Single-Cell Analysis Reveals EP4 as a Target for Restoring T-Cell Infiltration and Sensitizing Prostate Cancer to Immunotherapy

This article is featured in Highlights of This Issue, p. 433

Prostate cancer is the second most commonly diagnosed cancer and the fifth leading cause of cancer-related death in men worldwide (1). While patients diagnosed at a localized disease stage could receive potentially curative therapies, many ultimately progress to metastatic castration-resistant prostate cancer (CRPC) and experience fatal outcomes (2, 3). Recently, some immunotherapies, particularly immune checkpoint inhibitors, such as Opdivo, Keytruda, and Tecentriq, have been approved for marketing with valuable benefits in various patients with cancer (4). Unfortunately, few patients with prostate cancer responded to immunotherapy targeting T-cell–associated immune checkpoints, such as CTL-associated protein 4 (CTLA-4), programmed death-ligand 1 (PD-L1), and programmed cell death protein 1 (PD-1), which prompted growing interest in combinatorial immunotherapy for metastatic CRPC as the next critical frontier (5–7). Having a better understanding of the molecular mechanisms involved in immune resistance in prostate cancer could reveal effective therapeutic strategies for advanced prostate cancer (8, 9).

The nature of the tumor microenvironment (TME), regulated by dynamic tumor–immune cell interactions, can drastically impact tumor progression, tumor clearance, and immune response (10, 11). Specifically, prostate tumors are characterized by low mutational burden, low CD8⁺ T-cell infiltration, and upregulation of PD-L1 expression induced by castration (12–14). Furthermore, trafficking of diverse immunosuppressive cell populations in the TME, such as T-regulatory cells (Treg), tumor-associated macrophages (TAM), and myeloid-derived suppressor cells (MDSC), shapes a unique molecular profile and distinct features (15, 16). Consequently, prostate cancer was termed a “cold” tumor due to a lack of infiltrated T cells and likely hindered the clinical outcomes of immune checkpoint inhibitors. In particular, MDSCs at the tumor sites are believed to be conspicuous suppressors of T-cell activity (17). They inhibit the proliferation and antitumor activity of effector T cells through reactive oxygen species (ROS), arginase, and NO production (18), suppress the cytotoxicity of natural killer (NK) cells by inhibiting perforin production (19), and induce Treg differentiation by secreting IL10 and TGFβ (20). These support the idea that MDSCs might be a potential combinatorial target for immunotherapy (21). However, the potential mechanism of MDSCs involved in immunosuppression for prostate cancer is inadequately elucidated.

Prostaglandin E2 (PGE2) is a type of prostanoid and a principal metabolic product of arachidonic acid generated through the sequential actions of COX and PGE synthases (22–25). PGE2 acts on specific G protein–coupled receptors (GPCR), prostaglandin E (EP1–4), to regulate intracellular signaling pathways (26). EP4 is a high-affinity EP receptor, and its abnormally elevated expression is well characterized as a protumorigenic mediator in numerous cancers (27–30). Extensive studies demonstrate that EP4 promotes tumor migration and metastasis by activating different signaling pathways, and antagonist-targeting EP4 are available for preclinical and clinical applications (31–34). EP4 has been implicated in driving various immunosuppressive effects, including inhibiting NK activity in breast cancer and impairing CTL function during chronic viral infection (35–38), as well as in developing Tregs and MDSCs (39, 40). Nevertheless, EP4 is involved in diverse and complex cellular reactions related to immune responses, and its elaborate profile acting on the TME remains to be fully depicted.

Recent advances in single-cell transcriptomics have revealed new insights into distinct cellular subpopulations and intratumoral heterogeneity, which characterized TME and deepened our understanding of the intercellular communication critical for tumor survival and growth. Here, we employed single-cell RNA-sequencing (scRNA-seq) to dissect prostate cancer intra-tumoral heterogeneity and identify a specific membrane receptor, EP4 (PTGER4), as an immune suppressor in the immune microenvironment of prostate cancer. We then used a novel EP4 antagonist, YY001, which inhibited the differentiation, maturation, and immunosuppressive function of MDSCs. YY001 enhanced the proliferation and anticancer function of T cells and further reversed the infiltration level of MDSC and T cells in the TME by overturning the chemokine profile of tumor cells in vitro and in vivo. More importantly, YY001 strongly synergized with anti–PD-1 antibodies, converting completely unresponsive prostate tumors into responsive ones with marked tumor regression, long-term survival, and lasting immunologic memory, which may increase clinical benefits and possibly prevent tumor recurrence in patients with prostate cancer.

The mouse prostate cancer cell line RM-1; human prostate cancer cell lines C4–2, PC3, and LNCaP; and human umbilical vein endothelial cell line (HUVEC) were purchased from the Chinese Academy of Sciences Type Culture Collection Cell Bank. RM-1, C4–2, PC3, and LNCaP cells were cultured in an RPMI1640 medium supplemented with 10% (v/v) FBS and 1% antibiotics (penicillin and streptomycin) with 5% CO₂ at 37°C. HUVECs were cultured in Endothelial Cell Medium (Sciencell, 1001).

The orthotopic and bone metastatic tumor tissue samples of patients with prostate cancer or CRPC were obtained from the Department of Urology, Shanghai Changhai Hospital (Shanghai, China). The study was based on residual surgical tissue samples available after histopathologic analyses and not required for diagnosis without interference with routine clinical practice. This study was approved by the ethics committee at Shanghai Changhai Hospital and performed in accordance with local legislation and established ethical guidelines (Declaration of Helsinki). All patients provided written informed consent.

Primary tumors were dissociated using the human tumor dissociation kit (Miltenyi Biotec, 130–095–929) according to the manufacturer's instructions. Freshly isolated tumor tissues were collected, immediately minced into small pieces, and then placed into a gentleMACS C-tube containing the enzyme mix. The tubes were centrifuged briefly and placed onto the gentleMACS Octo Dissociator with Heaters, and then the program was run. Following completion of the program, the tube was briefly centrifuged. The homogenate was filtered through a 70-μm cell strainer into a new centrifuge tube and centrifuged; the supernatant was then removed, and the pellet was lysed with BD Pharm Lyse Lysing buffer, filtered through a 70-μm cell strainer, centrifuged, and resuspended with 1× PBS containing 0.04% BSA. The cell suspension was then filtered through a 40-μm strainer, which was subsequently rinsed with 1× PBS containing 0.04% BSA (2 mL) and centrifuged. The supernatant was then removed, and the pellet was resuspended in 1× PBS containing 0.04% BSA. Single-cell suspensions were achieved for subsequent sequencing or functional studies.

