAS-tdTomato^CYTO). See Supplementary Video 1 and Extended Data Fig. 7f–i for additional experiments. AL, antenna lobes. e–g, Blocking glial AP1 activity for 1 day per week reduces lipogenesis genes (e) and BODIPY⁺ LD with age (f), with quantification (g) (n = 55 brains). See Extended Data Fig. 8a,b for additional genes and images. h, Lipidomic analysis shows that blocking glial AP1 activity for 1 day per week reduces brain FFA and TAG abundance (n = 8 brains per replicate). See Extended Data Fig. 8c–e for additional data. i, Schematic of the de novo lipogenesis pathway; proteins in black and lipids in grey text. j, AP1⁺ glia express lipogenesis genes (cells as in Fig. 1i). k,l, Lipidomic analysis of FACS-isolated cells from 40-day-old brains (as in Fig. 1h) shows that AP1⁺ glia have a distinct composition by relative (k) and total lipid abundance (l). m,n, Analysis of differentially expressed lipids by log fold change (m) and summed intensity (n) shows that AP1⁺ glia have more FFA but fewer TAGs than AP1^neg glia (n = 100,000 neurons, n = 100,000 AP1^neg glia, n = 35,000 AP1⁺ glia per replicate). Bar graphs in e,j represent mean and in h,l,n represent summed values across 5–6 biological replicates. Each point in a microscopy experiment represents one brain, and in bulk RNA-seq experiments it depicts one biological replicate. Lipid class key: AC, acyl carnitine; CE, cholesteryl ester; CER, ceramide; PC, phosphatidylcholine; SM, sphingomyelin; PE, phosphatidylethanolamine; PG, phosphatidylglycerol; PI, phosphatidylinositol. Precise n and P values are provided in the Source Data. *P-adjusted <0.05 for sequencing data; false discovery rate (FDR) <0.10 for lipidomic data; ***P < 0.001; **P < 0.01, *P < 0.05 for all other data. Source Data

Fig. 5. JUN in senescent human cells promotes LDs in non-senescent cells.

a,b, Representative western immunoblot (a) and quantification (b) of JUN protein in proliferating IMR90 cells (left) compared to senescent IMR90 cells treated with siJUN or siNTC (non-targeting control). For gel source data, see Supplementary Fig. 1c. c,d, Quantification of BODIPY 493/503 intensity in proliferating IMR90 cells (c) treated with conditioned medium from indicated conditions with representative images (d). e, Model for interaction between neurons (blue), AP1^neg glia (green) and AP1⁺ glia (red); declining mitochondrial function in neurons triggers AP1 activity in AP1⁺ glia with age, which promotes LD (yellow) accumulation in AP1^neg glia. Each point represents one biological replicate from three independent experiments. One-way ANOVA with Tukey’s comparison (b,c). Precise n and P values are provided in the Source Data. *P-adjusted <0.05 for sequencing data; ***P < 0.001; **P < 0.01, *P < 0.05 for all other data. FC, fold change. Source Data

Extended Data Fig. 1. AP1 is active in a subset of glia with age.

a, Representatives high magnification images of a 40 d brain showing the co-localization of AP1 activity (dsRed) with glial (repo) or neuronal (elav) markers. b, Proportion of dsRed+ cells in 40 d brains (x-axis shows total number) positive for repo (green), elav (blue) or neither (grey), with the total number of dsRed+repo+ and dsRed+elav+ cells shown in (c). d, Lifespan of genotype used for FACS-isolation of neurons, AP1^neg glia, and AP1+ glia (n = 100 flies). e, Representative gating strategy used for FACS-based isolation of neurons (dsRed^negGFP^neg), AP1^neg glia (dsRed^negGFP + ) and AP1+ glia (dsRed+GFP + ) for bulk RNA-sequencing. f, Expression of neuronal and glial marker genes in FACS-isolated and bulk RNA-sequenced neurons, AP1^neg glia, and AP1+ glia (repo-GAL4 > TRE-dsRed,UAS-GFP; n = 500 cells per replicate). For all bar graphs, data are mean. Each point in a microscopy experiment represents one brain and in bulk RNA-sequencing experiments it depicts one biological replicate. All data were collected from two or three independent experiments. Precise n and P values are provided in the Source Data. *P-adjusted<0.05 for sequencing data; ***P < 0.001; **P < 0.01, *P < 0.05 for all other data. Source Data

