Senescent Stromal Cells Promote Cancer Resistance through SIRT1 Loss-Potentiated Overproduction of Small Extracellular Vesicles

Cytotoxic chemotherapy represents an effective modality for cancer treatment, inducing clinical responses associated with a significantly lowered risk of recurrence, but only in a limited fraction of patients (1–3). As selection of genes accurately predicting therapeutic responses might improve cancer outcomes, gene signatures aimed at predicting responses to specific anticancer agents and minimizing drug resistance are being actively evaluated (4). Multiple studies have reported side-effects generated by local and systemic treatments, and their therapeutic benefits may be restrained by tumor-promoting host responses induced by certain forms of cytotoxicity (5). Development of innate and/or acquired resistance are among the major mechanisms underlying diminished responsiveness to anticancer regimens, as disease relapse after an initial response can occur as a result of systemic release of multiple soluble factors, which frequently remodel the treatment-damaged tumor microenvironment (TME; refs. 6, 7).

Cellular senescence is a state of cell-cycle arrest that occurs upon exposure of cells to different stresses, usually resistant to cell death induction. While cellular senescence is beneficial for a few physiologic events such as tissue repair, wound healing, and embryogenesis (8–10), it accelerates organismal aging and is responsible for age-related disorders including cancer (11, 12). Senescent cells synthesize and secrete a plethora of extracellular proteins, covering growth factors, cytokines, chemokines, and proteases, a phenomenon referred to as the senescence-associated secretory phenotype (SASP; refs. 13–15). Expression of most of the SASP components are regulated by kinases including TAK1, p38, mTOR, and Jak2/Stat3, all of which indeed converge on the NF-κB complex and the c-EBP/β family of transcription factors (16). By means of the SASP expression, senescent cells remodel the local tissue via paracrine mechanisms and recruit immune cells in aging tissues (17). In response to damage signals, degradation of the nuclear lamina and elimination of the major structural component lamin B1 not only enhance histone depletion and chromatin remodeling but also forms microtubule-associated protein light chain 3 (LC3/ATG8)-containing cytoplasmic chromatin fragments (18, 19). If not eliminated via exosome secretion, such chromatin fragments can activate the cyclic GMP-AMP synthase and stimulator of IFNγ genes response, further promoting expression of the SASP and type I IFNs (20, 21).

Extracellular vesicles (EV) are membranous vesicles released by almost all cell types and contain a lipid bilayer that protects the luminal contents of proteins, nucleic acids, lipids, and metabolites against harsh environmental conditions (22). According to their distinctive biogenesis and release modes, EVs can be categorized as exosomes, microvesicles, and apoptotic bodies (AB; ref. 23). The cargos packaged within or associated with the EVs can reflect the pathophysiologic state of host cells (24). However, the capacity of senescent cell–derived EVs in orchestrating intercellular communications within the TME and changing pathologic track under therapeutic settings, remain largely unexplored. To this end, we investigated the biogenesis mechanisms of EVs generated by senescent human stromal cells developing the SASP, and disclosed that EVs can shape acquired resistance of cancer cells to anticancer treatments. We further demonstrated the feasibility of modulating EV production to control therapeutic resistance acquired from the treatment-damaged TME, thus presenting a potent avenue to improve therapeutic efficacy by harnessing senescent cell–derived EVs.

Primary normal human prostate stromal cell line PSC27 and breast stromal cell line HBF1203 were derived from fresh surgical samples and maintained in stromal complete medium as described previously (25). Stromal cells at a passage number of 2–5 were employed for experimental assays. Prostate cancer cell lines PC3, DU145, LNCaP, and M12 and breast cancer cell line MDA-MB-231 (ATCC) were routinely cultured with RPMI1640 (10% FBS). All cell lines were tested negative for Mycoplasma contamination and authenticated with short tandem repeat assays.

PSC27 cells were grown until 80% confluent (CTRL) and treated with bleomycin (50 μg/mL) for 12 hours. After treatment, the cells were rinsed thrice with PBS and allowed to stay for 7–10 days in media. Alternatively, HBF1203 cells were treated with doxorubicin (10 μmol/L) for 12 hours to induce senescence. To examine cellular senescence, senescence-associated β-galactosidase (SA-β-Gal) staining and bromodeoxyuridine (BrdU) incorporation were performed. DNA damage extent was evaluated by immunostaining for γH2AX or p-53BP1 foci by following a 4-category counting strategy as formerly reported (25). Random fields were chosen to show SA-β-gal positivity, BrdU incorporation or DNA damage response (DDR) foci, and quantified using CellProfiler (http://www.cellprofiler.org).

Lentiviral vector pLKO.1-Puro (Addgene) was used to clone small hairpin RNAs (shRNA), including those targeting human SIRT1 (1# sense strand 5′- CCGGTGCGGGAATCCAAAGGATAATTCTCGAGAATTATCCTTTGGATTCCCGCTTTTTG-3′; 2# sense strand 5′-CCGGTGCTGATGAACCGCTTGCTATCCTCGAGGATAGCAAGCGGTTCATCAGCTTTTTG-3′), human ABCB4 (#1 sense strand 5′-CCGGCCAATCGAGAATCAGCTATTACTCGAGTAATAGCTGATTCTCGATTGGTTTTTG-3′; #2 sense strand 5′-CCGGCCGAGCATTCTCTGATCCATTCTCGAGAATGGATCAGAGAATGCTCGGTTTTTG-3′), and human miR-34a (Accession NR_029610; #1 sense strand 5′-CCGGAGCAATCAGCAAGTATACTGCCCTACTCGAG TAGGGCAGTATACTTGCTGATTGCTTTTTTG-3′; 2# sense strand 5′- CCGGAATCAGCAAGTATACTGCCCTAGAACTCGAGTTCTAGGGCAGTATACTTGCTGATTTTTTTG-3′). For each target gene, the scramble sense strand was 5′-CCGGTTAGCGACTAAACACATCAATTCAAGAGATTGATGTGTTTAGTCGCTATTTTTTG-3′. Lentiviral vector pLenti-CMV-Puro-DEST (Invitrogen) was used for generation of expression constructs for target genes. Upon production by 293FT cells, lentiviral titers were adjusted to infect approximately 90% of cells. Stromal cells were infected overnight in the presence of polybrene (8 μg/mL), allowed to recover for 48 hours, and selected for 72 hours before analysis. For expression of target genes in either stromal or epithelial cells, total RNA was prepared and subject to qRT-PCR assays (primers listed in Supplementary Table S1).

