ApoE-TREM2 axis induces pathogenic senescent-like myeloid cells in prostate cancer

Tumour cells promote the expansion and intra-tumoural recruitment of Myeloid-derived suppressor cells (MDSCs), a subset of immature myeloid cells, that support tumour cell proliferation and confer treatment resistance. While immature myeloid cells have a very short lifespan, whether pathogenic MDSCs can persist in the tumour microenvironment remains unknown. Here, we report the identication of a subset of long-lasting MDSCs that upregulate markers of cellular senescence and the TREM2 receptor. Senescent-like MDSCs possess higher pro-inammatory capabilities compared to canonical MDSCs. Genetic and pharmacological elimination of senescent-like MDSCs decreases tumour progression in different mouse models of prostate cancer. Mechanistically, we nd that Apolipoprotein E (ApoE) secreted by prostate tumour cells binds TREM2 in senescent-like MDSCs, thereby regulating the survival of these cells. ApoE and TREM2 mRNA levels are upregulated in prostate cancers and correlate with poor patients’ prognosis. Taken together, these results reveal a novel mechanism by which the tumour microenvironment shapes the intra-tumoural immune response. Pathogenic senescent-like MDSCs persist longer in the tumour microenvironment and can be eliminated by histone deacetylase inhibitors enhancing the ecacy of standard therapy in prostate cancer. Plan-Apochromat; Olympus), using laser excitation at 405, 488, or 594 Images were processed using ImageJ Confocal images were obtained with the Leica TCS SP5 confocal microscope using × 10/1.25 oil.

