Cell aging causes mitochondrial DNA release from stress and increases inflammatory signals

Cellular senescence refers to the irreversible growth arrest that occurs as a response to different stressors^1,4. Senescent cells secrete multiple factors, collectively known as the SASP⁵. Senescent cells accumulate during ageing and chronic diseases and clearance of senescent cells alleviates several age-related pathologies in mice⁶. These cells therefore represent promising therapeutic targets to prevent age-related disorders.

Mitochondrial dysfunction is a hallmark of cellular senescence^2,7,8. Our earlier data revealed that clearance of mitochondria in senescent cells suppresses the SASP while preserving the cell-cycle arrest², which led us to propose that mitochondria may be promising targets for anti-senescence therapies⁹. Mitochondria also have a major role in apoptosis, a process that involves MOMP, dependent on BAX or BAK, causing rapid cell death³. Apoptosis is also accompanied by BAX/BAK-dependent release of mtDNA into the cytosol^10,11.

Notably, we found that MOMP occurring in a small subset of mitochondria without inducing cell-death, an event called miMOMP¹², is a feature of cellular senescence. During senescence, miMOMP releases mtDNA into the cytosol, which activates the cGAS–STING pathway—a major regulator of the SASP^13,14. Finally, we found that inhibition of miMOMP in vivo decreases inflammatory markers and improves multiple healthspan parameters in aged mice.

Our data support the concept that apoptosis and senescence are regulated by similar mitochondrial-dependent mechanisms and miMOMP is a major contributor to senescence and the SASP. Our results suggest that targeting miMOMP-induced inflammation may be a therapeutic route to improve healthspan.

To investigate whether miMOMP is a feature of cell senescence, we conducted 3D structured illumination microscopy (SIM) analysis of proliferating or senescent human fibroblasts and analysed the co-localization between the outer mitochondrial membrane protein TOM20 and cytochrome c (Cyt c). While Cyt c and TOM20 co-localized in proliferating cells, in senescent cells, we observed a subset of peripheral mitochondria (dissociated from the network and exhibiting a globular structure) that were positive for TOM20, but not for Cyt c (Fig. 1a). Pearson’s correlation coefficient analysis revealed decreased co-localization between Cyt c and TOM20 in radiation-induced senescent cells (Sen IR) and replicatively senescent cells (Fig. 1b). Supporting this, we found increased cytosolic Cyt c (Fig. 1c,d) and cleaved caspase-3 (Fig. 1e) in senescent cells. We next investigated whether miMOMP was accompanied by BAX oligomerization (indicative of BAX activation). BAX oligomerization was analysed using size-exclusion chromatography, which revealed that BAX forms oligomers in senescent, but not proliferating, cells (Fig. 1f). Using the antibody BAX6A7, which detects an active form of BAX¹⁵, we found using 3D-SIM microscopy that, in senescent cells, peripheral globular mitochondria dissociated from the main network show increased expression of activated BAX (Fig. 1g). Increased BAX activation was observed in two independent human fibroblast strains (MRC5 and IMR90) and in irradiation-induced (Sen IR) and ER-RAS-oncogene-induced (OIS) senescence (Fig. 1h,i). These data demonstrate that miMOMP, a consequence of sublethal apoptotic stress, occurs during senescence.

During apoptosis, MOMP leads to mtDNA release into the cytosol, but the inflammatory effects of this phenomenon are inhibited by caspase activity^10,11. We hypothesized that mitochondria undergoing miMOMP may also release mtDNA into the cytosol of senescent cells. Dual immunostaining of TOM20 and DNA using 3D super-resolution Airyscan confocal microscopy revealed that senescent fibroblasts contain an increased number of DNA nucleoids in the cytosol. This observation was consistent between different cell types and irrespective of senescence-inducing stimulus (Fig. 2a,b). Similarly, we observed increased cytosolic nucleoids in senescent cells by immunogold labelling transmission electron microscopy (Fig. 2c). To determine whether the DNA nucleoids were of mitochondrial origin, we performed quantitative PCR (qPCR) to specifically detect mtDNA (D-loop region) and found increases in the cytosolic fractions of Sen IR, OIS and replicative senescent cells (Fig. 2d). Further supporting the mitochondrial origin of cytosolic DNA, we found that the majority of cytosolic DNA co-localized with transcription factor A (TFAM, mitochondrial), one of the core components of the mtDNA nucleoid (Fig. 2e,f). Cytosolic fractionation followed by western blotting confirmed an enrichment of TFAM in the cytosol of senescent cells (Fig. 2g). These results demonstrate that TFAM-bound mtDNA nucleoids, which are a preferred substrate for cGAS¹⁶, are present in the cytosol of senescent cells.

