Main

The causative agent of the COVID-19 pandemic, the human pathogenic coronavirus SARS-CoV-2, causes upper respiratory infections and subsequently lung disease that may become further complicated by septic multiorgan failure and is associated with increased mortality10,11. Escalating immune activation with massive cytokine release seems to drive severe COVID-19 disease1,2,3, possibly more than the virus infection itself. Mechanisms of viral entry—primarily through the interaction of the SARS-CoV-2 spike (S) protein with the host cell receptor angiotensin-converting enzyme 2 (ACE2)—and consecutive steps of RNA replication, virus assembly and budding represent therapeutic targets12,13. Anti-inflammatory drugs are clinically being used or currently being tested; however, they have led to mixed results14,15. Therefore, more comprehensive and specific measures to control the cytokine storm are needed.

Viral entry reflects a major cellular insult and induces marked biological changes in infected host cells that potentially lead to cellular senescence as a virus-triggered state switch16,17. This can have a disparate impact on virus propagation18,19. Senescence is also acutely stress-induced by activated oncogenes (termed oncogene-induced senescence (OIS)) or by DNA-damaging cancer therapy, and physiologically occurs in embryonic development, wound healing and ageing6. Little is known about systemic ramifications of cellular senescence as an antiviral host defence. We hypothesize here that cellular senescence and its SASP, which is largely composed of pro-inflammatory cytokines, matrix-degrading proteases and pro-coagulatory factors5,6,7, underlie the cytokine storm, macrophage activation, the formation of neutrophil extracellular traps (NETs), endothelialitis and widespread pulmonary thrombosis frequently observed in patients with COVID-19 (refs. 1,8,9).

Properties of VIS

We first tested signs of senescence in human diploid fibroblast (HDF) models exposed to high-titre retrovirus. At day 5 after infection, most of the cells presented with characteristics typical of senescence. The cells had flattened, enlarged morphology and were stained positive in the senescence-associated β-galactosidase (SA-β-gal) assay20. They also exhibited 4′,6-diamidino-2-phenylindole (DAPI)-dense chromatin reminiscent of senescence-associated heterochromatin foci (SAHF)21 and overexpressed p16INK4A, which is an inhibitor of cyclin-dependent kinase 4 (CDK4) and CDK6, in the nucleus (Fig. 1a, Extended Data Fig. 1a). Gene set enrichment analysis (GSEA) of HDF RNA sequencing (RNA-seq) datasets showed that senescence- and SASP-related Gene Ontology terms were skewed towards the VIS state and overlapped with the OIS state of the same cells in Ras-induced senescence (Fig. 1b). GSEA also retrieved host responses to viral infection in OIS, which further underlines the overarching role of a senescent stress response in this context (Extended Data Fig. 1b). Specific analyses of VIS-associated expression changes by quantitative PCR with reverse transcription (RT–qPCR) showed transcriptional changes typical of senescence in VIS cells21,22 (Fig. 1c, Extended Data Fig. 1c). Notably, lower virus titres did not induce senescent cell cycle arrest, which is the goal in retrovirus- or lentivirus-assisted gene transfer experiments (Extended Data Fig. 1d). Multiplex protein analysis showed that VIS cells secreted a plethora of SASP factors, many of which are bona fide NF-κB targets23,24 (Fig. 1d). Importantly, these responses depended on gene activities that are essential in senescence. For example, HDFs expressing the H3K9me3-active demethylase JMJD2C or engineered with a small hairpin RNA (shRNA) to stably knockdown p53 (also known as TP53) to genetically disable OIS22,25, and mouse embryo fibroblasts without intact p53 alleles failed to enter VIS and lacked senescence-typical changes, as assessed using RT–qPCR and multiplex secretome analyses (Fig. 1e, Extended Data Fig. 1e–g). After viral infection, only VIS-capable cells showed activation of the cGAS–STING pathway (Extended Data Fig. 1h), which is a cytoplasmic DNA sensor driving SASP-mediated interferon responses26. Notably, blocking retroviral replication using the reverse transcriptase inhibitor azidothymidine prevented VIS and blunted cGAS–STING pathway activation (Extended Data Fig. 1i, j). cGAS–STING-mediated phospho-activation of the interferon regulatory factor IRF3 (ref. 27), the NF-κB transcription factor p65 (ref. 28) and other nuclear senescence markers, such as p21CIP1 and heterochromatin-associated H3K9me3, were strongly increased in VIS cells. Conversely, non-G1-phase-hyperphosphorylated retinoblastoma (Rb) protein was markedly decreased in these cells (Fig. 1f). Hence, viral infection induces a host cell response reminiscent of and genetically dependent on key features of other well-established types of cellular senescence and their associated secretome.

Fig. 1: Senescence is a universal host cell response to viral stress.
figure 1

a, SA-β-gal, SAHF formation by DAPI and p16INK4A staining of IMR90 cells 5 days after retroviral infection or mock control. b, GSEA probing senescence-associated genes in WI38 HDFs in VIS and OIS. A positive normalized enrichment score (NES) indicates enrichment compared to proliferating counterparts. Presented NES are significant (false discovery rate (FDR) q < 0.05; see Supplementary Information for individual values). c, Relative expression levels of core senescence and SASP transcripts in cells as in a by RT–qPCR compared to mock control. d, Multiplex bead-based protein analysis in the supernatant of HDF IMR90 (as in a), WI38 and Tig3 cells. MIP-1α is also known as CCL3; MIP-1β is also known as CCL4. e, SA-β-gal staining of IMR90 cells expressing JMJD2C, p53 shRNA (shp53) or control vector. f, Immunoblot analysis of nuclear senescence markers in samples as in d (α-tubulin was used as a loading control). g, Quantification of SA-β-gal positivity in human cell lines (RPE1, A549) infected with AAV, lentivirus, HCoV-NL63 or VSV. h, Quantification of SA-β-gal positivity in human cell lines and primary HNEpC cells infected with VSV-ΔG*-CoV-S, VSV-ΔG*-CoV-2-S or VSV-ΔG*/empty vector (VSV-ΔG*-emp). NIH3T3 cells were used as ACE2-negative, infection-resistant controls. i, SA-β-gal staining of stable hACE2-engineered primary nasal epithelial cells (HNEpC-hACE2) infected with SARS-CoV-2 and treated as indicated at 48 h after infection. GS, GS-441425; UT, untreated. j, Relative expression levels of the indicated transcripts by RT–qPCR of cells as in i, k, GSEA of RNA-seq-based GEP from AECs infected with SARS-CoV-2 compared with uninfected controls (n = 4). Preselected senescence-associated terms with FDR q ≤ 0.01 are shown (see Supplementary Information for individual values). Scale bars, 100 µm (a, e) or 5 µm (SAHF row in a). All quantifications in this figure refer to the mean ± s.d. of n = 3 independent experiments (except k) with individual values as dots in bar plots or representative photomicrographs shown.