Extended Data Fig. 2. Neuronal mitochondrial function declines in age.

a, Bulk RNA-sequencing of FACS-isolated neurons shows lactate dehydrogenase expression increases with age (repo-GAL4 > TRE-dsRed,UAS-GFP; n = 500 cells per replicate). b, Brain ATP, measured as a ratio of ATP to cytotoxicity markers, decreases with age (w¹¹¹⁸; n = 37 brains). c, Total brain mtDNA, measured by PCR, decreases with age (w¹¹¹⁸; n = 8 brains per replicate). d, Schematic showing the complexes associated with the (e) 33 inner complex genes reduced in aged neurons, with expression in AP1+ glia shown for comparison. See Supplementary Data 3 for full figure DE genes. For all bar graphs, data are mean. Each point in ATP/cyto toxicity experiment represents one brain, and in mtDNA experiment and bulk RNA-sequencing experiments it depicts one biological replicate. All data were collected from two or three independent experiments. Precise n and P values are provided in the Source Data. *P-adjusted<0.05 for sequencing data; ***P < 0.001; **P < 0.01, *P < 0.05 for all other data. Source Data

Extended Data Fig. 3. Neuronal knockdown of select inner complex genes activates glial AP1.

a, Gene expression by real-time qPCR of RNAi target gene levels (x-axis) relative to levels in TRE-dsRed;elav-GS > UAS-mCherry-RNAi control (black) (n = 8 brains per replicate). b, Proportion of dsRed+ cells in 10 d brains (x-axis shows total number) positive for repo (green), elav (blue) or neither (grey); genotypes as noted in figure. c, Representatives high magnification images of 10 d brains showing co-localization of AP1 activity (dsRed) with nuclear glial (repo; middle) and neuronal (elav) antibodies in four of the AP1 activating RNAi lines. d, Brain ATP, measured as a ratio of ATP to cytotoxicity markers, in 10 d brains with neuronal RNAi expression relative to UAS-mCherry-RNAi. For bar graph (d), data are mean. Each point in ATP/cytotoxicity experiment represents one brain, and in real-time qPCR experiments it depicts one biological replicate. All data were collected from two or three independent experiments. Precise n and P values are provided in the Source Data. *P-adjusted<0.05 for sequencing data; ***P < 0.001; **P < 0.01, *P < 0.05 for all other data. Source Data

Extended Data Fig. 4. Neuronal knockdown of mitochondrial genes triggers AP1+ glia.

a, Bulk RNA-sequencing of FACS-isolated neurons shows parkin, pink1, marf and Opa1 expression is unchanged with age (repo-GAL4 > TRE-dsRed,UAS-GFP; n = 500 cells per replicate). b, Quantification of brain dsRed intensity shows that knockdown of mitochondrial genes in neurons increases AP1 activity (black) relative to controls at 10 d age (white) (n = 5-11 brains per genotype). Representative images are shown in (c) with high-magnification images in (d) showing co-localization of AP1 activity with nuclear glial (repo) and neuronal (elav) antibodies. e, Quantification of UAS-RNAi efficiency by real-time qPCR relative to UAS-mCherry-RNAi control (black) (n = 8 brains per replicate). For all bar graphs, data are mean. Each point in a microscopy experiment represents one brain, and in real-time qPCR and bulk RNA-sequencing experiments it depicts one biological replicate. All data were collected from two or three independent experiments. Precise n and P values are provided in the Source Data. *P-adjusted<0.05 for sequencing data; ***P < 0.001; **P < 0.01, *P < 0.05 for all other data. Source Data