We next investigated whether miMOMP is sufficient to drive senescence and the SASP. Proliferating human fibroblasts were treated with a known inducer of miMOMP, ABT-737, the prototypic BH3-mimetic compound that inhibits anti-apoptotic BCL-2 proteins. We used low concentrations that cause caspase activation but do not induce cell death (Extended Data Fig. 1a,b). Chronic treatment with ABT-737 resulted in a significant increase in the secretion of IL-6 and IL-8, as well as increased expression of IL6, CXCL8 (encoding IL-8), IL1A, IL1B, IFNA and IFNB mRNAs in MRC5 and IMR90 human fibroblasts (Extended Data Fig. 1c–g). Notably, repeated ABT-737 treatment did not affect the rate of cell division, but cells reached replicative senescence a few population doublings earlier compared with the controls and showed a small, but significant, increase in the senescence-associated markers γH2A.X foci and SA-β-Gal (Extended Data Fig. 1h–j). This is consistent with previous findings that BH3-mimetic induced DNA damage can promote senescence¹⁷. Consistent with the reported role of MOMP in cytosolic mtDNA release, ABT-737 treatment resulted in an increased number of extramitochondrial nucleoids (Extended Data Fig. 1k).

The release of mtDNA can occur through BAX and BAK macropores^10,11. Given our observation that senescent cells display BAX activation, we sought to address whether BAX and BAK (two proteins that are essential for MOMP) can facilitate mtDNA release in senescent cells. First, we observed that BAK (but not BAX) protein levels significantly increased in senescent cells when normalized to α-tubulin, but not when normalized to the mitochondrial protein VDAC. Thus, it is possible that the increase in BAK in senescent cells is due to changes in mitochondrial content (Supplementary Fig. 2). We used CRISPR–Cas9 gene editing to generate human fibroblasts that are deficient for both BAX and BAK (Fig. 3a). Combined deletion of BAX and BAK suppressed mtDNA release in senescent cells induced by DNA damage (Fig. 3b). To investigate whether BAX and BAK impacted the SASP, we performed RNA-sequencing (RNA-seq) analysis and a cytokine array using conditioned medium and found that commonly expressed SASP genes were decreased after BAX and BAK deletion (Fig. 3c,d). We confirmed the BAX/BAK-dependent decrease in different pro-inflammatory cytokines by independent enzyme-linked immunosorbent assays (ELISAs) and qPCR analysis of relevant targets (Extended Data Fig. 2a–f). We investigated whether deletion of BAX and BAK impacted senescence-related cell-cycle arrest pathways. Deletion of BAX and BAK did not alter the expression of the cyclin-dependent-kinase inhibitors p21 and p16^INK4a (Fig. 3e–g,m) or increase SA-β-Gal activity (Fig. 3h,m) or the number of γH2A.X foci (Fig. 3i,m), or impact the senescence-associated loss of lamin B1 and HMGB1 (Fig. 3j). No changes in the expression of the proliferation marker Ki-67 (Fig. 3k,m) or proliferation genes that are normally downregulated in senescent cells were observed (Fig. 3l). These data suggest that BAX and BAK regulate the SASP, but not senescence-associated cell-cycle arrest.

The mitochondrial permeability transition pore (MPTP) can also be involved in the release of mtDNA¹⁸. To investigate the role of the MPTP, we treated senescent cells with cyclosporin A, which inhibits the MPTP by binding to mitochondrial cyclophilin D, and then measured mtDNA release and secretion of the SASP factors. Cyclosporin A treatment did not alter the percentage of cells containing cytoplasmic mtDNA and did not suppress the secretion of SASP components IL-6 and IL-8 (Extended Data Fig. 2g,h).

