Senescent Cell–Derived Extracellular Vesicles Inhibit Cancer Recurrence by Coordinating Immune Surveillance

Chemotherapeutic regimens such as paclitaxel and cisplatin exert their antitumor activity by switching cancer cells to a stable nonproliferative, metabolically active survival state generally referred to as cellular senescence (1). Senescent cells (SC) undergo characteristic changes, including an enlarged and flattened morphology and expression of distinct genes, including senescence-associated β-galactosidase (SA-β-Gal), p16 (CDKN2A), and p21 (CDKN1A) (2). Senescence induction leads to an immediate restriction in cancer cell propagation and tumor growth. However, pathologic accumulation of SCs leads to cancer recurrence (3, 4). How SCs interact with their microenvironment and the mechanisms that fail to prevent SC accumulation in aging tissues are still largely unknown.

The senescence-associated secretory phenotype (SASP) is a hallmark of senescence (5). The SASP is composed of several different families of soluble mediators involved in extracellular matrix proteolysis, chemotaxis, cell growth and differentiation, and immune regulation (5, 6). SASP factors are known to both induce paracrine senescence in neighboring cells and to elicit immune responses (1). Extracellular vesicles (EV) are important mediators of intercellular communication (7). SC-derived EVs (senEV) were recently identified as a component of the SC secretome that impacts the recruitment of immune cells (8–12). EVs are membrane-bound, nanometer-sized particles released by every cell type and carrying bioactive signaling molecules from the parental cells (13). Unfortunately, investigations on the roles of EVs in cancer and senescence have been hampered by the need to isolate them before intravenous reinfusion in animal models; these manipulations can introduce biases such as loss of EV diversity, misjudgment of the amount to reinfuse, and assumptions on EV biodistribution (14). Model systems that distinguish the contributions of senEVs from those of SASP factors would enable the study the EV functions during immune responses to cellular senescence.

Senescence surveillance is the immune-mediated clearance of SCs. Immune cell types involved in senescence surveillance include monocytes (15, 16), NK cells (16), and T cells (15). However, SASP-mediated recruitment of other immune cell subsets, such as myeloid-derived suppressor cells (16), can inhibit cytotoxic CD8 T-cell responses (3, 4) and promote tumor angiogenesis (1), thereby fostering tumor progression and recurrence. The dual role of the SASP on senescence surveillance may derive from confounding effects due to the presence of different amounts and types of senEVs in SASP studies. Approaches to distinguish between senEVs and the SASP would enable to clarify the paradoxical influence of cellular senescence in cancer.

In the present study, we hypothesized that senEVs play a key role in senescence surveillance. To test this hypothesis, we developed and validated an engraftment-based in vivo senescence model that minimizes the impact of senescence-inducing stimuli on immune cells while at the same time enabling functional studies of senEVs without impacting the SASP (17–19).

Paclitaxel from Taxus brevifolia >95% (cat. #T7402-1MG) and Cisplatin Pharmaceutical Secondary Standard (cat. #PHR1624-200MG), DMEM (cat. #D8062-6X500ML), dispase II (cat. #D4693-1G), BCA Protein Assay Kit (cat. #71285-3), insulin (cat. #I6634-50MG), hydrocortisone (cat. #H0888-10G), cholera toxin (cat. #C8052-.5MG), transferrin (cat. #T1147-100MG), Tri-Iodo-Thyronine (cat. #T5516-1 mg) were purchased from Sigma-Aldrich. Cellular Senescence Detection Kit (cat. #SG04-10) was purchased from Dojindo. Amicon ultra-15 centrifugal filters UFC910024 100 kDa (cat. #UFC910024), Licor IRDye 800CW streptavidin (cat. #NC0883593), trypsin–EDTA (cat. #25-200-056), and RIPA (10×) buffer (cat. #50-195-822) were purchased from Thermo Fisher Scientific. Phase-Flow Alexa Fluor 647 BrdU Kit (cat. #370706), 5-bromo-2-deoxyuridine (BrdU; cat. #423401), APC Anti-BrdU Antibody (cat. #364114), Annexin V Binding buffer (cat. #422201), FITC Annexin V (cat. #640945), True-Nuclear Transcription Factor Buffer Set (cat. #424401), and Zombie Aqua Fixable Viability Kit 500 tests (cat. #423102) were purchased from BioLegend. Cell lysis buffer (cat. #9803S) was purchased from Cell Signaling Technology. Anti-rabbit detection module for whole-exome sequencing, 12-230 kDa whole-exome sequencing separation module (cat. #DM-001), was purchased from ProteinSimple. Proteome Profiler Mouse Cytokine Array Kit, Panel A (cat. #ARY006) was purchased from R&D Systems. Penicillin/streptomycin solution (cat. #SV30010) was purchased from GE HealthCare. EGF (cat. #PHG0311) was purchased from Gibco. Flow count beads (cat. #C36950), Di 8 ANNEPS (cat. #D3167), pluronic F-127 (cat. #P3000MP), Halt Protease Inhibitor Cocktail (PI78430), High-Capacity cDNA Reverse Transcriptase Kit (#4368813), and Fast SYBR Green Master Mix (#4385616) were purchased from Thermo Fisher Scientific. Ultracentrifugation tubes 1 × 3 ½ (cat. #NC9146666) were purchased from Beckman Coulter. Flotillin-1 antibody (cat. #NBP2-75492) was purchased from Novus Biologicals). MACS buffer was prepared by adding 2 mmol/L EDTA (Thermo Fisher Scientific) and 0.5% BSA (Thermo Fisher Scientific) to 500 mL PBS. CD4 depleting antibody (cat. #GK1.5) was purchased from Bio X Cell. Oligos for RNAi (cat. #SR405972) were purchased from OriGene. A Key Resource Table is available as Supplementary Material (Supplementary Table S1).

Two oral epithelial squamous cell carcinoma cell lines were used in this study. MOC2 cells were obtained from Kerafast, Inc., and mEER⁺ cells were kindly provided by Dr. William Spanos (Sanford Research, Sioux Falls, SD). MOC2 carries the same mutations observed in human head and neck cancers, namely, Trp53, MAPK, and FAT (20), whereas mEER⁺ cells have been engineered to express Hras^G12V and HPV-E6/E7 (21). MOC2 and mEER⁺ cell lines were cultured in E-media as previously described (22). Cell lines were tested for Mycoplasma. We prepared E-media by mixing DMEM with Ham’s F12 3:1 and supplementing with 10% FBS (from VWR), 0.5% penicillin/streptomycin solution,10 µL of 25 µg/µL hydrocortisone 16.8 µL of 0.25 µg/µL cholera toxin, 100 µL of 25 µg/µL transferrin, 250 µL of 10 µg/µL insulin, 3.4 µL of 0.2 µg/µL Tri-Iodo-Thyronine, and 250 µL of 10 µg/mL EGF.

Lentiviral vector (LV) production and transduction of MOC2 and mEER⁺ were performed as previously described (17, 23). Briefly, third-generation, self-inactivating LVs were produced by calcium phosphate precipitation. To compare cells expressing Rab35 dominant-negative mutant (Rab35-DN) with those expressing wild-type Rab35 (Rab35-WT), MOC2 and mEER⁺ carrying one copy of LV per cell (1CpC) were generated by titration of LVs on cells and purification of transduced cells from the LV dilution that yielded less than 20% positive cells. Purification was performed by magnetic sorting using dLNGFR, a truncated receptor commonly used as a surface marker (24). Cells were used as bulk populations, without cloning.

To complement our findings with Rab35-DN–expressing cells, Rab27a was targeted using siRNAs. SCs were treated with scrambled control oligos (siScr) or oligos targeting Rab27a (siRab27a) for 48 hours, along with paclitaxel treatment (see “Induction of senescence”). For oligo transfection, the kit protocol was followed. Briefly, 900 µL of RNAiMAX was diluted with half of the cell media required. Oligos (145 µL for siScr oligo and 50 µL for siRab27a oligo) were added to 30 mL cell media. The oligo and RNAiMAX dilutions were combined in a 1:1 ratio and incubated for 10 minutes, after which, the mixture was added dropwise to SCs in a 15-cm dish. At the endpoint, cell pellets and conditioned media were collected to determine Rab27a gene expression and EV inhibition levels.

