Reciprocal Network between Cancer Stem-Like Cells and Macrophages Facilitates the Progression and Androgen Deprivation Therapy Resistance of Prostate Cancer

Intratumor heterogeneity promotes tumor evolution and contributes to disease progression, therapeutic failure, and patient survival (1). Cancer stem-like cells (CSC), a small subset of the hierarchical organization of cells, offer an explanation for heterogeneity among cancer cells (2). In different types of tumors, CSCs are functionally defined by their strong stem-like properties including self-renewal, chemoresistance, tumor initiation upon serial passages, and metastatic potential (3, 4). Thus, elucidating the molecular mechanisms responsible for CSCs would help to develop new and promising therapies for advanced tumors in clinical practice.

Prostate cancer, with its strong heterogeneity, remains one of the most common causes of male cancer–related deaths worldwide (5). Androgen deprivation therapy (ADT), when applied to advanced and recurrent prostate cancers, achieves short-term effectiveness, but ultimately induces drug resistance, leading to increased cancer-related deaths (6, 7). Increasing evidence has indicated that ADT induces reprogramming of prostate cancer and enriches a subpopulation of cells with CSC properties, which are resistant to ADT and drive prostate cancer progression (8, 9). Therefore, eliminating CSCs may be crucial for achieving a good response of prostate cancer to ADT.

Intracellular programs including pluripotency transcription factors (OCT4, Nanog, Sox2, etc.; refs. 10, 11) and aberrant signaling pathways (Wnt/β-catenin, STAT3, NFκB, etc.; refs. 12–14) play a crucial role in maintaining the stem-like properties of CSCs. Autophagy, a highly conserved catabolic process that functions as a cell survival mechanism under external stimuli such as chemotherapy or radiotherapy, has recently been shown to support tumor cell survival, differentiation, and the self-renewal of CSCs (15, 16). However, the related mechanisms need further study.

In addition to their internal characteristics, CSCs reside in niches that support their self-renewal (17). Recent studies indicate that CSCs remodel their specific niche by recruiting monocytes and educating them to become tumor-associated macrophages (TAM; ref. 18). Furthermore, the CSC–TAM cross-talk facilitates tumor growth, metastasis, and chemoresistance (19, 20). Thus, jointly targeting CSCs and their niche components may prove to prevent tumor drug resistance and progression more effectively than targeting the CSCs alone. Although our studies and others have shown that ADT-induced transdifferentiation attracts the infiltration of TAMs and that their reciprocal network influences ADT resistance in prostate cancer (21–23), the molecular mechanism underlying CSC–TAM interaction in ADT resistance and the progression of prostate cancer has not been elucidated.

In this study, we demonstrated that a subpopulation in prostate cancer harboring CSC characteristics could be identified by the epithelial marker OV6, which serves as a CSC marker in epithelium-derived malignant tumors and was associated with patient prognosis in our previous studies (24–27). We demonstrated that an intracellular pathway, autophagy-related gene 7 (ATG7)/β-catenin/OCT4, sustained the self-renewal of OV6⁺ cells and facilitated the progression and ADT resistance of prostate cancer. In addition, IL6/STAT3 was found to be required for the reciprocal network between OV6⁺ cells and TAMs. Finally, the joint targeting of OV6⁺ CSCs and their reciprocal network effectively ameliorated ADT resistance in an orthotopic prostate cancer model.

Clinical data, tissue specimen, and follow-up information were collected from cohort one for 78 prostate cancer patients who were diagnosed in Changhai Hospital (Shanghai, China) and cohort two for 67 prostate cancer patients who were diagnosed in Changzheng Hospital (Shanghai, China) between 2012 and 2013. Patients enrolled in these two cohorts were pathologically diagnosed as prostate cancer without distant metastasis and underwent radical prostatectomy. Patients who received additional treatment such as ADT, radiotherapy, or chemotherapy were not included. Follow-up time was 42 (6–62) months. Tumor stage and Gleason Scores (GS) were assessed in terms of the American Joint Committee on Cancer (AJCC) 2002 and the World Health Organization (WHO) classification guidelines. The time to biochemical recurrence (BCR; cutoff: PSA = 0.2 ng/mL) and disease progression identified by MRI, CT, or ECT were selected as the clinical endpoint of BCR-free survival and disease-free survival, respectively. Except for the two cohorts as above, this study also included patients who were pathologically diagnosed castration-resistant prostate cancer (CRPC; n = 10) in which 5 patients' samples were taken before and after ADT and NEPC (n = 6) in Changhai Hospital (Shanghai, China). The samples were obtained after writing informed consent from patients according to an established protocol approved by the Ethics Committee of Second Military Medical University.