BAX and BAK can trigger other mechanisms that may contribute to the SASP, besides enabling mtDNA release. For example, dysfunctional mitochondria in senescent cells lead to the formation of cytosolic chromatin fragments (CCFs) and the SASP¹⁹. However, loss of BAX and BAK did not affect mitochondrial respiration, generation of reactive oxygen species or the formation of CCF in senescent cells (Extended Data Fig. 2i–k). Moreover, there were no significant differences in mRNA expression of OXPHOS components after BAX and BAK deletion (Extended Data Fig. 2l). Finally, we investigated whether BAX and BAK deletion impacted the SASP in therapy-induced senescence. Combined BAX and BAK deletion reduced the SASP, but not expression of the cyclin-dependent-kinase inhibitors p21 and p16^INK4a when senescence was induced by chemotherapeutic agents doxorubicin and etoposide (Extended Data Fig. 3a–l).

We addressed the possibility that SASP may be influenced by BAX- and BAK-dependent caspase activation, which, during apoptosis, dampens mtDNA-driven inflammation^20–22. As APAF1 is required for mitochondria-dependent caspase activation (Extended Data Fig. 4a), we used CRISPR–Cas9 to delete APAF1 in human fibroblasts (Extended Data Fig. 4b). APAF1-deficient cells reach replicative senescence at the same population doubling as control cells (Extended Data Fig. 4c). Moreover, APAF1 deletion did not affect the expression of cyclin-dependent-kinase inhibitors, the proliferation marker Ki-67 or SASP components when senescence was induced by irradiation (Extended Data Fig. 4d–k). These results indicate that miMOMP-dependent caspase activation does not affect SASP activation under these conditions.

To investigate a role of BAX and BAK in senescence in vivo, we used Bax^fl/flBak^−/− mice and performed tail-vein injection with AAV-TBG-Cre virus to delete Bax in the liver. We exposed mice to 4 Gy of ionizing irradiation (IR), a dose that was previously shown to induce senescence and the SASP in the liver^13,19 (Fig. 4a). The absence of BAX in the liver was confirmed using immunohistochemistry (Fig. 4a). We next characterized the mRNA expression of different pro-inflammatory components 6 days after IR and found that, while these were significantly increased in Bak^−/− mice, deletion of both Bax and Bak suppressed their induction (Fig. 4b). Nonetheless, when we analysed the senescence marker telomere-associated foci, which denotes co-localization between the DNA damage response protein γH2A.X and telomeres, we found that it was induced by irradiation irrespectively of the genotype (Supplementary Fig. 3a,b). IR did not affect the percentage of caspase-3^high liver cells, which suggests that senescence—rather than apoptosis—was induced under these conditions (Supplementary Fig. 3c). To further test the hypothesis that BAX and BAK regulate the SASP, we aged Bax^fl/flBak^−/− mice to 20 months and administered either an AAV9-CAG-eGFP or AAV9-CAG-iCre/eGFP virus through tail-vein injection to delete Bax (Fig. 4c,d). Expression of pro-inflammatory factors known to significantly increase in the liver during mouse ageing were significantly decreased in the absence of both BAX and BAK (Fig. 4e). In agreement with our in vitro data, no differences in the mRNA expression of the cell-cycle-dependent-kinase inhibitors p21 and p16^INK4a were evident after Bax and Bak deletion (Fig. 4f), supporting a central role for BAX and BAK in regulating the SASP, but not the cell-cycle-arrest component of senescence. Consistent with attenuated inflammation, we observed a significant decrease in the number of infiltrating CD68⁺ and CD45⁺CD38⁺ immune cells in the liver of Bak^−/−Bax^−/− mice (Fig. 4g,h). Notably, Bax mRNA expression was positively correlated with the expression of the majority of proinflammatory factors analysed (Fig. 4i). We also examined the impact of Bax and Bak deletion in the bone of aged mice. Bax and Bak deletion in the bone also resulted in decreased mRNA expression of several pro-inflammatory factors that are known to increase during ageing (Supplementary Fig. 3d,e), many of which were positively correlated with Bax expression (Supplementary Fig. 3f). These data demonstrate that miMOMP contributes to the SASP in vitro and in vivo.