Full size image

Notably, VIS became detectable in response to a broad spectrum of viruses, including lentivirus, adeno-associated virus (AAV), vesicular stomatitis virus (VSV) and the low-pathogenic human alphacoronavirus NL63 (HCoV-NL63) in human lung carcinoma cells and in non-malignant epithelial cells (Fig. 1g, Extended Data Fig. 2a; see Extended Data Fig. 2b–d for human primary bronchial HBEpC and nasal HNEpC epithelial cells, and the relationship between multiplicity of infection (MOI) and SA-β-gal). To closely mimic SARS-CoV-2 infection, we exposed ACE2-expressing HNEpC cells to SARS-CoV spike-protein-pseudotyped VSV glycoprotein G-deletion mutants (VSV-ΔG*/CoV-S and VSV-ΔG*/CoV-2-S). These pseudotypes all induced senescence in the cells in a spike-protein-dependent manner, as ACE2-negative NIH3T3 cells29 did not show senescence (Fig. 1h, Extended Data Fig. 2e). Retrovirus- and VSV-induced senescent cells exhibited increased levels of reactive oxygen species (ROS) and γ-H2AX DNA damage foci (Extended Data Fig. 2f). Conversely, the ROS scavenger N-acetylcysteine reduced γ-H2AX foci and abrogated VIS. Similarly, GS-441524, the active metabolite of the viral RNA polymerase inhibitor remdesivir, reduced γ-H2AX foci and blunted cGAS–STING pathway activation in VSV-infected cells (Extended Data Fig. 2g, h).

Importantly, HNEpC cells infected with the authentic SARS-CoV-2 virus also displayed SASP-positive senescence that was prevented by p53-shRNA-mediated inhibition of senescence or by GS-441524 treatment (Fig. 1i, j, Extended Data Fig. 2i, j). Moreover, RNA-seq analysis of SARS-CoV-2-infected alveolar epithelial cells (AECs) and publicly available RNA-seq datasets of normal human bronchus epithelial (NHBE) cells and lung cancer cell lines4 demonstrated that they also show robust signs of senescence induction (Fig. 1k, Extended Data Fig. 2k). Thus, virus infection in general and SARS-CoV-2 infection of respiratory epithelial cells in particular induce full-featured cellular senescence, which underscores our hypothesis that VIS might play a critical pathogenic part in COVID-19 lung disease.

Cellular senescence in patients with COVID-19

Next, we asked whether VIS would be a typical feature detectable in patients infected with SARS-CoV-2. We specifically analysed nasopharyngeal and lung tissue specimens from patients with a fatal course of PCR-proven COVID-19. Immunohistochemical analysis (IHA) of the senescence markers lipofuscin, p16INK4A, p21CIP1 and H3K9me3, and interleukin-8 (IL-8; encoded by CXCL8) as a representative SASP marker, showed robust reactivity in samples from infected individuals with or without remaining SARS-CoV-2 RNA in situ but not in individuals without COVID-19 (Fig. 2a, Extended Data Fig. 3a). Single-cell RNA-seq (scRNA-seq) analyses of upper airway mucosa samples provided further evidence of a senescence switch in patients with COVID-19 (ref. 30). The nasopharyngeal specimens showed increased transcript levels of p16INK4A (encoded by CDKN2A) and multiple SASP factors not only in SARS-CoV-2-permissive ciliated respiratory epithelial cells but also in macrophages, which implies that paracrine senescence spread to SASP-attracted macrophages and to other less virus-susceptible p16INK4A-elevated mucosal epithelial cell types (Fig. 2b). Gene expression profiling (GEP) of upper and lower airway specimens showed SASP-reminiscent cytokines in COVID-19 specimens induced at the transcript level (Fig. 2c). scRNA-seq data further showed that predominantly ciliated epithelial cells and macrophages scored positive for p16INK4A transcripts (Extended Data Fig. 3b).

Fig. 2: Patients with COVID-19 exhibit features of VIS.
figure 2

a, IHA of senescence markers and BCL-w, and in situ hybridization of SARS-CoV-2 RNA in the respiratory mucosa of nasopharyngeal specimens from patients with COVID-19 (n = 6 for p16INK4A and IL-8, otherwise n = 5) compared with individuals without COVID-19 (non-COVID-19; n = 2). Quantification of affected areas is presented as the mean percentage of positive cells ± s.d. or the mean percentage of positive area ± s.d. (lipofuscin only). b, scRNA-seq-based GEP of upper airway tissue from patients with COVID-19 (n = 19) and healthy donors (n = 5). Colour-coded average-scaled (z scores) expression levels of the respective transcripts and cell types are shown. nrMa, non-residential macrophage; rMA, residential macrophage. c, GEP of SASP-reminiscent cytokines in upper (nasopharynx) and lower airway (lung) tissue from autopsy samples from patients with COVID-19 compared with individuals without COVID-19 (n = 3 each) by oligonucleotide-based transcript hybridization. d, CD86 IHA of macrophages in lung specimens from patients with COVID-19 (n = 13) compared with individuals without COVID-19 (n = 6). Mean number of positive non-epithelial cells in 0.1 mm2 ± s.d., P = 0.0005. e, Multiplex bead-based protein analysis of serum samples from patients with COVID-19 (n = 32) and healthy donors (n = 8). f, Mass spectrometry-based proteomics analysis of serum from patients with COVID-19 (n = 26) and from healthy donors (serum n = 3 and plasma n = 11). Heatmaps in e and f present scaled (z scores) protein intensities. g, Three-dimensional plot showing serum levels of D-dimer (clotting activity), SERPINA3 (coagulation-related SASP factor) and IL-6 (pro-inflammatory SASP factor) correlated to clinical COVID-19 severity per World Health Organization grades (g3–g6). n = 133 patients. The size of the sphere is proportional to the population size. Scale bars, 50 μm (a, d).