RNA was isolated from SC pellets after siRNA oligo transfection using Qiagen RNeasy Column Purification Kit and was quantified using a NanoDrop spectrophotometer. cDNA was synthesized from 1,000 ng of RNA using High-Capacity cDNA Reverse Transcriptase Kit. Relative quantification of Rab27a was performed in triplicate with Fast SYBR Green Master Mix (Thermo Fisher Scientific) using the ΔΔCt method, with GAPDH as reference gene. The following primers were utilized for amplification of Rab27a and GAPDH (FWD: GAG​CAA​AGT​TTC​CTC​AAT​GTC​CG, REV: CTT​TCA​CTG​CCC​TCT​GGT​CTT​C and FWD: CAT​CAC​TGC​CAC​CCA​GAA​GAC​TG, REV :ATG​CCA​GTG​AGC​TTC​CCG​TTC​AG, respectively).

To optimize the induction of senescence in MOC2 or mEER⁺ using chemotherapeutic agents, a matrix of paclitaxel (0–1,000 nmol/L) and/or cisplatin (0–48 µmol/L) was used. MOC2 and mEER⁺ cells were expanded in E-media, plated at 20,000 cells per well in a 96-well plate (in quadruplicate), and maintained at 37°C and under 5% CO₂ for 24 hours. Subsequently, the specified ranges of paclitaxel and/or cisplatin were added to the cells, and they were incubated for an additional 48 hours at 37°C in 5% CO₂. As the stock solutions of paclitaxel and cisplatin were dissolved in DMSO, an equivalent amount of DMSO was added to the control wells.

The analysis of cellular senescence was conducted using the cellular senescence detection kit SPiDER-β-Gal, and all steps were executed in accordance with the manufacturer’s instructions. Following the 48-hour treatment with paclitaxel and/or cisplatin, as previously outlined, the cells were washed with PBS and then incubated at 37°C with 5% CO₂ for 1 hour after the addition of bafilomycin A1 solution. Subsequently, the cells were further incubated for an additional 30 minutes after the addition of SPiDER-β-Gal solution. After washing the cells with PBS, they were digested with EDTA-free trypsin, and the resulting cell suspension was collected in a round-bottom 96-well plate containing MACS buffer following centrifugation at 800 rpm for 5 minutes. Finally, a flow cytometer (BD Biosciences) was used for high-throughput detection, with excitation and emission set at 488 and 530 nm, respectively. For flow cytometry analysis, samples that were not treated with SPiDER-β-Gal were utilized to establish the threshold for the negative control.

To assess the potential unwanted induction of apoptosis and/or cell death by paclitaxel and/or cisplatin, parallel plates were prepared. After removing the E-media containing paclitaxel and/or cisplatin, cells were washed with PBS, harvested by trypsin, and resuspended in 50 µL FITC Annexin V working solution [3% FITC Annexin V and 3% 7-aminoactinomycin D (7-AAD) in Annexin B Binding Buffer]. The cells were then incubated for 15 minutes at room temperature in the dark. Following this, the FITC Annexin V working solution was discarded, and the cells were resuspended in MACS buffer. Quantification was performed using a flow cytometer (BD Biosciences).

Six- to eight-week-old female and male C57BL/6J mice weighting between 18 and 25 g for females and 20 to 30 g for males were purchased from Charles River Laboratories, housed under conventional SPF conditions, and provided with food and water ad libitum. Animal care was handled in accordance with the Guide for the Care and Use of Laboratory Animals of the Oregon Health and Science University (OHSU) and approved by the OHSU Institutional Animal Care and Use Committee. All animal studies, including implantation, tumor size assessment, anesthesia administration, surgery, and euthanasia, were carried out in the OHSU animal facility with a priority on minimizing stress and discomfort. Mice were housed in clean cages with a maximum of five mice per cage, furnished with bedding material, a nutritionally balanced diet, and access to clean water ad libitum. The temperature was maintained within the range of 20 to 26°C, and lighting adhered to a regular light–dark cycle of 12 hours each. Regular visual checks were conducted for health monitoring to promptly detect any signs of distress or illness. Anesthesia was induced using isoflurane, and euthanasia was performed in a CO₂ chamber.

Samples were dissociated using the following method: tissues were minced with scissors in a 1.5-mL tube in 100 µL dissociation media containing 2% FBS, 1× penicillin/streptomycin, 1× glutamine, 0.2 mg/mL collagenase IV, 2 mg/mL dispase, and 5 U/mL DNase I in Iscove’s Modified Dulbecco's Medium; after transferring to 15-mL conical tubes containing 5 mL dissociation media, minced tissues were incubated at 37°C with agitation at 500 rpm for 30 minutes in a heater–shaker. After dissociation, the tissues were filtered through a 100-µm strainer with 50 mL of MACS buffer and centrifuged at 500 g for 10 minutes. The resulting pellets were transferred to a deep 96-well plate on ice and mixed with prediluted surface antibodies and flow count beads. A fraction of the pellets was pooled and stained with corresponding antibodies for fluorescence minus o control samples. The plates were sealed with adhesive film and incubated on ice for 30 minutes in the dark. After incubation, the cells were washed twice with PBS and centrifuged at 500 g for 5 minutes. Subsequently, they were stained with Zombie Aqua dye and incubated for 15 minutes on ice in the dark. Following this, the cells were washed with PBS and gently vortexed to dissociate the pellets. The cells were then fixed and permeabilized using True-Nuclear Transcription Factor Buffer Set following the manufacturer’s guidelines. Briefly, the cells were fixed and incubated at room temperature for 30 minutes. Permeabilization was achieved by adding 1× perm buffer. After three washes in 1× perm buffer, the cells were gently vortexed. A mixture of intracellular antibodies in 1× perm buffer was added to the cells, and they were incubated overnight at 4°C in the dark. Following antibody incubation, the antibodies were removed by washing with perm buffer, and the cells were resuspended in MACS buffer. The readout was performed using a Fortessa flow cytometer.

Flow cytometry analysis of immune cells was performed according to the following gating:

First in vivo experiment:

• Neutrophils: Zombie⁻, CD45⁺, Ly6G⁺, and F4/80⁻

• MHC-II⁺ macrophages: Zombie⁻, CD45⁺, Ly6G⁻, F4/80⁺, CD80⁻, and MHC-II⁺

• Macrophages: Zombie⁻, CD45⁺, Ly6G⁻, and F4/80⁺

• B cells: Zombie⁻, CD45⁺, Ly6G⁻, F4/80⁻, B220⁺,CD3⁻

• Dendritic cells (DC): Zombie⁻, CD45⁺, Ly6G⁻, F4/80⁻, B220⁻, CD3⁻, MHC-II⁺⁺, and NK1.1⁻

• NK cells: Zombie⁻, CD45⁺, Ly6G⁻, F4/80⁻, B220⁻, CD3⁻, MHC-II⁻, and NK1.1⁺

• NKT cells: Zombie⁻, CD45⁺, Ly6G⁻, F4/80⁻, B220⁻, CD3⁺, NK1.1⁺, and CD4⁻

• T cells: Zombie⁻, CD45⁺, Ly6G⁻, F4/80⁻, B220⁻, and CD3⁺

• T regulatory cells: Zombie⁻, CD45⁺, Ly6G⁻, F4/80⁻, B220⁻, CD3⁺, and FoxP3⁺

• T helper 1 cells: Zombie⁻, CD45⁺, Ly6G⁻, F4/80⁻, B220⁻, CD3⁺, CD4⁺, and Tbet⁺

Second in vivo experiment:

