Targeting YAP-Dependent MDSC Infiltration Impairs Tumor Progression

The tumor microenvironment (TME) is comprised of a complex mixture of tumor-associated fibroblasts, infiltrating immune cells, endothelial cells, extracellular matrix proteins, and signaling molecules, such as cytokines (1–3). Homotypic and heterotypic interactions between these cellular constituents play essential roles in cancer development and response to therapeutics (3, 4). Among the infiltrating immune cells, myeloid-derived suppressor cells (MDSC) represent a phenotypically heterogeneous population of immature myeloid cells that play a tumor-promoting role by maintaining a state of immunologic anergy and tolerance (5). In particular, activated MDSCs provide a source of secreted chemokines, cytokines, and enzymes, which suppress local T-cell activation and viability (5). In addition, MDSCs can suppress T-cell activity through deprivation of nutrients, such as l-arginine and l-cysteine, and interference with T-cell receptor functions via reactive oxygen species (ROS) and reactive nitrogen species.

Prostate cancer is the most common noncutaneous malignancy in men in the United States. Similar to many other solid tumor types, prostate cancer is characterized by a rich tumor–stroma interaction network that forms the TME (1–3). In prostate cancer, various signaling pathways have been implicated in the cross-talk between tumor and stroma, such as androgen receptor signaling, FGF, SRC, TGFβ, IGF, integrin, and Hedgehog pathways (1). Interestingly, MDSC abundance in the blood correlates with circulating PSA levels in patients with prostate cancer (6–8). MDSCs have been identified recently as a TME constituent in an indolent prostate cancer mouse model with conditional Pten deletion (9) and demonstrated to antagonize senescence during early tumorigenesis (10). However, the molecular mechanisms underlying the recruitment of MDSCs are not well understood, and the extent to which MDSCs facilitate prostate cancer progression has not been determined.

Previously, we have shown that deletion of Pten in the mouse prostate causes upregulation of SMAD4, which constrains cell proliferation and invasion, and, accordingly, dual deletion of Pten and Smad4 results in rapid prostate cancer progression, including metastasis (11). Comparative transcriptomic and cell profile analyses of PTEN- versus PTEN/SMAD4–deficient prostate cancer revealed a prominent immune signature and resident MDSCs as a major TME population in PTEN/SMAD4–deficient tumors. Biologic, molecular, and pharmacologic analyses established that a YAP1-mediated CXCL5–CXCR2 signaling axis recruits MDSCs into the TME and that MDSCs play critical roles in facilitating tumor progression. Our comprehensive analyses using a prostate cancer model coupled with clinical validation using patient samples support the view that targeting either MDSC recruitment or infiltrated MDSCs may represent a valid therapeutic opportunity in treating advanced prostate cancer.

We previously reported that conditional deletion of Smad4 bypassed the senescence barrier instigated by Pten loss in the prostate epithelia, resulting in a highly proliferative and invasive prostate adenocarcinoma characterized by an exuberant stromal reaction and frequent metastasis to distant organs (11). Correspondingly, ingenuity pathway analysis (IPA) revealed prominent representation of cell movement, cell proliferation, and antigen presentation as the top three categories represented in the Pten^pc−/−Smad4^pc−/− tumors (11). Further analysis revealed a prominent immune signature, including Granulocytes Adhesion and Diapedesis, Leukocytes Extravasation Signaling, and Agrandulocytes Adhesion and Diapedesis as three of the top four most activated pathways in Pten^pc−/−Smad4^pc−/− tumors compared with those present in Pten^pc−/− tumors (Fig. 1A; P value < 2.03E−7). Correspondingly, IHC staining highlighted conspicuous infiltration of CD45⁺ leukocytes in Pten^pc−/−Smad4^pc−/− tumors (Fig. 1B). To comprehensively audit the spectrum of infiltrating immune cells in tumors, we performed mass cytometry (CyTOF) immunophenotyping (12) to catalog tumor cell–type constituents from well-established tumors in 16-week-old Pten^pc−/− and Pten^pc−/−Smad4^pc−/− mice. Employing a 9-marker antibody panel (Supplementary Table S1), CyTOF confirmed a significant increase in CD45⁺-infiltrating leukocytes in Pten^pc−/−Smad4^pc−/− as compared with Pten^pc−/− tumors (Fig. 1C). Within the CD45⁺-infiltrating cells, CD11b⁺ myeloid cells represented a significantly increased immune population in Pten^pc−/−Smad4^pc−/− as compared with Pten^pc−/− tumors (Fig. 1D).

To obtain a dynamic view of peripheral and infiltrating immune cells as a function of tumor progression in the Pten^pc−/−Smad4^pc−/− model, which initiates tumor development at 6 to 8 weeks and progresses to early invasive carcinoma by 14 weeks of age, serial CyTOF analyses using an expanded antibody panel of 17 surface markers (Supplementary Table S1) were performed on single cells from primary tumors, peripheral blood, spleen, and draining lymph nodes at 5, 8, and 14 weeks of age. The detailed immunophenotyping profiles enabled construction of the spanning-tree progression analysis of density-normalized events (SPADE)–derived tree (12). SPADE is a computational approach to facilitate the identification and analysis of heterogeneous cell types. SPADE of the Pten^pc−/−Smad4^pc−/− model displays the complexity of the TME, which is composed of epithelial tumor cells (EpCAM⁺CD45⁻), nonimmune TME cells (EpCAM⁻CD45⁻), and infiltrating immune cells (EpCAM⁻CD45⁺) that can be further grouped into various immune cell subpopulations (Fig. 2A and Supplementary Fig. S1A). Among the infiltrating immune cells, there was a striking age-dependent increase of CD11b⁺Gr1⁺ cells in tumors (Fig. 2B) and peripheral blood from Pten^pc−/−Smad4^pc−/− mice (Fig. 2C); this trend was much less pronounced in the spleen or draining lymph nodes (Supplementary Fig. S1B; for gating strategy, see Supplementary Fig. S1C).