Neutrophils were isolated from the single-cell suspensions of tissue samples using EasySep Human Neutrophil Isolation Kit (STEMCELL, 17957) according to the manufacturer's manual. To achieve MDSC suspensions, the isolated neutrophils were subsequently FACS-sorted with anti–LOX-1 antibody (BioLegend, 358603) using a SH800S Cell Sorter (SONY).

PBMCs were isolated from human peripheral blood samples by centrifugation on a density gradient according to the manufacturer's instructions (TBD, LTS10771). CD3⁺ T cells were isolated from freshly isolated PBMCs using CD3 Positive Selection Magnetic Beads (Miltenyi Biotec, 130–050–101), according to the manufacturer's specifications. For evaluation of T-cell proliferation, isolated CD3⁺ T cells were labeled with 1 μmol/L CFSE or CellTrace Violet (Thermo Fisher Scientific, C34571) and activated in vitro with T Cell TransAct (Miltenyi Biotec, 130–111–160) in X-VIVO 15 (Lonza, 04–418Q) supplemented with human recombinant IL2(100 U/mL). PGE2, E7046, or YY001 was added into the culture medium for 72 hours. Cells were collected after 3 days of culture for FACS (BD, LSRFortessa). To evaluate the cytotoxicity of T cells, activated cells were pretreated with E7046 or YY001 for 1 hour, and then PGE2 was added to the culture for 6 hours. After treatment, expression of PD-1, Tim-3, and intracellular IFNγ was evaluated by flow cytometry by gating on CD3⁺CD8⁺ populations.

For in vitro induction of MDSCs, tibias and femurs bones from C57BL/6 mice were removed using sterile techniques, and bone marrow (BM) was flushed. Single-cell suspensions were prepared from BM, followed by the removal of red blood cells with ammonium chloride lysis buffer (ACK). Cells were seeded into a 6-well plate with density at 2 × 10⁶ cells/well in RPMI1640 medium supplemented with GM-CSF (20 ng/mL) plus IL4 (10 ng/mL). E7046 or YY001 was added to the culture medium in the presence or absence of PGE2 for 5 days in a 5% CO₂ and 37°C incubator. The generated MDSCs (CD11b⁺Gr-1⁺) were harvested and analyzed by flow cytometry (BD LSRFortessa).

As mentioned above, cells derived from the BM of C57BL/6 mice were seeded into a 6-well plate with density at 2 × 10⁶ cells/well in RPMI1640 medium supplemented with GM-CSF (40 ng/mL) plus IL6 (40 ng/mL) for 4 days, in a 5% CO₂ and 37°C incubator. Differentiated MDSCs (nonadherent cells) derived from BM was harvested and directly seeded into 6-well plates at 1.5 × 10⁶ cells/well and then treated with PGE2, E7046, or YY001 for 2 days. To evaluate the expression of Arg1 in MDSCs, MDSCs (nonadherent cells) were harvested and analyzed by flow cytometry (BD LSRFortessa). The expression of Arg1 was gated on the Arg1⁺ cells in MDSCs (CD11b⁺Gr1⁺). To evaluate the expression of immunosuppressive molecules in MDSCs at the transcriptional level, MDSCs (nonadherent cells) were harvested, and the total RNA was extracted using the TRIzol reagent, according to the manufacturer's instructions. The amplification of Arg1, iNOS, IDO, COX2, IL10, and CXCR4 in MDSCs was detected using real-time PCR. Primer sequences used are shown in Supplementary Table S2.

CD8⁺ T cells were isolated from mouse splenocytes using CD8-negative selection magnetic beads according to the manufacturer's specifications. Isolated CD8⁺ T cells were labeled with 1 μmol/L CFSE and activated in vitro with anti-CD3 and anti-CD28 beads (Invitrogen) according to the manufacturer's instructions. BM-derived MDSCs were added to the culture medium (1:1). After 3 days, the proliferation of CFSE-labeled CD8⁺ T cells was analyzed by BD Fortessa.

C57BL/6 mice were obtained from the Animal Center of East China Normal University (Shanghai, China). A total of 1 × 10⁶ RM-1 cells were subcutaneously injected into the dorsal part of mice. Tumor-bearing mice were randomly assigned, and YY001 or E7046 was formulated for treatment in 0.5% carboxymethylcellulose in sterile water and administered by oral gavage once a day. YY001 or E7046 was given at 150 mg/kg. Anti–PD-1 (200 μg per dose) or anti-CD8 (200 μg per dose) was administered intraperitoneally twice a week. In the combined treatment group, YY001 or E7046 was orally administered at daily doses of 150 mg/kg, and anti—PD-1 (200 μg per dose) or anti-CD8 (200 μg per dose) was administered intraperitoneally twice a week. Tumor size was calculated by length × width² × 0.52. Mice were euthanized on day 15 after treatment, and tumors were isolated, weighed, photographed, and fixed for IHC and immunofluorescence analysis.

For the orthotopic prostate cancer model, mice were anesthetized using tribromoethanol, and 1 × 10⁵ RM-1-luci cells, resuspended in PBS with 50% Matrigel, were orthotopically injected into the dorsum of the prostate. Tumor-bearing mice were randomly divided into six groups based on photon flux indexes. The control group was given 0.5% carboxymethyl cellulose in sterile water by oral gavage once a day and 1× PBS intraperitoneally twice a week. YY001 or E7046 was formulated for treatment in 0.5% carboxymethyl cellulose in sterile water and administered by oral gavage once a day. YY001 or E7046 was given at 150 mg/kg. Anti–PD-1 (200 μg per dose) was administered intraperitoneally twice a week. In the combined treatment group, YY001 or E7046 was orally administered at daily doses of 150 mg/kg, and anti–PD-1 (200 μg per dose) was administered intraperitoneally twice a week. Tumor growth and metastasis were monitored weekly by the IVIS Imaging System. The treatments were maintained until the mice died or there was an obvious difference between the untreated and combination groups. The survival of mice was observed and recorded.