• DCs: Zombie⁻, CD45⁺, B220⁻, F4/80⁻, CD3⁻, CD103⁺, XCR1⁺, and MHC-II⁺

• B cells: Zombie⁻, CD45⁺, B220⁺, F4/80⁻, MHC-II⁺, CD3⁻, NK1.1⁻, and Ly6G⁻

• Activated B cells: Zombie⁻, CD45⁺, B220⁺, F4/80⁻, MHC-II⁺, CD3⁻, NK1.1⁻, Ly6G⁻, CD80⁺, and CD86⁺

• MHC-II⁺ macrophages: Zombie⁻, CD45⁺, B220⁻, F4/80⁺, MHC-II⁺, CD206⁻, and CD80⁻

Ultrafiltration EV-depleted FBS was prepared as described elsewhere (25). In brief, regular FBS was subjected to centrifugation in Amicon Ultra-15 centrifugal filters for 55 minutes at 3,000 g. The resulting flow-through was collected and used to make E-media. Rab35-WT MOC2 (capable of EV production) and Rab35-DN MOC2 (exhibiting impaired EV production) were cultured in E-media containing EV-depleted FBS and paclitaxel for 72 hours. The absence of EVs in FBS did not impact senescence induction, nor did it change apoptosis or cell death. We used both E-media with and without FBS EVs, and we did not notice any difference. Supernatants were collected, normalized based on cell numbers, and then centrifuged at 2,000 rpm for 10 minutes to remove cell debris. The supernatant was transferred to ultracentrifugation tubes (Beckman Coulter) and subjected to centrifugation at 24,200 rpm at 20°C for 70 minutes in a SW28 rotor. Following centrifugation, the supernatant was discarded, and 35 mL of sterile, 0.2 µm filtered PBS was added to the pellet. The mixture was then centrifuged again at 24,200 rpm at 20°C for 70 minutes. After discarding the supernatants, the tubes were inverted and placed on tissue papers for 1 minute to ensure complete drainage of the tube walls. Subsequently, 50 µL of 0.2 µm filtered PBS was added to each tube. The contents were gently pipetted several times to collect EVs, which were then stored at −20°C until further use.

Electron microscopy of EV preparations was done as previously described (23). Briefly, a 200 mesh–formvar, carbon-coated nickel grids were glow-discharged just before use. Five μL of each EV sample (1–5 mg/mL of protein) were placed on the grids and incubate for 5 minutes at room temperature. After washing with 3 × 30 µL of filtered PBS (1 minute/wash), the grids were fixed with 2% glutaraldehyde for 5 minutes at room temperature. After washing five times with water 2 minutes each wash, the grids were stained with UA 2% 3 minutes at room temperature.

After successfully purifying EVs through ultracentrifugation, we proceeded with Western blot analysis using the Simple Western system (26) according to the manufacturer’s guidelines. For this purpose, EVs were lysed using 1× RIPA buffer containing proteinase inhibitor cocktail. The lysate was vortexed several times and mixed with 5× Fluorescent Master Mix. The resulting mixture was then denatured at 95°C for 5 minutes along with a ladder and subsequently transferred to ice. The final concentrations of EV lysate and rabbit anti–flotillin-1 were optimized at 0.2 mg/mL and 1:50, respectively.

The quantification of EVs was conducted using nanoparticle tracking analysis (NTA) instrument (ZetaView) using two independent methods: the first method utilized the properties of both light scattering and Brownian motion to obtain the nanoparticle size distribution of the sample in liquid suspension, and the second method evaluated the fluorescence of membrane-intercalating dyes (Di 8 ANNEPS) that have bound to the membrane of EVs. The sensitivity of ZetaView was calibrated to 70 using EV-free media diluted at a ratio of 1:1,000 in 0.22 µm filtered PBS. Subsequently, the number of particles in each sample was determined after diluting the samples at a 1:1,000 ratio in filtered PBS with the same calibration adjustment. For fluorescent-based quantification, a Di 8 ANNEPS working solution was prepared by combining 1% Di 8 ANNEPS and 0.25% pluronic F-127 in PBS. The supernatants containing EVs were then incubated with the Di 8 ANNEPS working solution in a 10:1 ratio for 10 minutes at room temperature. Following the incubation, the samples were diluted at a ratio of 1:1,000 in filtered PBS and subsequently applied to ZetaView, selecting the laser at 488 nm and the filter at 500 nm. Control samples (filtered PBS, filtered media, and diluted Di 8 ANNEPS working solution) were routinely used and yielded no particle counts.

EVs from senescent MOC2 cells were collected as described above. Senescent MOC2 cells expressing Rab35-DN were mixed immediately before injection with senEVs (100 µg/mouse, about 25 µL) and implanted intradermally as described above (N = 10 mice per group). Additional transfers of senEVs (same dosing) were performed every 5 days, intradermally and near the site of SC injection. Control mice received senescent MOC2 cells expressing Rab35-DN mixed with PBS and subsequent PBS injections. The dosing was established based on the amount of senEVs released per day in vitro. To avoid batch effects, three separate senEV preparations were pooled, each yielding ∼4 mg/mL in 250 µL. Tumor volumes were measured as described above.

Proteins were extracted using a modified Folch extraction method (27). Briefly, samples were extracted with chloroform/methanol (2:1, v/v) in a 4-fold excess to the sample volume to yield a hydrophobic layer containing lipids and a protein disc. Each layer was transferred to separate glass autosampler vials, and the protein pellet was washed with 1 mL of ice-cold methanol. Each sample was evaporated to dryness in vacuo and appropriately stored at −20°C until further processing and analysis.

Proteins were dissolved in 8 mol/L urea prepared in 50 mmol/L NH₄HCO₃, pH 8.0, containing 5 mmol/L dithiothreitol and reduced for 30 minutes at 37°C. Cysteine residues were alkylated for 30 minutes at 37°C with a final concentration of 10 mmol/L iodoacetamide from a 500 mmol/L stock solution. Samples were diluted 8-fold with 50 mmol/L NH₄HCO₃, pH 8.0, and 1 mol/L CaCl₂ was added to a final concentration of 1 mmol/L. Proteins were digested with sequencing-grade modified trypsin (Promega) overnight at 37°C. Samples were submitted to solid-phase extraction using C18 spin columns (UltraMicroSpin columns, C18, 3- to 30-µg capacity; Nest Group), as previously described (28).

Peptides were separated on a capillary column (60 cm × 50 μm ID packed with C18, 3-μm particles) using a gradient of acetonitrile (mobile phase B) in water (mobile phase A), both supplemented with 0.1% formic acid. The elution was carried with the following gradient: 2 minutes, 8% B; 20 minutes, 12% B; 75 minutes, 30% B; 97 minutes, 45% B; 100 minutes, 95% B; 105 minutes, 95% B; 115 minutes, 1% B; and 120 minutes, 1% B. Eluting samples were analyzed on a Q-Exactive mass spectrometer (Thermo Fisher Scientific). Full-scan spectra were collected in a range of 300 to 1,800 m/z with a resolution of 70,000 at m/z 400. The 12 most intense parent ions with multiple charges were submitted to high-energy collision dissociation with 33% normalized collision energy at a resolution of 17,500. Each parent ion was fragmented once and then excluded for 30 seconds.

LC-MS/MS data were processed with MaxQuant (29) by searching tandem mass spectra against the human sequences from UniProt Knowledgebase. The intensity-based absolute quantification method was used for quantification (30).

The expression levels of cytokines and chemokines in cell supernatants were quantified using Proteome Profiler Mouse Array Kit. Supernatants from 80% confluent cells were collected and centrifuged at 1,000 rpm for 5 minutes. Array membranes were incubated and developed with supernatants mixed with biotinylated detection antibodies following the manufacturer’s instructions. For the detection of cytokines/chemokines, IRDye 800CW streptavidin was used, and arrays were scanned using a Li-Cor Odyssey imaging system, which enables near-IR fluorescence detection. The mean pixel intensity for each signal was analyzed using ImageJ.