Full size image

Key components of severe COVID-19 are a cytokine storm and macrophage activation syndrome3. We reasoned that the SASP of senescent upper airway epithelial cells may attract and activate macrophages through paracrine senescence induction31. This then contributes to the escalation of macrophage activation syndrome, a condition in which macrophages subsequently infiltrate the lungs. Notably, lungs of patients with COVID-19 showed infiltration by pro-inflammatory CD86+ macrophages and an M1-macrophage polarization profile32,33 (Fig. 2d, Extended Data Fig. 3c). In serum samples taken from patients with COVID-19, SASP peptides such as IL-8, PAI-1 (also known as serpin E1), CCL2 (also known as MCP-1), MMP-1, MMP-9 and TIMP-1 were among the most strongly upregulated factors (Fig. 2e). Selective scanning for clotting-relevant factors by ultrahigh-throughput mass-spectrometry-based proteomics34 also showed that these factors were induced to a large degree in COVID-19 serum probes (Fig. 2f). This result is in line with the transcriptional skewing of the Gene Ontology terms ‘coagulation’ and ‘complement’ in SARS-CoV-2-infected NHBE cells and in retrovirally senescent and even OIS or irradiation-senescent HDF cells7 (Extended Data Figs. 1b, 2k). Further underscoring the pro-thrombogenic propensity in patients with COVID-19, we detected elevated CD62+-marked platelet activation in their blood samples (Extended Data Fig. 3d). Accordingly, clinically more severe cases of COVID-19 exhibited a tight association of high SASP serum levels, clotting-relevant factors and activated coagulation35 (Fig. 2g, Extended Data Fig. 3e, f). There was also a close correlation between senescence marker transcripts and microthrombosis in lungs affected by COVID-19 in situ (Extended Data Fig. 3g, h). Thus, the findings from samples of patients with COVID-19 revealed alterations consistent with VIS and its SASP in the upper airway epithelia. Moreover, the results suggest that SASP-mediated effects are critical contributors to the respective lung disease and associated pro-inflammatory, tissue-damaging and pro-thrombogenic features.

Linking COVID-19 features to VIS

Next, we mechanistically interrogated whether supernatant from VIS cells (SNVIS) might induce COVID-19-typical functionalities in critical target cell types. Indeed, human THP-1 macrophages turned into a CD86+ or CD13+ M1-like state32 when incubated with SNVIS, but not supernatant from non-infected proliferating cells (SNprol.) or genetically senescence-incapable virus-infected HDFs (Fig. 3a, Extended Data Fig. 4a). The HDFs exhibited a strong SASP-reminiscent secretory profile, which is indicative of paracrine senescence, as only replication-incompetent viruses were used here (Fig. 3b, Extended Data Fig. 4b). Notably, transcripts reminiscent of SNVIS-induced CD86+-associated THP1 were similarly detected in lungs of patients with COVID-19 (Fig. 3c). Indeed, SNVIS exposure enforced paracrine, SA-β-gal-positive senescence in macrophages (Fig. 3d), which is all in line with their contribution as a VIS-triggered SASP-amplifying cellular messenger in COVID-19. Underscoring the critical role of an altered endothelium as a cytokine-releasing relay36, SNVIS also induced paracrine senescence in HUVEC human endothelial cells, which—in turn—launched a pro-inflammatory and pro-thrombogenic SASP (Fig. 3e, Extended Data Fig. 4c). Consistent with the microangiopathic thrombosis observed in severely affected lungs of patients with COVID-19, transfer of SNVIS, but not SNprol. or supernatant from virus-infected senescence-incapable HDFs, prompted platelet activation37 (Fig. 3f), accelerated clotting time (Fig. 3g) and provoked NET formation in vitro9 (Fig. 3h). Moreover, we observed enhanced formation of the C5b–C9 membrane attack complex and its ultimate cytotoxic action on HUVEC cells in response to SNVIS-activated human serum (Fig. 3i, j). This result mirrored the strongly elevated complement-activating factors in samples from patients with COVID-19. In essence, the pathogenesis of COVID-19 lung disease appears to be closely linked to senescence-governed immune escalation and pro-thrombotic effects.

Fig. 3: The VIS-associated secretome drives key immune-thrombotic components of COVID-19.
figure 3

a, The percentage of CD86+ cells in the THP-1 macrophage population exposed to supernatant from retroviral VIS compared with proliferating (Prol.) HDF IMR90 (as in Fig. 1a) and WI38 (as in Extended Data Fig. 1a) (left). Same experiment but using the supernatant of senescence-incapable IMR90 cells after mock or virus infection, as in Fig. 1e (right). b, Multiplex bead-based protein analysis of the supernatant of THP-1 exposed to supernatant as in a. SNVIS without THP-1 cells is shown on the right. c, GEP of M1-macrophage polarization-related transcripts by RNA-seq of THP-1 cells exposed to SNprol. or SNVIS and sorted for CD86 positivity (n = 2 each, left) and by oligonucleotide-based transcript hybridization of lung tissue from autopsy samples of patients with COVID-19 compared with individuals without COVID-19 (n = 3 each, right). d, SA-β-gal staining (top) and the CD86+ fraction (bottom) of THP-1 cells exposed to the supernatant of RPE1-infected VSV-ΔG* variants as indicated. e, SA-β-gal staining (top) and transcript expression by RT–qPCR (bottom) of HUVEC endothelial cells incubated with the supernatant of IMR90 cells. f, Platelet activation measured by ADP release from healthy donor platelets incubated with the supernatant of IMR90 cells as in a. n = 5 donors. g, Clotting time in whole-blood samples (n = 4 healthy donors) mixed with the indicated supernatant volumes (WI38, as in b). INTEM captures intrinsic, platelet-dependent clotting, whereas EXTEM captures extrinsic, non-platelet-dependent clotting. h, NET formation after exposing neutrophils to supernatant from HDFs as in b. n = 4 healthy donors, two measurements each. MPO, myeloperoxidase. i, Quantification of the cytolytic complement complex C5b–C9 on HUVEC cells incubated with human serum and supernatant as indicated. ina, inactivated. j, Cytotoxicity assay of HUVEC cells as in i. All quantifications (mean ± s.d. with individual values as dots) and representative photomicrographs are from n = 3 independent experiments unless otherwise noted. *P  < 0.05 by unpaired two-tailed t-test (a, f, h–j), unpaired one-tailed t-test (d) or paired two-tailed t-test (g). Scale bars, 100 µm.