A subcutaneous tumor model was developed by injecting 1 × 10⁶ RM-1 cells into the left flank of C57BL/6 mice. Tumor-bearing mice were randomly assigned to vehicle, anti–PD-1 antibody, YY001, and combination treatment groups. YY001 was formulated for treatment in 0.5% carboxymethyl cellulose in sterile water and administered by oral gavage once a day. YY001 was administered at 150 mg/kg. Anti–PD-1 (200 μg per dose) was given intraperitoneally twice a week. In the combined treatment group, YY001 was orally administered at daily doses of 150 mg/kg, and anti–PD-1 (200 μg per dose) was administered intraperitoneally twice a week. After 10 days of treatment, tumors were surgically removed, and 1 × 10⁶ RM-1 cells were injected into the right flank of the same group of C57BL/6 mice. Treatments continued until the mice died, and the tumor recurrence on the left flank and tumor formation on the right flank were observed and recorded.

Humanized CD34⁺ mice were purchased from ALLCELLS Biotechnology Co., Ltd., and NPSG mice with hCD45⁺/(hCD45⁺⁺mCD45⁺)> 20% were subcutaneously transplanted with MSK-PCa2 organoid tumor (presented by Professor Gao Dong, Shanghai Institute of Biochemistry and Cell Biology, Chinese Academy of Sciences). Treatment was initiated after tumor nodules grew to 100 mm³, and mice were randomly assigned to vehicle, anti–PD-1 antibody, YY001, and combination treatment groups. Mice received intraperitoneal injection of 100 μg anti–PD-1 antibody twice a week and received oral administration of YY001 at the dose of 150 mg/kg/day. The tumor volume and mouse body weight were measured twice per week. The tumor volume was calculated as described above. Mice were sacrificed 40 days after treatment; tumors were removed and imaged, and then processed for IHC and immunofluorescence analysis.

At the end of the experiment, mice were sacrificed, and tumors were harvested to characterize cell-mediated antitumor responses. Tumors were minced and digested in PBS supplemented with collagenase I, collagenase VI, and DNase I on a rocking platform at 37°C for approximately 15 to 20 minutes. Single-cell suspensions were prepared by mechanically disrupting the tissues through a 70-μm cell strainer. Cells were then subjected to ACK and then counted and kept on ice for the subsequent analyses.

After counting, cells or tissue single-cell suspensions were stained for approximately 10 to 15 minutes at room temperature in the dark using live/dead staining kits according to manufacturer's instructions and then incubated with Fc-blocking reagent (1 μL for every million cells) for 10 minutes at 4°C. Cells were then stained with the indicated antibodies as prearranged panels for 30 minutes at 4°C. For the analysis of intracellular IFNγ expression, cells were first stained for surface markers, followed by fixation and permeabilization using the BD Cytofix/Cytoperm kit (BD Biosciences) according to the manufacturer's instructions, and then stained with an anti-IFNγ antibody. For the analysis of the transcription factor Foxp3 expression, cells were first stained for surface markers, then preprocessed with Foxp3/Transcription Factor Staining Buffer Set (eBioscience) according to the manufacturer's instructions, and then stained with anti-Foxp3 antibody. Stained cells were washed with 1× MACS buffer and analyzed using a digital flow cytometer (BD LSRFortessa, BD Biosciences). Antibodies used for flow cytometry are listed in Supplementary Table S1. The data were analyzed with FlowJo version 10.3 (Tree Star). The results of cell experiments were recorded as average values of duplicate measurements. The results of animal experiments were expressed as the mean ± SEM for at least five mice per group.

For polychromatic flow cytometry (PFC), an antibody panel up to 18 antibodies was designed in advance according to the immune cell types. Samples were first stained for surface markers, then fixed using the Foxp3/Transcription Factor Staining Buffer Set, and then stained with anti-Foxp3 antibody. Cells were collected on a 5-laser, 18-color flow cytometer (BD LSRFortessa X-20, and analyzed using FlowJo version 10.3 (Tree Star).

scRNA-seq data analysis was performed on the NovelBrain Cloud Analysis Platform (see Supplementary Methods for details). Detailed information for bulk RNA-seq analysis and immune checkpoint blockade (ICB) trial analysis is provided in Supplementary Methods.

To quantitatively assess the CD8⁺ T cells in tumor samples, the scanned images were cropped into areas that were approximately the size of a 20× view. Representative panels were selected to simultaneously visualize CD8, MHC-II, TCF1, and cellular nuclei (DAPI). Quantitative immunofluorescence analysis was conducted using CellProfiler 3. The “IdentifyPrimaryObjects” module was used to define all cell objects (by DAPI staining), antigen-presenting cell (APC) objects (by MHC-II staining), and CD8⁺ T cell objects (by CD8 staining). The “IdentifySecondaryObjects” module was used to extend the border of the identified CD8⁺ T-cell objects by 2 pixels. The expression level of TCF1 in individual CD8⁺ T cells was measured by the “MeasureObjectIntensity” module. Immunomaps were plotted using custom R scripts.

All experiments were performed with three or more replicates. Statistical analyses were conducted using Student t test. P values <0.05 were considered significant. Differences between the control and experimental groups were determined by one-way ANOVA. Because both treatment and time-course experiments were investigated, data were analyzed by two-way ANOVA followed by a post hoc test. Data were expressed as means with a 95% confidence interval, and P <0.05 was considered significant. All statistical analyses were performed using Microsoft Excel 2019 (RRID: SCR_016137), GraphPad Prism 5 (RRID: SCR_002798), or R 3.6.0 software (RRID: SCR_001905).

Requests for resources and reagents should be directed to and will be fulfilled by the lead contact Shancheng Ren ([email protected]). The accession number for the processed scRNA-seq data used in this study is GEO: GSE141445. Processed scRNA-seq data can be accessed at http://www.pradcellatlas.com, where gene expression in each cell cluster can be queried and visualized in a user-friendly single-cell browser application. Customized scripts for reproducing the analyses and plots will be available upon request.

To investigate the role of EP4 in T-cell–mediated immunity of prostate cancer, we sought to evaluate the expression of EP4 mRNA (PTGER4) in T cells at a single-cell resolution. We have recently performed scRNA-seq on 13 primary prostate cancer samples using 10X Genomics platform, and another four castration-resistant prostate cancer (CRPC) samples using the BD Rhapsody platform (41). Unsupervised clustering using GraphClust algorithm, followed by visualization in Uniform Manifold Approximation and Projection (UMAP) space, identified diverse cellular identities based on known canonical cell markers in primary prostate cancer (Fig. 1A) and CRPC (Fig. 1B) tissue samples. PTGER4 expression was observed in the epithelial cells and cells within the TME, including T cells, monocytic cells, and granulocytes (Supplementary Fig. S1A and S1B).