Bulk RNA sequencing (RNA-seq) libraries were prepared using TruSeq Stranded mRNA Library Prep Kit (Illumina). Briefly, RNA was profiled for integrity on a 2200 Bioanalyzer (Agilent). Poly(A)+ RNA was isolated using Oligo-dT magnetic beads and then fragmented using divalent cations and heat. Fragmented RNA was converted to double-stranded cDNA by random hexamer priming. The second strand was synthesized with dUTP in place of dTTP to prevent priming from this strand during amplification. The cDNA was ligated to adapters with dual unique indexes and amplified by limited rounds of PCR. The amplified material was cleaned using AMPure XP Beads (Beckman Coulter). The final library was profiled on a 4200 TapeStation (Illumina) and quantified by RT-PCR using NGS Library Quantification Kit (Roche/Kapa Biosystems) on a StepOnePlus RT-PCR workstation (Thermo Fisher Scientific/ABI).

Single-cell RNA-seq libraries were prepared using Chromium Next GEM Single Cell 3′ Reagent Kit v 3.1 with dual indexes (10x Genomics). Cells were counted and then processed through the Chromium Controller (10x Genomics) to produce amplified cDNAs with cell-specific tags. The cDNA was enzymatically fragmented and then size-selected using solid-phase reversible immobilization (SPRI) beads. Adapters were ligated to the cDNA, and samples were indexed using indexed amplification primers. Libraries were profiled on a 4200 TapeStation (Agilent) and quantified using RT-PCR with NGS Library Quantification Kit (Roche/Kapa Biosystems) on a StepOnePlus RT-PCR workstation (Thermo Fisher Scientific/ABI). Libraries were sequenced on a NovaSeq 6000 (Illumina) running RTA v3.4.4. BCL files were demultiplexed using bcl2fastq v2.20.0.422 (Illumina).

A custom reference genome was constructed using GRCm38 (Ensembl) and the reporter gene dLNGFR with genome annotations from GRC38.89.gtf. Briefly, the dLNGFR FASTA file was added to the end of the GRCm38 FASTA as a virtual chromosome using the Linux cat command. The GTF file was manually edited to add dLNGFR annotation. The FASTA and GTF files were then used to construct a reference genome using the STAR genomeGenerate runMode command. Cell Ranger 7.1.0 uses STAR 2.7.2a aligner for building and aligning to the reference genome. Bulk RNA libraries were aligned using STAR 2.5.0a.

Bulk RNA-seq analysis was performed with the DESeq2 (v1.42.0) R package (31). Raw count data were normalized using the variance stabilization transformation function, and differential expression was computed between sets of triplicate samples using DESeq2 with LFCshrinkage via apeglm (32).

All analyses conducted in this study were carried out using R software (version R 4.1.2) and RStudio (version 2023.12.0+369). Single-cell RNA-seq data underwent analysis utilizing the Seurat R package (version 4.3) to convert matrix files into Seurat objects (33). We excluded cells of substandard quality (minimum number of unique genes = 350 and maximum mitochondrial RNA = 15%). After hashtag demultiplexing (Seurat), doublets were removed, and the data from the two independent experiments were merged. Both singlets and hashtag negative cells were retained for analysis, with hashtag negative cells used just for dimensionality reduction and unsupervised clustering but excluded from differential expression analysis. Unique molecular identifier (UMI) counts were normalized with the SCTransform approach to correct for depth of sequencing between cells (34). The total number of the quality control–passed cells in this study was 27,219. The top 2,000 highly variable genes were selected for dimensionality reduction using principal component analysis and uniform manifold approximation and projection. Initial cell clustering was performed with the top N principal components using the Louvian algorithm (“FindClusters” function) across a sweep of resolutions. We computed the silhouette width and root mean square error for each clustering resolution and used the value that maximized silhouette width and minimized root mean square error (33). The “FindAllMarkers” function was utilized to identify marker genes upregulated in each cluster and define their identity, resulting in the identification of 23 clusters, including 7 stromal clusters, 5 epithelial clusters, and 9 immune clusters. This approach failed to resolve DCs and B cells, so the cluster containing both was separated and reprocessed (variable feature identification and principal component analysis) to identify them, increasing the total number of clusters to 24. Additionally, the Seurat function “FindAllMarkers” was used to identify differentially expressed genes (DEG) for each cluster."

We used three metrics to differentiate between epithelial clusters composed of endogenous cells and the implanted MOC2 SCs; (i) imputed copy-number variation (CNV) distance from diploid, (ii) correlation to bulk MOC2 RNA-seq profiles, and (iii) expression of human transgene dLNGFR. For CNV analysis, we subset the scRNA-seq data to just the cells with either a “stromal” or “epithelial” lineage label and used the CopyKAT R package (v1.1.0) to infer CNVs from gene expression (35). The CopyKat algorithm was applied with “stromal” cells set as expected diploid using the “mm10” genome and default parameters. To identify outlier cells, we computed the distance from diploid as the logarithm of summed squared distance from diploid. To compute expression correlation to the bulk profile, we found the intersect of genes expressed in the scRNA-seq and bulk-RNAseq data and computed the single-cell Pearson correlation to mean expression bulk profiles for triplicate samples of paclitaxel-treated MOC2-25 (EV-proficient) and MOC2-27 (EV-deficient). For assessment of dLNGFR expression, this gene was added to our reference transcriptome prior to alignment, and dLNGFR was normalized with the SCTransform approach along with all endogenous mouse genes.

For intercellular communication analysis and network visualization of scRNA-seq data, we used the R package “CellChat” (v1.6.1), a public repository of ligands, receptors, cofactors, and their interactions (36). We utilized CellChatDB.mouse, a database containing literature-supported ligand–receptor interactions in mice. This database distinguishes between three classes of ligand–receptor interactions: secreted signaling, ECM–receptor, and cell–cell contact interactions. Seurat normalized gene expression was used to create cell chat objects from the pooled tumors for each condition (senEV-proficient and senEV-deficient), and then these interactions were compared to identify differences across the conditions. To identify signaling pathways between senEVs and their microenvironment, we focused our analysis specifically on membrane-bound ligand–receptor interactions (i.e., cell–cell contact interactions). A heatmap was used to detail the differential interaction numbers and strengths between the different subsets identified within the senescence microenvironment. Chord diagrams were used to focus on top membrane-bound ligand–receptor interactions after specifying the sender and recipients of interest. A bubble plot was used to compare communication probabilities mediated by specific ligand–receptor pairs (based on proteomic data) toward any immune cell subset. SCs and stem-like SCs displayed overlapping interaction partners, including themselves, IFN-activated neutrophils, basal epithelial cells, and Schwann cells, when expressing Rab35-WT. For this reason, we considered them as a single subset in following analyses.

To unbiasedly define SASP factors, we filtered the BioMart database (37) for genes containing a signal peptide and lacking a transmembrane domain.

To deplete CD4 T cells, 200 µg of anti-CD4 antibody (GK1.5) was administered intraperitoneally once per week. The first administration was performed a day before challenge with senescent MOC2 cells. More than 99% of CD4 T cells were depleted in circulation, as confirmed 4 days after the first GK1.5 treatment (using a different anti-CD4 clone).

Statistical analyses of the results were performed using appropriate tests in GraphPad Prism version 10. Proteomic and RNA-seq data were analyzed using R. Adjusted P values <0.05 were defined as significant. Biological and technical replicates are reported for each experiment.

Sequencing data has been deposited to Gene Expression Omnibus under accession numbers GSE262139 (bulk RNA-seq) and GSE262141 (scRNA-seq). Other data generated in this study are available upon request from the corresponding author.