To evaluate the potential immunosuppressive activity of intratumoral CD11b⁺Gr1⁺ cells from Pten^pc−/−Smad4^pc−/− tumors, we examined T-cell proliferation using a standard cell coculture system. These CD11b⁺Gr1⁺ cells strongly suppressed CD3 and CD28 antibody-induced T-cell proliferation and activation (Fig. 3A and B; see Supplementary Fig. S2 for cell isolation strategy), establishing that CD11b⁺Gr1⁺ cells are indeed functional MDSCs.

MDSCs can be further classified as a Ly6G⁺Ly6C^lo subset with polymorphonuclear phenotype (PMN-MDSC) and a Ly6G⁻Ly6C^hi subset with monocytic phenotype (M-MDSC; ref. 13). PMN-MDSCs represented the major MDSC population in Pten^pc−/−Smad4^pc−/− tumors (Fig. 3C and D), consistent with previously observed preferential expansion of PMN-MDSCs in tumor-bearing mice of various syngeneic models (5, 9, 13). The abundance of PMN-MDSCs was further confirmed by IHC for Ly6G, as shown by quantification of both intraepithelial and stromal Ly6G⁺ cells in tumors from Pten^pc−/−Smad4^pc−/− mice and Pten^pc−/− mice (Fig. 3E and F). It has been shown previously that ROS production by PMN-MDSCs is one of the mechanisms driving immune suppression (5, 14–16). Correspondingly, IPA revealed that pathways involved in ROS and nitric oxide (NO) production are among the top pathways activated in Pten^pc−/−Smad4^pc−/− tumors (Fig. 1A, arrow). Consistent with the increased infiltration of PMN-MDSCs in the Pten^pc−/−Smad4^pc−/− tumors, the expression of several subunits of NADPH oxidase (Nox2, p40^phox, and p47^phox), which are responsible for ROS production in PMN-MDSCs (5), was significantly upregulated in Pten^pc−/−Smad4^pc−/− tumors relative to Pten^pc−/− tumors (Fig. 3G). Moreover, Arg1, but not Nos2, was highly upregulated in the Pten^pc−/−Smad4^pc−/− tumors (Fig. 3G). Together, MDSCs in autochthonous Pten^pc−/−Smad4^pc−/− tumors display strong T-cell–suppressive activity and are predominantly the PMN-MDSC subtype.

Enrichment of MDSCs in advanced Pten^pc−/−Smad4^pc−/− tumors prompted us to explore the possible role of MDSCs in tumor progression. Using a well-characterized anti-Gr1 neutralizing monoclonal antibody (clone RB6-8C5; ref. 17), MDSCs were depleted in Pten^pc−/−Smad4^pc−/− mice at 14 weeks of age, a point coincident with progression to the early invasive carcinoma stage (see Supplementary Fig. S3A for treatment scheme). The potent MDSC depletion activity of anti-Gr1 monoclonal antibody was evidenced by significantly decreased PMN-MDSCs and M-MDSCs in peripheral blood as early as day 2 after treatment (Supplementary Fig. S3B). In addition, a systemic reduction of MDSCs in spleen, bone marrow, and prostate tumors was documented following a 30-day treatment regimen of anti-Gr1 monoclonal antibody (Fig. 4A and Supplementary Fig. S3C). This MDSC depletion was accompanied by an increase of CD8⁺ T cells (so-called killer T cells; Fig. 4A), consistent with elimination of the T-cell suppression activity of MDSCs. Importantly, in line with the CD8⁺ T-cell expansion, we observed that the Gr1-treated prostate displayed remarkable weight reduction in ventral and dorsolateral prostates (VP and DLP; Fig. 4B). The lack of difference in the weight of the anterior prostate (AP) is likely due to the fact that the AP tends to develop cysts with fluid accumulated inside the gland (18, 19), which also prevents the accurate measure of the prostate weight (Supplementary Fig. S3D). Histopathologic analysis revealed adenocarcinoma was the predominant pathology in mice treated with the control IgG, whereas mouse prostatic intraepithelial neoplasia (mPIN) was the predominant morphologic presentation in prostates from mice treated with anti-Gr1 monoclonal antibody (Fig. 4C and Supplementary Table S2). In addition, by immunohistochemical staining for CD45, Ki67, vimentin, smooth muscle actin (SMA), and Trichrome staining, we observed that tumor remnants in mice treated with anti-Gr1 monoclonal antibody displayed markedly reduced levels of cellular proliferation, stromal reaction, and inflammation as compared with those tumors treated with control IgG antibody (Supplementary Fig. S4A).