Pseudotime analysis of the CD8⁺ T cells in primary prostate cancer samples revealed two divergent cell fates, where Cluster 0 and 3 served as starting points, Cluster 1 as a transitioning state, and Clusters 2 (cytotoxic T cells) and 4 (exhausted T cells) as two alternative endpoints (Fig. 1C). Along the trajectory to Cluster 2, the expression level of cytotoxic T-cell markers, including GZMB and PRF1, increased; the expression of classic immune checkpoint proteins, including PDCD1, CTLA4, CD44, and AHR, decreased (Fig. 1C; Supplementary Fig. S1A). Conversely, along the trajectory to Cluster 4, decreased expression of multiple immune checkpoints and PTGER4 was observed (Fig. 1C; Supplementary Fig. S1A). Pseudotime analysis of T cells in CRPC samples revealed a progressive loss of cytotoxicity (Fig. 1D). Along this trajectory, expression of the cytotoxic markers GZMB and PRF1 gradually decreased, while the expression of PTGER4 and multiple immune checkpoints increased (Fig. 1D; Supplementary Fig. S1B). Taken together, these results suggest that the higher PTGER4 expression level in T cells is associated with an exhausted phenotype marked by impaired cytotoxicity.

To validate these findings at the pan-cancer level, we examined the PTGER4 expression in infiltrating T cells in other tumor samples. Unsurprisingly, we observed a significantly higher expression of PTGER4 in LAYN-positive exhausted CD8⁺ T cells than CX3CR1-positive effector memory T cells in multiple scRNA-seq cohorts (Supplementary Fig. S1C), including hepatocellular carcinoma (HCC; ref. 42), colorectal cancer (CRC; ref. 43), and non—small cell lung carcinoma (NSCLC; ref. 44). In addition, among the four tumors, the expression of PTGER4 was positively correlated with multiple immune checkpoint molecules (Supplementary Fig. S1D), further confirming the role of PTGER4 as a universal T-cell exhaustion marker.

Next, we performed gene-set enrichment analysis (GSEA) on the T cells in single-cell sequencing cohorts. Among the enriched pathways with the highest normalized enrichment scores were immune-related pathways such as TCR signaling and NK cell–mediated cytotoxicity (Fig. 1E). Differential expression analysis in primary prostate cancer samples revealed that PTGER4-high T cells expressed a number of immune-inhibitory factors, such as CCL4, AREG, and CCL4L2, indicating a dysfunctional phenotype. In contrast, PTGER4-low T cells expressed a series of genes related to the HSP family that are associated with antigen presentation and immune stimulation (Fig. 1F). In CRPC samples, CCL4 was also highly expressed in PTGER4-high CD8⁺ T cells. These results indicate that an altered expression pattern of cytokines and immunomodulatory molecules in T cells with high PTGER4 expression drives the shift toward an exhausted phenotype (Fig. 1F).

We then validated these findings in bulk sequencing data. Consistent with single-cell sequencing data, bulk RNA-seq data of both primary (Fig. 1G) and metastatic prostate cancer samples (Supplementary Fig. S1E) demonstrated a positive correlation between PTGER4 and T-cell exhaustion signature genes. Furthermore, PTGER4 was significantly positively correlated with T-cell exhaustion signature score across most solid tumors in TCGA PANCAN dataset (29/33, 88%), indicating that PTGER4 is a universal T-cell exhaustion marker irrespective of tumor origin (Fig. 1H; Supplementary Fig. S1F). The correlations between the expression level of PTGER4 and other genes were subjected to GSEA, and significantly enriched pathways for each cancer type were acquired. The most commonly enriched pathways were those related to adaptive immune responses and T-cell differentiation (Fig. 1I). Genes in these pathways that were positively correlated with PTGER4 included anti-inflammatory cytokines and cytokine receptors such as IL33, markers of immune suppression such as FOXP3, and multiple immune checkpoints, highlighting a potential role of PTGER4 as a negative regulator of the adaptive immune response and T-cell differentiation (Fig. 1J). In addition, we leveraged results from clinical trials investigating ICB for treating advanced solid tumors. Pooled analyses of these trials demonstrated that patients with higher PTGER4 expression levels exhibited more favorable overall survival and progression-free survival (PFS), suggesting that these patients are more sensitive to ICB (Supplementary Fig. S1G). This is in line with our findings that PTGER4 is associated with T-cell exhaustion because recent work has revealed that ICB primarily works by reversing T-cell dysfunction in tumors with high CTL infiltration (45).

We then used our single-cell data to further investigate the role of PTGER4 in the various compartments of TME. Among the monocytic cells detected in the primary prostate cancer samples, which were further classified into seven subclusters, five TAM clusters (Clusters 0, 2, 4, 5, and 6) were identified (Supplementary Fig. S2A and S2B). We applied the Monocle 2 algorithm to map the TAMs in pseudotime and discovered two divergent trajectories to M2-like TAMs; one of two trajectories revealed an increasing expression level of PTGER4 (Fig. 2A). Using canonical signatures for classic (M1) and alternative (M2) types of macrophage activation (46), we classified individual TAMs as M1- or M2-like and performed differential gene expression analysis. PTGER4-positive, M2-like TAMs highly expressed inhibitory cytokines and chemokines (Fig. 2B). Pathway enrichment analysis against the KEGG knowledgebase indicated that highly expressed genes in PTGER4-positive, M2-like TAMs were involved in immune regulatory pathways, including cytokine—receptor interaction, NFκB signaling, TNF signaling, and IL17 signaling (Supplementary Fig. S2C).

In CRPC samples, single-cell sequencing using BD Rhapsody platform allows for unbiased analysis of the function of granulocytes, which are highly heterogenous and have previously been reported to play a key role in prostate cancer progression and castration resistance. Granulocyte clusters were further identified as six subclusters of neutrophils and two subclusters of PMN-MDSCs (Fig. 2C), the latter of which characterized by high expression of LOX-1 (OLR1) (Fig. 2D). To validate these MDSCs, tumor-infiltrating neutrophils in prostate tumor samples were isolated and a subgroup of cells with high expression of LOX-1 was identified using flow cytometry (Fig. 2E). Furthermore, T cells cocultured with these LOX-1⁺ neutrophils exhibited inhibited proliferation ability (Supplementary Fig. S2D) and diminished cytotoxicity, as exemplified by reduced expression of IFNγ and granzyme B (Fig. 2F). These results provided cellular and functional evidence for the PMN-MDSCs identified by single-cell sequencing in CRPC samples.