Extended Data Fig. 5. Neuronal mitochondrial health triggers senescent AP1+ glia.

a, Bulk RNA-sequencing of brains shows that neuronal loss of ND42 increases AP1 subunits and AP1-target genes (n = 20 brains per replicate). Supplementary Data 7 for DE genes. b, ND42 and ND30 levels are selectively reduced with neuronal UAS-RNAi. c, Neuronal loss of ND42 increases γH2Av protein levels at 10 d age (n = 8 brains per replicate). d, Real-time qPCR of 10 d brains shows that neuronal knockdown of mitochondrial genes increases Irbp while reducing Vglut and Dop1R (n = 20 brains per replicate). e, Feeding TRE-dsRed flies the anti-oxidant drug AD4 reduces brain AP1 activity by dsRed at 20 d with representative images in (f). FACS-isolated and bulk RNA-sequenced 10 ddsRed+ and dsRed^neg cells from two AP1-activating inner complex RNAi lines showing (g) expression of neuronal and glial marker genes, (h) AP1 subunits and AP1 target genes and (j) RNAi-targeted genes (n = 500 cells per replicate). See Supplementary Data 8 for DE genes. For all bar graphs, data are mean. Each point in a microscopy experiment represents one brain, and in western immunoblot, real-time qPCR and bulk RNA-sequencing experiments it one biological replicate. All data were collected from two or three independent experiments. Precise n and P values are provided in the Source Data. *P-adjusted<0.05 for sequencing data; ***P < 0.001; **P < 0.01, *P < 0.05 for all other data. Source Data

Extended Data Fig. 6. Blocking glial AP1 activity extends lifespan and mitigates SA-β-Gal activity in age.

a, Lifespan of flies with glial AP1 activity blocked by UAS-puc or UAS-dFosDN for 1 d/wk relative to UAS-GFP controls (n = 100 flies per genotype and condition). Grey lines indicate days animals were fed the geneSwitch activating drug, RU-486. b, Representative images from one experimental replicate showing SA-β-Gal activity decreases when glial AP1 activity is blocked for 1 d/wk (RU-486). c, Quantification of brain SA-β-Gal activity with glial AP1 activity blocked for 7 d/wk by expression of puc (left) or dFosDN (right), with representative images shown in (d). e, Quantification of brain SA-β-Gal activity from flies (TRE-dsRed) maintained on vehicle or RU-486 7d/wk from eclosion with representative images (right). Each point in a microscopy experiment represents one brain. All data were collected from two or three independent experiments. Kaplan-Meier Survival with pairwise comparison by Log-Rank test (a). Two-way ANOVA with Tukey’s comparison (c,e). One-way ANOVA with Tukey’s comparison (d). Precise n and P values are provided in the Source Data. *P-adjusted<0.05 for sequencing data; ***P < 0.001; **P < 0.01, *P < 0.05 for all other data. Source Data

Extended Data Fig. 7. Blocking glial AP1 for 1 d/wk reduces senescence biomarkers in the aging brain, but does not improve mitochondrial health.

a, Hypergeometric comparison showing there is a significant overlap of DE genes between the two AP1 targeting complexes, UAS-puc or UAS-dFosDN. b, Blocking glial AP1 activity reduces age-onset expression of (b) AP1 subunits, AP1-target genes and (c) senescence biomarkers in age while (d) inner complex genes are unchanged. See Supplementary Data 10 for DE genes. e, Age-onset reduction in brain ATP, measured as a ratio of ATP to cytotoxicity markers, persists age in the setting of blocking glial AP1 activity (n = 27 brains). TAG lipase (f, Lip4 or h, bmmr) expression starting from age 20 d reduces BODIPY + LD at 30 d when expressed in glia (repo-GS >) but not neurons (elav-GS >), with representative images shown in (g,i). For all bar graphs, data are mean. Each point in a microscopy and ATP/cyto toxicity experiments represents one brain, and in bulk RNA-sequencing experiments it depicts one biological replicate. All data were collected from two or three independent experiments. Precise n and P values are provided in the Source Data. *P-adjusted<0.05 for sequencing data; ***P < 0.001; **P < 0.01, *P < 0.05 for all other data. Source Data