Proteins were separated using SDS-PAGE gels and transferred onto nitrocellulose membranes. The blots were blocked with 5% nonfat dry milk at room temperature for 1 hour and incubated overnight at 4°C with desired primary antibodies, followed by incubation with horseradish peroxidase–conjugated secondary antibodies (Santa Cruz Biotechnology) and membrane exposure with ECL Reagent (Millipore).

For immunofluorescence (IF) staining, cells were cultured in dishes and preseeded for at least 24 hours on coverslips. Cells were fixed with 4% paraformaldehyde in PBS, blocked with 5% normal goat serum, and incubated with primary antibodies. Alexa Fluor 488 or 594-conjugated secondary antibodies (Invitrogen, 1:400) were used. Nuclei were stained with Hoechst 33342 before fluorescence imaging (Nikon Eclipse Ti-S). Alternatively, a confocal microscope (Zeiss LSM 780) was applied to acquire confocal images (antibodies listed in Supplementary Table S2).

All animals were maintained in a specific pathogen-free facility, with NOD/SCID (Nanjing Biomedical Research Institute of Nanjing University) mice at an age of approximately 6 weeks (∼20g body weight) used. Each group comprised 10 mice, and xenografts were subcutaneously implanted at the hind flank. Stromal cells (PSC27) were mixed with cancer cells (PC3) at a ratio of 1:4 (25 × 10⁴ stromal cells admixed with 1 × 10⁶ cancer cells to make tissue recombinants). Animals were sacrificed at end of the 8th week after tumor xenografting as described previously (26).

For chemoresistance studies, animals received subcutaneous implantation of tissue recombinants and were given standard laboratory diets for 2 weeks to allow tumor uptake and growth initiation. Starting from the 3rd week (tumors reaching 4–8 mm in diameter), mitoxantrone (0.2 mg/kg doses) or vehicle control was administered via intraperitoneal injection on the 1st day of 3rd, 5th, and 7th weeks, respectively. Upon completion of the 8-week regimen, animals were sacrificed, with tumor volumes recorded and tissues processed for histologic evaluation. For SIRT1 activation, SRT2104 (Selleck) was dissolved in suspension (30% PEG, 400/0.5% Tween, and 80/5% Propylene glycol at 30 mg/mL), administered via oral gavage at a 100 mg/kg dose every other week, starting from the 1st day of the 3rd week after xenograft implantation and provided either alone or together with mitoxantrone, until completion of the 8-week regimen.

All animal experiments were performed in compliance with NIH Guide for the Care and Use of Laboratory Animals (National Academies Press, 2011) and the ARRIVE guidelines, and were approved by the Institutional Animal Care and Use Committee of Shanghai Institutes for Biological Sciences, Chinese Academy of Sciences (Shanghai, China).

Randomized control trial protocols and all experimental procedures were approved by the Ethics Committee and Institutional Review Board of Shanghai Jiao Tong University School of Medicine (Shanghai, China) and Zhongshan Hospital of Fudan University (Shanghai, China). Relevant methods were applied following official guidelines. Written informed consent was obtained from all subjects, with experiments conformed to principles set out in the WMA Declaration of Helsinki and the Department of Health and Human Services Belmont Report.

All in vitro experiments were performed in triplicates, and animal studies were conducted with n ≥ 8 mice per group. Animals were distributed into groups of equal body weight, and no animals were excluded from analysis. Sample sizes were estimated on the basis of an 80% power to detect a 50% reduction in tumor volumes observed in mice subject to chemotherapy and/or targeted therapy compared with control mice, accepting a type I error rate of 0.05.

When comparing two conditions, Student t tests with Mann–Whitney tests were used to determine statistical significance. More than two comparisons were made using one-way ANOVA followed by Dunn or Tukey post hoc tests. Two-way ANOVA with Bonferroni correction was used for grouped analysis. Nonlinear dose–response fitting curves were generated with GraphPad Prism 7.0. P < 0.05 was considered significant, and two-sided tests were performed. Unless otherwise indicated, all data in the figures were presented as mean ± SD. Univariate and multivariate Cox proportional hazards model analysis were performed with statistical software SPSS.

The raw RNA-sequencing (RNA-seq) data have been deposited in the Gene Expression Omnibus database (accession codes GSE128282 and GSE128405). The authors declare that all other data supporting the findings of this study are available within the article or its Supplementary Materials and Methods and from the corresponding author upon reasonable request.

As stromal cells represent the major noncancerous mesenchymal components providing structural architecture within the tissue of most solid tumors, we chose to employ a primary normal human prostate stromal cell line, namely PSC27, to initiate the study. Composed of predominantly fibroblasts but with a minor percentage of nonfibroblast cell lineages including endothelial cells and smooth muscle cells, PSC27 cells are primary in nature and develop a typical SASP after exposure to stressful insults such as cytotoxic chemotherapy or ionizing radiation (25–27). We treated these cells with a preoptimized sublethal dose of bleomycin, resulting in enhanced SA-β-Gal staining positivity, decreased BrdU incorporation, and elevated DNA damage foci several days afterwards (Supplementary Fig. S1A–S1C). Data from Illumina sequencing (RNA-seq) indicated that PSC27 cells were strongly expressing multiple SASP factors including but not limited to CXCL8, CCL20, CSF3, IL1α, and CCL3 after treatment (Supplementary Fig. S1D). We isolated stromal cell–derived vesicles with sequential ultracentrifugation before assessment of their size and distribution. Strikingly, quantitative nanoparticle tracking analysis (NTA) showed that the number of senescent (SEN) PSC27 cell released vesicles between 30 nm and 1.0 μm, a size range characterized of major eukaryotic EV subtypes including exosomes and microvesicles (28), was approximately 7 times of their proliferating (PRO) counterparts (Fig. 1A). Furthermore, we noticed a substantial difference of EV size distribution between proliferating and senescent stromal cells. In contrast to EVs from proliferating cells, which were mainly enriched at 72, 102, 121, and 170 nm, EVs from senescent cells displayed an evident redistribution in particle diameter, as reflected by several major peaks at 33, 47, 64, 122, 156, 182, 266, and 314 nm, respectively, although in each case the refereed EVs fell in the category of small extracellular vesicles (sEV) as defined by recent literatures and excluding their identity as other EV subtypes such as ABs (Fig. 1B; refs. 28, 29). The average size shifted from 116 to 161 nm, with a statistical significance between proliferating and senescent cell–derived sEVs, a pattern suggesting potential difference of their cargo contents (Fig. 1C and D). Although there was a small number of EVs displaying sizes beyond such a principal window, they accounted for a limited percentage of overall stromal EVs and were not the main focus of subsequent investigations.