The IHC was done as reported previously (24). Primary antibodies were used as follows: mouse anti-OV6 (MAB2020, R&D Systems), rat anti-F4/80 (ab6640, Dako), mouse anti-CD68 (M0876, Dako), and rabbit anti-ATG7 (ab52472), mouse anti-β-catenin (ab22656), mouse anti-OCT4 (ab184665), and rabbit anti-STAT3 (ab68153) from Abcam, respectively. The protein expression was score by staining intensity and percentage of positively stained cells as reported previously (28). Hematoxylin and eosin (H&E)-stained sections of the prostate cancer specimens were reevaluated by two experienced pathologists (Jun-hui Ge and Yong-wei Yu, Second Military Medical University, Shanghai, China) to identify representative areas in double blind procedure. The percentage of positive cells (% of PPs) and the staining intensity (SI value) were determined and multiplied (IRS value), and the score range is from a minimum score of 0 to a maximum score of 12. An IRS value more than one was considered as positive (weak expression); an IRS value more than three, moderate expression; and an IRS value more than eight, strong expression. The quantitative cutoff for each marker was defined: low expression (IRS value 0–3 including negative and weak expression) and high expression (IRS value 4–12 including moderate and strong expression).

The immunofluorescence analysis was carried out as reported previously (23). The primary antibodies were used as follows: mouse anti-OV6 (MAB2020, R&D Systems), rabbit anti-CD44 (ab51037, Abcam), rabbit anti-CD133 (ab19898, Abcam), rabbit anti-STAT3 (ab68153, Abcam), rabbit anti-β-catenin (ab32572, Abcam). Nuclei were stained by DAPI (E607303, Sangon Biotech) staining. Fluorescence images were observed and collected under a Leica DM5000B fluorescent microscope (Leica).

THP-1, obtained from the Cell Bank of Type Culture Collection of the Chinese Academy of Sciences (Shanghai, China), was maintained in RPMI1640 medium (C11875500CP, Gibco) supplemented with FBS (10%, 10099-141, Gibco) and 0.05 mmol/L β-mercaptoethanol (07604, Sigma). C4-2B and C4-2 were kindly provided and authenticated by Dr. Leland Chung (Cedars-Sinai Medical Center, Los Angeles, CA). C4-2B, C4-2, or C4-2B Cas9-ATG7-gRNA (ATG7^KO) cells were cultured in RPMI1640 medium (11835093, Gibco). C4-2B Cas9-gRNA NC cell line was maintained in RPMI1640 medium (11835093, Gibco), 10 μg/mL blasticidin (S7419, Selleck Chemicals), and 0.5 μg/mL puromycin (A610593, Sangon Biotech). All cell lines were supplemented with 1% penicillin/streptomycin (15140122, Gibco) and cultured at 37°C and 5% CO₂.

The cell lines in this study were authenticated by short tandem repeat (STR) profiling and tested for mycoplasma contamination by Mycoplasma Detection Kit (B39032, Selleck Chemicals). The latest test was performed in January 2018. The cell lines used in this study were within 40 passages. Permanent stocks of the cell lines were prepared and stored in liquid nitrogen until use.

Spheroid formation assay was carried out as described in our previous study (24), Briefly, after magnetic sorting, single-cell suspensions with 5,000 cells were seeded in 6-well ultra-low attachment culture plates (Corning) cultured in serum-free DMEM/F12 (Gibco), supplemented with B27 (1:50, Invitrogen), 20 ng/mL EGF (Invitrogen), 10 ng/mL basic fibroblast growth factor (Invitrogen), and ITS (1:100, Gibco) for 5 days. The number of spheroids formed was counted under a microscope and the representative pictures were taken.