In another therapeutic trial, we also utilized the recently developed MDSC-specific peptide–Fc fusion protein (i.e., peptibodies) that has been shown to effectively eliminate MDSCs in vivo through targeting the S100A9 surface protein (20). Employing a hydrodynamic injection approach for nucleic acid delivery (21), intravenous injection of either Pep-H6 peptibody expression vector or irrelevant control peptibody vector was initiated at 14 weeks every 4 days in Pten^pc−/−Smad4^pc−/− mice. Strikingly, a single injection of the Pep-H6 peptibody significantly reduced the MDSCs in the peripheral blood, whereas such effect was not observed using the irrelevant control peptibody (Supplementary Fig. S4B). Pep-H6 peptibody treatment for 1 month led to a dramatic decrease in cancer cell content in the prostate tumors (Fig. 4D) and provided significant survival benefit for tumor-bearing mice (Fig. 4E). Together, our data strongly support the view that MDSC depletion blocks prostate tumor progression in the Pten^pc−/−Smad4^pc−/− model.

To elucidate the cellular origins and signaling molecules governing MDSC recruitment to prostate tumors, we incorporated the mTmG dual fluorescence reporter allele into the Pten^pc−/−Smad4^pc−/− model where signaling events between tumor cells and stroma can be precisely delineated. The mTmG allele (22) allows Cre-dependent GFP expression in prostate epithelial cells and ubiquitous tdTomato expression in all other non–Cre-expressing cells (Fig. 5A). Transcriptomic and IPA analyses of FACS-sorted GFP⁺ tumor cells and Tomato⁺ stromal cells showed distinct expression patterns by hierarchical clustering (Fig. 5A) with tumor cells enriched for pathways involved in cell adhesion molecules and tight junctions (consistent with their epithelial nature) and stromal cells displaying activation of more diverse pathways involved in chronic inflammation, such as cytokine/cytokine receptor interaction, chemokine, JAK–STAT, T-cell repector, and B-cell receptor signaling (P < 0.01, data not shown). This result is consistent with the immunopathologic and histopathologic analyses showing a massive infiltration of immune cells in the Pten^pc−/−Smad4^pc−/− tumors.

Employing this new model, we sought to identify genes that were upregulated in Pten^pc−/−Smad4^pc−/− cancer cells relative to Pten^pc−/− cancer cells that might illuminate mechanisms involved in the recruitment of MDSCs by classifying the upregulated genes into either stroma- or tumor-enriched genes. To this end, our previously generated list of 242 genes with greater than 2-fold increased expression in Pten^pc−/−Smad4^pc−/− relative to Pten^pc−/− tumors (11) was intersected with 486 genes preferentially expressed in Pten^pc−/−Smad4^pc−/− GFP⁺ cancer cells relative to Tomato⁺ stroma cells (fold change ≥4; Supplementary Tables S3 and S4), yielding 28 genes that are markedly enriched in Pten^pc−/−Smad4^pc−/− cancer cells (Supplementary Table S5). Among these 28 genes, Cxcl5, which encodes a key cytokine involved in MDSC recruitment (23, 24), is the most significantly upregulated cancer cell–specific cytokine in Pten^pc−/−Smad4^pc−/− tumors as compared with Pten^pc−/− tumors (Fig. 5B and Supplementary Fig. S5A). Notably, CXCR2, the cognate receptor for CXCL5, is also upregulated in Pten^pc−/−Smad4^pc−/− tumors as compared with Pten^pc−/− tumors and is significantly enriched in Pten^pc−/−Smad4^pc−/− Tomato⁺ stroma cells (Fig. 5B). The upregulation of CXCL5 expression in Pten^pc−/−Smad4^pc−/− prostate tumors was further confirmed by IHC (Fig. 5C). In addition, we performed FACS analysis of CD11b⁺Gr1⁺ cells and CD11b⁻Gr1⁻ cells from bone marrow, spleen, peripheral blood, and tumors for CXCR2 expression. As shown in Supplementary Fig. S5B, CD11b⁻Gr1⁻ cells (largely lymphocytes) are devoid of CXCR2 expression, whereas a large fraction of CD11b⁺Gr1⁺ cells express CXCR2. When CXCR2 expression was further separated into CXCR2^hi and CXCR2⁺, we observed an enrichment of the CXCR2^hi subpopulation in the CD11b⁺Gr1⁺ cells in prostate tumors compared with CD11b⁺Gr1⁺ cells from bone marrow, spleen, or blood (Supplementary Fig. S5B). This is consistent with the model of active recruitment of MDSCs by tumors through CXCR2-mediated chemoattraction.

To validate the CXCL5–CXCR2 axis in the recruitment of MDSCs to the TME of Pten^pc−/−Smad4^pc−/− tumors, we assessed the impact of pharmacologic inhibition of CXCL5 and CXCR2 in MDSCs using a transwell migration assay (23). First, anti–CXCL5-neutralizing antibody pretreatment of conditioned medium (CM) derived from Pten^pc−/−Smad4^pc−/− prostate cancer cell line resulted in decreased migration of MDSCs (Fig. 5D). Second, CXCR2 inhibitor SB255002 or anti-CXCR2 neutralizing antibody pretreatment also impeded migration of MDSCs (Fig. 5D). Third, in vivo blockade of the CXCL5–CXCR2 axis using SB255002 in 14-week-old Pten^pc−/−Smad4^pc−/− mice over a 14-day daily dosing schedule revealed a dramatic reduction in infiltration of MDSCs in the prostate tumors (Supplementary Fig. S5C and S5D). Notably, similar to mice treated with anti-Gr1 neutralizing antibody, these SB225002-treated Pten^pc−/−Smad4^pc−/− mice also showed significant reduction in tumor burden (VP and DLP) as compared with the vehicle-treated controls (Fig. 5E and Supplementary Fig. S5E). Strikingly, all SB225002-treated tumors presented with prostatic intraepithelial neoplasia (PIN) pathology, whereas the control group uniformly possessed advanced adenocarcinoma (Fig. 5F and Supplementary Table S2). Furthermore, SB225002 treatment significantly prolonged the overall survival of the Pten^pc−/−Smad4^pc−/− mice as compared with the vehicle control (Fig. 5G). Thus, we conclude that the CXCL5–CXCR2 axis plays a prominent role in the recruitment of MDSCs to the Pten^pc−/−Smad4^pc−/− prostate TME and that inhibition of this axis profoundly impairs tumor progression.