MDSCs were not detected alive in the spleen, bone marrow and blood of tumour-bearing mice (Extended Data Figure 1a-c). The long-lasting presence of donor MDSCs in the tumour of Pten pc-/mice was validated by the infusion of MDSCs differentiated from bone marrow precursors (BMPs) collected from mCherry mice. In line with the previous experiment, mCherry + donor MDSCs were still present in the prostate tumours of Pten pc-/mice several weeks after their infusion (Extended Data Figures. 1d-e).
Intrigued by this nding, we next checked whether persistent MDSCs could upregulate markers of cellular senescence. Senescent cells have the capability to inde nitely survive in vivo in the tissue microenvironment and resist to different apoptotic stimuli by remaining metabolically active 15 . In line with this hypothesis, multi-parametric ow cytometry analysis showed that CD45.1 persistent MDSCs stained positive for C 12 FDG which detects increased SA-Beta-Galactosidase (b-Gal) activity at acidic pH 16 , a hallmark of cellular senescence 15 (Figure 1b-c). Additionally, persistent MDSCs stained positive for both p16 and p21, two key markers of cellular senescence 15  In order to validate this nding, C 12 FDG positive and negative PMN-MDSCs were sorted from Pten pc-/prostate tumours and tested for markers of cellular senescence by qPCR (Figure 1f). Tumour-in ltrating C 12 FDG + PMN-MDSCs expressed higher mRNA levels of p16ink4a (p16), p21Cip (p21) and PAI-1 ( Figure   1g) and stained positive for SA-b-Galactosidase when analysed ex vivo, thereby validating the C 12 FDG marker analysed by ow cytometry (Figure 1h).
Whole gene expression pro le analysis identi ed a signature enriched in C 12 FDG positive PMN-MDSCs (Senescent-like PMN-MDSCs signature). Differential expression analyses between C 12 FDG + and C 12 FDG -PMN-MDSCs showed that C 12 FDG + MDSCs upregulate genes involved in mitochondrial respiration and oxidative phosphorylation (Figure 1i) and possess a peculiar senescence-associated secretory phenotype (SASP), enriched in factors regulating in ammation, angiogenesis and granulocyte chemotaxis ( Figure   1k; Extended Data Figures 1h). Of note, senescent-like MDSCs were also detected in human prostate tumour biopsies by using the C 12 FDG staining ( Figure 1l). These data were further validated using available single-cell RNA sequencing (scRNA-seq) of human prostate tumours 17 . For this analysis, PMN-MDSCs were identi ed using a published gene signature 18 , whereas senescent like-PMN-MDSCs were spotted by using the Senescent-like PMN-MDSCs signature (Extended Data Figure 1i-k). In sum, these results demonstrate that tumour-in ltrated PMN-MDSCs can acquire features of senescent cells and that the C 12 FDG staining can be used to identify this cell population in prostate tumours.
To assess whether factors secreted by tumour cells impact on MDSCs vitality and induce the upregulation of markers of senescence, we co-cultured BM-derived MDSCs in presence or absence of conditioned medium (C.M.) derived from TRAMP-C1 cells, a murine prostate cancer cell line 19  We, therefore, assessed the levels of TREM2 expression on tumour-in ltrating PMN-MDSCs in Pten pc-/mice. Flow cytometry analyses con rmed that senescent-like PMN-MDSCs expressed high TREM-2 levels ( Figure 2e). Accordingly, CD45.1 + donor MDSCs infused in Pten pc-/mice and analysed 4 weeks after the infusion expressed both C 12 FDG and high level of TREM2 (Figure 2f). BM-MDSCs co-cultured in vitro with C.M. from TRAMP-C1 cells also co-expressed high levels of both C 12 FDG and TREM2 whereas canonical BM-MDSCs culture in absence of C.M. did not (Figure 2g). Finally, senescent-like BM-MDSCs upregulated the downstream signalling pathway of TREM2 as detected by increased phosphorylation of Syk and Erk1/2 ( Figure 2h). We next performed a protein pro le analysis using different fractions of C.M. obtained from TRAMP-C1 prostate tumour cells. C.M. was concentrated on 100 kDa centrifuge lters. Then, proteins were denatured by heat inactivation in order to obtain different fractions of C.M. containing macromolecules above or under 100kDa threshold. ApoE was detected in the total C.M. fraction before and after heat inactivation and in the fraction containing high molecular weight protein complexes (>100kDa) (Figure 2i). Co-culture experiments with different fractions showed that only the fractions containing ApoE were capable to prolong the survival of BM-MDSCs and trigger the senescence-like state (Figure 2j). To functionally validate the ApoE-TREM2 axis for the induction of senescence in MDSCs, we cultured BM-MDSCs in the presence of C.M. of TRAMP-C1 infected with shApoE (C.M. TRAMP-C1 shAPOE ) or shEV cells (C.M. TRAMP-C1 ShEV ). While the C.M. of TRAMP-C1 ShEV promoted the survival and increased C 12 FDG positivity in BM-MDSCs, the C.M. of TRAMP-C1 shAPOE was ineffective (Figure 2k). Mechanistically, we found that the C.M of ApoE de cient prostate tumour cells did not increase the levels of pSyk and of its downstream regulators in BM-MDSCs ( Figure 2l). These data were further validated in vitro and in vivo by using Trem2 wt (Trem2 wt ) and Trem2 mutant BM-MDSCs (BM-MDSCs Trem2mut ) that lack the capability to fully activate the downstream TREM2 signalling 24 (Figure 2m-n). Pten pc-/mice were reconstituted with these BMPs to generate the Pten pc-/-; Trem2 wt and Pten pc-/-; Trem2 mut mice ( Figure 2o). Pten pc-/-; Trem2 mut mice were less in ltrated by C 12 FDG + PMN-MDSCs and developed prostate tumours of smaller size than control mice thereby demonstrating that TREM2 is required for the survival of senescent-like MDSCs ( Figure 2 o-q).
Intriguingly, single cell analysis in human prostate tumour samples showed that the majority of senescent-like MDSCs were also positive for TREM2 and that the senescence-like signature strongly correlated with TREM2 expression (Figure 2r, s). This was also validated in different human prostate cancer datasets using bulk RNA-seq data (Extended data Figure 3a, b).
In human prostate tumour biopsies, we found a cluster of PMN-MDSCs that express high TREM2 levels as detected by both multi-parametric ow cytometry analysis and multiplex immune uorescence Since TREM2 + senescent-like PMN-MDSCs persist in the tumour microenvironment, we next attempted to identify compounds capable to eliminate these pathogenic cells. To identify small molecules that selectively kill senescent-like PMN-MDSCs, we screened compounds from the NCI drug repository team of NIH (DTP program) (Figure 4a; Extended Data Figure 5a). All the drugs included in these libraries are FDA approved and used in the clinic for different diseases. Within the compounds tested only ve compounds showed activity against these cells signi cantly affecting their viability (>50% at 10μM concentration).
Between these drugs, we found that Vorinostat and Romidepsin, two HDAC inhibitors 26 , were capable to eliminate senescent-like MDSCs at nanomolar concentrations as assessed in a dose-response assay ( Figure 4b). Pathway analysis showed that HDAC inhibition in senescent-like MDSCs impacted the TREM2 pathways decreasing the mRNA expression of TREM2, Dap12 and Syk already 6 hours after the administration of the compound (Figure 4 c-e). Accordingly, western blot analysis of senescent-like BM-MDSCs treated with Romidepsin showed increased acetylation of H3K9 and a downregulation of the TREM2 signalling as assessed by decreased total and phosphorylated-Syk and Erk levels ( Figure 4f).
Given that senescent-like MDSCs were enriched in CR prostate cancers (Figure 1e), we next assessed whether Romidepsin could enhance the e cacy of enzalutamide (ENZA) by promoting the elimination of senescent-like PMN-MDSCs in vivo. ENZA, a small molecule that binds the AR and suppresses the androgen receptor-signalling axis, is a standard of care for patients insensitive to rst-line ADT 27,28 . CR Pten pc-/mice were treated with Romidepsin in combination with ENZA for four weeks at the indicated dosages. Co-treatment of CR prostate tumours with ENZA and Romidepsin strongly decreased the percentage of senescent-like MDSCs and lead to a reduction in tumour cells proliferation as detected by decreased Ki67 staining, and percentage of glands affected by invasive prostate cancer ( Fig. 4g-j). Next, we assessed whether this combination of compounds could be further improved by the addition of a CXCR2 inhibitor, a compound under clinical evaluation in prostate cancer. We and others have previously shown that CXCR2 inhibitors partially decrease, without abolishing, the recruitment of MDSCs in prostate tumours blocking tumour progression 5,1,7 . Triple combination of ENZA, Romidepsin and anti-CXCR2 (aCXCR2) lead to the strongest inhibition of prostate cancer when compared to mice treated with the double combination (Figure 4g-j and Extended Data Figure 5c). These changes were associated with a robust reduction of both senescent-like and canonical PMN-MDSCs in ltration in these tumours ( Figure   4g). These data were also validated in RM1 allograft 29 , a model of aggressive prostate cancer driven by Ras and Myc overexpression (Figure 4k-o). Taken together, these data demonstrate that Romidepsin kills senescent-like PMN-MDSCs and when administered in combination with a CXCR2 inhibitor further impacts prostate cancer progression enhancing the e cacy of ENZA.
In conclusion, we have identi ed a novel mechanism driven by the tumour microenvironment to reinforce its pool of tumour-in ltrating PMN-MDSCs adding novel insights on the mechanism by which ApoE produced by prostate tumour cells promotes immunosuppression in the tumour microenvironment 30 . These data also add novel knowledge on the role played by MDSCs in cancer, describing a new unexpected feature for this immune subset. Previous evidence demonstrates that MDSCs can support tumourigenesis in a number of tumours through different mechanisms 3,7,10 . However, the discovery that MDSCs can persist into prostate cancers, express markers of senescence, live longer and remain metabolically active was unexplored. This discovery also opens a novel scenario for the therapeutic targeting of MDSCs proving through preclinical studies that Romidepsin can target this persisting protumourigenic immune population.