Full size image

Eliminating VIS cells using senolytic drugs

On the basis of these findings, we considered VIS cells as a central therapeutic target in SARS-CoV-2 infection, and the early elimination of these cells might mitigate the course of disease. Previously, we and others presented evidence that senolysis (that is, the selective depletion of senescent cells) by drugs such as bafilomycin, ABT-263 (a BCL-2 family inhibitor; also known as navitoclax) or fisetin and quercetin (flavonoids that inhibit multiple kinases, with the latter typically in combination with the tyrosine kinase inhibitor dasatinib), reduced undesirable local and system-wide effects that persistent senescent cells might otherwise exert38,39,40,41,42. We first profiled transcript and protein expression of anti-apoptotic BCL-2 family members in VIS cells and found that BCL-xL (encoded by BCL2L1), BCL-w (encoded by BCL2L2) and MCL-1 (encoded by BCL2L3) were induced (Fig. 4a, Extended Data Fig. 5a; see also Fig. 2a, Extended Data Fig. 3a, b for elevated BCL2L1 and BCL2L2 expression in patients with COVID-19). This result suggests that navitoclax can selectively kill VIS cells. We also found that the PI3K–AKT–p70S6K axis, p38MAPK and the tyrosine kinase SRC were deregulated in VIS; therefore these serve as potential targets, especially for fisetin and for the combination of quercetin plus dasatinib (D/Q), according to predicted protein–chemical interactions43,44,45 (Fig. 4a, Extended Data Fig. 5b). Indeed, navitoclax, fisetin and D/Q were selectively cytotoxic in retrovirus-driven and VSV-driven VIS cells, but not in equally infected senescence-incapable HDFs in vitro (Fig. 4b, Extended Data Fig. 5c–e). By contrast, the BCL-2-specific venetoclax (also known as ABT-199) did not have these effects in the VIS cells. After senolysis, the release of infectious virions is a potential concern; however, we did not find enhanced infectivity after lysis of VIS cells (Extended Data Fig. 5f, g).

Fig. 4: VIS cells are sensitive to senolytic targeting.
figure 4

a, Immunoblots of BCL-2 family members (left) and kinases (right) in HDF lysates as indicated (α-tubulin was used as a loading control). b, The relative viability of IMR90 cells, as in Fig. 1e, 48 h after treatment with senolytics as indicated or with solvent-only (dimethyl sulfoxide (DMSO)). c, Viability of HNEpC-hACE2 cells infected with SARS-CoV-2 or mock, and treated with senolytics as indicated for c–f. d, The relative change of the CD86+ fraction in the THP-1 cell population after receiving supernatant from HNEpC-hACE2 cells as in c. Of note, supernatant transfer experiments were carried out in the presence of SARS-CoV-2-neutralizing anti-spike antibodies, which prevented the transmission of infection (as evidenced by negative target cell SARS-CoV-2 PCR, data not shown). e, Quantification of SA-β-gal staining of HUVEC endothelial cells after receiving supernatant from HNEpC-hACE2 cells as in d. f, Quantification of C5b–C9 on HUVEC cells exposed to human serum with supernatant from HNEpC-hACE2 cells as in d. All bar plots in this figure show the mean ± s.d. of n = 3 independent experiments with individual values as dots. *P < 0.05 by unpaired two-tailed t-test.

Full size image

Importantly, infection by the authentic SARS-CoV-2 virus rendered HNEpC cells susceptible to navitoclax, fisetin and D/Q (Fig. 4c). Moreover, as seen above with other viruses, supernatant transfer from SARS-CoV-2-infected cells expanded CD86+ THP-1 macrophages and led to paracrine senescence or complement-mediated lysis of HUVEC cells. These effects were not observed when supernatants of epithelial cells pretreated with p53 shRNA or GS-441524 or exposed to any of the three senolytic treatment options were used for transfer (Fig. 4d–f, Extended Data Fig. 5h; see Extended Data Fig. 5i–l for VIS, SASP and senolytic susceptibility in the context of the SARS-CoV-2 variants Alpha and Beta, which have mutations that affect the spike protein).

Senolytic targeting of COVID-19 in vivo

To determine the impact of senolytic treatment on the clinical severity of COVID-19-like lung disease, we first tested early intervention with navitoclax in the well-established Syrian golden hamster model of SARS-CoV-2-driven lung disease46,47. After intranasal infection with SARS-CoV-2, animals developed COVID-19-like pneumonia within the first 3–5 days. scRNA-seq profiling of SARS-CoV-2-infected and mock-infected hamster lungs revealed upregulated p16INK4A, core senescence and SASP factors, as well as Src transcript expression in alveolar epithelial cells, macrophages and neutrophils due to primary virus infection or paracrine senescence induction during days 3–5 after infection (Extended Data Fig. 6a). This analysis also demonstrated induction of BCL2L2 transcripts in these cell types at day 5 after infection (Extended Data Fig. 6b), which provides support for the use of navitoclax as a suitable senolytic. Animals on the navitoclax protocol exhibited moderate adverse effects (Extended Data Fig. 6c, d). However, they had a marked reduction in the number of senescent cells and improvements in the histopathological features compared with the solvent-only-treated but SARS-CoV-2-infected cohort on day 7 after infection (Fig. 5a, Extended Data Fig. 6e–g, Extended Data Table 1). We also detected much lower concentrations of key pro-inflammatory SASP factors in the serum of navitoclax-treated hamsters on day 7 after infection, values that were almost indistinguishable from uninfected controls (Extended Data Fig. 6h).

Fig. 5: Senolytic targeting mitigates disease features of COVID-19 in animal models and patients.
figure 5

a, H3K9me3 IHA in nasopharyngeal (top) and lung (middle) specimens, and lipofuscin-based detection (bottom) of senescent cells in the lungs of golden hamsters infected with SARS-CoV-2 and treated with navitoclax or solvent-only (n = 5 animals each), or mock-infected and untreated (n = 3 animals) at day 7 after infection. b, H3K9me3 IHA and lipofuscin staining in nasopharyngeal sections of Roborovski dwarf hamsters infected with SARS-CoV-2 or mock at day 4 after infection and treatment as indicated. c, Lung tissue of dwarf hamsters as in b on day 4 after infection stained with haematoxylin and eosin. Uninfected healthy hamsters (mock) are shown for comparison ((xiv)–(xvii)). (i), (v) and (x) are overviews of lung sections. Signs of COVID-19-like pneumonia and haemorrhage are indicated: hyperplasia of the bronchial epithelium (double-headed arrows), haemorrhage (asterisks) and necrosis (black arrowheads) of alveolar epithelial cells with neutrophil infiltration (white arrowheads), and endothelialitis (black arrows). d, Quantification of the findings in c. *P < 0.05 by unpaired t-test (one-tailed). e, Multiplex bead-based protein analysis of dwarf hamster serum samples on day 4 after infection (as in b–d). n = 5 each (b–d; n = 4 each in e) for SARS-CoV-2-infected and treated (with solvent (Sol.) or navitoclax (Navito.) animals. n = 2 for healthy (mock) control. f, Constellation plots of patients with COVID-19 (ref. 52) before (day 0) and after 7 days of standard care (SC) only or quercetin plus SC. Variables (Extended Data Table 2) were simultaneously analysed by agglomerative hierarchical clustering to visualize similarity (distance) between patients and separation of the cohorts after treatment. Multi-dimensional analysis of individual patient data, with each red or green dot indicating a patient, the line length represents the distance between patients, and axes showing arbitrary units. n = 21 for each group. All quantifications are mean ± s.d. Scale bars, 20 µm (c, middle two columns), 50 µm (a, b, c (right column)) or 1 mm (c, left column).