SenEVs are a novel and largely underappreciated component of the senescence secretome (38). To enable in vivo assessment of senEVs, we developed an engraftment-based approach that allows us to genetically control EV release specifically in SCs. We induced senescence in vitro by treating murine squamous cell carcinoma cell lines (MOC2 and mEER⁺) with chemotherapeutic agents (Fig. 1; Supplementary Fig. S1, respectively). We chose cisplatin and paclitaxel because they are standard-of-care for head and neck squamous cell carcinoma (HNSCC; ref. 39). To optimize the dose, we tested different concentrations of paclitaxel (0–750 nmol/L) and cisplatin (0–35 µmol/L) in a combinatorial manner, with each condition replicated in quadruplicate (Fig. 1A). After 48 hours of treatment, we quantified cellular senescence by measuring the senescence marker SA-β-Gal activity (40). A defining feature of cellular senescence is the exit from the cell cycle while maintaining cell viability. To confirm that paclitaxel treatment generates bona fide SCs, we assessed proliferation (by BrdU incorporation), apoptosis (by Annexin V), and cell death (by 7-AAD), or lack thereof. These experiments determined the optimal doses of paclitaxel and cisplatin to maximize induction of cellular senescence by minimize proliferation without inducing significant cell loss by apoptosis or cell death (Fig. 1A; Supplementary Fig. S1). Although maximum senescence was induced when treating MOC2 or mEER⁺ with 600 nmol/L paclitaxel and 4 µmol/L cisplatin, the addition of cisplatin increased cell death and reduced cell recovery, which would have a negative impact on the model. Therefore, cisplatin was excluded from the senescence-inducing treatment in subsequent experiments. Bulk RNA-seq data from cultured senescent and proliferating cells (MOC2) confirmed increased expression of the senescence markers Cdkn1a (p21), Cdkn2a (p16), and Glb1 (SA-β-Gal) and decreased expression of the proliferation marker Ki67 in paclitaxel-treated versus untreated MOC2 (Fig. 1B). SCs exhibit an enlarged and flattened morphology (41). SA-β-Gal activity correlated with morphologic changes of the cells after exposure to chemotherapeutic drugs (Supplementary Fig. S1). These results indicate that 600 nmol/L paclitaxel triggers treatment-induced senescence in vitro while preserving cell recovery.

We next quantified the proliferative ability of SCs to evaluate the durability of the paclitaxel-induced senescent phenotype. To this end, we analyzed the in vitro proliferation rate of MOC2 cells treated with 600 nmol/L paclitaxel (Fig. 1C–E). Continuous paclitaxel treatment blocked cell proliferation, with only a ∼10% cell loss on day 3 as compared with days 1 to 2 (Fig. 1C). To quantify the residual proliferative potential of SCs, we measured BrdU incorporation at saturation and after a short pulse while culturing SCs in presence of paclitaxel (Fig. 1D and E). Over a 24-hour period, BrdU incorporation plateaued at ∼15%, suggesting that paclitaxel alone was sufficient to reduce maximum proliferation of MOC2 ∼7-fold (down from 100% BrdU incorporation). To assess the rate of proliferation, we compared BrdU incorporation between paclitaxel-treated and untreated MOC2 cells after a short pulse with BrdU (2 hours; Fig. 1E). We observed an 8-fold reduction in the rate of BrdU incorporation, from 40% to 5%. These analyses indicate that paclitaxel treatment alone can reduce the residual proliferation potential of a fast-growing squamous carcinoma model by ∼2% (see “Materials and Methods”).

Because we aimed to implant SCs in vivo, without continuing paclitaxel treatment in mice, we assessed whether in vitro–generated SCs retain the senescence phenotype after removal of paclitaxel. To this end, we simulated in vivo conditions by removing paclitaxel after 2 days of treatment and quantified the persistence of senescence over time (Fig. 1F). We determined that over 7 days, SA-β-Gal activity did not decrease and remained present in ∼80% of the cells. These results suggest that the senescent state may be retained in vivo. To confirm this, we measured the tumor formation ability of our SC preparation in immunocompetent syngeneic mice (Fig. 1G). After implanting the same numbers of viable or senescent MOC2 cells, we observed that SCs were unable to form tumors for at least 2 weeks, indicating that paclitaxel treatment induces a functional state of cellular senescence. These results indicate that although our senescence model does not fully recapitulate all the events that take place in patients before and during recurrence (i.e., mutagenesis, immune escape, tumor growth, surgery, chemotherapy, and recurrence), it can faithfully model the final steps, and therefore it represents a useful tool to study tumor recurrence.

The removal of SCs is mediated by the immune system (42). The immune response against SCs is different from that elicited within the tumor microenvironment (43). To test whether our engraftment model of senescence recapitulates this difference, we studied the immune infiltrate recruited within 3 days of orthotopic injection of SCs (Fig. 1H). When we compared senescence and tumor models, we observed significant differences in their immune composition. As expected (15, 16), the fractions of T and NK cells were significantly increased within senescent tissues 3 days after challenge. Although not statistically significant, we observed a trend in macrophage subsets, with a relative increase in the fraction of MHC-II⁻ CD80⁺ CD206⁺ macrophages and a corresponding decrease in the MHC-II⁺ CD80⁻ CD206⁻ macrophages in presence of SCs. On the contrary, neutrophils and regulatory (FoxP3⁺) and type-1 helper (Tbet⁺) T cells were similarly recruited by both tumor cells and SCs. These results suggest that our senescence model recapitulates the immune responses observed in different in vivo models of cellular senescence (15).

EVs are key mediators of intercellular communication (7). SenEVs may carry unique signals that impact the recruitment of immune cells (Fig. 1H; refs. 8, 9). To explore the role of senEVs in senescence surveillance (i.e., the immune-mediated removal of SCs; ref. 15), we characterized the protein composition of senEVs in our model system. We isolated EVs from senescent and proliferating squamous carcinoma cells using ultracentrifugation and compared their protein content, quantities, and size by spectrometry, NTA, and electron microscopy (Fig. 2A–C; Supplementary Fig. S2A). We found that the protein content released into EVs from SCs was ∼6-fold higher compared with nonsenescent squamous carcinoma cells (Fig. 2A). These results may be explained by either a higher rate of senEV release, larger average senEV size, higher density of protein cargo, or a combination of the above. To discriminate between these alternatives, we performed NTA. Whereas EV numbers were similar, we observed a 16% increase in median diameter (Fig. 2B), which corresponds to a 56% increase in EV volume (assuming a spherical shape). Electron microscopy analysis confirmed that senEVs are 16% larger than tumor EVs in both MOC2 and mEER⁺ cells (Fig. 2C). Immunoblot analysis for a pan-EV marker (flotillin-1; ref. 44) confirmed the presence of bona fide EVs (Fig. 2D). These findings suggest that senEVs exhibit a larger size and carry a greater protein payload in comparison with EVs from proliferating squamous carcinoma cells.

To determine the protein composition of senEVs, we performed proteomic analyses on EVs isolated from senescent and proliferating MOC2 and mEER⁺ cells (Fig. 2E–G; Supplementary Table S2). We found 150 and 120 proteins significantly different in senEVs from MOC2 and mEER⁺ cells, respectively (as compared with EVs from proliferating squamous carcinoma cells; Fig. 2E). Pathway analysis showed that proteins enriched in senEVs were significantly associated with core cellular functions, including ribosomes, spliceosomes, and DNA replication but also with type-1 immunity (systemic lupus erythematosus; Fig. 2F). When we determined the common proteins enriched in senEVs from both cell lines (as compared with EVs from the respective proliferating cells), we found a 30% overlap (Fig. 2G). Among these proteins, ribosomal proteins were the largest family.

Several studies have reported a dual role for the SASP in both promoting and inhibiting cancer progression and recurrence (3, 4, 15, 16). These conflicting results may derive from the presence of different amounts and types of senEVs in SASP studies. To distinguish the role of senEVs in senescence surveillance from that of the SASP, we aimed to specifically inhibit senEV release. Inhibition of Rab27a is a commonly used method to block EV release (45). To test whether inhibition of Rab27a would stop release of EVs in the context of senescence, we used siRNA oligoes to successfully inhibit expression of Rab27a in SCs, as measured by qPCR (Supplementary Fig. S2B). However, two independent methods to assess inhibition of EV release from SCs showed that EVs were instead significantly increased (Supplementary Fig. S2C and S2D). These observations, together with the fact that the total protein content of senEVs was significantly decreased by Rab27a knockdown (Supplementary Fig. S2E), suggest that Rab27a in SCs may contribute to loading EV cargo proteins and that SCs attempt to compensate the diminished EV cargo by releasing higher numbers of senEVs.