Having identified cancer cell–derived CXCL5 as a key signaling molecule governing recruitment of MDSCs into the TME, we sought to define the molecular mechanisms underlying the strong induction of CXCL5 expression in the Pten^pc−/−Smad4^pc−/− cancer cells. As CXCL5 expression is not significantly upregulated in the Pten^pc−/− tumors (Fig. 5C), we performed unbiased Gene Set Enrichment Analysis (GSEA) to identify pathways that were activated in the Pten^pc−/−Smad4^pc−/− tumors as compared with Pten^pc−/− tumors, aiming to identify potential regulators for Cxcl5 in Pten^pc−/−Smad4^pc−/− tumors. The YAP oncogenic signature emerged as the second most hyperactivated pathway (Fig. 6A and Supplementary Fig. S6A). Although it is known that the Hippo–YAP pathway plays an important role in development and cancer in organs such as the liver, skin, intestine, and pancreas (25–27), the role for the Hippo–YAP pathway in prostate cancer biology is emerging. Specifically, Hippo pathway components LATS1/2 have been implicated in anoikis and metastasis in prostate cancer (28), and ERG-induced YAP1 activation can promote age-related prostate tumor development (29). However, beyond the cancer cell–specific functions, the Hippo–YAP1 pathway has not been linked to signaling communication between cancer cells and immune cells in the TME. Consistent with the in silico analysis, IHC analysis documented a dramatic increase in the nuclear localization of YAP1 in Pten^pc−/−Smad4^pc−/− cancer cells as compared with Pten^pc−/− cancer cells (Fig. 6B). As YAP1, a transcriptional coactivator and the downstream mediator of Hippo signaling, is regulated posttranscriptionally by either kinase-mediated degradation or cytoplasmic sequestration (25), our findings of increased nuclear localization of YAP1 are consistent with the hypothesis that the Hippo–YAP pathway is activated in the Pten^pc−/−Smad4^pc−/− tumors. In addition, unbiased oPOSSUM analysis (30) indicated that TEAD1, a member of the TEAD transcription factor family that is required for YAP1 function, ranked second among the top 10 transcription factors with overrepresented binding sites in the 70 cancer-specific genes that were upregulated in the Pten^pc−/−Smad4^pc−/− tumors as compared with the Pten^pc−/− tumors (≥1.5 fold, Z-Score = 13.362; Supplementary Fig. S6B and S6C), an observation reinforcing the relevance of the Hippo–YAP pathway. Furthermore, we identified six YAP/TEAD binding motifs in the promoter of Cxcl5 gene (Supplementary Fig. S6D), suggesting YAP1 could be directly involved in the recruitment of MDSCs through regulating Cxcl5 expression. This hypothesis was supported by chromatin immunoprecipitation (ChIP) assay showing that YAP1 binds to Cxcl5 promoter (Fig. 6C) and that shRNA-mediated knockdown of Yap1 in Pten^pc−/−Smad4^pc−/− cancer cells drastically reduced the expression of Cxcl5 mRNA (Fig. 6D). In addition, overexpression of a constitutively active YAP1^S127A mutant dramatically increased Cxcl5 mRNA expression in the Pten^pc−/−Smad4^pc−/− cell line (Fig. 6E), whereas overexpression of a TEAD binding defective YAP1 mutant S127A/S94A compromised its ability to activate Cxcl5 transcription (Fig. 6F). To examine the effect of YAP1-dependent cytokine signaling in the regulation of MDSCs recruitment, we first prepared CM from the Pten^pc−/−Smad4^pc−/− cell line either infected with shRNA against Yap1 or pretreated with verteporfin (25), a small-molecular inhibitor that disrupts YAP1–TEAD interaction. We then tested the effect of various CM on the migration of MDSCs in vitro. As shown in Fig. 6G and H, we observed significantly decreased MDSC migration in vitro when CM was from cells with either YAP1 knockdown or verteporfin treatment.

Finally, to test if targeting YAP1 in vivo can impair the infiltration of MDSCs and inhibit tumor growth, we used our recently isolated syngeneic murine prostate cancer line PPS, which is derived from the backcrossed Pten^pc−/−Smad4^pc−/−Trp53^pc−/− model (31) and can form subcutaneous or orthotopic tumors robustly in C57BL/6 hosts. Doxycycline-dependent shRNA knockdown of Yap1 (two independent shRNA designs #1 and #3) was established in PPS (Fig. 6I) and injected subcutaneously in C57BL/6 mice. YAP1 knockdown induced by switching to doxycycline-containing drinking water resulted in a reduction of MDSCs in the intratumoral CD45⁺ population (Fig. 6J and K) and impaired tumor progression (Fig. 6L). Although the observation supports the hypothesis that targeting YAP1-dependent MDSC infiltration impairs tumor growth, we acknowledge that the tumor growth impediment by YAP1 silencing is likely due to a combined effect of both cell-nonautonomous and cell-autonomous mechanisms. Together, these findings reveal a novel function for YAP1 in the recruitment of MDSCs through direct upregulation of Cxcl5 transcription in prostate tumor cells.