Full size image

To evaluate the potential of senolytic intervention in model systems with a more severe COVID-19-like clinical course, we next considered the Roborovski dwarf hamster model48. On day 4 after infection and treatment with navitoclax, D/Q or solvent-only, animals with senolytic intervention presented with a substantial reduction of H3K9me3- or lipofuscin-marked senescent cells in their respiratory epithelium and a profound reduction in lung disease features, especially in the D/Q-treated group (Fig. 5b–d). Both senolytic regimens also produced a reduction in SASP-reminiscent serum cytokines on day 4 after infection in this model, to levels close to healthy controls (Fig. 5e). Effects in the D/Q-treated group were particularly meaningful: 5 out of 5 animals survived, displayed no apparent signs of disease and no or only moderate weight loss (Extended Data Fig. 7a). By contrast, 4 out of 5 animals in the navitoclax-treated group and 3 out of 5 animals in the placebo-treated group had to be euthanized due to weight loss and clinical impairment by day 4 after infection. Last, we investigated the K18-hACE2-transgenic mouse model, in which SARS-CoV-2 infection causes severe and typically lethal disease49,50. Mice were treated in a similar manner to the dwarf hamster model, with oral administrations of navitoclax, D/Q or fisetin, and scheduled for lung analysis on day 6 after infection. For the solvent-only-treated mice, 2 out of 5 died on or before day 6, whereas 1 out of 5 of the navitoclax-treated mice died and the weight of another fell underneath the accepted weight limit on day 6. By contrast, all the D/Q-treated and fisetin-treated mice were alive, had limited weight loss and presented without clinical symptoms on day 6 (Extended Data Fig. 7a, b). Although overall histopathological analyses did not show overt differences regarding COVID-19-like parameters (Extended Data Fig. 7c; see Extended Data Fig. 7d regarding virus loads in this model and the hamster models), markers of senescent cells were significantly lower in the lungs of senolytic-treated mice across all treatment regimens (Extended Data Fig. 7e).

Quercetin was recently tested as an oral formulation with sunflower lecithin against standard care in two randomized clinical trials (ClinicalTrials.gov identifiers NCT04578158 and NCT04861298) for patients with confirmed SARS-CoV-2 infection and mild COVID-19-associated symptoms. In both trials, based on 152 patients51 and 42 patients52, senolytic intervention led to clinical improvement51,52. Agglomerative hierarchical clustering of data from the second trial52 across numerous parameters separated the patient population—which was indistinguishable at baseline—into quercetin and standard care at day 7 (Fig. 5f, Extended Data Table 2). Moreover, collective analysis of all 194 patients concluded that the quercetin-treated group showed significant risk reductions regarding the needs of hospitalization and oxygen therapy compared to standard care, particularly in terms of length of hospitalization, referral to the intensive care unit and number of deaths (Extended Data Fig. 8a, b). Together, early senolytic intervention during SARS-CoV-2 infection significantly attenuates COVID-19-like lung disease and systemic inflammation.

Our investigation marks VIS as a central pathogenic principle and a valid therapeutic target for SARS-CoV-2 infection to prevent severe COVID-19 lung disease. The primary virus–host cell encounter induces a senescence response associated with massive pro-inflammatory cytokine secretion that complements virus spreading to the lower airway tract. Our data pinpoint macrophages as pivotal cellular messengers in this process. Attracted to and primed by upper airway VIS cells, macrophages undergo a secondary, SASP-amplifying senescent-state switch. Subsequently deployed in the lungs, they further induce senescence, particularly in endothelial cells. System-wide and locally active pro-coagulatory and complement-activating SASP factors contribute to hallmark histopathological features and the clinical severity of COVID-19. Hence, early senolytic intervention is an appealing strategy to interrupt such an escalating immune cascade at its beginnings. The effects of D/Q, fisetin and navitoclax treatment that we observed here are encouraging, but optimized dose scheduling to maximize their clinical potential and limit potential toxicities is needed. Owing to their good safety profiles, compounds such as the flavonoids quercetin and fisetin are particularly attractive candidates as VIS-targeting senolytics. As already demonstrated for quercetin51,52, fisetin is currently being investigated in COVID-19 trials. Robbins and colleagues53 recently reported enhanced SASP production of pre-existing senescent cells in old mice by pathogen-associated molecular pattern factors such as lipopolysaccharide or the SARS-CoV-2 spike protein. In turn, pharmacological removal of senescent cells by fisetin or D/Q in β-coronavirus-infected mice was clinically effective. This adds to the rationale of senolytic VIS cell elimination, as proposed here, for SARS-CoV-2-induced senescence. Moreover, it will be interesting to explore whether early senolytic intervention can mitigate chronic post-COVID-19 impairment, which is known as long COVID54. Finally, we highlight that although effective vaccinations are available worldwide and will hopefully make COVID-19 history soon, our approach remains conceptually valid in the context of other viral epidemics or pandemics in the future.

Methods

Cell culture and virus production

Cell lines were maintained according to the guideline of the suppliers. The HDF cell lines IMR90, WI38 and Tig3 have been previously described22. The human cell lines A549 and HUVEC were purchased from the American Type Culture Collection (ATCC) and RPE1 (hTERT RPE1)  was provided by A. Schiebel. Primary human nasal or bronchial epithelial cells (HNEpC and HBEpC) were purchased from PromoCell. HNEpC cells with exogenous ACE2 expression (HNEpC-hACE2) were generated by lentiviral infection (hACE2 lentivirus, BPS Bioscience). THP-1 cells were provided by A. Walter. Senescence-incapable cell lines were generated by transduction with lentiviruses encoding JMJD2C or p53 shRNA as previously described22,55.