Preliminary in-gel analysis indicated that protein components of sEVs from proliferating and senescent cells apparently differ from each other, when sample loading was normalized to the number of sEVs (Fig. 1E). However, when loading was normalized to the number of parental cells that secreted these sEVs, we noticed substantially enhanced amounts of sEV protein cargoes (Fig. 1F). The findings were supported by immunoblots, which suggested that expression of sEV-specific markers such as syntenin-1, TSG101, and ALIX was markedly increased in total sEV lysates upon cellular senescence (Fig. 1G). We then analyzed the whole lysates of parental cells, and found considerably elevated expression of sEV markers in senescent cells, in contrast to proliferating samples (Fig. 1H). Upon IF staining, we found the expression levels of CD63 (a tetraspanin protein) and TSG101 apparently enhanced in senescent cells, implying the possibility of active synthesis of sEVs in multivesicular bodies (MVB; Fig. 1I). Electron microscopy imaging showed expanding MVBs with an increase in both volume and the number of their internal vesicles in senescent cells, relative to their proliferating counterparts (Fig. 1J–K). Indeed, there was a concomitant upregulation of the vast majority of sEV-associated molecules as indicated by transcript assays, although the expression level of SMPD2/3, two neutral sphingomyelinases associated with biogenesis of both sEVs and microvesicles, appeared increased, as well (Fig. 1L). Further analysis with successive differential ultracentrifugation of pellets from conditioned media, a strategy that effectively separates sEVs from microvesicles (14,000 × g and 100,000 × g of sedimentation, respectively; ref. 30), indicated enhanced secretion of sEVs by senescent cells, while microvesicle release remained largely unchanged, implying engagement of a mechanism specifically responsible for production of sEVs, but not all EV subtypes (Fig. 1M). Notably, appearance of such a distinct expression pattern was accompanied by pronounced upregulation of several hallmark SASP factors including, but not limited to, IL6, IL8, IL1α, WNT16B, MMP3, and GM-CSF (Fig. 1L).

To validate the data, we chose HBF1203, another stromal cell line that was isolated from human breast tissue and consists mainly of fibroblasts upon primary culture. We treated HBF1203 with doxorubicin, a chemotherapeutic agent frequently used in breast cancer clinics. The results suggested that senescence-associated changes including those in cell phenotypes, sEV marker and SASP factor expression, and sEV production, can be readily reproduced in these cells (Supplementary Fig. S1E–S1J). Thus, these data uncover a common feature potentially shared by stromal cells of multiple tissue and/or organ origin.

Given the substantial change of the number and cargo composition of sEVs generated by stromal cells upon senescence, we interrogated the mechanism supporting these alterations. It was suggested that dysfunctional lysosome or autophagy can promote EV biogenesis through modifying the fate of MVBs, the direct cytoplasmic source of exosomes (31, 32). A recent study further revealed that reduced SIRT1 expression in breast cancer cells can modify lysosomal activity, resulting in enhanced release of exosomes from these cells and significant changes in their exosome composition (33).

We first assessed the expression level of human SIRT family, a group of NAD⁺-dependent deacetylases, in proliferating and senescent cells. Whole-transcriptomic analysis by RNA-seq suggested that all seven members of this family were ubiquitously downregulated, although a statistical significance was observed mainly in SIRT1 and SIRT2 (Fig. 2A and B). Downregulation of SIRT1/2 was accompanied by upregulation of IL8 and MMP3, hallmark SASP factors (Fig. 2C). Upon cellular senescence, DNA damage–associated concomitant decline of SIRT1/2 was indeed also observed in a human primary fibroblast line BJ (34), essentially supporting our data. SIRT1 regulates diverse cellular targets and correlates with organismal aging, while SIRT2 has been reported to be marker of cellular senescence in certain cancer types such as osteosarcoma (35, 36). Although loss of SIRT1 enhances the secretion of exosomes by breast cancer cells, whether or not this occurs in stromal cells remains unclear. In our work, knockdown of SIRT1 with shRNAs caused significant upregulation of not only the typical sEV markers including TSG101, Syntenin-1, ALIX, and tetraspenins (CD9/63/81), but also downregulation of HSP70 (HUGO symbol HSPA4), the latter reported to be responsible for accumulation of ubiquitinated proteins in mammalian cells (Fig. 2D; ref. 37). However, expression of IL6, IL8, and MMP3, a set of canonical SASP factors, remained unaffected upon SIRT1 depletion, suggesting that SIRT1 loss itself was insufficient to induce the SASP development (Fig. 2D). We further noticed that DNA damage treatment caused reduction of HSP70, accompanied by increased TSG101, IL8, and MMP3, as evidenced by both transcript and protein assays (Supplementary Fig. S2A and S2B). To the contrary, exogenous expression of HSP70 caused decreased levels of TSG101 and ALIX, suggesting a negative correlation between HSP70 expression and sEV production in these cells (Supplementary Fig. S2C).

The SASP is functionally modulated by several master regulators, including the NF-κB complex, which controls both cell-autonomous and noncell-autonomous activities of cellular senescence program (38). However, whether SIRT1 loss is subject to NF-κB regulation, remains unknown. We used Bay 11-7082 (BAY) to treat PSC27 cells, and found that SIRT1 decrease appeared reversed upon NF-κB suppression, in sharp contrast to many of the SASP factors such as IL8 and MMP3, whose expression was markedly diminished after BAY treatment (Supplementary Fig. S2D; ref. 39). More importantly, NF-κB inhibition substantially reduced expression of TSG101, ALIX, CD63, and CD81 after bleomycin treatment, indicating compromised sEV biogenesis even in the setting of DNA damage. We next investigated the mechanism underlying SIRT1 reduction in damaged cells. As miRNA 34a (miR-34a) can inhibit SIRT1 expression through targeting the 3′ untranslated region of SIRT1 upon cellular stress (40), we first examined the expression profile of miR-34a. Of note, miR-34a was significantly upregulated in PSC27 cells upon genotoxic treatment (Supplementary Fig. S2E). Further analysis indicated that miR-34a expression was mainly driven by NF-κB, while targeting miR-34a essentially prevented SIRT1 decline in damaged stromal cells (Supplementary Fig. S2E and S2F). Furthermore, chemical inhibition of the PARP1, an enzyme functionally involved in cellular senescence, SASP development, and NF-κB activation upon DNA damage events (41), failed to affect the expression pattern of either SIRT1 or ALIX, suggesting double-strand breaks, which underlie ATM- and/or ATR-dependent DDR, instead of single-strand lesions, are indeed responsible for reduced SIRT1 expression and enhanced sEVs biogenesis in DNA-damaged cells (Supplementary Fig. S2G).