To examine the role of EP4 in MDSCs, we first confirmed the expression of EP4 on PMN-MDSCs using flow cytometry (Fig. 2G), which is in line with our previous analysis revealing granulocytic expression of PTGER4 (Fig. 1B). We then performed pseudotime analysis on the PMN-MDSCs and identified two trajectories from early-stage MDSCs (e-MDSC) to maturation, resulting in two types of PMN-MDSCs with differential PTGER4 expression along with distinct molecular characteristics. PTGER4-high PMN-MDSC subpopulations exhibited high levels of CCL4 and VEGF1, while PTGER4-low subpopulations highly expressed ARG2, IL1R2, TNFSF8, and PTGS2 (Fig. 2H). GSEA of PMN-MDSCs showed notable changes in pathways associated with leukocyte migration, IL10 production, and chemokine-mediated signaling, which were biased toward PTGER4-high PMN-MDSC subgroups (Fig. 2I), suggesting that these cells are more actively involved in immune modulation of the TME.

In view of the results of single-cell sequencing and bioinformatics analysis, we further explored the function of PGE2 on T cells, PMN-MDSCs, and prostate cancer cells through the EP4 receptor. PGE2 inhibited the production of the effector cytokine IFNγ (Supplementary Fig. S3A) and induced exhaustion accompanied by the upregulation of T-cell immunoglobulin and mucin domain–containing protein 3 (TIM-3) and PD-1 in T cells (Supplementary Fig. S3B and S3C). In addition, PGE2 promoted the development and maturation of MDSCs from BM cells (Supplementary Fig. S3D) and caused MDSCs to secrete more ARG1, iNOS, and indoleamine 2,3-dioxygenase (IDO), which play roles in inhibiting T-cell function (Supplementary Fig. S3E and S3F). Furthermore, PGE2 induced an increase in COX-2 expression in MDSCs, where positive feedback promoted the production of PGE2 (Supplementary Fig. S3G). The production of more IL10 and CXCR4 by MDSCs under PGE2 stimulation also led to the recruitment of more MDSCs in the TME (Supplementary Fig. S3H). EP4 was also highly expressed in prostate tumor cells, and while PGE2 stimulation significantly changed the chemokine profile secreted by tumor cells, it did not significantly affect cell viability (Supplementary Fig. S3I). PGE2 suppressed the secretion of CXCL9, CXCL10, and CXCL11, which usually induce the migration of T cells into the TME (Supplementary Fig. S3J). Conversely, the secretion levels of CCL2, CCL3, CCL4, CCL5, CXCL5, CXCL12, and CXCL16, which resulted in elevated MDSC recruitment to TME, were specifically upregulated (Supplementary Fig. S3K). Taken together, we confirmed the immune function of PGE2 in T cells, PMN-MDSCs, and prostate cancer cells through EP4, and identified EP4 as a specific target for immunotherapy.

Synthetic small molecule YY001 is a novel, high potent EP4 antagonist developed in our lab with an IC₅₀ value of 3.91 nmol/L in cAMP accumulation assay (47). YY001 reversed the inhibition of the effector cytokine IFNγ production induced by PGE2 with a similar level of effectiveness as E7046, an EP4 antagonist undergoing a clinical trial investigation (Fig. 3A; ref. 48). Furthermore, YY001 reversed the T-cell proliferation inhibition induced by PGE2 (Supplementary Fig. S4A) and promoted the expansion of T cells into naïve T and central memory T cells (Supplementary Fig. S4B). As shown in Supplementary Fig. S4C, the YY001 antagonism of EP4 significantly blocked the T-cell expression of PD-1 induced by PGE2, consistent with YY001 reversing the exhaustion of T cells. When T cells isolated from OT1 mice (which possess a modified TCR that targets a peptide from the protein ovalbumin (OVA)) were cocultured with tumor cells overexpressing OVA, PGE2 significantly attenuated the killing ability of T cells, while YY001 or E7046 restored and strengthened the tumor killing ability of these T cells (Fig. 3B).

PGE2 significantly stimulated the secretion of ARG1 and iNOS, which YY001 inhibited, and therefore, YY001 inhibits the function of MDSCs (Fig. 3C; Supplementary Fig. S4D). MDSC differentiation induced by PGE2 or conditional medium was also restored by YY001 treatment (Supplementary Fig. S4E–S4G). Furthermore, coculture of T cells and MDSCs demonstrated the blockade of immunosuppressive function of MDSCs by YY001 (Fig. 3D; Supplementary Fig. S4H).

In tumor cells, YY001 treatment significantly reversed the chemokine profile induced by PGE2, which impairs the migration of T cells and promotes the recruitment of MDSCs into the TME (Supplementary Fig. S5A–S5D). In addition, we found that YY001 had no significant effect on tumor cell proliferation (Supplementary Fig. S5E) but inhibited cell migration induced by PGE2 (Supplementary Fig. S5F). In addition, YY001 inhibited angiogenesis, impairing vascular endothelial cell tube formation (Supplementary Fig. S5G).

A syngeneic prostate cancer model was built using a C57BL/6-derived prostate tumor cell line, RM-1, to evaluate the antitumor activities of YY001 in vivo (Supplementary Fig. S6A). RM-1 tumors had minimal T-cell infiltration (Supplementary Fig. S6B) and significant MDSC infiltration, which has a similar immune microenvironment to that seen in patients with CRPC (Supplementary Fig. S6C). Subcutaneous tumor-bearing animal experiments demonstrated that YY001 inhibits tumor growth in a dose-dependent manner, with the maximum dose of 150 mg/kg/day, inhibiting tumor growth by approximately 50% (Fig. 3E; Supplementary Fig. S6D and S6E). Tumors were dissociated and the immune infiltrates were examined by flow cytometry. There was a significant increase in the proportion of CD45⁺ cells following the treatment with YY001 (Fig. 3F), accompanied by <10% increase in the number of CD3⁺ T cells (Fig. 3G). Moreover, the increased number of CD3⁺ cells predominantly involved CD3⁺CD8⁺ T cells, which increased from 22% of the population in control tumors to 37% in YY001-treated tumors (Fig. 3H), which was confirmed by immunofluorescence (Supplementary Fig. S6H). In addition, YY001 treatment significantly inhibited the infiltration of Gr-MDSCs (CD11b⁺LY6G⁺LY6C^low) as observed through FACS (Fig. 3J and K) and immunofluorescence (Supplementary Fig. S6I) analyses. Further analysis of gene expression in tumor tissues revealed that YY001 treatment induced a significant increase in the secretion of cytokines typically secreted by CTLs (Fig. 3I; Supplementary Fig. S6F), while the immunosuppressive cytokines produced by MDSCs were significantly reduced (Fig. 3L; Supplementary Fig. S6G).