To determine whether YAP1 is overexpressed and activated in human prostate cancer, we performed IHC staining of a human prostate cancer tissue microarray (TMA) for YAP1. Interestingly, YAP1 is expressed in basal cells, but not in the luminal cells of the normal human prostate (Fig. 7A). In addition, we observed that YAP1 is overexpressed in a subset of human prostate cancers (Fig. 7A and B and Supplementary Table S6), consistent with a recent report (29). Given the lack of validated antibodies for human MDSCs for TMA analysis, we generated a list of 39 MDSC-related genes curated from literature analysis (Supplementary Table S7) to generate evidence of a link between YAP1 activation and MDSC prominence in human prostate. Using the prostate RNA-sequencing data from The Cancer Genome Atlas (TCGA), unsupervised clustering with the 39-gene MDSC signature categorized 498 TCGA primary prostate tumors into three subtypes: MDSC-high (n = 139), MDSC-medium (n = 158), and MDSC-low (n = 201; Fig. 7C), suggesting that a subset of human prostate tumors may have prominent infiltration of MDSCs. In addition, using GSEA, we found that several YAP1 signature genes are significantly overexpressed in MDSC-high samples as compared with MDSC-low samples (Fig. 7D; P value < 0.005), reinforcing the link between MDSC-high prostate tumors and YAP1 transcriptional activities. Furthermore, CXCL6, the human homolog of murine Cxcl5, is expressed at higher levels in the MDSC-high samples as compared with MDSC-low samples (Fig. 7E; P = 9.40E−29). Similar analysis was performed in a published dataset focused on tumor immunobiological differences in prostate cancer between African-American and European-American men (32). The 39-gene MDSC signature can cluster the 69 primary prostate tumors into MDSC-high (n = 40) and MDSC-low groups (n = 29), and YAP1 signatures were prominent in the MDSC-high groups (Supplementary Fig. S7A and S7B). Together, these human prostate tumor findings, which parallel our murine observations, suggest that activated YAP1 is integral to MDSC infiltration in both mouse and human prostate cancer, thus enhancing the translational value of the study.

Although a large number of studies have demonstrated a direct relationship between MDSC frequency and tumor burden (5), our understanding of the role of MDSCs in tumor progression, particularly prostate cancer, remains largely speculative. Here, using a highly invasive PTEN/SMAD4-deficient prostate cancer model, we established the signaling circuits involved in the recruitment of MDSCs to the TME and demonstrated a critical role of these cells in facilitating tumor progression.

Homozygous deletion of Pten in murine prostate elicited a strong senescence response that restricts tumor progression (33); thus, Pten-deficient prostate tumors are largely indolent and progress slowly to invasive prostate adenocarcinoma without metastasis to distant organs (11, 33). Recently, it was shown that infiltrating Gr1⁺ myeloid cells suppress Pten loss–induced cellular senescence through a paracrine signaling mediated by myeloid-secreted IL1RA (10). We have previously reported that deletion of Smad4 leads to bypass of Pten loss–induced senescence in prostate cancer progression, resulting in aggressive cancer cell proliferation and invasion/metastasis (11). Using the state-of-the-art CyTOF technology, we revealed that progression in the Pten^pc−/−Smad4^pc−/− model is associated with abundant immune cell infiltration characterized by prominent representation of CD11b⁺Gr1⁺ MDSCs, which display potent immunosuppressive activities as shown by their strong antagonistic effect on T-cell proliferation (Fig. 3A and B).

The basis for the increased frequency of MDSCs in the TME and, specifically in the Pten^pc−/−Smad4^pc−/− model, was not known and presumably could derive from either active chemoattraction or passive nonspecific responses to tissue stress associated with expanding tumor burden. Taking an unbiased approach to identify pathways that may recruit MDSCs, we deconvoluted cancer versus stromal cell transcriptomes by exploiting a Cre-dependent dual fluorescence lineage tracing system in the Pten^pc−/−Smad4^pc−/− model. This approach identified unique immune regulatory molecules that are activated prominently in Pten^pc−/−Smad4^pc−/− cancer cells, most prominently CXCL5. We established that the CXCL5 chemokine plays a key role in the efficient recruitment of MDSCs which enables tumor progression, as blocking CXCL5–CXCR2 signaling with a CXCR2 inhibitor led to reduced MDSC infiltration with associated antitumor effects. It should be noted that the human homolog for murine CXCL5 is CXCL6, and CXCL6 has been shown to be upregulated in prostate cancer as compared with normal prostate and significantly associated with high Gleason scores 8 to 9 (34). Interestingly, it was shown that CXCL5 promotes recruitment of MDSCs to primary melanoma, resulting in epithelial–mesenchymal transition (EMT) and cancer cell dissemination (35). Thus, the possible role of CXCL5/CXCL6 in prostate cancer metastasis merits further study.