The CRISPR-targeting sequence in this study, as shown in Supplementary Table S14, was designed on the basis of the Optimized CRISPR Design web tool (http://crispr.mit.edu/) for knockout of ATG7. C4-2B cells were transduced with sgATG7 cloned into lenti Cas9-Blast (52962, Addgene), lentiGuide Puro vector (52963, Addgene). Cells were selected with blasticidin puromycin. First, C4-2B cells were transduced with lentivirus Cas9 and were selected with 20 μg/mL blasticidin for four days. In each transduction, 100-mm dish was used to seed 500 blasticidin-resistant cells and cultured until cell colony formation. Individual colonies were shifted to 12-well plates as one colony per well and grown to confluence. Second, C4-2B Cas9–stable cell line was transduced with lentiviruses sgRNA and were selected with 1 μg/mL puromycin. PCR or RT-PCR was used to identify clones. A TOPO vector was used to package genomic PCR products to sequence alleles with indels or deletions. Finally, Western blot analysis was used to validate the ATG7 protein expression.

Gene knockdown, RNA interference, and plasmid transfection were done as reported previously (23). The C4-2B or C4-2 was cultured in 6-well plates, inoculated at a density of 5 × 10⁴ cells/mL, and transfected with the shRNAs expressing lentivirus (sh-ATG7, shβ-catenin, sh-OCT4, sh-STAT3) or control lentivirus at a multiplicity of infection (MOI) of 45. After 72-hour transfection, they were observed and photographed under microscope. The sequences for shRNA were presented in Supplementary Table S14.

siRNA and plasmid transfection was carried out using Lipofectamine 3000 reagents (L3000015, Invitrogen) according to the manufacturer's protocols. The sequences of si-Beclin1, si-ATG5, and ATG7 plasmid are shown in Supplementary Table S14.

The isolation of circulating monocytes was done as reported previously (23), The C4-2B or C4-2 cells' coculture with circulating monocytes or THP-1 was the same as in our previous study (23). After 5-day coculture in the 37°C, 5% CO₂ condition, both the supernatant and cells were collected for subsequent analysis.

The nano-scale liquid chromatography tandem electrospray ionization mass spectrometry (NanoLC-ESI-MS/MS) analysis was performed as reported previously (23). For the identified proteins reported here, the certainty should be >98% if the identification is based on LC/MS-MS sequencing of one peptide and >99.9% if based on the sequencing of two or more peptides.

The magnetic-activated cell sorting (MACS) assay was performed with a MiniMACS Cell Sorter (Miltenyi Biotec) as reported in previous study (24). Flow cytometric assay was performed with a Cyan ADP Sorter (Beckman Coulter). Prostate cancer cell lines were labeled with APC-conjugated-OV6 antibody (FAB2020A, R&D Systems), FITC-conjugated CD44 antibody (130-095-195, Miltenyi Biotec), PE-conjugated-CD133 (130-098-826, Miltenyi Biotec), PE-conjugated-CD14 antibody (REA599, Miltenyi Biotec).

Apoptotic cells were evaluated by ANXA5 and PI staining (Invitrogen) according to the manufacturer's instructions, and analyzed by flow cytometry with a Cyan ADP Sorter (Beckman Coulter)

RNA was isolated from OV6⁻, OV6⁺ without and with ATG7-knockdown C4-2B cells with TRIzol reagent (Invitrogen), The total RNA was purified by the Qiagen RNeasy Mini Kit (Qiagen) and then the purified RNA was checked to determine the quantity. Single and double stranded cDNA were synthesized from mRNA samples. The double-strand cDNA was then purified for dA tailing, end repair, adaptors ligation, and DNA fragment enrichment. Quantify the libraries using Qubit (Invitrogen) based on the Qubit user Guide. The constructed library was sequenced on Illumina Hiseq 4000 sequencer.

The TopHat version 1.2.0 was used to align the paired-end raw reads that allows two mismatches in the alignment. Cufflinks version 2.0.0 was used to assemble the aligned reads into transcripts using. The distribution of the reads and the alignment quality were estimated using SAM tools. From the aligned reads, the gene and transcript expression was performed using cufflinks version 1.3.036. The differential transcripts were analyzed via cuffdiff. Finally, GO and pathway functional analyses were performed for the differentially expressed transcripts.

The RT-PCR was done as reported previously (24). Briefly, RNAiso Plus (9109, Takara) and PrimeScript One Step RT reagent Kit (RR037B, Takara) were used to extraction and reverse transcription of RNA. The results were normalized by the expression of the β-actin gene and each measurement was performed in triplicate. Fold change relative to control group was determined by 2^−ΔΔC_t. The primer sequences were presented in Supplementary Table S14.