Although we observed that YY001 monotherapy affected the balance of infiltrated CD8⁺ T cells and MDSCs in the microenvironment, the effect of YY001 on tumor growth remains limited. To explore the possible effects, we performed additional experiments. The results presented in Fig. 3M revealed that although YY001 monotherapy increased the number of infiltrated T cells, almost the same proportion of T cells existed in the PD-1–positive exhausted state, and this was confirmed by PD-1 IHC staining in the RM-1 tumor tissue (Supplementary Fig. S6J). We also used the RM-1 tumor model to test the therapeutic effect of PD-1 antibodies. As predicted, tumors showed resistance to PD-1 antibody monotherapy, and the inhibition of tumor growth was less than 50% even at doses of up to 200 μg twice a week (Supplementary Fig. S6K). Immunofluorescence staining of immune cells also showed that PD-1 antibody monotherapy could not change the frequency of CD8⁺ T cells and slightly reduced the infiltration of MDSCs in the TME, which led to the noted resistance (Supplementary Fig. S6M and S6N). The limited therapeutic effect of PD-1 antibody monotherapy may be achieved by the functional activation of the few infiltrated T cells in the TME (Supplementary Fig. S6L). Currently, many studies aim to identify some combination therapies involving PD-1 antibodies to enhance the therapeutic effect. On the basis of our findings, we speculated that the combination of an EP4 antagonist and PD-1 antibodies might have a good effect on prostate cancer.

On the basis of the previous findings, we explored the combination of an EP4 antagonist and anti–PD-1 antibodies to treat prostate cancer. The results of the syngeneic model of murine prostate cancer showed that the combination therapy overcame the tumor's resistance to anti–PD-1 antibodies with an excellent combined therapeutic effect comparable with that of E7046, which is currently undergoing clinical investigation (Fig. 4A and B; Supplementary Fig. S7A). Flow cytometry analysis of tumor tissue showed that PD-1 alone inhibited the infiltration of MDSCs but did not increase the infiltration of CD8⁺ T cells; however, when combined with YY001, the infiltration of MDSCs was significantly reduced (Fig. 4C and E). The combined effect of EP4 antagonism and anti–PD-1 therapy also increased the infiltration of CD8⁺ T cells (Fig. 4D and F), with the effect of YY001 equivalent to that of E7046. The same conclusions were obtained from immunofluorescence analysis of MDSCs and CD8⁺ T cells in tumor tissues (Fig. 4G and H). Changes in the distribution of macrophages and dendritic cells (DC) cells also contributed to the inhibition of tumor progression (Supplementary Fig. S7B–S7D). As described above, treatment with YY001 alone led to a massive infiltration of CD8⁺ T cells but limited tumor inhibition, while there was obvious resistance to treatment with PD-1 antibodies alone. In combination, however, the infiltrated CD8⁺ T cells were greatly activated, increasing IFNγ and TNFα secretion to kill tumor cells (Fig. 4I; Supplementary Fig. S6F–S6H). In contrast, the immunosuppressive function of MDSCs was further weakened by the combination treatment with YY001 and anti–PD-1 antibodies (Fig. 4J). Subsequently, chemokine profiles associated with MDSC and T-cell recruitment were examined. Consistent with our in vitro cell results, the YY001 monotherapy and combination therapy further strengthened the chemokine changes, with greater CXCL9, CXCL10, and CXCL11 secretion in cotreated tumors to recruit T cells (Fig. 4K). On the other hand, the secretion levels of CCL2, CCL3, CCL4, CCL5, CXCL5, CXCL12, and CXCL16 were more strongly inhibited, reducing the infiltration of MDSCs (Fig. 4L), when compared with control group. Pathologic analysis of the main organs from mice receiving different treatments suggested that neither of the single drugs nor their combination presented significant toxicity (Supplementary Fig. S7E). The downregulation of p-ERK in tumor tissue by treatment with YY001 or its combination with anti–PD-1 antibodies definitively demonstrated the blockage of the PGE2–EP4 signaling pathway in vivo (Supplementary Fig. S7F, S7I, and S7J).

To further validate the mouse model and assess the pattern and composition of immune cells in the TME with EP4 antagonist treatment alone or in combination with PD-1 antibodies, a 15-antibody PFC panel was developed and used to analyze the tumor samples from mouse models. All results were visualized using t-distributed stochastic neighbor-embedding (t-SNE) analyses. First, whole cells were separated into two major populations, cancer cells (CD45⁻) and leukocytes (CD45⁺). Compared with the untreated control group, the number of leukocytes was significantly increased after treatment with anti–PD-1 antibodies, YY001, or E7046 and continued to increase significantly after combination treatment with YY001 and PD-1 or E7046 and PD-1 antibodies (Fig. 5A). All leukocytes were then selected and analyzed individually to uncover the heterogeneity of the immune cell population. As shown in Fig. 5B, there were few CTLs present in the control group, and a 6-fold increase in the number of CTLs was observed after treatment with YY001 or E7046. Notably, there was a 10-fold increase in CD8⁺ T-cell number in the YY001 and anti–PD-1 antibody combination treatment group. Among MDSCs, a approximately 50% reduction in cell numbers was achieved after treatment with EP4 antagonist monotherapy or anti–PD-1 antibodies (Fig. 5C). Meanwhile, other immune cell populations were also visualized, including DCs (Supplementary Fig. S8A), macrophages (Supplementary Fig. S8B), Tregs (Supplementary Fig. S9A), and NK cells (Supplementary Fig. S5D). NK cells appeared as a separate cell population only after treatment with EP4 antagonist or combined treatment with PD-1 antibodies, implying potential tumor suppression with the combination treatment.

High CD8⁺ T-cell infiltration depends on stem-like CD8⁺ T cells residing within antigen-presenting niches in the tumor to generate terminally differentiated effector T cells. Furthermore, a higher number of intratumoral niches containing MHC-II⁺ APCs and TCF⁺CD8⁺ T cells correlated with better postsurgery PFS (49, 50). Therefore, we also examined the difference in the immune niche between tumor samples in each group. We used immunofluorescence staining to image MHC-II⁺ APCs and TCF⁺CD8⁺ T cells. As shown in Supplementary Fig. S9B–S9D, regardless of the level of CD8⁺ T-cell infiltration, we identified a few dense regions of MHC-II where TCF1⁺CD8⁺ T cells resided (APC niches) in the treatment groups, especially the combination treatment group. Segmentation and quantification of the digital slides using CellProfiler revealed higher proportions of MHC-II⁺-dense, CD8⁺-dense, and shared MHC-II⁺- and CD8⁺-dense regions in the tumor after combination treatment (Fig. 5E; Supplementary Fig. S9E), suggesting that combination therapy promoted the formation of intratumoral immune niches and prevented tumor cell escape from destruction by CD8⁺ T cells.