mtDNA activates cGAS–STING signalling during apoptosis and after other stresses^10,11 and cGAS–STING signalling regulates the SASP^13,14. However, investigation of the link between cytoplasmic mtDNA release, cGAS–STING and the regulation of the SASP is lacking. To study this, we generated cells lacking mitochondria²³. Human fibroblasts were stably transduced with YFP-Parkin and senescence was induced by X-ray irradiation. Parkin-expressing senescent cells were treated with the mitochondrial uncoupler CCCP, which triggers widespread mitophagy and generates mitochondria-depleted cells²³. Cells were then transfected with isolated mtDNA (Fig. 5a). We confirmed by western blotting that the mitochondrial proteins NDUFB8 and UQCRC2 were absent after Parkin-mediated mitophagy and were not affected by transfection with mtDNA (Fig. 5b). mtDNA was undetectable by qPCR in mitochondria-depleted senescent cells and significantly increased after mtDNA transfection (Fig. 5c). Secretion of the common SASP factors IL-6 and IL-8 was suppressed after Parkin-mediated mitochondrial clearance and partially restored after reintroduction of mtDNA (Fig. 5d). RNA-seq analysis confirmed that Parkin-mediated mitochondrial clearance suppressed the expression of commonly known SASP components and that these were subsequently upregulated after transfection with mtDNA (Extended Data Fig. 5a–c).

Aberrant mtDNA packaging due to TFAM deficiency enhances the expression of a subset of interferon-stimulated genes (ISGs)²⁴. We used our RNA-seq dataset to investigate whether this subset of mtDNA-stress-induced ISGs was upregulated during senescence and whether its expression was dependent on the presence of mitochondria or mtDNA. The majority of mtDNA-stress-induced ISGs was upregulated in senescence, reduced after mitochondrial clearance and restored after reintroduction of mtDNA (Fig. 5e). Using a more comprehensive database of NF-κB-pathway-regulated genes and ISGs, we found a similar pattern (Extended Data Fig. 5b). To further investigate whether mtDNA could have a role in the regulation of the SASP, we induced senescence by X-ray irradiation in Rho⁰ cells lacking mtDNA by prolonged treatment with ethidium bromide. The absence of mtDNA-encoded MT-ND1 and MT-ND5 confirmed that the Rho⁰ cells do not have any mtDNA (Fig. 5f). Aligned with our hypothesis, we found that the absence of mtDNA led to a significant reduction in the secretion of major SASP components IL-6 and IL-8 (Fig. 5g). To corroborate the involvement of mtDNA in the regulation of the SASP, we modified MRC5 fibroblasts to express a tamoxifen-inducible viral DNase (HSV-1 UL12.5) that specifically targets mitochondria, causing mtDNA depletion²⁵. We found that the induction of HSV-1 UL12.5 expression significantly reduced the amount of mtDNA and suppressed the SASP in senescent cells (Extended Data Fig. 6a–e).

Given our data supporting a role for cytosolic mtDNA in cellular senescence, we cultured both wild-type and Tfam^+/− mouse embryonic fibroblasts (MEFs), the latter of which is enriched for cytosolic mtDNA²⁴, until they reached senescence. We found that Tfam^+/− MEFs reached replicative senescence earlier than wild-type MEFs (Supplementary Fig. 4a) and had increased SA-β-Gal activity and mRNA expression of Cdkn2a (encoding p16^INK4a; Supplementary Fig. 4b,c). Tfam^+/− MEFs also showed a passage-dependent increase in both NF-κB-dependent activation of pro-inflammatory cytokines Il6 and Cxcl15 (Supplementary Fig. 4d,e) and genes associated with a type I interferon response (Supplementary Fig. 4f–h). This is consistent with a recent report indicating that Tfam deficiency in T cells accelerates senescence, inflammation and ageing in mice²⁶. Administration of the STING inhibitor SN011 did not prevent the senescence-associated growth arrest in Tfam^+/− MEFs nor the expression of the senescence-associated markers SA-β-Gal and p16^INK4a (Supplementary Fig. 4i–k); however, it prevented the expression of pro-inflammatory factors (Supplementary Fig. 4l–n), consistent with the established role of cGAS–STING signalling in the regulation of the SASP^13,14.