The Western blot analysis was done as described in our previous study (24) and the following primary antibodies were used rabbit anti-Beclin1 (3495), rabbit anti-ATG5 (12994), rabbit anti-LC3B (3868), rabbit anti-β-actin (4970) from Cell Signaling Technology, and rabbit anti-STAT3 (ab68153), mouse anti-OCT4 (ab184665), rabbit anti-P-STAT3 (ab76315), mouse anti-β-catenin (ab22656), rabbit anti-β-TrCP (ab71753) from Abcam, overnight in the blocking solution at 4°C. Then, the membranes were incubated with anti-rabbit IgG-HRP–linked antibody (7074S) or anti-mouse IgG-HRP–linked antibody (7076S) from Cell Signaling Technology at room temperature for 1–2 hours (1: 5,000), respectively. To analyze ATG7 and β-catenin protein interactions, coimmunoprecipitation experiments were performed using OV6⁺ C4-2B cells according to previously published protocols (24) and the following primary antibodies were used: rabbit anti-ATG7 (1:100, ab52472) and mouse anti-β-catenin (ab22656) from Abcam. β-Actin protein level was used as an internal control.

The cell migration ability of sorted monocytes and THP-1 was evaluated by cell migration assay as we reported in previous study (23). The proliferation of C4-2B and C4-2 cells in indicated condition was evaluated using a CCK-8 kit (CK-04, Dojindo) as we reported in previous study (23). The proliferation rates were presented as a proportion of the control value which was detected at the first time point.

The number of autophagosomes in OV6⁻ and OV6⁺ prostate cancer cells was analyzed by electron microscopy (Hitachi-7650). The mRFP-GFP-LC3B adenovirus construct was provided by Hanbio Inc. (autophagy-adv-GFP-RFP-LC3B-1000). This construct fluorescence in terms of the difference in pH between the neutral autophagosome and the acidic autolysosome, and exhibited red fluorescence or the red/green (yellow) makes it possible to monitor progression of autophagic flux. Confocal fluorescence microscopy was used to scan and assess the fluorescence.

To determine the association of transcription factor β-catenin and OCT4 promoter, ChIP assays were performed according to previously published protocols (23). Primers complementary to the promoter region of OCT4 (forward: 5′- GCCCATTCAAGGGTTGAGCACT-3′; reverse: 5′- GGTTCAAAGAAGCCTGGGAGGG -3′) were used to detect OCT4 genomic DNA and primers specific to human GAPDH promoter was used as control (kit supplied). Enrichment of the targets was calculated as follows: fold enrichment = 2^(C_t^{[OCT4−ChIP] −C}_t^[IgG]).

We detected the interaction between β-catenin and ATG7 using the Duolink PLA according to previously published protocols (23).

The interaction between β-catenin and the OCT4 promoter was evaluated by using a luciferase reporter assay as we reported in previous study (23). The β-catenin–binding sites of the OCT4 promoter (sequence: CTTTGAA, −1302 to −1308 relative to the OCT4 transcription site) or its deletion mutant was cloned into the pGL3-basic luciferase reporter vector (Promega). Cells were collected 48 hours after transfection and OCT4 transcription activity was evaluated by measuring luminescence with the Dual-Luciferase Assay Kit (E1910, Promega). Fold induction was derived relative to normalized reporter activity.

The antibody–microarray experiment was done as we previously reported (23). Cytokine profiles were measured by Quantibody Human Inflammatory Array 3 (QAH-INF-G3, RayBiotech) which permitted detection of 40 inflammation-associated cytokines.

The IL6 or PSA levels in cell culture medium or serum were measured using ELISA Kit for IL6 (SEA079Hu, Cloud-Clone Corp), PSA Quantikine ELISA Kit (DKK300, R&D Systems), respectively, according to the manufacturer's instructions.

All experimental animal procedures were approved by the Animal Care and Use Committee of the Second Military Medical University, Shanghai, China. NOD-SCID mice (male, 7 weeks old) were purchased from the Shanghai Laboratory Animal Center (SLAC) and housed under specific pathogen-free conditions. The orthotopic prostate cancer xenograft assay was done as previously reported (23). A total of 5 × 10⁵ OV6⁺ C4-2B cells or C4-2 cells mixed with Matrigel (1:1) were orthotopically injected into both anterior prostates of NOD-SCID mice through an abdominal incision. Three weeks after the cell injections, the animals were grouped so that the average bioluminescence index was similar in the indicated groups. These mice in treatment group were treated with castration/enzalutamide (30 mg/kg, twice a week) by intraperitoneal injection, combined with or without tocilizumab (5 μg/mL) and ATG7^KO, while the control group was treated with vehicle. Bioluminescence data were quantified using IVIS Lumina II imaging system (PerkinElmer).