All viruses used in this work are tagged with green fluorescent protein (GFP), with the exception of HCoV-NL63 and SARS-CoV-2, and infection efficiency was monitored by detecting fluorescence levels in infected cells. Retrovirus was produced by transfecting Phoenix packaging cells with a murine stem cell virus (MSCV) GFP plasmid56. Lentivirus was produced by transfecting 293T cells with psPAX2, pMD2.G (Addgene, 12260 and 12259) and pCDH-CMV-MCS-EF1-GFP plasmids22. AAV was obtained from the Charité Viral Core Facility. Caco-2 cell-adapted wild-type HCoV-NL63 virus at a titre of 3 × 106 PFU ml–1 was provided by the M.L. laboratory. VSV-GFP was a gift from W. Brune. VSV-ΔG* GFP (Kerafast) and plasmids encoding SARS-CoV spike (pKS SARS St19; provided by S. Fukushi), SARS-CoV-2 spike (Wuhan SARS-CoV-2 spike deltaCT28; provided by C. Ratswohl and K. de la Rosa), SARS-CoV-2 spike variants Alpha and Beta (pCDH-CMV-spike-B117-EF1-puro and pCDH-CMV-spike-B1351-EF1-puro synthesized by GENEWIZ) proteins were used to generate mutant pseudotypes. SARS-CoV-2 (hCoV-19/Germany/BY-ChVir-984/2020, sequence reference EPI_ISL_406862) for in vitro applications was propagated in Vero E6 cells (ATCC, CRL-1586) and genome-sequenced to exclude stocks with mutations from further experiments. Infection was monitored by PCR, to detect viral genomic RNA in the supernatant 48 or 72 h after infection, as previously described57. Experiments were conducted under appropriate biosafety conditions with enhanced respiratory personal protection equipment in the BSL-3 facility at the Institute for Virology, Charité, Berlin. In-house generated virus stocks were concentrated ten times using virus concentration kits (TaKaRa Bio and Abcam). High-titre virus infection to induce VIS reflect MOI of 50 for replication-incompetent retrovirus and lentivirus, 10 for replication-incompetent AAV and VSV-ΔG*, 10 for replication-competent HCoV-NL63, 1 for replication-competent VSV, and 0.1 for replication-competent SARS-CoV-2. VIS features were assessed on day 5 (retrovirus and lentivirus), day 3 (HCoV-NL63, VSV-ΔG* and SARS-CoV-2) or day 2 (VSV) after infection.

Patients with COVID-19, individuals without COVID-19 and healthy donor material

Upper (nasopharyngeal) and lower airway (lung) tissue formalin-fixed paraffin-embedded (FFPE) samples were derived from deceased patients infected with SARS-CoV-2 and from archive material from patients who passed away without notable clinical signs of a respiratory infection before October 2019 (that is, individuals without COVID-19), blood samples of patients with SARS-CoV-2 and hospitalized patients in the course of their COVID-19 disease, or blood samples from healthy volunteers were used based on approval by the local ethics commission of the Johannes Kepler University (reference number 1070/2020). All patients with COVID-19 were PCR-proven as SARS-CoV-2-positive; however, SARS-CoV-2 RNA was detectable in 4 out of 5 cases of COVID-19 shown in Fig. 2a. All participants gave informed consent. In addition, we conducted re-analyses of additional patient and control materials that were previously obtained and published elsewhere30,34,51,52.

Flow cytometry, immunoblotting, immunofluorescence and immunohistochemical analysis

Flow cytometry, immunoblotting (IB), immunofluorescence (IF) and IHA were performed as previously described24,58. Surface antigen detection by flow cytometry was carried out using fluorescence-conjugated antibodies against human CD86-PE-Cy7 (BioLegend, 374209; 1:200), CD13-APC-Cy7 (BioLegend, 301709; 1:200) and CD62P-PE (BioLegend, 304905; 1:200). Antigen detection by IB was carried out using the following antibodies: BCL-2 (Cell Signaling Technology (CST), 15071; 1:1,000); BCL-xL (CST, 2764, 1:1,000); BCL-w (CST, 2724; 1:1;000); MCL-1 (CST, 39224; 1:1,000); phospho-p65-Ser536 (CST, 3033; 1:1,000); phospho-IRF3-Ser386 (CST, 37829; 1:1,000); phospho-Rb-Ser807/811 (CST, 8516; 1:1,000); p21CIP1 (Santa Cruz Biotechnology (SCBT), sc-6246; 1:200); p16INK4A (SCBT, sc-1661; 1:200); H3K9me3 (Abcam, ab8898; 1:2,000); phospho-Akt (Thr308) (CST, 13038T; 1:2,000); phospho-p38 MAPK (Thr180/Tyr182) (CST, 4511T; 1:1,000); phospho-p70 S6 kinase (Thr389) (CST, 9234T; 1:1,000); phospho-SRC (Tyr416) (CST, 6943T; 1:1,000); and α-tubulin (Sigma, T5168; 1:2,000). Anti-mouse or anti-rabbit horseradish-peroxidase-conjugated antibodies were used as secondary antibodies (GE Healthcare, RPN4301 and NA931V, respectively). Antigen detection by IF was performed using the following antibodies: phospho-H2A.X (CST, 9718; 1:400); p16INK4A (Abcam, ab211542; 1:250); C5b–C9 (SCBT, sc-58935; 1:250); anti-mouse-Cy3 secondary antibody (Thermo Fisher, A10521; 1:1,000); and anti-rabbit-594 secondary antibody (Thermo Fisher, A11012; 1:1,000). Antigen detection by IHA was performed using a HiDef Detection HRP Polymer System (Cell Marque, 954D-30) with antibodies against p21CIP1 (SCBT, sc-6246; 1:200), IL-8 (Abcam, ab18672; 1:200), H3K9me3 (Abcam, ab8898; 1:2,000), BCL2L2 (Abcam, ab190952; 1:1,000) and CD86 (Antibodies-online, ABIN736701; 1:100).

Original scans of the immunoblots presented in Figs. 1f, 4a, with molecular mass markers and crop area indicated, are shown in the Supplementary Information.

In situ hybridization

In situ hybridization was performed as previously described46 using a ViewRNA Tissue Assay Core kit (Invitrogen, 19931) and a ViewRNA Tissue Assay Blue Module (Invitrogen, 19932) according to the manufacturer’s instructions. For the detection of SARS-CoV-2, ViewRNA Type 1 Probe Set COVID19 polyprotein (Invitrogen, CVX-06, Assay ID VPNKRHH) and ViewRNA Type 6 Probe Set Human ACTB (Invitrogen, VX-06, Assay ID VA6-10506) were used. Following hybridization, sections were counterstained with haematoxylin and mounted with ProLong Glass Antifade Mountant (Invitrogen, P36982). Images were taken using an Olympus Slide Scanner VS200.

Quantification of gene transcription

RT–qPCR

RNA isolated from cell pellets using TRIzol (Invitrogen) was transcribed to cDNA using SuperScript II reverse transcriptase (Invitrogen). Individual gene expression was analysed by RT–qPCR using Taqman assays from Applied Biosystems. A list of the Taqman assays used is available in the Supplementary Information.

RNA-seq

Total RNA was isolated using a RNAeasy Mini kit (Qiagen) and sequenced at BGI Genomics. scRNA-seq was conducted as previously described59. Further processing of the raw sequencing data is described in the section ‘Statistics and data analysis’.