For further verification, we blocked CD8⁺ T cells with anti-CD8 antibodies in the animal model treated with YY001 alone or in combination with anti–PD-1 antibodies. Results indicated that when T cells were blocked by CD8 antibodies, antitumor activities were significantly impaired regardless of whether YY001 alone or in combination with PD-1 antibodies was used (Fig. 6A–C). The immune cell infiltration in the TME was analyzed, and we found that CD8 blockade did not affect the infiltration of leukocytes but affected CD8⁺ CTLs (Fig. 6D–F). These results confirmed the key role that CD8⁺ CTLs played in the treatment of prostate cancer with YY001 and its combination with anti–PD-1 antibodies. However, we also noticed that the anti-CD8 blockade did not completely inhibit the antitumor effects of YY001 alone or in combination with anti–PD-1 antibodies, suggesting that there are some immune cells, such as the NK cells discovered by our PFC analysis (Fig. 5D), that play an important role in the treatment.

To further verify the therapeutic effect of YY001 combined with anti–PD-1 antibodies on prostate tumors, we employed three additional animal models. The first model was an orthotopic prostate carcinoma model where RM-1 cells were implanted in the prostate of C57BL/6 mice, leading to the formation of primary prostate tumor and metastasis to para-aortic lymph nodes (51, 52). The second model was a rechallenged cancer model consisting of the reimplantation of RM-1 cells subcutaneously into C57BL/6 mice after tumor removal by surgery accompanied by the first-stage treatment, leading to lasting immunologic memory (53). The third model was a humanized CD34⁺ model produced by injecting CD34⁺ hematopoietic stem cells to irradiated NPSG mice and subcutaneously implanting MSK-PCa2 organoid tumor, which is derived from a patient with metastatic prostate cancer (54, 55).

As shown in Fig. 6G and H, RM-1 cells grew into a primary prostate tumor with metastasis to adjacent tissues and organs. PD-1 antibodies and YY001 partially regressed this tumor growth, while their combination almost completely inhibited the tumor progression, with a 100% survival rate for combination treatment (Fig. 6I). In the primary challenge, anti–PD-1 antibodies and YY001 partially regressed subcutaneous tumor growth, while their combination completely inhibited tumor growth after 10 days of treatment (Fig. 6J and K). Next, primary tumors were removed by surgery, and RM-1 cells were implanted for the rechallenge experiment. As shown in Fig. 6L, the rechallenge of mice with subcutaneous tumor cells resulted in 100% tumor growth in naïve control mice but only 20% in mice previously treated with the combination of YY001 and anti–PD-1 antibodies. Anti–PD-1 antibody alone had no significant effect on RM-1 tumor growth, whereas YY001 monotherapy decreased tumor growth by 50%. The survival rate was also significantly extended with continuous combination therapy (Fig. 6M). A humanized CD34⁺ mouse model was built (Supplementary Fig. S10A) to evaluate the antitumor activities of YY001 for patient-derived MSK-PCa2 organoid tumor in NPSG mice which the ratio of hCD45⁺/(hCD45⁺⁺mCD45⁺) was >20% before and after the treatment (Supplementary Fig. S10B). As shown in Fig. 6N, the combination of anti–PD-1 and YY001 exhibited an excellent combined therapeutic effect on the growth of patient-derived MSK-PCa2 prostate cancer organoid tumors, which was significantly better than any single treatment. Flow cytometry (Fig. 6O) and immunofluorescence analyses (Supplementary Fig. S10C) of the tumor tissue also showed that YY001 and combination therapy distinctly increased the infiltration of human CD45⁺ lymphocytes with a simultaneous increase in the CD8⁺ cell population. Meanwhile, the infiltration of human MDSCs was evidently prevented by YY001 or combination treatment (Supplementary Fig. S10D). These data suggest that the combination of YY001 and PD-1 antibodies provided lasting immunologic memory, prolonged the survival of mice, and protected most surviving mice from tumor rechallenge.

In this study, we have unbiasedly analyzed the single-cell transcriptomes of prostate cancer cells and the surrounding tumor microenvironment to investigate the immune system and response to immunotherapy. We demonstrated that high expression of PTGER4 in T cells is indicative of an exhausted phenotype, and discovered that PTGER4-high TAMs and MDSCs play important roles in immune suppression. Novel EP4 antagonist YY001 was used to target EP4. On the basis of the results of single-cell analysis, cytokines associated with T-cell recruitment (CXCL9, CXCL10, and CXCL11; refs. 56, 57) and MDSC recruitment (CCL2, CCL4, CCL5, CXCL5, CXCL12, and CXCL16; refs. 58–61) were selected for functional experiments. We observed significant changes in the levels of these cytokines in our study in relation to the EP4 antagonists. By altering the expression levels of these chemokines, YY001 can help tumor tissues attract T cells and reject MDSCs in vivo.

EP4 has been regarded as a potential antitumor target, which arouses interests in the development of specific small-molecule compounds for targeted therapy. Our finding that targeting EP4 is effective against cancer is consistent with results of latest clinical trials and mechanistic research on EP4. Two phase I clinical trials showed that E7046, a specific antagonist of EP4, was well-tolerated when used alone or in combination with chemoradiation and exhibited encouraging preliminary efficacy results (62, 63). Taking advantage of Ptger4 conditional knockout mice, Bonavita and colleagues (64) established a key role for PGE2 acting on EP4 in modulating the intratumoral inflammatory response and identified NK cells as drivers of a profound TME switch conducive to adaptive immunity, enabling immune evasion. Sajiki and colleagues (65) discovered that the serum PGE2 concentration was increased in bovine leukemia virus (BLV)-infected cattle that were administered the anti–PD-L1 blocking Ab, which was mediated through TNFα signaling and associated with T-cell dysfunction. Dual blockade of PD-L1 and EP4 may overcome resistance to anti–PD-1/PD-L1 therapy and candidates for a novel immunotherapy combination.