Animals
All mice were maintained under speci c pathogen-free conditions in the IRB facility and experiments were performed according to state guidelines and approved by the local ethics committee. Male C57BL/6 or NSG mice 6-8 weeks of age were purchased from Jackson Laboratories (Envigo) and acclimated for at least a week before use. C57BL/6 3MRp16 (3MRp16) mice 31 were kindly provided by Prof. Demaria (Groningen, Netherlands). C57BL/6 Trem2 mutant (Trem2 mut ) mice 32 were kindly provided by Prof. Mike Sasner (The Jackson Laboratory). Pten pc-/mice were generated and genotyped as previously described 17 . Female Pten loxP/loxP mice were crossed with male PB-Cre4 transgenic mice and genotyped for Cre using following primers: primer 1 (5'-AAAAGTTCCCCTGCTGATGATTTGT-3') and primer 2 (5'-TGTTTTTGACCAATTAAAGTAGGCTGTG-3') for PTEN loxP/loxP ; primer1 (5' TGATGGACATGTTCAGGGATC 3') and primer2 (5'CAGCCACCAGCTTGCATGA 3') for Probasin-CRE. Surgical castration was performed under anesthesia with iso urane. Mice were monitored postoperatively for recovery from anesthesia and checked daily for 2 days postoperatively. Surgical skin clips were removed on postoperative day 5. Mice undergoing treatment were administered control vehicle or therapeutic doses of the appropriate agents. Any mouse suffering distress or greater than 15% weight loss during treatment was euthanized by CO 2 asphyxiation. At the completion of study, mice were euthanized by CO 2 asphyxiation and tissue was collected for histology, mRNA analysis, protein analysis, and single cell suspensions for ow cytometry.
For allograft experiments, 2,5x10 6 TRAMP-C1 cells or 2x10 5 RM1 cells were injected subcutaneously into the ank of male respectively NSG or C57BL/6 mice. For TRAMP-C1 allograft when tumours were approximately 100 mm 3 , mice were randomized to the treatment groups. For RM1 allograft, castration was performed 3 days after injection and mice were randomized to the treatment groups. Tumour growth was monitored daily by measuring the tumour size with caliper. The tumour volume was estimated by calculating R1*R2*R3*4/3π, where R1 and R2 are the longitudinal and lateral radii, and R3 is the thickness of tumour protruding from the surface of normal skin. Animals were sacri ced when the tumour reached approximately 600 mm 3 . Treatments αCXCR2 (AZD5069; Astrazeneca) was administered with daily intraperitoneal injections at a nal concentration of 100mg/kg on a Monday through Friday schedule. Control animals received vehicle.
Enzalutamide (APExBio) was administered daily by oral gavage with a dose of 30mg/kg/day on a Monday through Friday schedule. Romidepsin (0,03mg/kg per mouse; MedChemExpress) was administered twice per week via intraperitoneal injection.