For in vivo limiting dilution assay, sorted OV6⁺, OV6⁻, OV6⁺CD44⁺, OV6⁺CD44⁻, OV6⁻CD44⁺ tumor cells were diluted in appropriate cell dose and injected in NOD/SCID mice, the number of tumors formed from each cell dose injected was scored. According to the ELDA software (http://bioinf.wehi.edu.au/software/elda/index.html; the Walter and Eliza Hall Institute). The frequency of CSCs was calculated.

Categorical data were presented as number (%) and continuous data were presented as median (range). Numerical data were expressed as the mean ± SD. Statistical differences between variables were analyzed by two-tailed Student t test, categorical/binary measures were assessed by χ² test or Fisher exact test and ANOVA for continuous measures. Survival curve was plotted by the Kaplan–Meier method and compared using the log-rank analysis. Difference was considered significant at P < 0.05. All experiments for cell cultures were performed independently at least three times and in triplicate each time. Data analysis was performed by the GraphPad Prism 5 (GraphPad Software, Inc.) and SPSS 22.0 (IBM Corporation).

On the basis of our previous studies (24–27), we postulated that OV6 was associated with the heterogeneity and progression of prostate cancer. By IHC, the positive staining for OV6 in normal prostate basal cells suggests that OV6 is an epithelial marker (Fig. 1A), which is consistent with the results of our previous studies (24–27). Samples from locally advanced prostate cancer (LAPC), CRPC, and neuroendocrine prostate cancer (NEPC) exhibited higher OV6 expression than localized prostate cancers (Fig. 1A). In prostate cancers with different Gleason Scores (GS), higher OV6 expression was found in samples with GS > 7 than in samples with GS < 7 (Fig. 1B). Even in the same sample, staining showed stronger OV6 expression in worse differentiated areas (Fig. 1B). In addition, a positive correlation between the percentage of OV6 and GS was observed in fresh clinical samples assessed by immunofluorescence analysis (Fig. 1C; Supplementary Fig. S1A). These results indicate that the differential expression of OV6 predicts the degree of differentiation and the progression of prostate cancer.

We next examined the relationship between the clinical outcomes of prostate cancer patients and OV6 expression. Patients from two independent clinical centers were respectively divided into OV6^high and OV6^low groups according to OV6 expression in samples of prostate cancer (Supplementary Fig. S1B; Supplementary Table S1 and S2). The OV6^high group exhibited worse BCR and disease-free survival (DFS; Fig. 1D and E; Supplementary Table S1 and S2). In addition, we investigated whether OV6 was associated with androgen deprivation therapy (ADT) resistance in prostate cancer. As expected, higher OV6 expression was observed in samples from the same patient with metastatic prostate cancer complicated by ADT failure than in the corresponding tissues before ADT (Fig. 1F; Supplementary Fig. S1C). Accordingly, OV6 expression was higher in ADT-treated xenografts than in naïve ones from a mouse model of orthotopic prostate cancer (Fig. 1G; Supplementary Fig. S1D–S1F), which was described in our previous study (23).

To summarize, our findings indicate that OV6 is a potent marker for poor differentiation of prostate cancer and effectively evaluates the prognosis and disease progression of prostate cancer patients.

Given that CSCs offer an explanation for tumor heterogeneity and progression, we next examined whether OV6 served as a putative CSC marker in prostate cancer. First, higher expression of OV6 was observed in sphere-forming prostate cancer cells (with CSC properties; ref. 29) than in adherent cells, which was assessed by flow cytometry and immunofluorescence analysis (Fig. 2A; Supplementary Fig. S2A and S2B). Second, multimarker analyses revealed that OV6⁺ cells in prostate cancer cell lines C4-2B also expressed CD44 or CD133, and the percentage of double positive cells (OV6⁺CD44⁺ or OV6⁺CD133⁺) was increased in spheres (Supplementary Fig. S2C). Third, the colocalization of OV6 with CD44 or CD133 was observed in spheres from C4-2B (Supplementary Fig. S2D). These data suggest that the pattern of OV6 expression in prostate CSCs may be similar to that of CD44 or CD133.