We previously validated an alternative genetic tool to inhibit EV release from parental cells (17–19). In this approach, Rab35-DN is expressed in parental cells. As control, Rab35-WT is used. To confirm that expression of Rab35-DN efficiently and specifically reduces EV release in both proliferating and senescent MOC2 and mEER⁺ cells, we performed immunoblot assays for the pan-EV marker flotillin-1 on purified EVs. We observed almost complete absence of EVs in both proliferating and senescent MOC2 expressing Rab35-DN (Fig. 3A; Supplementary Fig. S3A and S3B). We further validated inhibition of EV release using an independent technique and confirmed a 4- to 6-fold decrease of EV release from SCs expressing Rab35-DN (Fig. 3B). Comparison of EV diameters using membrane-intercalating fluorescent dyes (which allows distinguishing membrane-bound particles from nonvesicular particles) suggests that, notwithstanding some variability due to low EV numbers, the increase in size is associated with the senescence phenotype and not with the expression of Rab35-DN (Supplementary Fig. S3C). These results indicate that the profound inhibition of senEV release is suitable for in vivo functional studies.

A caveat of interfering with vesicular trafficking via Rab35 is that some SASP factors may be decreased as well, thereby confounding the interpretation of in vivo data. To test whether Rab35-DN expression in SCs inhibits SASP components, we quantified 40 different secreted factors by cytokine array on SC culture supernatants (Fig. 3C). Of these 40 mediators, 8 were expressed at detectable levels (G-CSF, GM-CSF, CXCL10, CXCL1, M-CSF, CCL2, CXCL2, and TIMP1), and are all well-established SASP factors (46). Surprisingly, two of these eight SASP factors (CCL2 and CXCL2) were expressed at higher levels in Rab35-DN SCs. An additional six mediators (IL1α, IL5, IL17, CCL5, CXCL12, and TNFα) were expressed at low but detectable levels and were unaffected by expression of Rab35-DN. To confirm these observations, we performed bulk RNA-seq of proliferating and senescent MOC2 from cultures (Fig. 3D; Supplementary Fig. S3D; Supplementary Table S3). We focused our analysis on those genes exclusively featuring signal peptides and lacking transmembrane domains, categorizing them as secretory genes. The analysis of DEGs between SCs expressing Rab35-DN and SCs expressing Rab35-WT revealed that most secretory genes were not differentially expressed, with only three being downregulated more than 2-fold (Ccl28, Inhbb, and Chrd; Fig. 3D). Noting that these three genes are not commonly found among SASP factors, these findings indicate that expression of Rab35-DN does not impede the release of SASP factors.

Whereas the role of the SASP in intercellular communication is well established (6), the specific role of senEVs in communicating with the SC microenvironment is still largely unclear. Although current approaches using the isolation and injection of exogenous EVs (derived from cell cultures or biofluids) in animal models allow precise control over pharmacokinetic and pharmacodynamic parameters, it remains uncertain whether the insights gained from exogenously administered EVs sufficiently address key aspects of EV biology (47). To overcome these limitations, we took advantage of our engraftment-based senescence model to investigate native senEVs, i.e., senEVs released by SCs living within a tissue. To this end, we challenged mice orthotopically with Rab35-WT or Rab35-DN SCs (Fig. 4A). We observed that in the absence of senEVs, tumor recurrence was significantly accelerated, as compared with mice receiving Rab35-WT SCs (Fig. 4A). To assess whether expression of Rab35-DN interferes with paclitaxel-induced senescence, we compared SA-β-Gal activity and cell expansion in vitro (Supplementary Fig. S4A and S4B). We found that Rab35 manipulation did not influence senescence induction (Supplementary Fig. S4A) and did not rescue the blockade of cell expansion at the population level (Supplementary Fig. S4B), pointing to the fact that inhibition of senEV release does not interfere with induction of senescence by paclitaxel treatment. To test whether inhibition of tumor recurrence by EVs is specific to senescence, we tested whether expression of Rab35-DN impacts in vivo growth of untreated, nonsenescent MOC2 cells. As compared with MOC2 expressing Rab35-WT, we observed no differences in tumor growth of Rab35-DN expressing MOC2 (Supplementary Fig. S4C), indicating that the inhibition of disease progression is a property specific of senEVs. These findings suggest that the release of senEVs inhibits tumor recurrence, possibly via non–cell-autonomous mechanisms.

To better understand the initial host responses to cellular senescence, we studied our engraftment-based model 3 days after challenge with SCs (i.e., before tumor formation). We took advantage of a genetic marker (dLNGFR) expressed along with Rab35 variants to track SCs in vivo (Fig. 4B). We found that interfering with senEV release led to 4-fold higher accumulation of SCs on day 3, as compared with Rab35-WT SCs (Fig. 4B). This observation was confirmed by two independent single-cell sequencing experiments, also performed on day 3 after SC implantation (Fig. 4C). To test whether senEVs have a role in SC removal through recruitment of immune cells, we repeated the experiment above and quantified the immune infiltrate early after SC engraftment by flow cytometry (Fig. 4D; Supplementary Fig. S4D). Among the immune subsets we queried were CD4 and CD8 T cells, their type-1 effector subsets expressing T-bet, T regulatory cells, NK and NKT cells, myeloid cells (including neutrophils, DCs, macrophages, and their subsets), and B cells. We observed that in the presence of senEVs, specific immune cell subsets were significantly increased (∼50%; P < 0.05), including infiltrating DCs, B cells, and MHC-II⁺ macrophages (Fig. 4D; Supplementary Fig. S4D). These immune cell subsets were related only by the fact that they are all professional antigen-presenting cells (APC), indicating that senEVs are necessary to recruit APCs and that the absence of senEVs increases accumulation of SCs in tissues.

To test whether senEVs are necessary and sufficient to counter tumor recurrence, we performed a complementation experiment in which we reintroduced senEVs along with SCs unable to release them. To this end, we implanted senescent MOC2 cells expressing Rab35-DN intradermally in two groups of mice. One group received the SCs mixed with 100 µg of senEVs and additional intradermal administrations of senEVs (100 µg) every 5 days near the SC injection site. We selected the amount of senEVs based on how many EVs SCs release daily in vitro. Control mice received the same treatment with PBS (Fig. 4E). Strikingly, provision of senEVs restored tumor recurrence to levels observed in untransduced SCs (Fig. 1G) and in SCs expressing Rab35-WT (Fig. 4A). Together, these results indicate that the presence of senEVs is sufficient and necessary to counter tumor recurrence.

To better understand the signals that senEVs deliver to immune cells during senescence surveillance, we profiled the senescence microenvironment at the single-cell level, 3 days after engraftment of Rab35-WT or Rab35-DN SCs, and prior to tumor formation. To collect enough cells for profiling, we performed the experiment twice by pooling three mice per group every time. We corrected for differences in sequencing depth using SCTransform (48, 49), which successfully removed batch effects (Supplementary Fig. S5A). We classified three major lineages, namely, immune (Ptprc, S100a8, S100a9, and Cd14), epithelial (Epcam, Krt5, Krt19, and dLNGFR), and stromal (Pdgfra, Pdgfrb, and Pecam1; Supplementary Fig. S5B), based on the DEGs in each lineage compared with the others (Fig. 5A). To distinguish between different immune cell subsets, we subclustered immune lineage cells separately (Fig. 5B). This procedure allowed us to identify a diverse combination of lymphoid and myeloid immune cells recruited to the site of senescence, including two distinct subsets of macrophages characterized by expression of C1q and Spp1, DCs, monocytes, T_H17 cells, four subsets of neutrophils, and B cells (Supplementary Table S4). The Spp1⁺ macrophage subset corresponds to the MHC-II⁺ macrophages identified by flow cytometry (Supplementary Fig. S6). These results confirmed the presence of the APC subsets we observed by flow cytometric analyses (Fig. 4D; Supplementary Fig. S4D).