Further strengthening our hypothesis that mtDNA regulates the SASP through cGAS and STING, we found increased co-localization between cGAS–GFP fusion protein and cytosolic TFAM in senescent human fibroblasts (Extended Data Fig. 7a,b). Moreover, we confirmed that CRISPR–Cas9-mediated deletion of cGAS or STING significantly reduced the secretion of SASP proteins IL-6 and IL-8 (Extended Data Fig. 7c–h). Lastly, when we transfected mtDNA into WT human fibroblasts, we observed increased secretion of the SASP components IL-6 and IL-8. However, this increase was significantly reduced in cells in which STING was absent (Extended Data Fig. 7i,j).

Having established that miMOMP results in cytosolic mtDNA release and contributes to the SASP, we next examined why MOMP is limited to a subset of mitochondria in senescent cells. Activated BAX was immunoprecipitated from mitochondrial fractions of senescent and proliferating cells followed by mass spectrometry (MS) analysis to identify regulators of miMOMP. Supporting earlier data, we were able to immunoprecipitate activated BAX in senescent cells, but not in proliferative cells, indicating that miMOMP is increased in senescent cells. MS analyses uncovered multiple BAX6A7-interacting mitochondrial proteins with Gene Ontology term functions in mitochondrial dynamics, energy metabolism, lipid homeostasis and other functions (Supplementary Fig. 5). This prompted us to analyse mitochondrial dynamics in senescent cells, finding that most senescent cells contain hyperfused mitochondrial networks, with relatively few dissociated/fragmented mitochondria (Supplementary Fig. 6a). Confocal live-cell imaging of cells labelled using the CellLight Mitochondria-RFP system confirmed previous reports that the rates of mitochondrial fission are lower in senescent cells compared with in proliferating cells²⁷ (Supplementary Fig. 6b). To investigate whether the few fragmented mitochondria in senescent cells are the ones undergoing MOMP, we immunostained proliferating and senescent cells for TOM20/Cyt c and BAX6A7. We observed that only circular, fragmented mitochondria (TOM20⁺) in senescent cells were negative for Cyt c and positive for BAX6A7 (Supplementary Fig. 6c,d) consistent with the hypothesis that this is the subset of mitochondria undergoing MOMP.

Our observations, as well as our recent research demonstrating a key role for mitochondrial dynamics in regulating miMOMP²⁸, led us to hypothesize that mitochondrial hyperfusion during senescence could be a mechanism to prevent miMOMP and mtDNA release. To test this, we used shRNA to knockdown mitofusin-2 (MFN2) in MRC5 human fibroblasts. Knockdown of MFN2 alone was sufficient to induce significant mitochondrial fragmentation in proliferating and senescent cells (Extended Data Fig. 8a–d). We induced senescence through X-ray irradiation in cells expressing control shRNA or MFN2 shRNA. MFN2 knockdown significantly increased the frequency of mitochondria displaying activated BAX (BAX6A7⁺) (Extended Data Fig. 8a,e) and the number of cytosolic mtDNA nucleoids (Extended Data Fig. 8b,f) in senescent cells. Senescent cells expressing MFN2 shRNA had increased expression of the SASP components IL-6, IL-8 and IL-1β (Extended Data Fig. 8g), but no change in the expression of the cyclin-dependent-kinase inhibitors p16^INK4a and p21 (Extended Data Fig. 8h,i). To validate these findings, we generated MFN2-depleted IMR90 fibroblasts using CRISPR–Cas9 genome editing. As before, we observed hyperfragmentation of mitochondria in proliferating and senescent MFN2-depleted cells (Supplementary Fig. 7a–c). However, only after induction of senescence did we observe a significant increase in the frequency of mitochondria positive for BAX6A7 (Supplementary Fig. 7d), the number of cytosolic DNA nucleoids (Supplementary Fig. 7e,f) and increased SASP (Supplementary Fig. 7g) in MFN2-depleted cells compared with in the controls. To investigate a role for mitochondrial fission using an alternative method, we examined whether treatment with carbonyl cyanide m-chlorophenyl hydrazine (CCCP), which causes extensive mitochondrial fission, exacerbated mtDNA release and the SASP in senescent cells. Indeed, senescent cells treated with CCCP showed increased mitochondrial fragmentation, frequency of cytosolic mtDNA foci and mRNA expression of SASP factors (Supplementary Fig. 8a–d). These data demonstrate that mitochondrial dynamics regulate miMOMP-induced SASP in senescent cells (Supplementary Fig. 8e).