Our finding that CXCL5 is the main chemoattractant in the Pten^pc−/−Smad4^pc−/− model also provided a framework to determine the cancer cell signaling pathways driving Cxcl5 upregulation. By integration of bioinformatic analysis and experimental validation, we identified that YAP1 is activated in Pten^pc−/−Smad4^pc−/− prostate tumors and that YAP1 directly regulates Cxcl5 transcription and MDSC recruitment. In addition, we showed that YAP1 is overexpressed in a subset of human prostate cancers, which is consistent with a recent publication showing a correlation of ERG and YAP1 coexpressed in a subset of human prostate cancers (29). Importantly, a 39-gene MDSC signature clusters the prostate TCGA samples into three subtypes. By comparing the samples with high and low abundance of MDSC-related gene expression, YAP1 signatures and higher expression of CXCL6 are identified in the MDSC-high samples, which is consistent with our findings in the mouse model. Furthermore, the 39-gene MDSC signature can cluster primary prostate tumor samples from a published dataset (32) into two subtypes using MDSC-high and MDSC-low, with YAP1 signatures identified in the MDSC-high subtype. The Hippo–YAP signaling pathway is widely deregulated in human solid neoplasia and often associated with enhanced cancer cell proliferation and cancer stem cell phenotypes (25), and is implicated in the regulation of anoikis and metastasis in prostate cancer (28) and the development of age-related prostate cancers driven by ERG overexpression (29), yet how the Hippo–YAP pathway regulates the TME in prostate cancer has hitherto not yet been elucidated. Our finding of a novel non–cell autonomous function for Hippo–YAP signaling in MDSC recruitment in TME complements well the recently elucidated roles of YAP1 in promoting cell-autonomous functionality of cancer cells, including enhanced tumor survival, EMT, and bypass mechanism for oncogene addiction (26, 27).

Pharmacologic depletion of MDSCs using Gr1 antibody, Pep-H6 peptibody, or CXCR2 inhibitor arrested prostate progression at the high-grade PIN stage whereas controls exhibited full-fledged adenocarcinoma in Pten^pc−/−Smad4^pc−/− model. Given that treatment commences at 14 weeks of age (Supplementary Fig. S3A), when prostate tumors have uniformly advanced to the invasive adenocarcinoma stage (11) with significant MDSC infiltration (Fig. 2B), our findings support the view that anti-MDSC treatment provokes regression of advanced tumors. In addition, both Pep-H6 peptibody and CXCR2 inhibitor treatment significantly prolonged the overall survival of the Pten^pc−/−Smad4^pc−/− tumor-bearing mice. Therefore, our preclinical data suggest that pharmacologic depletion of MDSCs may offer potential therapeutic benefits for patients with advanced prostate cancer, particularly those deficient for PTEN and SMAD4. In line with our findings, others have demonstrated that depletion of G-MDSCs promotes the intratumoral accumulation of activated CD8⁺ T cells and apoptosis of tumor epithelial cells in a Kras/Trp53mouse pancreatic cancer model (36).

MDSCs are of myeloid cell lineage, and their coordinated regulation represents one of the most complex aspects of cancer–host interactions (37). The involvement of the myeloid compartment of the hematopoietic system in innate immunity, adaptive immunity, as well as in regulation of TME through nonimmune mechanisms highlights the need to understand more deeply how modulating different myeloid populations, including MDSCs, can positively or negatively affect tumor growth.

Pep-H6 peptibody, targeting S100A9 expressed on MDSCs, has been shown to have minimal toxicity in treated mice (20) and potent antitumor activity (Fig. 4D and E; ref. 18). Interestingly, tasquinimod, a small-molecular inhibitor for S100A9, has been shown to increase progression-free survival and overall survival for metastatic castration-resistant prostate cancer in a phase II clinical trial and has entered phase III clinical trials (38). Importantly, similar to the peptibody treatment in mice, tasquinimod is well tolerated and causes only minor adverse effects in human patients (38), suggesting that tasquinimod or similar drugs targeting S100A9 could potentially be used as chemopreventive agents for patients with high-risk primary prostate cancer. The antiproliferative mechanism may explain why targeting CXCR2 in prostate cancer with abundant preexisting MDSC infiltration can lead to MDSC depletion, as MDSCs have been shown to undergo active proliferation inside the prostate tumor of the Pten^pc−/− model (9). The effectiveness of targeting CXCR2 in our model suggests targeting mechanisms that specifically regulate MDSC recruitment as well as their proliferative and survival potential in human cancers would provide therapeutic benefit for patients with prostate cancer.

Targeting MDSCs as a cooperative approach for immunotherapy is clinically relevant, as increasing evidence indicates MDSCs represent a bona fide immunosuppressive cell population in patients with various solid tumors (39, 40). Immunosuppressive mechanisms by MDSCs in mice have been validated in humans, which include l-arginine depletion, NO and ROS production, TGFβ secretion, blocking T_eff cells and inducing T_reg cells, among others (39). Future studies are warranted to evaluate if combining MDSC depletion with immune checkpoint inhibitors, such as anti–CTLA-4, anti–PD-1, and anti–PD-L1 antibodies, may elicit synergistic efficacy in preclinical models of prostate cancer and eventually benefit patients with prostate cancer.

Pten^pc−/− and Pten^pc−/−Smad4^pc−/− models were developed previously (11) and were backcrossed to the C57BL/6 background for more than four generations. B6.129(Cg)-Gt(ROSA)26Sortm4(ACTB-tdTomato,-EGFP)Luo/J (“mTmG”) strain was obtained from The Jackson Laboratory. Mice were maintained in pathogen-free conditions at the MD Anderson Cancer Center. All manipulations were approved under the MD Anderson Cancer Center Institutional Animal Care and Use Committee.