To identify the implanted SCs among the single cell pool, we used three different approaches: correlation with bulk RNA-seq profiles of SCs from culture, expression of the transgenic marker dLNGFR (coexpressed along with WT or DN Rab35), and CNV analysis (Supplementary Fig. S7). This examination identified three clusters of epithelial lineage cells as originating from the engrafted SCs (Fig. 5C). We named these three clusters based on their relative upregulation of markers related to “senescence,” “stemness,” and “proliferation” (Fig. 5D). Notably, all three SC clusters were found in both senEV-proficient and -deficient microenvironments, with SCs constituting the largest fraction (62%), followed by SCs expressing the stemness markers Sox9, Hmga2, Ly6a (Sca1), and Krt19 (Fig. 5D; refs. 50–53). These results agree with reports showing that SCs can transform into cancer stem cells (54–57) and highlight the usefulness of this model system to further study that process.

The presence of senEVs had a significant impact on which signaling pathways were engaged in immune cells (Supplementary Fig. S6). To query immune cell–SC communications via EVs, we utilized the R package CellChat to impute likely ligand–receptor signaling from gene expression and distinguish secreted and membrane-bound ligand–receptor interactions (36). As compared with senEV-deficient, the senEV-proficient senescence microenvironment displayed significant changes in the strength and number of putative ligand–receptor interactions among the 24 cell types (Supplementary Fig. S8A). When expressing Rab35-DN, SCs and stem-like SCs mostly signaled to monocytes, T cells, neutrophils, B cells, luminal epithelial cells, and lymphatic endothelial cells (Supplementary Fig. S8A). These results indicate that the senescence microenvironment is highly affected by senEVs.

To focus on membrane-bound ligands (i.e., those on the surface of SCs and senEVs), we excluded SASP signaling (see Materials and Methods). When we assessed senescence microenvironments in the presence of senEVs, we observed that SCs signaled via membrane-bound ligands mainly with MHC-II⁺ APCs, i.e., B cells, DCs, and Spp1⁺ macrophages (Fig. 5E). In these interactions, the top membrane-bound ligand–receptor pair we identified was App-Cd74, followed by homotypic interactions via Cdh1. Strikingly, when senEVs were absent, the interaction partners of SCs shifted to neutrophil subsets (Fig. 5E). SC-Neu interactions in the absence of senEVs were based upon homotypic Pecam1 (CD31) interactions and Cdh5-Cdh1, followed by App-Cd74 and homotypic Esam interactions. Consistently, IFN-activated neutrophils and C1q⁺ macrophages (i.e., MHC-II–negative) were both increased in senEV-deficient senescence microenvironments (Supplementary Fig. S8B and S8C). These results suggest that senEVs promote SC engagement with APCs.

To distinguish between membrane-bound ligands on senEVs versus those on their parental cells (i.e., SCs), we took advantage of our senEV proteomic profile (Fig. 2E–G). First, we calculated the communications originating from SCs that were enriched in the presence of senEVs. This analysis was restricted to membrane-bound ligands (Supplementary Fig. S9A). Then, we intersected that list with the list of proteins detected on senEVs (Fig. 5F). The resulting short-list represented putative ligands specifically present on senEVs and not on the SC surface. This analysis yielded six membrane-bound ligands present on senEVs and engaging with cells within the senescence microenvironment: Cdh1, Dsg2, F11r, Ncam1, Ocln, and Pvr. These proteins are involved in adhesion and were found on both SCs and stem-like SCs. Interaction analysis suggested that senEVs are capable of binding to both MHC-II⁺ APC and T_H17 cells through these adhesion-related proteins (Fig. 5F). These results suggest that senEVs may enhance the likelihood of endocytosis of SC antigens, potentially supporting T-cell engagement by APCs.

Given the observed shift in SC–immune cell interactions away from APCs when senEVs are absent (Fig. 5E), and the resulting increase in tumor formation (Fig. 4A), we hypothesized that adaptive immune responses may have been impacted as well. To test this hypothesis, we queried the signaling pathways from APCs to T_H17 cells (Fig. 5G; Supplementary Fig. S8D). We found that in senEV-proficient senescence microenvironments, T_H17 cells have the potential to interact with all three MHC-II⁺ APC subsets via cell surface signaling (Fig. 5G). Moreover, T_H17 cells expressed the activation marker CD69 (Supplementary Table S4), suggesting that they received antigen-specific stimulation (58). In senEV-deficient senescence microenvironments, DC cross-talk with T_H17 cells used different membrane-bound ligand–receptor interactions (Fig. 5G), whereas B cells and Spp1⁺ macrophages displayed an immunoregulatory role by targeting T_H17 cells with secreted signals (Supplementary Fig. S8D). These analyses suggest that APCs recruited by senEVs may prime and activate T_H17 cells.

B–T cell coactivation is central to adaptive immune responses (59). In senEV-proficient senescence microenvironments, B-cell activation was restrained by the inhibitory coreceptors CD22 and Siglecg and by the release of Galectin-9 (Lgals9) from T_H17 cells (Supplementary Fig. S8E and S8F; refs. 60, 61). B–T cell communication via soluble mediators was inferred to be mostly unidirectional, from T_H17 cells to B cells (Supplementary Fig. S8F). These analyses suggest that T_H17 cells may inhibit germinal center formation and that B cells may act mainly as APCs during the initial phases of senescence surveillance.

To experimentally validate the role of CD4 T cells in senescence surveillance, we performed antibody-mediated CD4 T-cell depletion (Supplementary Fig. S9B). CD4 depletion removes all CD4 T-cell subsets, including T_H17 cells. We observed that CD4 T cells were necessary to control tumor recurrence in senEV-proficient senescence microenvironments (Fig. 5H). These results are in line with the reported role of this T-cell subset in senescence surveillance (15) and suggest that CD4 T cells are downstream of senEVs in the immune response to senescence (Fig. 5I).

Murine models have been instrumental in studying cellular senescence across various diseases and physiologic processes (62). They led to the identification of key senescence markers, understanding the role of senescence in disease progression, and exploring potential therapeutic interventions (1). However, challenges remain in identifying and characterizing SCs in vivo due to the lack of shared markers (63). Transgenic models of senescence are particularly susceptible to the uncertainty around SC biomarkers and require lengthy characterization of the expression patterns of the gene construct utilized. Nongenetic methods used to induce senescence in wild-type animals have systemic effects; for example, hydrodynamic (tail vein) injection of large volumes of plasmids expressing oncogenes (such as mutant Ras; refs. 15, 16) or administration of cytotoxic drugs such as bleomycin (64) cannot control which cell types become senescent. Of note, administration of cytotoxic/cytostatic drugs induces broad damage to tissues and alters immune cell development (65), which complicates the study of immune responses to SCs. In this study, we addressed these limitations by validating an engraftment-based senescence model that recapitulates therapy-induced senescence and tumor recurrence in patients with HNSCC. Although our model system does not fully recapitulate the pathogenesis of HNSCC, we think that our approach specifically models the recurrence stage, in which residual tumor cells accidentally left behind after surgery become senescent in vivo, during chemotherapy cycles.

Our approach allows to track the fate of implanted SCs in vivo by insertion of a genetic marker (dLNGFR) before senescence induction. The engraftment-based senescence model described here enables tracking of SC fates and functional studies via gene overexpression or knockdown; we took advantage of this feature to inhibit the release of senEVs. Lastly, our model allows us to tightly control which cells are induced into senescence without affecting immune cells.