To substantiate the correlation of SIRT1 with sEV production, we used suberoylanilide hydroxamine acid and nicotinamide (NAM), a pan-histone deacetylase (HDAC) inhibitor (at least against HDAC 1/2/4), and a SIRT-specific inhibitor (against HDAC3), respectively, to treat PSC27 cells. Immunoblots indicated that expression of SIRT1, IL8, and MMP3 remained unchanged, HSP70 decreased, while sEV markers including ALIX and CD63 increased when cells were treated by either agent even in the absence of DNA damage, suggesting that the activity of SIRTs and other HDACs responsible for deacetylation of certain targets is critical for the synthesis of sEV central molecules, but not for development of the SASP (Supplementary Fig. S2H). To the contrary, activation of SIRT1 by SRT2104, a selective SIRT1 activator, substantially diminished sEV signals and protein ubiquitination, alterations caused by NAM, regardless of DNA damage treatment, although expression of TSG101, ALIX, CD63, IL8, and MMP3 was remarkably higher when cells were exposed to bleomycin (Fig. 2E). Thus, SIRT1 is a critical but negative modulator of sEV biogenesis, and functionally responsible for restrained proteome-wide ubiquitination.

Enhanced production or reduced degradation of ubiquitinated proteins by the proteasome or autophagy can result in elevated amount of ubiquitinated proteins (37). To assess proteasome integrity in damaged cells, we used the proteasome inhibitor MG132 to treat PSC27, and found pronouncedly increased protein ubiquitination after genotoxic exposure (Supplementary Fig. S2I). Upon treatment with bafilomycin A1, a selective inhibitor of the late phase of autophagy that prevents maturation of autophagic vacuoles by targeting vacuolar H+-ATPases (V-ATPases) and restraining fusion between autophagosomes and lysosomes (42), we noticed elevated levels of autophagy markers p62 and LC3 II, regardless of cell exposure to genotoxic stress, indicating the integrity of autophagy even with SIRT1 loss and protein ubiquitination in these senescent cells (Supplementary Fig. S2J). Upon treatment by cycloheximide, a protein synthesis inhibitor, we observed markedly reduced protein ubiquitination, but not SIRT1 level (Supplementary Fig. S2K). The total protein of EGFR, a transmembrane receptor subject to polyubiquitination and proteasome-mediated degradation (43), remained largely unaffected when bleomycin-damaged cells were exposed to cycloheximide. Indeed, the relative constant amount of EGFR protein was in-line with reduced level of ubiquitination in cycloheximide-treated cells, which presumably had decreased capacity in mediating protein turnover. Thus, SIRT1 was not subject to significant protein degradation in senescent cells, with its decrease mainly attributable to downregulation of SIRT1 starting from the transcription level, although accumulation of ubiquitinated proteins implies functional deficiency of lysosomes.

We next interrogated whether SIRT1 loss affects lysosomal function by assaying the pH of lysosomes with LysoSensor yellow/blue dextran ratiometric probe, which accumulates in acidic organelles as the result of protonation. In contrast to proliferating cells whose lysosome pH was 4.8, consistent with the reported pH of physiologically normal lysosomes (44), senescent cells exhibited a lysosomal pH approaching 6.0 (Fig. 2F). To determine whether the increase of senescent cell lysosomal pH results from a defective proton pump, we performed lysosomal reacidification assays by first enhancing the pH by bafilomycin A1 treatment, then adding LysoTracker to assess the rate of lysosomal pH recovery. While the pH of proliferating cell lysosomes basically recovered within 60 minutes, the pH recovery of senescent cell lysosomes appeared markedly slower (Fig. 2G). Interestingly, upon SIRT1 depletion from these cells, a similar tendency of lysosomal pH increase and post-stress pH recovery deficiency was observed, which largely resembled that of senescent cells (Fig. 2F and G).

As the data suggested defective proton pump in senescent cells, we analyzed V-ATPases, a set of multi-subunit enzymes functionally supporting the acidification of late endosomes and lysosomes (45). Immunoblots showed that expression of subunit A of V1 (ATP6V1A) substantially decreased upon cellular senescence, in parallel with SIRT1 reduction and pan-ubiquitination (Fig. 2H). Further analysis showed enhanced ubiquitination when ATP6V1A was eliminated, even with SIRT1 level unaffected, suggesting ATP6V1A is responsible for controlling protein ubiquitination, a process downstream of SIRT1 regulation (Fig. 2I). While SIRT1 elimination remarkably enhanced ubiquitination, ectopic expression of ATP6V1A was able to rescue such an effect (Fig. 2J). We then assessed the capacity of sEV secretion, and found ATP6V1A knockdown caused a significant increase in the number of sEVs released from stromal cells, an effect that largely resembled that of SIRT1 depletion (Fig. 2K). To the contrary, overexpression of either SIRT1 or ATP6V1A prominently restrained sEV production in senescent cells, although no changes were observed in proliferating cells (Fig. 2L). Thus, our data consistently demonstrated that SIRT1 loss causes lysosomal deacidification and is responsible for sEV overproduction via ATP6V1A downregulation in senescent cells, a mechanism reported for certain cancer cell types (33).