Bone marrow reconstitution
Bone marrow was ushed from the femurs of male C57BL/6 or 3MRp16 mice under sterile conditions with RPMI 1640 using a 21-gauge needle. Mononuclear cells were ltered, collected and checked for viability using trypan blue. Recipient Pten pc-/mice were lethally irradiated (900 rad) and transplanted i.v. two hours after with 1 × 10 7 viable bone marrow cells from either C57/BL6 or 3MRp16 mice.
TRAMP-C1 conditioned media collection TRAMP-C1 cells were cultured and expanded in the appropriate medium. When the cells were at about 40-50% con uence, the medium was replenished with fresh complete medium. After 3 days the conditioned medium was centrifuged at 500g for 5 min, 2,000 g for 10 min and 4,600 g for 20 min at 4 °C to remove dead cells and debris. The medium was then aliquot in single use tubes and store at -80°C or fractionated as followed. Fractions of conditioned media were collected using Centrifugal Filter Unit 100 KDa cutoff (Amicon® Ultra, merckmillipore). The fractions were reconstituted to initial volume with PBS. Heat inactivation was performed heating the samples at 65°C for 30 min.

Viral infection and establishment of stable cell lines
TRAMP-C1 cells were infected with lentivirus encoding mouse ApoE shRNA (in TRC1.5 Vector: pLKO.1puro). Infected cells were selected in 3 µg/ml puromycin-containing medium for 7 days, and the selected cells were collected and used for further experiments.

Differentiation of senescent-like BM-MDSCs in vitro
Murine MDSCs were differentiated in vitro as previously described 33 . Brie y, bone marrow precursors were ushed from the femurs of C57/BL6 or TREM2 mut mice with RPMI 1640 medium. The cell pellet was resuspended in RPMI 1640 containing 10% heat-inactivated FBS, and the cells were cultured in vitro in the presence of 40 ng/ml GM-CSF and 40 ng/ml IL-6. On day 4, the cells were washed and resuspended with TRAMP-C1 conditioned medium or TRAMP-C1 ShApoE. After 3 days, the cells were analysed by ow cytometry or used for in vitro experiments.
Measurement of SA-β-Gal activity by ow cytometry SA-β-Gal activity was measured according to the method described previously by Debacq-Chainiaux et al. 16 . Cells were pre-treated with 100 nM ba lomycin A1 (Ba lomycin A1 from Streptomyces griseus, Calbiochem cat.196000) in fresh cell culture medium at 37 °C for 1 h to induce lysosomal alkalinization. Thereafter, the uorogenic substrate 5-Dodecanoylamino uorescein Di-β-D-Galactopyranoside (C 12 FDG Thermo Fisher Scienti c, cat. D2893) was added to the cell culture medium to yield a nal concentration of 25 μM. After incubation, the cells were used to performed different staining. The expression of different molecules was then analysed with the multi-parameter ow cytometer (BD Fortessa ow cytometer, BD Biosciences). FITC channel was used to detect C 12 FDG.