Oligonucleotide-hybridization-based transcriptome profiling

RNA was extracted from FFPE tissue sections using a PureLink FFPE RNA Isolation kit (Thermo Fisher Scientific, K156002). RNA was then hybridized with the probe set of the NanoString nCounter Human Tumor Signaling 360 Panel (XT-CSPS-H-T360-12) and analysed on a NanoString nCounter SPRINT.

Quantification of protein expression

Multiplex bead-based protein detection (MAGPIX)

MAGPIX assays were conducted according to the manufacturer’s manual with following bead panels: custom Procartaplex 19-plex (Thermo Fisher Scientific, PPX-19, Assay ID MXAACDR), custom Procartaplex 4-plex (Thermo Fisher Scientific, PPX-04, Assay ID MXCE4XN) and Bio-Plex Pro Rat Cytokine 23-Plex Assay (Bio-Rad, 12005641). The assay was measured using a BioPlex MAGPIX multiplex reader (Bio-Rad) and analysed with a 5PL algorithm provided by the Bio-Plex Manager software (v.6.1). Detailed procedures for sample preparation and measurement are described in the Supplementary Information. Note that owing to a different normalization procedure, values cannot be cross-read between different panels (for example, Fig. 1d compared with Fig. 3b).

Mass-spectrometry-based proteomics

Detailed protocols of sample preparation, mass spectrometry and data processing have been previously described34; for further details see the Supplementary Information. In brief, serum samples were reduced, alkylated, digested and conditioned on a Biomek i7 workstation, and approximately 5 µg of peptides per injection was used for the gradient liquid chromatography–tandem mass spectrometry analysis. The raw data were processed using DIA-NN 1.7.12.

VIS assays in vitro

Growth curve analysis

The HDF IMR90, WI38 or Tig3 cells were infected with MSCV retrovirus at day 0 and day 3, and SA-β-gal activity was measured at day 5. For growth curve analyses, 2 × 104 cells were seeded into 12-well plates and cell numbers were counted at days 3, 6 and 9.

2′3′-cGAMP ELISA assay

Cell cytoplasmic 2′3′-cGAMP concentration was measured using a Caymanchem ELISA kit. In brief, cells were lysed in M-PER extraction reagent (Thermo Fisher) and 10 μg protein lysate was used to determine the 2′3′-cGAMP content.

Analysis of cell viability and senescence

Viability and cell numbers were analysed by Trypan blue dye exclusion or Guava Viacount (Millipore). SA-β-gal activity at pH 6.0 (for human cells) or pH 5.5 (for rodent cells) as a senescence marker in fresh or cryopreserved cells was assessed as previously described60. In situ detection of lipofuscin as a senescence marker in FFPE samples was carried out using SenTraGor reagent (Lab Supplies Scientific; provided by V. G. Gorgoulis) according to the manufacturer’s instructions.

ROS detection

Dihydroethidium (20 μM, MedChemExpress) was directly added to cells and incubated at 37 °C for 20 min. Dihydroethidium was then washed away with PBS. Cells were fixed in 4% paraformaldehyde and staining intensity was analysed using a Nikon fluorescence microscope.

VIS supernatant transfer and secondary, paracrine senescence

Conditioned medium was collected from proliferating and VIS cells after 24 h of incubation and centrifuged at 5,000 r.p.m. for 5 min. The supernatant (SNprol. or SNVIS, respectively) was transferred to HUVEC or THP-1 cells for further analysis. For the supernatant transfer from SARS-CoV-2-infected HNEpC cells, custom-made neutralizing antibodies against SARS-CoV-2 (provided by the H.S. laboratory) were mixed in to block secondary infection.

NET formation assay

NET formation was measured according to the concentration of NET DNA bound to myeloperoxidase (MPO). Neutrophils were isolated from whole blood of four healthy donors. NET formation was quantified as previously described61. In brief, 10 µg ml–1 anti-MPO antibody (Abcam, ab25989) was immobilized on the surface of black MaxiSorp-treated plates (Thermo Scientific). The plate was subsequently washed with PBS and blocked with 5% non-fat milk for 2 h at room temperature. Neutrophils incubated for 4 h at 37 °C in RPMI medium from proliferating or VIS WI-38 or IMR-90 cells were pipetted onto the MPO-coated plate and incubated for 1 h at room temperature. NET formation quantification was performed using a MPO-DNA PicoGreen assay according to the manufacturer’s instructions (Quant-iT PicoGreen dsDNA Assay kit, Invitrogen). Sample fluorescence was measured using a GloMax Discover microplate reader (Promega).

Rotational thromboelastometry

To measure clotting time, a rotational thromboelastometry assay was performed on a ROTEM delta (Werfen) device according to the manufacturer’s instructions. INTEM captures intrinsic, platelet-dependent clotting and EXTEM extrinsic, non-platelet-dependent clotting. Whole-blood samples from healthy donors were spiked with the supernatant from WI38 cells in proliferation or VIS. Exclusion criteria for the healthy donors were coagulation disorders and clotting-affecting medication. Sex distribution was equal among the donors.

Platelet activation and ADP release measurement by VIS supernatant

Platelet activation was measured by flow cytometry as the percentage of CD62/P-selectin-positive cells in whole blood. To measure VIS-induced ADP release, 2 × 106 platelets from healthy donors were resuspended in 200 µl conditioned medium and incubated at 37 °C for 1 h. Supernatant was collected by centrifugation at 2,000g for 5 min, and the ADP concentration was measured using an ADP Colorimetric Assay Kit II (BioVision) according to the manufacturer’s instruction.

Complement-mediated cytotoxicity in HUVEC endothelial cells

Serum from healthy donors (500 µl) was incubated with proliferating or VIS HDF (Fig. 3i, j) or HNEpC cells (Fig. 4f, Extended Data Fig. 5h) in 5 ml medium at 37 °C for 30 min. Heat-inactivated (56 °C for 30 min) serum was used as a control. Supernatant was collected and incubated with HUVEC endothelial cells at 37 °C. After 2 h of incubation, HUVEC cells were fixed in 4% paraformaldehyde, and C5b–C9 immunofluorescence staining was carried out. Alternatively, 24 h after incubation, a cytotoxicity assay was performed using a CyQUANT LDH Cytotoxicity kit (Invitrogen).