We next assessed whether mtDNA release and the SASP could be suppressed using pharmacological inhibitors of MOMP. We investigated the small-molecule BAX inhibitor BAI1, which inhibits conformational events in BAX activation preventing BAX mitochondrial translocation and oligomerization^29–31. Aligning with published data, BAI1 treatment had a protective effect against BH3-mimetic induced cell-death in BAK-knockout cells, but not in BAX-knockout cells, consistent with a BAX-specific inhibitory effect (Extended Data Fig. 9a,b). We next investigated whether BAI1 impacted senescence. BAI1 was effective at preventing mtDNA release, BAX activation and the SASP in MRC5 and IMR90 senescent human fibroblasts (Fig. 6a,b and Extended Data Fig. 9c–f). Senescent cells treated with BAI1 were also more resistant to ABT263-induced cell death (Extended Data Fig. 9g). Further supporting that the anti-inflammatory effect of BAI1 in senescence was due to BAX inhibition, we found that inflammation induced by transfection with herring testes DNA in human fibroblasts occurred independently of BAX and BAK and was not affected by BAI1 (Extended Data Fig. 9h–j). Finally, treatment of senescent fibroblasts with eltrombopag, an FDA-approved drug that directly inhibits BAX through a different mechanism from BAI1³², was also effective at suppressing the expression of several SASP factors in senescent fibroblasts (Extended Data Fig. 9k–n).

These results prompted us to investigate pharmacological inhibition of miMOMP as a therapeutic approach targeting senescent cells during ageing. The prevalence of frailty-related characteristics increases with ageing. To investigate whether MOMP inhibition can improve frailty phenotypes during ageing, we treated aged (20 months old) mice with BAI1 for 3 months (Fig. 6c). We found that treatment with BAI1 ameliorated age-related decline in neuromuscular coordination as demonstrated by a significant increase in rotarod latency (Fig. 6c) and improved performance in the pole test, in which BAI1-treated animals were able to maintain balance for a significantly longer duration on a raised rod (Fig. 6d and Extended Data Fig. 10a). Moreover, BAI1 treatment improved forelimb grip strength in old mice and delayed the progression of age-associated frailty symptoms³³ (Fig. 6e,f). Notably, BAI1 treatment improved the healthspan of aged mice without affecting the lifespan (Extended Data Fig. 10b). We also performed a cytokine array in the plasma from vehicle-treated and BAI1-treated mice and, although not statistically significant, we found tendencies for decreased levels of several circulating SASP factors, including IL-1α, IL-6, MCP-1 and TNF (Extended Data Fig. 10c).

Given the improvements in musculoskeletal phenotypes observed in BAI1-treated aged mice, and that senescent cells and the SASP have been shown to have a role in age-related bone loss³⁴, we performed micro-computed tomography (µCT) analysis to investigate the effects of BAX inhibition on bone microarchitecture (Fig. 6g). BAI1 treatment effectively improved spine and femur trabecular bone microarchitecture (Fig. 6h), such that animals treated with the BAX inhibitor had a significantly increased spine and femur trabecular number and decreased trabecular separation (Fig. 6g–i), as well as a higher bone volume fraction in the femur (Fig. 6j). Analysis of femur cortical bone showed that BAI1-treated animals have increased femur cortical thickness and polar moment of inertia, which is a measure of the bone’s resistance to torsion, although these differences were not statistically significant (Extended Data Fig. 10d). Consistent with the hypothesis that miMOMP is a driver of the SASP, we found that BAI1 treatment was effective at significantly reducing the mRNA expression of several pro-inflammatory SASP factors in the bone (Fig. 6k) without affecting genes that are responsible for senescence-associated cell-cycle arrest, such as Cdkn2a (encoding p16^Ink4a), Cdkn1a (encoding p21) and Trp53 (encoding p53) (Extended Data Fig. 10e).