We next asked whether sEVs secreted by senescent stromal cells can exert profound impacts on cancer cells. We generated a stable PSC27 subline (PSC27-CD63) with a construct (pCT-CD63-eGFP), permitting to utilize the eGFP-fusion CD63 to mark cellular compartment, organelles, and structures to enable long-term and in-depth tracing of sEV biosynthesis, secretion, and uptake. Upon collection of PSC27-derived sEVs, we applied them directly to treat prostate cancer cells in culture (Fig. 3A). Not surprisingly, eGFP-marked CD63 was easily detected, although mainly in the perinuclear region of prostate cancer cells, suggesting active uptake of these vesicles by recipient cancer cells (Fig. 3B).

Subsequent in vitro assays demonstrated that senescent stromal sEVs can significantly enhance the proliferation of several prostate cancer cell lines including PC3, DU145, LNCaP, and M12, in contrast to sEVs released from proliferating stromal cells, which although also generated detectable effects (Fig. 3C). We further observed markedly increased migration and invasion of these cells upon exposure to senescent stromal sEVs (Supplementary Fig. S3A and S3B). More importantly, prostate cancer cells showed significantly enhanced resistance to mitoxantrone, a type II topoisomerase inhibitor administered to treat several types of malignancies including prostate cancer in clinical medicine (Fig. 3D). Although proliferating stromal sEVs did not confer significant benefits on the survival of the majority of prostate cancer cell lines we assayed, senescent stromal sEVs pronouncedly enhanced the resistance of cancer cells to mitoxantrone-induced cytotoxicity (Fig. 3D). We further assessed the influence of stromal sEVs on cancer cell survival with mitoxantrone, which was specifically prepared in a range of doses (0.01–>1.0 μmol/L) designed to resemble its serum concentrations in clinical conditions. The data suggested that proliferating stromal sEVs did not significantly change PC3 survival when cells were exposed to mitoxantrone, whereas senescent stromal sEVs did (Fig. 3E). We noticed that the two groups of stromal sEVs differed most dramatically in the range of 0.1–1.0 μmol/L mitoxantrone, although further augment of drug concentration consistently resulted in rapid clearance of cancer cells presumably due to cell intolerance against overwhelming cytotoxicity.

Mechanistic dissection revealed that mitoxantrone induced cleavage of caspase-3 in cancer cells, a process that was modestly affected by proliferating stromal sEVs but remarkably weakened by senescent stromal sEVs (Fig. 3F). Although alternative cell survival pathways may be operative, our data suggest that senescent stromal sEVs drive cancer resistance likely via a caspase-counteracting mechanism. We further applied QVD-OPH and ZVAD-FMK, two potent pan-caspase inhibitors, as well as PAC1 and gambogic acid, two typical caspase activators, to individually treat PC3 cells shortly before mitoxantrone exposure. Cell apoptosis was substantially attenuated in the presence of QVD-OPH or ZVAD-FMK (P < 0.001; Fig. 3G). However, once the procaspase-activating compound PAC1 or gambogic acid was used, apoptosis index was markedly elevated, thus offsetting the antiapoptosis effect of senescent stromal sEVs (P < 0.01). The data were largely reproduced when docetaxel, another chemotherapeutic drug that interferes with microtubule depolymerization, was applied to the system (Supplementary Fig. S3C and S3D). Thus, our results consistently demonstrate that senescent stromal sEVs restrains caspase-dependent apoptosis of cancer cells, a process that underlies its resistance-boosting capacity via nanoparticle secretion–based paracrine influence on cells exposed to cytotoxic agents.

Given the remarkable changes of prostate cancer cell phenotypes induced by senescent stromal sEVs, we next sought to determine their influence on the expression pattern of cancer cells. We first chose to perform RNA-seq to quantitate gene expression modifications and profile the transcriptomics after treatment of prostate cancer cells with senescent stromal sEVs. Bioinformatics output showed that 344 transcripts were upregulated or downregulated significantly (≥2-fold; P < 0.05) in PC3 cells (Supplementary Fig. S4A and S4B), while the expression of 422 transcripts was modified in senescent stromal sEV-affected DU145 cells (Supplementary Fig. S4C and S4D). Among the upregulated gene products, many are correlated with tumor development, such as CEACAM5, KRT15, IFI6, KIF20A, CEMIP, and SERPINB3 in PC3, or DHRS2, IL21R, FGFBP1, DKK1, PRSS2, and KLK6 in DU145 (Supplementary Fig. S4E and S4F). We noticed that these proteins are mostly involved in key biological processes such as signal transduction, cell communication, cell metabolism, energy pathways, cell growth, and maintenance (Supplementary Fig. S4G and S4H). Therefore, data from bioinformatics analysis generally support the findings derived from in vitro assays, which showed enhanced cell proliferation, migration, and invasive activities of cancer cells upon exposure to senescent stromal sEVs (Fig. 3C; Supplementary Fig. S3A and S3B).

Next, we performed comparative assessment by scrutinizing genes whose expression was co-upregulated in both PC3 and DU145 cells (≥2-fold; P < 0.05). The data presented 22 genes that fell in this category, with 199 and 169 genes being uniquely upregulated in PC3 and DU145, respectively (Supplementary Fig. S4I). Interestingly, we noticed the ATP-binding cassette subfamily B member 4 (ABCB4), a full transporter and member of the p-glycoprotein family of membrane proteins with phosphatidylcholine as its substrate, showed up at the top of the PC3-DU145 co-upregulated gene list (Supplementary Fig. S4J). Gene ontology (GO) analysis indicated that ABCB4 is involved in multiple activities including glycoside transport, carbohydrate export, response to drug, fenofibrate, external biotic stimulus, and cellular hyperosmotic salinity (Supplementary Fig. S4K). Alternatively, Kyoto Encyclopedia of Genes and Genomes (KEGG) appraisal underscored its biological implications as a typical ABC transporter (Supplementary Fig. S4L).

After mapping ABCB4-involved protein–protein interaction (PPI) curated with the STRING database interactome (46), we generated a network highlighting the interaction of ABCB4 with a handful of proteins such as PPARA, FABP1, CREBBP, CHD9, and CARM1 in human cells (Supplementary Fig. S4M). The PPI network encompassed protein target nodes (n = 11) connected by edges (n = 55) with an average node degree of 10 (local clustering coefficient of 1, PPI enrichment P < 1.0e-16). We calculated the evidence based on GO and other databases including KEGG, PFAM, and InterPro, with binary human PPI data and enrichment results validated (Supplementary Table S3).