Extended Data Fig. 8. Intermittent glial AP1 blockade reduces lipogenesis and lipid accumulation with age.

a, Blocking glial AP1 activity for 1 d/wk reduces age-onset increase of lipogenesis and LD biosynthesis genes by bulk RNA-sequencing of whole brains (n = 20 brains per replicate). b, Representative images from one experimental replicate showing that blocking glial AP1 activity for 1 d/wk reduces age-onset increase in BODIPY + LD. c, Total intensity of all measured lipids in brains according to class. Analysis of differentially expressed lipids by (m) log fold-change and (n) summed intensity shows that blocking glial AP1 for 1 d/wk reduces FFA and TAG at 42 d age (n = 8 brains per replicate). Bar graphs in (a) represent mean; bars graphs in (c,e) represent summed values across 5-6 replicates. Each point in a microscopy experiment represents one brain, and in bulk RNA-sequencing experiments it depicts one biological replicate. See methods for lipid key. Precise n and P values are provided in the Source Data. *P-adjusted<0.05 for sequencing data; FDR < 0.10 for lipidomic data; ***P < 0.001; **P < 0.01, *P < 0.05 for all other data. Source Data

Extended Data Fig. 9. FFAs are abundant in AP1+ glia while TAGs are abundant in AP1^neg glia.

a, Blocking SREBP activity by SREBP^DN expression from age 20 d reduces BODIPY + LD at 30 d when expressed in glia (repo-GS >) but not neurons (elav-GS >), with representative images shown in (b). c, Lipogenesis pathway genes (FFA to lipid droplet) in bulk RNA-sequenced FACS-isolated cells from aged brains (data as in Fig. 1h–j). d, Total number of lipid species screened in FACS-isolated cells and whole brains, according to lipid class. e, Principal component analysis of lipidomic profiling of FACS-isolated cells, where each point represents one biological replicate (n = 100,000 neurons, n = 100,000 AP1^neg glia, n = 35,000 AP1+ glia per replicate). f, Log₂ fold change of summed intensity of DE lipids by class between FACS-isolated cell populations with the comparison as indicated on y or x axis. g, Summed intensity of DE lipids by lipid class, with differential comparison defined as FDR < 0.10 for AP1+ glia vs neurons or AP1+ glia vs AP1^neg glia. h, Maximum summed intensity, where maximum value was defined by the highest summed intensity among three cell populations for a given lipid class. i, Representative image showing BODIPY + LD do not co-localize with AP1+ glia by dsRed. Bar graphs in (c) represent mean; data in (g-h) represent summed values across biological replicates. Each point in a microscopy experiment represents one brain, and in bulk RNA-sequencing experiments it depicts one biological replicate. Precise n and P values are provided in the Source Data. *P-adjusted<0.05 for sequencing data; FDR < 0.10 for lipidomic data; ***P < 0.001; **P < 0.01, *P < 0.05 for all other data. Source Data

Extended Data Fig. 10. Blocking glial AP1 activity for 1 d/wk increases oxidative markers and reduces H2O2 resilience.

a, Expressing Lip4 (left) or SREBP^DN (right) in glia for 7 d/wk reduces animal lifespan (n = 100 flies per condition). b, Brain DHE staining at 42 d increases when glial AP1 blocked for 1 d/wk (RU-486), with quantification in (c). d, Blocking glial AP1 activity for 1 d/wk reduces animal resilience to 1% H2O2 feeding. Data represent mean survival (n = 20, two independent experiments). Each point in a microscopy experiment represents one brain. Precise n and P values are provided in the Source Data. *P-adjusted<0.05 for sequencing data; ***P < 0.001; **P < 0.01, *P < 0.05 for all other data. Source Data