Apoptosis Assays
Apoptosis assays were performed using Annexin V staining according to the manufacturer's instructions.
The cells were stained with PE-conjugated/eFluor™ 450-conjugated Annexin V (eBioscience) according to manufacturer's recommendations. Annexin V-FITC was diluted in the manufacturer's Hepes-buffer (containing 2.5 mM CaCl2), added to the cultures, and incubated for 15 min at room temperature. Further, the cells were then stained with other antibodies. The cells stained were subjected to FACS.

In vitro T cell suppression assay
In vitro suppression assays were carried out in RPMI/10% FCS in 96-well U-bottom plates (Corning, NY).

Arginase I Activity Assay
For in vitro experiment, Arginase I Activity Assay was performed using Arginase Activity Assay Kit (Sigma-Aldrich, Cat. No MAK112) according to the manufacturer's instructions.

Metabolic Phenotyping
Oxygen consumption rate (OCR) and extracellular acidi cation rate (ECAR) were measured using the Seahorse XFp bioanalyser. 1x10 5 cells per were spun onto poly-D-lysine coated Seahorse XFp Cell Culture Miniplates. Before the assay the plates were centrifuged 1min 300g w/o break and preincubated in Seahorse XF media (non-buffered DMEM + 10 μM L-glutamine + 10 μM sodium pyruvate + 25 mM glucose) at 37°C for a minimum of 30 min in the absence of CO 2 . OCR and ECAR were measured under basal conditions, and after the addition of the following drugs: 2 μM oligomycin, 2 μM urorcarbonyl cyanide phenylhydrazone (FCCP) and 2.5 μM rotenone + antimycin A as indicated. Measurements were taken using a Seahorse XFp Analyzer (Seahorse bioscience).

ROS detection
Oxidation-sensitive dye Dichlorodihydro uorescein diacetate (DCFDA, Molecular Probes/Invitrogen, Carlsbad, CA), was used to measure ROS production by MDSC. Cells were incubated at 37°C in prewarmed PBS in the presence of 2.5 µM DCFDA for 30 min. For induced activation, cells were simultaneously cultured, along with DCFDA, with 120ng/ml phorbol 12-myristate-13-acetate (PMA) (Sigma, St. Louis, MO) and 1 µg/ml ionomycin. Analysis was then conducted by ow cytometry as described above. mtROS detection Superoxide Detection Agent from Mitochondrial Superoxide Indicator (Red) Assay Kit (ab228567, abcam), was used to measure mtROS production by MDSC, according to the manufacturer's instructions. Cells were incubated at 37°C in prewarmed HBSS buffer in the presence of the suggested amount of the Superoxide Detection Agent working solution for 10 minutes in the dark. Analysis was then conducted by ow cytometry as described below. PE-TexasRed channel was used to detect Mitochondrial Superoxide.
All the antibodies were purchased from eBioscience or Biolegend or RnD or BD. Samples were acquired on a BD Fortessa ow cytometer (BD Biosciences). Data were analyzed using FlowJo software (TreeStar, Ashland, OR).
High-Dimensional Single-Cell Data Preprocessing and Analysis by UMAP and FlowSOM 27-Parameter Flow Cytometry Standard (FCS) 3.0 les were imported into FlowJo software version 9 or 10 and left untreated or biexponentially transformed (the same transformation for all les, performed in version 10) prior to UMAP analysis.
Following transformation, 15-Flow Cytometry Standard (FCS) 3.0 les were assigned with a computational barcode for their unique identi cation, concatenated and visualized with UMAP 34 in FlowJo. The following parameters were used: Euclidean; nearest neighbors: 10; minimum distance: 0.01; number of components: 2. All parameters except for CD45 and Aqua dead cell marker, were included in the analysis.
The clusterization was automatically de ne by the FlowSOM algorithm 35 . The default parameters were used to run the algorithm except for the number of meta clusters. The identity of the clusters was determined by the heatmap generated by FlowSOM.