Given the substantial impact of senescent stromal sEVs on transcriptome-wide expression of recipient cancer cells, we sought to explore the mechanism(s) supporting stromal sEV–enhanced malignancies, particularly drug resistance of prostate cancer cells. RNA-seq data showed that upon treatment with senescent stromal sEVs, PC3 and DU145 cells exhibited significantly (P < 0.05 for both lines) upregulated expression of ABCB4, while other members of the ABCB subfamily remained largely unchanged (Fig. 4A and B). Subsequent transcript assays essentially supported these results (Fig. 4C and D). We further confirmed the senescent stromal sEV–inducible pattern of ABCB4 by immunoblots (Fig. 4E).

Next, to address the relevance of ABCB4 upregulation to cancer cells, we eliminated this multidrug resistance protein of an ATPase-associated function responsible for drug efflux with gene-specific shRNAs (Supplementary Fig. S5A). Of note, the gain of function in cell proliferative rate after exposure to senescent stromal sEVs was largely diminished upon depletion of ABCB4 from representative prostate cancer cell lines (Fig. 4F). We observed significantly weakened migration and invasion when ABCB4 was eliminated from these cells (Supplementary Fig. S5B and S5C). The effects caused by ABCB4 removal from cancer cells were further observed in chemoresistance assays, which showed a similar tendency among the prostate cancer cell lines examined at their individual mitoxantrone concentration of IC₅₀ (Fig. 4G and H). To expand, we further examined cell survival capacity across a wide window of mitoxantrone concentrations with PC3, and found ABCB4 knockdown caused a remarkable reduction in cell survival upon treatment by mitoxantrone in the range of 0.1–1.0 μmol/L (Fig. 4I). Thus, our data consistently support the key function of ABCB4 in mediating resistance to a chemotherapeutic agent, a property acquired upon uptake of senescent stromal sEVs by cancer cells.

To validate the generality of these findings, we collected sEVs from HBF1203, the breast stromal cell line, which enters cellular senescence upon doxorubicin treatment (Supplementary Fig. S1E–S1H). Following a similar procedure, we eliminated ABCB4 from MDA-MB-231, a breast cancer cell line, before exposure to senescent stromal sEVs. The data from breast stroma cancer assessments closely resembled those derived from prostate cancer assays by showing the effect of ABCB4 depletion on breast cancer cell survival in a doxorubicin concentration range of 0.1–10 μmol/L, which approaches its plasma level observed in breast cancer patients in clinical oncology (Fig. 4J; ref. 47).

Because SIRT1 expression is reduced in senescent stromal cells, a process responsible for increased sEV production, we reasoned the possibility of minimizing sEV production by senescent stromal cells although targeting SIRT1. To this end, we chose SRT2104 and SRT1720, two potent selective activators of SIRT1, to treat stromal cells. Simultaneous exposure of PSC27 cells to bleomycin and either SIRT1 activator significantly decreased the number of sEVs released by senescent PSC27 cells (Fig. 5A; Supplementary Fig. S6A). SIRT1 activation caused remarkable deacetylation of HSF1 in senescent cells, a transcription factor that is functionally involved in proteostasis upon cellular stress via SIRT1-mediated deacetylation, paralleled by upregulation of its target HSP70, which reversely correlates with sEV production (Supplementary Fig. S2B, S2C, and S2H; ref. 48). More importantly, we observed markedly declined ABCB4 expression and diminished chemotherapeutic resistance of prostate cancer cells (PC3 as a representative) upon incubation with sEVs secreted by senescent stromal cells, which were simultaneously exposed to bleomycin-delivered genotoxicity and a SIRT1 activator (SEN/SRT2104 or SEN/SRT1720), in contrast to the senescence-naïve group (SEN or SEN/DMSO; Supplementary Figs. S5B and S6B). The data derived from PSC27 and PC3 were largely reproducible by a similar set of assays performed with HBF1203 and MDA-MB-231, both lines of human breast origin (Supplementary Fig. S6C and S6D). Thus, targeting SIRT1 with pharmacologic activators can restrain sEV production, a treatment strategy that is effective in governing the phenotype of senescent stromal cells but indeed correlated with significantly weakened drug resistance of cancer cells.

Given the prominent efficacy of SIRT1 activator–mediated control of sEV production by senescent stromal cells, we asked whether the in vitro findings can be technically repeated by in vivo studies. To precisely address tumor–stroma interactions in a TME context with structural and functional integrity, we generated tissue recombinants by admixing PSC27 with PC3 at a preoptimized ratio before subcutaneously injecting them to the hind flank of experimental mice with SCID. Prior to agent-mediated therapeutics, target-specific genetic assays were performed. We noticed that expression of ectopic SIRT1 can substantially restrain sEV secretion from senescent stromal cells, largely resembling effects observed upon SIRT1 activation (Supplementary Fig. S6E). In vivo data indicated that the presence of stromal cells markedly increased tumor volumes (52.6%; P < 0.001; Supplementary Fig. S6F). In contrast to proliferating stromal cells, senescent stromal cells further promoted tumor growth (45.2%; P < 0.001), a tendency that was essentially abolished upon SIRT1 overexpression in PSC27 (36.4%; P < 0.001; Supplementary Fig. S6F). We then initiated preclinical trials with therapeutic agents. Mitoxantrone and SRT2104 (chosen for preclinical studies due to its slightly higher efficacy than SRT1720; SRT2104 currently in phase II trials, while SRT1720 failed) were administered as mono or dual agents starting from the 3rd week post-tumor implantation, after which, animals received treatment in a metronomic manner to mimic clinical conditions (Fig. 5C; Supplementary Fig. S6G). Data from endpoint measurement of tumor size indicated that mitoxantrone administration caused remarkably delayed tumor growth, validating the efficacy of mitoxantrone as a cytotoxic agent (41.3%; P < 0.001; Fig. 5D). Although treatment with SRT2104 did not significantly change tumor volume, coadministration of mitoxantrone and SRT2104 caused maximal reduction of tumor mass (63.9% shrinkage in contrast to SRT2104, or 44.0% decrease in relative to mitoxantrone given in the mono-treatment; Fig. 5D). We observed substantially enhanced cellular senescence in tumors dissected from animals that underwent therapeutic regimens involving mitoxantrone, regardless of SRT2104 administration (Fig. 5E). Resembling the profiles captured with cultured cells, genotoxic treatment caused remarkable expansion of stromal cell MVBs, which displayed augment in both size and enclosed vesicle number, a tendency restrained upon SRT2104 administration (Supplementary Fig. S6H).