Pten^pc−/−Smad4^pc−/− prostate cell lines, which have been described previously (11), were generated in 2010. PPS, a C57BL/6-syngeneic cell line isolated from prostate tumors of Pten^pc−/−Smad4^pc−/−Trp53^c−/− mice, was generated in 2013. All cell lines tested for Mycoplasma were negative within 6 months of performing the experiments. Cell line authentication was not performed.

Prostate tumor single cells were isolated using the Mouse Tumor Dissociation Kit (Miltenyi Biotec). Single cells were isolated from spleen, lymph node, and peripheral blood using standard protocol. All isolated cells were depleted of erythrocytes by hypotonic lysis. For CyTOF analysis, cells were blocked for FcγR using CD16/CD32 antibody (clone 2.4G2, BD Biosciences) and incubated with CyTOF antibody (DVS Sciences, used at 0.5 test/1 million cells) for 30 minutes at room temperature. Cells were washed once and incubated with MAXPARNucleic Acid Intercalator-¹⁰³Rh (DVS Sciences) for 20 minutes for viability staining. Cells were fixed with 1.6% formaldehyde for 1 hour and incubated with MAXPARNucleic Acid Intercalator-Ir (DVS Sciences) at 4°C overnight to stain the nuclei. The samples were analyzed with CyTOF instrument (DVS Sciences) in the Flow Cytometry and Cellular Imaging Core Facility at the MD Anderson Cancer Center. Flow cytometry was performed using standard protocol on LSRFortessa analyzer (Becton Dickinson) and analyzed with FlowJo software (Tree Star).

T-cell suppression assay was performed as described (9) using FACS-sorted MDSCs and CFSE (Invitrogen)-labeled MACS-sorted (Miltenyi Biotec) CD8⁺ or CD4⁺ T cells in anti-CD3– and anti-CD28–coated 96-well plates at an MDSC/T-cell ratio of 0:1, 1:1, 1:2, 1:4, with 3.0 × 10⁵ to 5.0 × 10⁵ MDSCs used in each ratio. Cells were analyzed after 72 hours by flow cytometry, and the suppression of T cells is calculated as described (41). The percentage of CFSE⁺ cells divided in the presence of MDSCs was compared with the percentage of CFSE⁺ divided cells in the absence of any added MDSCs. For the MDSC migration assay, an equal number of FACS-sorted MDSCs, untreated or pretreated with neutralizing antibody or inhibitor, were placed on the upper chamber of a transwell system (BD Falcon), and conditioned media from PTEN/SMAD4-deficient cells under various conditions were added to the bottom chamber. Cells were allowed to migrate to the bottom well for 6 hours at 37°C with 5% CO₂. Migrated cells were then analyzed by flow cytometry using BD Fortessa X20. Migrated FITC-positive cells were gated to count the absolute number of cells migrated through the transwell.

Anti-Gr1 (clone RB6-8C5) and isotype control (clone LTF2) were purchased from BioXcell and dosed at 200 μg/mouse (i.p.) every other day. Endotoxin-free plasmids (15 μg) for irrelevant control peptibody (Irr-pep) and MDSC-specific Pep-H6 peptibody were injected into mice through tail vein using the established protocol (21) in TransIT-EE Delivery Solution (Mirus Bio LLC) every 4 days. SB225002 (Cayman Chemical) in DMSO was diluted in vehicle (0.9% NaCl, 0.3% Tween 80) for in vivo administration every other day (5 mg/kg).

Inducible Yap1 knockdown was constructed by cloning the two Yap1 shRNAs used previously (26) from the pLKO.1 into a doxycycline-inducible plasmid. Lentivirus was packaged in 293T and was used to infect PPS, a C57BL/6-syngeneic cell line isolated from prostate tumor of Pten^pc−/−Smad4^pc−/−Trp53^c−/−mice. Stable sublines were selected with puromycin (2 μg/mL) and injected subcutaneously to the flank of 5-week-old male C57BL/6 mice (Jackson Laboratory). Two weeks after injection, mice were fed with doxycycline water (2 g/L), a method used to execute doxycycline-inducible expressionin vivo (42). Tumors were measured and extracted 6 days later to analyze for MDSC percentage in infiltrating immune cells.

RNA was isolated from FACS-sorted GFP⁺ and Tomato⁺ cells using Pten^pc−/−Smad4^pc−/−mTmG⁺ prostate tumors, followed by microarray analysis at the MD Anderson Microarray Core facility using the Mouse Genome 430 2.0 Array (Affymetrix) to generate a Pten^pc−/−Smad4^pc−/− tumor/stroma dataset GSE71319. Dataset GSE25140 was downloaded from the NCBI Gene Expression Omnibus (GEO) database. Differentially expressed genes between two conditions (GFP⁺ vs. Tomato⁺ or PTEN/SMAD4 vs. PTEN) were subjected to IPA, GSEA, and oPOSSUM analysis. For analysis of human prostate data, we first generated a list of 39 human MDSC signature genes by literature mining (Supplementary Table S7). The gene expression data of 498 TCGA prostate samples were downloaded from the Broad GDAC Firehose (http://gdac.broadinstitute.org), which is the RSEM expression estimates normalized to set the upper quartile count at 1,000 for gene level and then with log₂ transformation. The 498 TCGA prostate samples were clustered using the 39 MDSC genes into MDSC-high, MDSC-low, and MDSC-medium (distance between pairs of samples was measured by Manhattan distance, and clustering was then performed using complete-linkage hierarchical clustering). Sixty-nine samples from Wallace and colleagues (32) were clustered into MDSC-high and MDSC-low. Differentially expressed genes between MDSC-high and MDSC-low were analyzed by GSEA. The expression of CXCL6 in MDSC-high samples is compared with MDSC-low samples using the Wilcoxon test.