Then, we compared the stem-like properties of OV6⁺ and OV6⁻ cells separated from prostate cancer cell lines via MACS (Supplementary Fig. S2E). First, higher expression of stem-associated genes was observed in OV6⁺ cells than in OV6⁻ cells (Fig. 2B; Supplementary Fig. S2F). Second, OV6⁺ cells formed more spheres in serial passages than OV6⁻ cells (Fig. 2C; Supplementary Fig. S2G). Third, OV6⁺ cells initiated tumors with a higher incidence and frequency in NOD/SCID mice in two serial generations (Fig. 2D; Supplementary Fig. S2H and S2I; Supplementary Table S3). In addition, OV6⁺CD44⁺ C4-2B cells exhibited higher incidence and frequency of tumor initiation than OV6⁻CD44⁺ cells in NOD/SCID mice, but there is no statistical difference in tumorigenicity between OV6⁺CD44⁺ and OV6⁺CD44⁻ cells (Supplementary Fig. S2J; Supplementary Table S4). The results indicate that OV6 may play a stronger role than CD44 in the tumor initiation of CSCs. Furthermore, we examined whether OV6⁺ cells exhibited heightened resistance to chemotherapy or ADT. After the administration of docetaxel (a chemotherapy drug) or enzalutamide (an androgen receptor antagonist), the ratio of OV6⁺ cells in prostate cancer was elevated (Supplementary Fig. S2K). We also observed that OV6⁺ cells presented heightened cell survival or decreased apoptosis during docetaxel or enzalutamide treatment, compared with OV6⁻ cells (Fig. 2E–H; Supplementary Fig. S2L–S2O). Collectively, these data demonstrate that OV6⁺ cells exhibit strong stem-like abilities in prostate cancer.

To examine the mechanisms underlying the regulation of OV6⁺ cells, RNA-seq was employed. Among the differentially expressed genes, we noticed that many stem or autophagy-related genes exhibited increased expression in OV6⁺ cells, which possessed relative quiescence (Fig. 2I; Supplementary Fig. S2P–S2T; Supplementary Table S5–S7). As autophagy has been reported to play crucial roles in CSCs, we examined whether autophagy mediated the stem-like properties of OV6⁺ CSCs. First, electron microscopy and Western blot analysis demonstrated that more autophagosomes and increased LC3BII was observed in OV6⁺ CSCs than in OV6⁻ cells (Fig. 2J and K). In addition, an autophagosomal–lysosomal fusion process was activated to a greater extent in OV6⁺ CSCs, suggesting a heightened autophagic flux (Fig. 2L). Second, after the treatment of autophagy inhibitor 3-methyladenine (3-MA), OV6⁺ CSCs exhibited decreased stem-associated genes and self-renewal compared with the control naïve OV6⁺ CSCs (Fig. 2M and N; Supplementary Fig. S2U and S2V). Furthermore, OV6⁺ CSCs subjected to 3-MA treatment initiated tumors in NOD/SCID mice with a lower efficiency and lower percentage of OV6 in two serial generations than OV6⁺ CSCs without 3-MA treatment (Fig. 2O–P; Supplementary Fig. S2W).

In addition to above results that autophagy was required in OV6⁺ CSCs, RNA-seq and Western blot analysis both exhibited that autophagy-related gene 7 (ATG7) was elevated in OV6⁺ CSCs (Fig. 2I and K). Thus, we established ATG7-deficient cell lines (C4-2B-ATG7^KO) using CRISPR/Cas9 technology to examine whether ATG7 regulated OV6⁺ CSCs (Supplementary Fig. S3A). Notably, silencing ATG7 reduced the expression of stem-associated genes and the self-renewal and tumorigenicity of OV6⁺ CSCs (Fig. 2M–O; Supplementary Fig. S2U–S2W). In contrast, the upregulation of ATG7 enhanced the stem-like properties of OV6⁺ CSCs (Fig. 3A–C; Supplementary Fig. S3B–S3F). In addition, after knockdown of other autophagy-related genes Beclin1 or ATG5 in prostate cancer, the expression of stem-associated genes and the self-renewal of OV6⁺ CSCs were not inhibited (Supplementary Fig. S3G–S3I). Therefore, ATG7 was required for sustaining the expansion and self-renewal of OV6⁺ CSCs in prostate cancer.