Treatment-induced cellular senescence was accompanied by in-tissue development of the SASP, as evidenced by increased expression of SASP hallmark factors such as IL6, IL8, IL1α, and MMP3 (Supplementary Fig. S6I). Although stromal cells exhibited consistently increased SASP expression, cancer epithelial cells did not show a typical senescence profile, including expression of p16, which remained largely unchanged after treatment (Supplementary Fig. S6I). This was likely due to the emergence of treatment-resistant cancer cells and expansion of resistance colonies during the chemotherapeutic regimen, a process responsible for tumor relapse posttreatment. Importantly, we observed significantly decreased signals of SIRT1 in stromal cells posttreatment, which was in-line with our in vitro data (Fig. 2A–C). Reduction of SIRT1 in stroma was accompanied by consistently enhanced expression of sEV markers including TSG101, Syntenin-1, CD9, CD63, and CD81 (Fig. 5F). In contrast, cancer cells did not seem to develop such a distinct pattern.

Loss of SIRT1 expression was also revealed by IHC staining of mouse tissues. In contrast to stromal cells, which showed a generally declining tendency, adjacent cancer cells appeared largely unchanged in SIRT1 expression (Fig. 5G). The data suggested the presence of a differential expression mechanism that demarcates stromal cells from their cancer epithelial counterparts. IF staining of tissues indicated that there was a remarkable increase of ABCB4 in cancer cells, but not stromal cells, after exposure of animals to mitoxantrone, although the signals were minimized when SRT2104 was delivered alongside mitoxantrone (Supplementary Fig. S6J and S6K).

As sEVs can foster premetastatic niche formation and mediate distant metastasis of cancer cells, we interrogated whether there are metastatic events in these animals. Bioluminescence imaging (BLI) of xenografts generated with PC3 cells stably expressing luciferase (PC3-luc) and PSC27 stromal cells excluded activities of cancer cell dissemination from the primary sites, while the relative BLI signal intensities essentially supported tumor growth patterns observed in xenografted mice (Fig. 5H). The data suggest that classic chemotherapy combined with a SIRT1-activating agent can achieve tumor regression more effectively than chemotherapy alone, a therapeutic strategy that holds a prominent potential to circumvent the senescent stroma–induced pathologic exacerbation, specifically acquired resistance.

As a topoisomerase II inhibitor, mitoxantrone induces apoptosis or premature senescence in the TME niche, depending on the cell type and agent dose (49, 50). Tissue assessment indicated that mitoxantrone administration to these animals caused dramatically increased DNA damage and apoptosis in cancer epithelial cells, as evidenced by DDR foci measurement and caspase-3 cleavage appraisal (Fig. 5I). In stark contrast, stromal cells manifested development of DDR foci, limited apoptosis, but markedly enhanced senescence, indicating a response and cell fate distinct from their adjacent cancer counterparts after exposure to mitoxantrone-delivered cytotoxicity (Fig. 5I; Supplementary Fig. S6L). Although SRT2104 alone did not induce typical DDR or apoptosis, significantly elevated indices of both DNA damage and apoptosis were observed in tissues of animals that underwent mitoxantrone/SRT2104 combinatorial treatment. In contrast to the mitoxantrone-only group, there was a pronounced increase of DDR and apoptosis in mitoxantrone/SRT2104 animals (40.3% and 112.5%, respectively), suggesting a remarkable potential of SRT2104 in promoting cancer cell clearance and achieving tumor regression when synergized with classic chemotherapy in vivo (Fig. 5I and J).

We next interrogated the pathologic relevance of SIRT1 reduction in stroma and ABCB4 elevation in cancer cells to patient survival after chemotherapy. Upon IHC staining, we observed enhanced p16^INK4a expression in stromal cells of patients who experienced chemotherapeutic intervention, while most cancer cells remained largely negative (Fig. 6A). The data revealed comprehensive occurrence of cellular senescence in the stroma, while cancer cells seemingly resisted in the gland, presumably through expansion from colonies that survived anticancer treatments via development of drug resistance. We noticed remarkably decreased expression of SIRT1 in stromal, rather than their adjacent cancer cells, although the latter showed elevated ABCB4 expression in posttreatment patients (Fig. 6A).

Upon precise isolation of stromal and cancer cell subpopulations with laser capture microdissection (LCM) and evaluation by qRT-PCR, we noticed SIRT1 reduction and ABCB4 elevation in stromal and epithelial cells, respectively (Fig. 6B and C). Thus, transcript expression data were essentially consistent with those acquired from tissue staining appraisal of these clinical specimens.

We further performed a sample-based clinical investigation by randomly choosing two subgroups of patients with prostate cancer (10 for each) from the overall cohort, including one that did not undergo chemotherapy and the other that experienced chemotherapy, with patient-derived peripheral blood samples available from both subgroups. TSG101-specific ELISA tests suggested significantly enhanced levels of circulating sEVs in the serum of posttreatment patients, as compared with those untreated individuals (Fig. 6D). Surprisingly, immunoblots even showed clearly visible signals of circulating sEVs (Syntenin-1, TSG101, and ALIX-positivity) in the serum of posttreatment patients (3/5 cases), in sharp contrast to samples from those untreated, which were generally negative (Fig. 6E). Therefore, sEVs in the circulating system of posttreatment patients hold the potential to be evaluated as blood-borne substance for clinical monitoring via examination with routine biotechniques.

We next sought to establish the consequence of enhanced sEV biogenesis in these patients. After tissue staining against SIRT1 and ABCB4, we assessed these patients by associating the results with their survival. Strikingly, there was a significant and positive correlation between stromal SIRT1 expression and patient disease-free survival in the posttreatment prostate cancer cohort (Fig. 6F). To the contrary, a significant but negative correlation was observed between cancer cell ABCB4 expression and patient survival (Fig. 6G). Thus, SIRT1 decline–mediated sEV production and release by stromal cells, combined with ABCB4 upregulation in cancer cells, can provide a novel and precise avenue for prognosis of advanced malignancies in future cancer medicine. Specifically, given the prominent relevance of SIRT1 loss in sEV biogenesis of senescent stromal cells and its hitherto well-documented implications in human pathologies, we would propose SIRT1 as a functionally multifactorial target in clinical oncology.