Immune tumour microenvironment characterization of tumours from patients with prostate cancer
Tumours were disaggregated and digested in collagenase I and DNase for 30 min at 37 °C to obtain single-cell suspensions. Single-cell suspensions were stained with speci c monoclonal antibodies (primary antibodies directly conjugated) to assess the phenotype. The antibodies used were: CD45 (clone HI30), CD33 (clone WM-53), CD11b (clone M1/70), HLA-DR (clone L243), TREM2 (Clone 237920), CD66b (G10F5). To draw the gates, we used isotype controls or uorescence minus one. All antibodies were purchased from BD Bioscience, eBioscience or Biolegend. Samples were acquired on a BD Fortessa ow cytometer (BD Biosciences). Data were analysed using FlowJo software (TreeStar, Ashland, OR).
Images were acquired by a confocal microscope Leica SP5, with an oil-immersion objective (63×/1.4 NA Plan-Apochromat; Olympus), using laser excitation at 405, 488, or 594 nm. Images were processed using ImageJ software. Confocal images were obtained with the Leica TCS SP5 confocal microscope using ×

Tissue image acquisition and analysis
After staining, slides were scanned using the VS200 slide scanner (Olympus). Quanti cation of the immune cell densities was performed using Halo v3.0 software (Indica Labs). Tissue segmentation algorithm based on PanCK positivity was used to separate tumour from adjacent stroma. Phenotype determination was based on positivity for CD15 and TREM2. Immune cell densities are presented as number of cells per mm2. All tissue segmentation, cell segmentation and phenotype determination maps were reviewed by a pathologist (BG).

Human prostate samples
Patients were identi ed from a cohort of men with CRPC treated at the Royal Marsden NHS Foundation Trust. All patients had given written informed consent and were enrolled in institutional protocols approved by the Royal Marsden hospital (London, UK) ethics review committee (reference no. 04/Q0801/60). Eligible patients (n = 13) had matched histologically con rmed formalin-xed para nembedded (FFPE) diagnostic (archival) and metastatic castration-resistant prostate cancer (mCRPC) biopsies. Castration-sensitive prostate cancer (CSPC) samples were obtained from primary prostate tumours and included needle core biopsies (n = 9) and transurethral resections of the prostate (TURP) (n = 4). mCRPC biopsies were obtained from primary (n = 2) and metastatic sites including lymph nodes (n = 5), bone (n = 4), liver (n = 1) and soft tissue (n = 1). Patients had at diagnosis a median age of 64 years. Tissue samples were collected from prostatic needle biopsies, transurethral resections of the prostate or prostatectomies. All tissue blocks were re-sectioned and reviewed by a pathologist who con rmed adequacy of the material.