Senolytic treatment in vitro and secondary viral infection

VIS cells were treated with venetoclax (1 µM), navitoclax (1 µM), fisetin (20 µM), D/Q (20 µM, reflecting 2 µM dasatinib plus 20 µM quercetin) or AT-101 (1 µM) for 48 h, and cell viability was determined by Trypan blue staining using a Countess II FL Automated Cell Counter (Thermo Fisher). To generate the supernatant used for secondary viral infection, wild-type mouse embryo fibroblasts were infected with VSV for 6 h and treated with 1 µM navitoclax for 24 h. Supernatant was centrifuged and filtered through a 0.22-µm PVDF filter and added to THP-1 cells. After 24 h, secondary infectivity was measured by counting GFP+ THP-1 cells.

SARS-CoV-2-driven COVID-19 animal models

In vitro and in vivo studies related to hamsters were conducted under appropriate biosafety conditions in the BSL-3 facility at the Institute for Virology, Freie Universität Berlin, Germany. All hamster experimentation was approved by the relevant state authority (Landesamt für Gesundheit und Soziales Berlin, Germany; permit number 0086/20), and performed in compliance with national and international guidelines for the care and humane use of animals. Preparation of SARS-CoV-2 virus stock and intranasal infection of Syrian hamsters (Mesocricetus auratus) as a faithful in vivo model of COVID-19 lung disease have been previously described46. In brief, a SARS-CoV-2 wild-type isolate (BetaCoV/Munich/BavPat2-ChVir984-ChVir1017/2020) was grown in Vero E6 cells. To ensure genetic integrity, passage 3 stocks were genome-sequenced, and results showed conformity with the published sequence (GenBank MT270112.1) and confirmed the presence of the furin cleavage site essential for in vivo pathogenesis. Anaesthetized male and female hamsters at 6–10 weeks of age were inoculated with 1 × 105 PFU SARS-CoV-2 in a total volume of 60 µl cell culture medium. Control animals were mock-infected with the same volume of cell culture medium. One day after infection, animals were randomized and half of the animals (n = 5) received a single 85 mg kg–1 body-weight dose of navitoclax in solvent (60% Phosal PG, 30% PEG 300, 20% ethanol) intraperitoneally every day for 6 consecutive days. The other half of infected animals (n = 5) received the same volume of solvent only. The control group (n = 3) did not receive virus or treatment. On day 7 after infection, 1 day after the last treatment, all hamsters were euthanized, and tissue and peripheral blood samples were prepared for further analyses.

The Roborovski dwarf hamster (Phodopus roborovskii), a species highly susceptible to a severe course of SARS-CoV-2 infection, was used under the permit mentioned above. Infections were performed as previously described48. Female Roborovski dwarf hamsters at 5–7 weeks of age underwent the procedure described above for Syrian hamsters with the exception that the total volume of the inoculum was reduced to 20 µl. The same virus stock was used. Animals were randomly distributed into three treatment groups (n = 5) and one non-infected, non-treated group (n = 2). After 6 h of infection, animals were treated orally once a day with navitoclax (85 mg kg–1) or D/Q (5/50 mg kg–1) dissolved in 60% Phosal PG, 30% PEG 300, 20% ethanol and mixed 1:1 with strawberry syrup. The placebo-treatment group received the same solvent/syrup combination without drug. On day 4 after infection, animals were killed, and tissue and peripheral blood specimens were collected for downstream analyses.

Infection of K18-hACE2 mice (Jackson Laboratory strain 034860) with SARS-CoV-2 was performed according to approval by the Institutional Animal Care and Use Committee of the Icahn School of Medicine at Mount Sinai as previously reported62. In brief, 6-week-old female mice were infected with 1 × 104 PFU SARS-CoV-2 (USA-WA1/2020, BEI resources NR-52281) grown in Vero E6 cells and validated by genome sequencing, and randomly distributed into three treatment groups (n = 5 each) and one solvent-only control group (n = 5). From 1 day after infection, randomly chosen animals were orally treated with oral navitoclax (85 mg kg–1), D/Q (5/50 mg kg–1) or fisetin (100 mg kg–1) dissolved in 60% Phosal PG, 30% PEG 300, 20% ethanol every other day. On day 6 after infection (1 day after the last treatment), animals were killed, and tissue specimens were collected for histopathological analyses.

Hamster and mouse lung tissues were processed for histopathology as previously described47 and analysed by board-certified veterinary pathologists (blinded about the treatment) following suggested standard criteria63. Transmission electron microscopy was performed with trachea tissue from infected and treated Syrian golden hamsters. Fixation and further processing of the samples and image acquisition procedures are described in the Supplementary Information.

Statistics and data analysis

Based on previous experience22,24,41,47,50,53,54,57, sample sizes typically reflected three to five replicates. In some experiments, much higher numbers of individual primary material were used as biological replicates. All quantifications from staining reactions were carried out by an independent and blinded second examiner and reflect at least 3 samples with at least 100 events counted (typically in 3 different areas) each. For patients and for animal models, IHA quantification was carried out in SARS-CoV-2-affected areas. Regarding Fig. 5, Extended Data Figs. 6 and 7, quantification of the indicated features was based on the following numbers of individual photomicrographs from animals as indicated in the legends and text: Fig. 5a: nasopharynx H3K9me3 (n = 3 for mock, n = 6 for solvent and navitoclax), lung H3K9me3 and lipofuscin (n = 3 for mock, n = 5 for solvent and navitoclax); Fig. 5b: H3K9me3 (n = 5 for solvent and navitoclax, n = 6 for D/Q), lipofuscin (n = 5 for all three groups); Extended Data Fig. 7e: p16INK4A (n = 5 for solvent, navitoclax and D/Q, n = 4 for fisetin), H3K9me3 (n = 4 for solvent and navitoclax, n = 5 for fisetin and D/Q), lipofuscin (n = 5 for solvent and navitoclax, n = 4 for fisetin, n = 7 for D/Q).

Data analysis was performed with GraphPad Prism 9. For statistical comparisons, a t-test was applied. Unless otherwise stated, a P value of <0.05 was considered statistically significant. For multiple-testing corrections, the Benjamini–Hochberg method was used to control for the false discovery rate64. Bioinformatics analysis was performed in R, v.4.0.3, and Bioconductor 3.12 using various R packages as described in the Supplementary Information in greater detail. High-performance computation was performed in the Research/Clinic cluster of the Berlin Institute of Health. For further analyses of the clinical data from patients with COVID-19 (refs. 51,52) as presented in Fig. 5f and Extended Data Fig. 8a, b, JMP Pro 14 (SAS Institute) was used for agglomerative hierarchical clustering, GraphPad Prism 9 was used for Fisher’s exact test for P values, and a Haldane–Anscombe correction was applied to calculate relative risks. Lipofuscin (SenTraGor) staining was quantified with the pixel counting function of the ImageJ program (bundled with Java 1.8.0_172, NIH)65. No statistical methods were used to predetermine sample size.

Reporting summary

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