Mutual interaction between cancer cells and the surrounding TME is crucial for malignant progression. Numerous studies have unraveled the paracrine signaling of various cytokines, chemokines, and growth factors through their receptors as a key means of intercellular communication in the TME. In contrast, EVs have recently emerged as another mechanism to mediate cell–cell interactions, which can be secreted from various cell types (51). As a subclass of EVs, sEVs contain various proteins, mRNAs, miRNAs, lipids, and even DNA fragments. To date, most cancer-associated sEV studies focus on cancer cells or proliferating cells. In contrast, here, we present evidence to show that senescent stromal cells, which abundantly reside in host tissues, are responsible for cancer progression by producing sEVs to shape drug resistance acquired from the treatment-damaged TME (Fig. 6H).

Sirtuins (SIRT1-7) belong to the third class of HDACs, whose activities are dependent on NAD⁺ (52). In this study, we revealed SIRT1 as the most dramatically declined sirtuin molecule in senescent stromal cells. Protein quality control is critical for maintenance of cell physiology, while damaged proteins need to be restored or degraded, a process mainly supported by molecular chaperones and the ubiquitin-proteasome system. Interestingly, both proteasome and autophagy machineries remained largely intact in these cells. SIRT1 reduction correlates with enhanced expression of multiple sEV markers and is responsible for ATP6V1A downregulation, resulting in defective lysosomal acidification and sEV overproduction. The data are reminiscent of a recent study, which demonstrated that genetic or pharmacologic inhibition of SIRT1 destabilizes the mRNA of ATP6V1A, leading to deficient lysosomal acidification, MVB enlargement, and enhanced exosome production (33). While our observation was mainly regarding senescent stromal cells, distinct from the discovery recently made in human breast cancer cells, a similarly occurring MVB expansion suggests that senescent stromal cells somehow resemble cancer cells with regard to sEV biogenesis. Furthermore, SIRT1 loss–driven ATP6V1A downregulation results in defective lysosomal acidification and increase of pH to approximately 6.0, partially explaining the comprehensively observed SA-β-Gal staining positivity of senescent cells, a phenomenon reported long time ago but rarely addressed for underlying mechanism(s) (53).

There is a growing interest in the discovery of small molecules modifying sirtuin activities, although most sirtuin activators have been described intensively for SIRT1 (54). Specifically, SIRT1 has been suggested to play multiple and, in some cases, contradictory roles in cancer, by functioning as either a tumor suppressor or tumor promoter, depending on tissue type and pathologic context (33). In vivo stimulation of SIRT1 either with agonist or NAD⁺ precursors can extend the lifespan of mice and protect against diverse aging-related diseases (55). Resveratrol is a natural polyphenol compound, which activates SIRT1, holding potential in treatment or prevention of tumorigenesis and aging-related diseases, but it has limited bioavailability (56). Molecules structurally unrelated to resveratrol, including SRT1720 and SRT2104, are typically sirtuin-activating compounds developed to stimulate sirtuin activities more potently than resveratrol, with markedly improved bioavailability (57). In our study, we disclosed that SRIT1 activators including SRT2104 can effectively restrain sEV biogenesis in senescent stromal cells. Although cancer cells exposed to the stromal conditioned media (CM) containing sEVs developed remarkably enhanced malignancy, addition of SRT2104 to the system was able to counteract such a tendency. Alternatively, treatment with nicotinamide mononucleotide, a key precursor of NAD⁺, represents another potential option to enhance stromal cell SIRT1 activity and minimize cancer resistance, with future work warranted to demonstrate the efficacy.

Our study support that senescent stromal cells produce an increased number of sEVs, which can be conceptually considered as part of the full SASP spectrum. However, one of the limitations is the lack of a dataset showing the distinct profile of proteomic composition in senescent stromal cells, which may be exploited as a reservoir for screening novel targets. Second, other bioactive contents such as miRNAs, which specifically merit future investigation, may serve as novel prognostic markers for evaluation of treatment consequence in cancer clinics. Third, sEV size redistribution in proliferating versus senescent stromal cells deserves attention, as this potentially correlates with cargo alterations, a case awaiting future studies to disclose the relevant mechanism(s). Fourth, the impact of pan-ubiquitination on senescent cell metabolism and long-term survival remains unexplored and represents an intriguing question. Despite these issues, as a special note, combination of conventional chemotherapy theoretically designed to target cancer cells, with a SIRT1 activator like SRT2104, can significantly improve therapeutic outcome by restraining sEV biogenesis in senescent stromal cells, which frequently arise as a part of side-effects of anticancer agents and are responsible for multiple age-related pathologies including, but not limited to, the most lethal form, cancer.

No potential conflicts of interest were disclosed.

Conception and design: Y. Sun

Development of methodology: L. Han, L. Cao, Y.E. Chin, J. Jiang, Y. Sun

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L. Han, Q. Long, S. Li, Q. Xu, B. Zhang, X. Dou, M. Qian, Y. Jiramongkol, J. Guo, L. Cao, Y.E. Chin, J. Jiang, Y. Sun

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L. Han, L. Cao, Y.E. Chin, E.W.-F. Lam, J. Jiang, Y. Sun

Writing, review, and/or revision of the manuscript: E.W.-F. Lam, Y. Sun

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y. Sun

Study supervision: Y. Sun

This work was supported by grants from National Key Research and Development Program of China (2016YFC1302400 and 2019YFC2002801), National Natural Science Foundation of China (81472709, 31671425, and 31871380), the Strategic Priority Research Program of Chinese Academy of Sciences (XDB39010500), CAS Key Laboratory of Tissue Microenvironment and Tumor of Chinese Academy of Sciences (201506, 201706), the National 1000 Young Talents Research Program of China, Anti-Aging Collaborative Program of SIBS and BY-HEALTH (C01201911260006), the University and Locality Collaborative Development Program of Yantai (2019XDRHXMRC08), and the US Department of Defense (DoD) Prostate Cancer Research Program (Idea Development Award PC111703, all to Y. Sun), National Science and Technology Major Project of China (2019ZX09732002-14 to J. Jiang), CRUK (A12011), Breast Cancer Now (2012MayPR070 and 2012NovPhD016), the Medical Research Council of the United Kingdom (MR/N012097/1), and Imperial ECMC and NIHR Imperial BRC to E.W.-F. Lam.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.