The role of EVs in cellular senescence is currently understudied and poorly understood. Some studies reported that senEVs promote tumor cell proliferation in vitro (66, 67). Whether these effects are also in place in vivo and how the senescence microenvironment modulates tumor cell–autonomous effects is unclear. Our results suggest that the non–cell-autonomous effects of senEVs may overrule their direct signaling to (pre)cancerous cells, ultimately establishing senEVs as key to inhibit tumor formation and recurrence. This conceptual advance in our understanding of the senescence microenvironment is in line with the large body of knowledge accumulated around the impact of the tumor microenvironment in cancer (68). Our results suggest that senEVs are a distinct subset of EVs and differ from them in both morphology and protein content. Importantly, the unique set of proteins carried by senEVs may be involved in recruiting and activating immune cells.

In this study, we engrafted genetically engineered SCs at orthotopic locations to investigate the specific role of senEVs in senescence surveillance without interfering with the SASP. Although small changes in five soluble mediators (Ccl2, Cxcl2, Ccl28, Inhbb, and Chrd) were detected after Rab35-DN expression, our complementation experiment indicates that senEVs are sufficient and necessary to counter tumor recurrence. Nonetheless, T_H17 cells expressed CCR2, which likely mediated their recruitment into the senescence microenvironment (69). These results suggest that senEVs and SASP factors play complementary but nonoverlapping roles in senescence surveillance.

The persistence of SCs in epithelial tissues is detrimental; it is commonly found in precursor lesions, leading to HNSCC in patients (70, 71), and it can drive cancer recurrence after chemoradiation regimens (3, 4). Our functional studies indicate that senEVs have a key role in the processes that mediate removal of SCs. This is important because our work shows that accumulation of SCs in the absence of senEV signaling leads to increased cancer recurrence. Our analyses suggest that specific immune cells (i.e., MHC-II⁺ APCs) need to be recruited to promptly clear SCs and avoid propagation of the senescence phenotype to neighboring cells. Our data suggest that a potential mechanism for this phenomenon may involve neutrophils, which were increased in the absence of senEVs and have been reported to spread senescence to neighboring cells (72). On the other hand, APCs like macrophages and DCs, along with IFN signaling, contribute to antitumor immunity (73). Our data show that APCs release cytokines (including Ccl2, Ccl6, and Ccl7) able to recruit T_H17 cells via CCR2. Once in the senescence microenvironment, T_H17 cells signaled with APCs via CD86-CD28, likely leading to T-cell activation. Consistently, depletion of CD4 T cells (the parental subset of T_H17 cells) fully abolished senEV protective ability, suggesting that T_H17 cells are downstream of senEVs in the immune response to senescence. Additional studies using depletion strategies more specific for T_H17 cell will shed more light on their mechanism of action in senescence surveillance.

SenEVs carried several cell adhesion molecules that may mediate the uptake of SC antigens by MHC-II⁺ APCs and may strengthen the immunologic synapse between APCs and T cells during senescence surveillance. As compared with tumor microenvironments, we observed that NK and T cells were increased in senescence microenvironments, which is consistent with previous reports (15, 16). While the microenvironment of SCs expressing Rab35-DN quickly becomes a tumor microenvironment, the early time point chosen for analyses (3 days) means that both Rab35-WT and Rab35-DN groups largely represent senescent microenvironments. It is likely that NK and T cells may follow these initial immune responses. Further studies will clarify the time course of SC populations and immune surveillance during tumor recurrence. Overall, these results indicate that senEVs play a major role in senescence surveillance. The ability of SCs to release EVs complements the activity of the SASP in recruiting and activating distinct immune cell subsets for efficient removal of SCs. More work will be necessary to understand whether, after uptake, presentation of senEV-bound antigens and generation of antigen-specific immune responses are required to remove SCs.

Virus-like particles from endogenous retrovirus can be found within EV preparations obtained from several mouse tumor cell lines (74). Our proteomic analysis of senEVs detected high levels of envelop protein from murine leukemia virus, which was present in both senEVs and tumor EVs. Still, it is possible that virus-like particles may have contributed to the observed effects on tumor recurrence.

Although NTA may suggest that there may be a small percentage of EVs that are still released by SCs expressing Rab35-DN, immunoblot analysis for flotilin-1, a bona fide pan-EV marker (44), indicates a nearly complete (>95%) suppression of EV release. Unfortunately, NTA methods are affected by high background due to the inherent inability of scatter-based measurements to distinguish membrane-bound particles from nonvesicular particles and the potential presence of micelle-forming dyes during fluorescent-based measurements (44). We concluded that even if EV secretion was not completely abated, the level of inhibition achieved was suitable for in vivo functional studies.

Rab35 is involved in cytokinesis (75), which may impact residual tumor cells’ fitness. The reported role of Rab35 in human tumorigenesis is associated with constitutive-active mutants, which activate the PI3K/AKT pathway (76). Although this study used a DN mutant of Rab35 (i.e., a constitutively inactive GTPase that also interferes with the normal function of WT Rab35), and therefore our observations should not be affected by cell-autonomous mechanisms, there is also the possibility that DN mutants of Rab35 may possess tumorigenic potential. More studies are needed to clarify this point.

Single-cell sequencing data can display significant variability from batch to batch. To account and compensate for this variability, we repeated the single-cell analysis twice and performed data transformation, which is a process that removes batch effects. Nonetheless, residual differences may have impacted the results and masked smaller differences.

A limitation of this work is that, in patients, the immune system is also exposed to senescence-inducing stimuli like chemoradiation therapy. To better understand these dynamics, it is important to study the interactions between SCs and host immune cells in stages, beginning with how immune cells respond to SCs under normal physiologic conditions, as we have done in this study. Additional work will explore how senescence-inducing therapies may impact these baseline immune reactions (77).

Although not fully capturing the complete pathogenesis of human HNSCC and its clinical course (de novo carcinogenesis, growth and clinical manifestation, surgical removal, treatment of minimal residual disease, and recurrence), our conclusions may still be directly relevant for understanding senescence during de novo carcinogenesis and recurrence, as senescence can be induced by either oncogene activation or chemotherapy. Therefore, our findings could also help develop novel cancer early-detection strategies based on the biology of cellular senescence (78).

M. Hoare reports grants from AstraZeneca and Pfizer and personal fees from Boston Scientific, Sirtex, Quotient Therapeutics, and Ensocell outside the submitted work. No disclosures were reported by the other authors.

T. Ziglari: Data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–original draft. N.L. Calistri: Data curation, software, formal analysis, visualization, methodology, writing–review and editing. J.M. Finan: Formal analysis, investigation, writing–review and editing. D.S. Derrick: Software, visualization. E.S. Nakayasu: Investigation, writing–review and editing. M.C. Burnet: Investigation. J.E. Kyle: Investigation, writing–review and editing. M. Hoare: Conceptualization, supervision, funding acquisition. L.M. Heiser: Supervision, writing–review and editing. F. Pucci: Conceptualization, resources, formal analysis, supervision, funding acquisition, investigation, visualization, methodology, writing–original draft, writing–review and editing.

We thank Drs. Sheila Stewart (Washington University, St. Louis, MO) for useful discussion and Robert Searles, Director of the Massively Parallel Sequencing Shared Resource at the OHSU, for help and advice with sequencing and genome alignment. We thank the OHSU Multiscale Microscopy Core for help with electron microscopy. This work was funded by the department of Otolaryngology startup grant (F. Ferdinando), the International Alliance for Cancer Early Detection, an alliance between Cancer Research UK (C52489/A29681 to F. Ferdinando, M. Hoare), Canary Center at Stanford University, the University of Cambridge, OHSU Knight Cancer Institute, University College London, and the University of Manchester, a NIH T32 Fellowship in “Integrated Training in Quantitative and Experimental Cancer Systems Biology” CA254888 (T. Ziglari), and Exploratory Research Seed Grant funding from the OHSU School of Medicine. This research is affiliated with the PMedIC joint research collaboration between the OHSU and Pacific Northwest National Laboratories (PNNL). The proteomics mass spectrometry analysis was performed at the Environmental Molecular Sciences Laboratory, U.S. Department of Energy National Scientific User Facility, located at the Pacific Northwest National Laboratory, operated under contract DE-AC05-76RL01830.