We also found evidence for reduced expression of some inflammatory factors in whole brains from aged mice that were treated with BAI1, including significant reductions in Il6, Mmp13 and Cxcl14 (Extended Data Fig. 10f). Brain pharmacokinetics analysis revealed that BAI1 penetrated the blood–brain barrier in aged mice, reaching concentrations of nearly 1,000 ng per g after 24 h (Supplementary Fig. 9). To elucidate the effects of BAI1 treatment, we conducted single-cell RNA-seq (scRNA-seq), which demonstrated a significant decrease in the frequency of p16^INK4a-positive cells across all brain cell populations after BAI1 treatment (Fig. 6l,m). Moreover, analysis revealed that BAI1 treatment led to a significant decrease in the senescence gene panel SenMayo³⁵ in microglia and oligodendrocytes in which senescence markers have been shown to increase in the ageing brain³⁶ (Fig. 6n). Given our previous observations indicating that miMOMP is a main driver of the SASP, it is conceivable that the decrease in p16^INK4a-positive brain cells mediated by BAI1 could be linked to an inhibition of paracrine senescence³⁷. These findings highlight the potential of targeting miMOMP as a therapeutic strategy for mitigating inflammation and cellular senescence in the ageing brain.

Mitochondrial dysfunction is a hallmark of ageing and cellular senescence and has been shown to regulate the SASP^2,19. Our work demonstrates that senescent cells, despite reported resistance to apoptosis^38,39, display features of mitochondrial apoptotic stress without cell death¹². Our results indicate that BAX and BAK macropores in a small subset of mitochondria are responsible for leakage of TFAM-bound mtDNA nucleoids that contribute to the SASP. Despite the release of mtDNA into the cytoplasm and activation of the cGAS–STING pathway, caspase activation during apoptosis allows this mode of cell death to be largely immunologically silent^20,21. However, we found that caspase activation due to miMOMP, a process that is dependent on APAF1, is insufficient to dampen cGAS–STING signalling and activation of the SASP in senescent cells.

Several reports indicate that nucleus-derived CCFs and activation of retrotransposons have a role in senescence and the SASP by activating the DNA-sensing cGAS–STING pathway^13,14,40. We demonstrate that mtDNA is present in the cytosol of senescent cells and can also engage this pathway. We found that pharmacological inhibition of miMOMP (through BAX inhibition) inhibits the SASP and improves various parameters of healthspan. Although inhibitors have potential off-target effects, we found that genetic inhibition of miMOMP also inhibits the SASP in vivo.

miMOMP in senescent cells is probably underpinned by several factors. Senescent cells often display high apoptotic priming⁴¹. Considering this, we have previously shown a correlation between apoptotic priming and miMOMP²⁸. Our recent research has shown that mitochondrial dysfunction, as observed in senescent cells, serves as a mitochondrial intrinsic signal to prime selective MOMP²⁸. Finally, we found that mitochondrial fission promotes miMOMP-mediated mtDNA release and the SASP. Activated BAX inhibits mitochondrial fusion⁴² most likely limiting the propagation of miMOMP to other mitochondria.

In summary, MOMP is often essential for apoptotic cell death, a terminal cell fate that is considered to be independent from cellular senescence and immunologically silent. Our findings indicate that miMOMP occurs during cellular senescence and can drive the SASP through the release of mtDNA into the cytosol. Importantly, we show that inhibition of miMOMP may be a therapeutic target to counteract age-associated sterile inflammation and improve healthspan.

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41586-023-06621-4.