Bioinformatics analysis
Different bulk RNA-seq experiments were performed: senescent MDSCs compared to not-senescent MDSCs; senescent MDSCs treated with or without Romidepsin. For both, the overall quality of sequencing reads was evaluated using a variety of tools, namely FastQC (Andrews S., 2010), RSeQC 36 , AfterQC 37 and Qualimap 38 . Sequence alignments to the reference mouse genome (GRCm38) was performed using STAR 39 (v.2.5.2a). Gene-expression was quanti ed at gene level by using the comprehensive annotations made available by Gencode 40 . Speci cally, we used v20 release of the Gene Transfer File (GTF). Raw counts were further processed in the R Statistical environment and downstream differential expression analysis was performed using the DESeq2 pipeline. Genes characterized by zero expression were removed and genes with low mean normalized counts were ltered out by the Independent Filtering feature embedded in DESeq2 41 (alpha = 0.05). Two different types of gene-set analysis were performed: gene-set enrichment analysis was performed using Camera 42 ; while over-representation analysis was performed using enrichGO (clusterPro ler package) and egsea.ora function. Statistical enrichments were determined for gene-sets obtained from the KEGG collection and Gene Ontology, which are curated by the Molecular Signature DataBase (MSigDB) 43,44 . For RNA-seq data of senescent and canonical MDSCs, two different runs were performed and to remove the batch effect, removeBatchEffect of limma 45 package was used. The database used for secreted factors derived from protein atlas secretome (https://www.proteinatlas.org). For the transmembrane protein-encoding genes we refer to Martinez-Martin et al work 46 .
To investigate the ligand-receptor interactions, we used as database CellTalkDB 47 . 9 couples were selected: for the ligands we ltered based on secreted factors up-regulated in PTEN pc-/compared to WT mice (E-MTAB-9624) 48 ; for the receptors we considered transmembrane proteins-encoding genes found up-regulated in senescent MDSCs. Only couples with both ligand and receptor signi cantly differentially expressed were considered. All graphs were produced using ggplot2 package 49 .
For single cell-RNA sequencing data, we used published data from "http://www.pradcellatlas.com" 17 . The original strategy was kept identifying the different cell populations. By expression of PMN-MDSCs signature 18 , calculated using AddModuleScore function, we de ned this population inside the "Monocytic"cluster.
Senescent-like signature of C12FDG+ MDSCs signature was composed by genes up-regulated in C12FDG+ MDSCs from our RNA-seq data. Correlation analysis were performed between Senescent-like signature and TREM2 expression. Pearson's correlation coe cient is the test statistics used for the analysis.
The human RNA-Seq datasets used were obtained from the TCGA database, which includes 481 primary prostate cancer patients and 51 normal prostate patients, and from SU2C effort 50 , which includes metastatic castration resistant patients (SU2C/PCF Dream Team, PNAS 2019; cBioPortal). Correlation analysis were performed between expressions of different genes of interest. Pearson's correlation coe cient is the test statistics used for the analysis. Plots of correlation were designed using ggscatter function for single correlation and corrplot function for multiple comparisons. Expression of APOE and TREM2 in cancer patients was analysed based on Gleason score and based on prostate cancer subtypes (normal tissue, primary tumour, metastatic tumour). Senescence PMN-MDSCs signature was composed by Senescent-like signature of C12FDG+ MDSCs signature and PMN-MDSCs signature and was calculated using gsva function 51 (method = ssgsea). Senescence PMN-MDSCs signature was correlated with expression of TREM2 in primary tumours (TCGA) and metastatic tumours (SU2C effort). Statistical signi cance between two groups was determined using wilcox test, while comparison among three or more groups was evaluated with two-way ANOVA followed by Tukey's post-hoc test.
Analysis of survival was performed using survival R package and, speci cally, using Kaplan-estimator and Cox-regression model. Log-rank test was used to calculate statistical signi cance in survival curves.
The groups compared in survival analysis (both disease free and overall survival) were classi ed into quartiles based on mRNA expression level of APOE: high level corresponds to forth quantile, low level to rst quantile while medium levels include second and third quantiles. The datasets used for survival analysis were TCGA database, SU2C effort 50 and data already reported 52 .
Statistical analysis and reproducibility. Data analyses used GraphPad Prism version 9. The data are presented as mean ± standard error of the mean, individual values as scatter plot with column bar graphs and were analyzed using Student's t-tests (unpaired) by a two-sided and, when indicated, followed by Wilcoxon posttest. One-way ANOVA was used to compare three or more groups in time point analyses.
Differences were considered signi cant when P < 0.05 and are indicated as NS, not signi cant, *P < 0.05, **P < 0.01, ***P < 0.001. Non-parametric tests were applied when variables were not normally distributed using the SPSS statistical software. N values represent biological replicates. For animal studies, sample size was de ned on the basis of past experience with the models 7 , to detect differences of 20% or greater between the groups (10% signi cance level and 80% power). For ethical reasons the minimum number of animals necessary to achieve the scienti c objectives was used. Animals were allocated randomly to each treatment group. Different treatment groups were processed identically and animals in different treatment groups were exposed to the same environment. For bioinformatic analyses, the data were considered statistically signi cant with FDR<0.05.   with BM precursors from C57 or TREM2mut mice. p, Tumour MDSCs frequencies determined by ow cytometry. q, Tumour volume of the anterior prostate lobes is reported. r, Representative dot plot of TREM2 expression within MDSC populations in single cells (sc) RNA-seq data from human biopsies. s, Correlation plot between TREM2 expression and senescence-like signature within MDSC populations in sc RNA-seq data from human biopsies e, p, q, Each dot represents an individual mouse. h, i, j, k, m, Aggregated data from at least three independent experiments are reported as mean ± s.e.m. Each dot represents an individual sample. e, h, i, p, q, Statistical analyses: Unpaired t test. j, k, m Statistical analyses: Two-way ANOVA followed by Tukey's multiple comparisons test.  Romidepsin increases the e cacy of prostate cancer standard of care and anti-CXCR2 therapy in vivo. a, Graphical representation of the senolytic screening cascade using senescent-like BM-MDSCs. The assay is based on the assessment of cellular viability using compounds administered at single dose. b, Doseresponse of the indicated drugs. Viability was measured after 3 days. The IC50 for each drug and condition is shown in the respective boxes. Data from at least two experiments are shown. c, Heatmap of