Recent advances in cancer immunology have unraveled two basic mechanisms of tumor immune evasion (45). Immunosuppressive cell populations, such as TAMs and MDSCs, can exclude T cells from infiltrating tumor tissues. In addition, infiltrating effector immune cells become exhausted and exhibit a decline in cytotoxicity. During tumor progression, tumor cells upregulate the expression of PD-L1, which leads to the exhaustion of CTLs upon PD-1 binding to the ligand (13, 14). Therefore, blocking the PD-1 pathway using mAbs against PD-1 or PD-L1 can reeducate the immune system against tumors, representing a new breakthrough in immunotherapy and promoting the discovery of more novel immune checkpoints and the development of checkpoint inhibitors (66–68). A significant fraction of patients with prostate cancer treated with androgen deprivation therapy (ADT) experience relapse and would gradually progress to lethal CRPC. However, results of clinical trials investigating the efficacy of ICBs on prostate cancer have been largely unsatisfactory (5, 6, 59, 60). Currently no ICBs are approved for prostate cancer specifically, other than cancers with high microsatellite instability, in which pembrolizumab showed modest efficacy. Lu and colleagues (69) discovered the combination therapy of ICB plus dasatinib and ICB plus cabozantinib led to lower tumor mass and fewer metastases compared with ICB alone. Moreover, they observed a decrease in the percentages of intra-tumoral PMN-MDSCs and an increase in CD8⁺ T cell/Treg ratio. This finding showed promise for MDSC suppression in combination with immunotherapy.

Our data have limitations. Given the modest number of clinical samples, clinical heterogeneity poses a challenge. All patients with CRPC had exposure to androgen deprivation therapy (ADT), and the observed difference between primary prostate cancer samples and CRPC samples could be driven by medications. However, ADT is the stand of care for prostate cancer and is predominantly used throughout the disease course, thus our findings are highly relevant for clinical translation and future studies. Different sequencing platform used for primary prostate cancer samples and CRPC samples resulted in the detection of PMN-MDSCs in only CRPC samples and defied the data integration and direct comparison between the two disease stages. Many results from single-cell analysis are speculative and warrants future experimental validation. Finally, immunomodulation is a complex process that may involve establishing and maintaining a balance between multiple factors. Blockade of EP4 may decrease the expression of certain cytokines that promote T-cell recruitment or increase the expression of others that promote MDSC recruitment. Further research and a better understanding of the nature of the balance are required.

In summary, our results highlight the importance of studying immunomodulatory pathways in prostate cancer apart from the PD-1 axis. Targeting the PGE2/EP4 axis with small-molecule antagonists, in conjuction with anti–PD-1 antibody therapy, may sensitize prostate cancers to immunotherapy, which results in lasting immunologic memory and marked tumor regression (Supplementary Fig. S10E). Our findings provide a path toward identifying therapeutic targets for immunologically “cold” tumors to reverse de novo resistance to immunotherapy.

S. Peng reports grants from National Natural Science Foundation of China, Fundamental Research Funds for the Central Universities and China Postdoctoral Science Foundation during the conduct of the study; in addition, S. Peng has a patent for CN112390814A pending and issued. W. Yu reports a patent for CN112390814A pending and issued. B. Zhang is the cofounder of Novel Bioinformatics Co., Ltd. W. Chen is the cofounder of Novel Bioinformatics Co., Ltd. T. Ding reports grants from Science and Technology Commission Fund of Shanghai Fengxian District (no. 20160907) during the conduct of the study. H. Zhang reports a patent for CN112390814A pending and issued. Z. Yi reports a patent for CN112390814A pending and issued. M. Liu reports grants from National Key R&D Program of China, National Natural Science Foundation of China, Innovation Program of Shanghai Municipal Education Commission, The Science and Technology Commission of Shanghai Municipality, and Shenzhen Municipal Government of China during the conduct of the study; in addition, M. Liu has a patent for CN112390814A pending and issued. S. Ren reports grants from National Key R&D Program of China, The National Science Fund for Distinguished Young Scholars, National Natural Science Foundation, and Shanghai Municipal Science and Technology Commission during the conduct of the study. No disclosures were reported by the other authors.

S. Peng: Conceptualization, funding acquisition, investigation, methodology, writing–original draft. P. Hu: Investigation, methodology, writing–original draft. Y.-T. Xiao: Data curation, formal analysis, visualization, writing–original draft, writing–review and editing. W. Lu: Investigation. D. Guo: Investigation. S. Hu: Investigation. J. Xie: Investigation. M. Wang: Investigation. W. Yu: Investigation. J. Yang: Investigation. H. Chen: Investigation. X. Zhang: Data curation. Y. Zhu: Investigation. Y. Wang: Investigation. Y. Yang: Data curation. G. Zhu: Data curation, formal analysis, visualization. S. Chen: Data curation, software, formal analysis, visualization. J. Wang: Investigation. B. Zhang: Data curation, formal analysis, visualization. W. Chen: Data curation, formal analysis, visualization. H. Wu: Investigation. Z. Sun: Funding acquisition, writing–review and editing. T. Ding: Investigation. H. Zhang: Investigation. Z. Yi: Conceptualization, resources, supervision, funding acquisition, project administration, writing–review and editing. M. Liu: Conceptualization, funding acquisition, project administration, writing–review and editing. S. Ren: Conceptualization, funding acquisition, project administration, writing–review and editing.

This study was supported by National Key R&D Program of China (2018YFA0507001 to M. Liu; 2017YFC0908002 to S. Ren), National Natural Science Foundation of China (81830083 to M. Liu; 81872105 to S. Ren; 81773204, 81472788, and 82073310 to Z. Yi; 81802970 to S. Peng; 81872418 to Z. Sun), Innovation Program of Shanghai Municipal Education Commission (2017–01–07–00–05-E00011 to M. Liu), the Fundamental Research Funds for the Central Universities, China Postdoctoral Science Foundation (2018M632065 to S. Peng), Shanghai Municipal Commission of Science and Technology (20YF1448100 to Y. Zhu, 11DZ2260300 to M. Liu, and 20JC1417900 to Z. Yi), Shenzhen Municipal Government of China (KQTD20170810160226082 to M. Liu), Science and Technology Commission Fund of Shanghai Fengxian District (no. 20160907 to T. Ding), and ECNU Public Platform for Innovation (011). We would like to thank Professor Dong Gao (Shanghai Institute of Biochemistry and Cell Biology, Chinese Academy of Sciences) for supplying the MSK-PCa2 organoid tumor and all those who contributed to the development, writing, and revision of this paper.

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