Tissues were fixed in 10% formalin overnight and embedded in paraffin. IHC was performed as described earlier (11). For Western blot analysis, cells were lysed on ice using RIPA buffer (Boston BioProducts) supplemented with protease and phosphatase inhibitors (Roche). YAP1 antibody was obtained from Novus Bio and Cell Signaling Technology. CXCL5 antibodies were obtained from Bioss and R&D Biosystems. CXCR2 antibody was obtained from Bioss and R&D Biosystems. CD45 and Ly6G antibodies were obtained from Biolegend. Prostate tissue microarray was obtained from Folio Bioscience.

ChIP was performed as described (26) using YAP1 antibody from Novus. Briefly, 5 μg of rabbit IgG (Santa Cruz) or YAP1 antibody was incubated with Protein A Dynabead magnetic beads (Invitrogen) for 4 hours, followed by extensive wash to remove unbound antibody. Antibody beads were then added to the chromatin and incubated overnight. The following primers were used for qPCR analysis: CXCL5_S: 5′-CTCCAGTTTCCTGCCTGAAG-3′ and CXCL5_as: 5′-GTGTGGAGATTGGGGCTCTA-3′.

RNA was isolated by the RNeasy Kit (Qiagen) and reverse transcribed using the Superscript III cDNA Synthesis Kit (Life Technology). Quantitative PCR was performed using the SYBR-GreenER Kit (Life Technology). The following primers were used: CXCL5_Fwd: GCATTTCTGTTGCTGTTCACGCTG, CXCL5_Rev: CCTCCTTCTGGTTTTTCAGTTTAGC; β-actin_Fwd: GAAATCGTGCGTGACATCAAAG, β-actin_Rev: TGTAGTTTCATGGATGCCACAG; YAP1_Fwd: TGAGATCCCTGATGATGTACCAC, YAP1_Rev: TGTTGTTGTCTGATCGTTGTGAT.

Data are presented as mean ± SD unless indicated otherwise. The Student t test assuming two-tailed distributions was used to calculate statistical significance between groups. Animal survival benefit was determined by the Kaplan–Meier analysis. P < 0.05 was considered statistically significant.

The expression array data used in this article were in GEO with accession numbers GSE25140 (11) and GSE71319.

A. Kapoor is a research investigator at Novartis. C.J. Logothetis has received commercial research grants from Astellas, BMS, J&J, Excelixis, Pfizer, Novartis, Bayer, AstraZeneca, and Helsinn HC. No potential conflicts of interest were disclosed by the other authors.

Conception and design: G. Wang, X. Lu, Z. Ding, C.J. Logothetis, Y.A. Wang, R.A. DePinho

Development of methodology: G. Wang, X. Lu, P. Dey, P. Deng, V. Ramamoorthy, Q. Chang

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): G. Wang, X. Lu, P. Dey, P. Deng, Z. Fang, S. Hua, E.M. Li-Ning-Tapia, Z. Guo, X. Shang, S. Khadka, E.-J. Jin, X. Pan, L. Li, Q. Chang, P. Troncoso, M.J. McArthur

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): G. Wang, X. Lu, P. Dey, C.C. Wu, J. Zhang, C.-J. Wu, N.B. Patel, Y. Shi, Q. Chang, C.J. Logothetis, M.J. McArthur, Y.A. Wang

Writing, review, and/or revision of the manuscript: G. Wang, X. Lu, P. Dey, C.C. Wu, C.-J. Wu, C.J. Logothetis, M.J. McArthur, Y.A. Wang, R.A. DePinho

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): P. Dey, P. Deng, C.C. Wu, S. Jiang, R. Konaparthi, E.M. Li-Ning-Tapia, A. Kapoor, N.B. Patel, Z. Guo, T. Heffernan, D. Zhao, X. Shang, P. Hou, B. Hu, W. Yao, Y. Shi, L. Chin

Study supervision: Y.A. Wang, R.A. DePinho

Other (mouse work): K. Zhao

Other (assessed immunohistochemical stains): E.M. Li-Ning-Tapia

Other (supported molecular biology needs, i.e., cloning): T.N. Tieu

The authors thank Drs. Larry W Kwak and Hong Qin for the peptibody plasmids; Drs. Willem Overwijk and Yared Hailemichael for advice in hydrodynamic injection; Samirkumar Amin and members of the DePinho laboratory for helpful suggestions and technical support; and the Flow Cytometry and Cellular Imaging Core Facility (Jared Burks, Duncan Mak, and Karen Dwyer) and the Sequencing & Non-coding RNA Core Services (Chang-Gong Liu) at The University of Texas MD Anderson Cancer Center (Cancer Center Support Grant CA16672).

The project was supported by U01CA141508 (to L. Chin and R.A. DePinho), Prostate Cancer Research Program (PCRP) W81XWH-13-1-0202 (to G. Wang) and W81XWH-14-1-0429 (to P. Dey); the Idea Development Award–New Investigator Option (W81XWH-14-1-0576; to X. Lu) from the Department of Defense; and the Clayton & Modesta Williams Cancer Research Fund. Additional support was provided by the Jane Coffin Childs Memorial Fund Postdoctoral Fellowship to X. Lu. E.-J. Jin is supported by Korean Governments (MISP) grant number 2011-0030130.