This work was funded by NIH grants R01AG068048 (to J.F.P.), UG3CA268103 (to J.F.P.), P01 AG062413 (to S.K., J.N.F., N.K.L. and J.F.P.), P01 AG073084 and R01 AR069876 (to G.S.S.), F31 AG062099 (to A.G.S.), R01 AG068182-01 (to D.J.), R01 CA225996 (to S.H.K. and H.D.), R01 DK128552 (to J.N.F.), R01 AG071861-01 (to X.L. and P.D.A.), P01 AG073084 (to P.D.A. and X.L.), R01 AG076515 (to S.K.), U54 AG079754 (to S.K.), R01 AG061875 and RF1 AG070391 (to T.Y.H.), R01AG071861 (to P.D.A.) and R33AG61456-4 and R01AG064165; Department of Defense grant W81XWH-20-1-0792 (to E.G.); U54 AG79754 (to S.K. and N.K.L.); Cancer Research UK grants C40872/A2014, DRCNPG-Jun22\100011 (to S.W.G.T.); the Ted Nash Long Life Foundation (to J.F.P. and D.J.); The Glenn Foundation For Medical Research (to J.F.P. and N.K.L.); Hevolution/AFAR (to D.J.); a Robert and Arlene Kogod Center on Aging Career Development Award (to S.V.); a fellowship from Coordenação de Aperfeiçoamento de Pessoal de Nível Superior—Brasil (Capes)—Finance Code 001 and by the Universidade Federal do Rio de Janeiro (to C.A.M.); Cancer Research UK grant DRCPFA-Nov22/100001 and Wellcome grant 203105/Z/16/Z0 to L.C.G., BBSRC PhD studentship BB/R506345/1 (to G.K. and V.I.K.); UKRI cross-council Newcastle University Centre for Ageing and Vitality Ph.D. studentship MR/L016354/1 (to J.C.) and CRUK Awards C18342/A23390 and DRCRPG-Nov22/100007 and MRC MR/R023026/1 (to D.A.M.). G.S.S. holds the Audrey Geisel Chair in Biomedical Science. MS interactome analysis was performed at the Proteomic Core Facility of IMol Polish Academy of Sciences and supported by ReMedy International Research Agenda (Foundation of Polish Science, MAB/2017/2) and EMBO Installation Grant 5040-2022 (to K.S.). scRNA-seq was performed by the Sanford Burnham Prebys Genomics Core. G.S.S. holds the Audrey Geisel Chair in Biomedical Science and is supported by P30AG068635 as director of the San Diego Nathan Shock Center. We thank G. Nelson and the members of the Newcastle University Bioimaging Unit for technical support.

S.V. was involved in mouse and cell culture experiments, analysis of end points, conceptualization, interpretation and editing the manuscript. H.S. and J.C. were involved in cell culture experiments, analysis of end points, conceptualization and interpretation of data. H.M., M.G.V., E.H.-Y., A.B.L., G.K., M.E., C.A.M. and A.G.S. were involved in cell culture experiments and analysis of end points. J.S.R., C.C., J.M.E.-N., L.S.G. and N.P. were involved in animal experiments and analysis of end points. R.R. conducted Seahorse analysis. J.N.F. and S.K. were involved in conceptualization and functional assessment of bone function. H.D. conducted analyses of BAX oligomerization through FPLC under the supervision of S.H.K. K.S. conceptualized and supervised BAX6A7 co-IP and subsequent proteomics analysis. D.J. oversaw animal experiments and analysis of end points. R.A.P., H.H. and T.Y.H. performed scRNA-seq analysis of the mouse brain. E.Z. conducted pharmacological analyses of BAI1 in the brain under the supervision of E.G. D.S. and X.L. conducted analyses of RNA-seq datasets. L.C.G. and D.A.M. were involved in the supervision of J.C. N.K.L. was involved in the supervision of H.S. V.I.K. oversaw cell culture and microscopy experiments performed by H.S. and G.K. G.S.S. supervised cell culture studies involving Tfam^+/− fibroblasts performed by A.G.S. Z.D. provided the CGAS shRNA lentiviral construct. P.D.A. oversaw RNA-seq analyses. J.F.P. and S.W.G.T. conceived the idea, oversaw the study and wrote the manuscript. All of the authors contributed to reviewing or editing the manuscript.

The RNA-seq datasets generated and analysed during this study are available at the Gene Expression Omnibus (GSE196610↗ and GSE235225↗). The MS proteomics data have been deposited at the ProteomeXchange Consortium through the PRIDE partner repository under dataset identifier PXD040018↗. Source data are provided with this paper.

V.I.K. is a scientific advisor for Longaevus Technologies. S.W.G.T. consults for Exo Therapeutics. The other authors declare no competing interests.

Stephen W. G. Tait, Email: Stephen.Tait@glasgow.ac.uk

João F. Passos, Email: passos.joao@mayo.edu

is available for this paper at 10.1038/s41586-023-06621-4.

The online version contains supplementary material available at 10.1038/s41586-023-06621-4.