Next, to elucidate the mechanisms underlying ATG7-mediated OV6⁺ CSCs, RNA-seq was applied to ATG7-knockdown and control OV6⁺ CSCs (Fig. 3D; Supplementary Fig. S3J–S3L; Supplementary Table S8–S10). We noticed that the expression levels of stem-associated genes and the pathways they participate in were reduced in ATG7-knockdown CSCs, and in particular, OCT4 was significantly downregulated (Fig. 3D; Supplementary Fig. S3L). In addition, OCT4 knockdown alleviated the ATG7-promoting stem-like characteristics of OV6⁺ CSCs (Fig. 3A–C; Supplementary Fig. S3B–S3E and S3M). Furthermore, ATG7 upregulation enhanced OCT4 expression and transcriptional activity (Fig. 3A, E, and F). Thus, OCT4 was required for the regulation of the stem-like characteristics of OV6⁺ CSCs by ATG7.

Furthermore, we next examined how ATG7 regulated OCT4 expression and transcription in OV6⁺ CSCs. Knowing that ATG7 was not a transcription factor, we assumed that OCT4 transcription was mediated by an indirect mechanism. Then, we immunoprecipitated the ATG7 protein with an anti-ATG7 antibody and analyzed the results by Nano LC-ESI-MS/MS (Fig. 3G; Supplementary Table S11). Among the ATG7-interacting proteins (Fig. 3G; Supplementary Table S11), β-catenin had been reported to promote OCT4 transcription (30). Co-immunoprecipitation (co-IP) and Duolink proximity ligation assays also demonstrated that the direct interaction between ATG7 and β-catenin occurred in OV6⁺ CSCs (Fig. 3H and I). But ATG7 knockout interrupted their interaction (Fig. 3I). These results prompted us to examine how ATG7 facilitated OCT4 transcription via β-catenin in OV6⁺ CSCs. First, β-catenin knockdown alleviated the ATG7-promoting stem-like characteristics of OV6⁺ CSCs (Fig. 3A–C; Supplementary Fig. S3B–S3E, S3N). Second, β-catenin knockdown alleviated the ATG7-induced increase in OCT4 expression (Fig. 3A and E). Third, a ChIP assay showed that β-catenin bound to OCT4 promoter (the sequence was reported in previous studies; ref. 30) in OV6⁺ cells, and this binding was enhanced by ATG7 overexpression (Fig. 3J). In addition, although ATG7 enhanced OCT4 transcriptional activity, mutated variants of the β-catenin–binding sites on OCT4 promoter abolished this effect (Fig. 3F). These results indicate that ATG7 mediates OCT4 expression and transcription via β-catenin.

Given that β-catenin is phosphorylated by a cytoplasmic destruction complex, then ubiquitinated by β-transducin repeat-containing E3 ubiquitin protein ligase (β-TrCP) and finally degraded by proteasomes, and given that stabilized β-catenin enters the nucleus to perform its biological function (31), we next determined how ATG7 mediated β-catenin stabilization in the cytoplasm. As shown in Supplementary Fig. S3O, ATG7 promoted the interaction between β-catenin and Axin1 or β-TrCP, while ATG7 silencing achieved the opposite effect. In addition, ATG7 facilitated β-catenin to enter the nucleus, but ATG7 knockdown decreased β-catenin expression in the nucleus of OV6⁺ cells, as shown in Fig. 3K. The results indicate that ATG7 facilitates stem-like properties and OCT4 transcription via β-catenin stabilization.

On the basis of the above results, we next examined whether ATG7 or OCT4 inhibition could reverse the ADT resistance of OV6⁺ CSCs in prostate cancer. As expected, in vitro experiments showed that ATG7 or OCT4 inhibition enhanced the effects of enzalutamide or abiraterone (another ADT drug) on OV6⁺ CSCs (Fig. 4A and B). However, the same intervention measures did not effectively inhibit ADT resistance of OV6⁺ CSCs in an orthotopic prostate cancer model although enzalutamide induced ATG7 expression (Fig. 4C–E; Supplementary Fig. S4A–S4D). The difference between in vivo and in vitro experiments might ultimately lie in the role of the microenvironment, which has been reported to support the self-renewal of CSCs (17). In addition, our studies and others have demonstrated that tumor-associated macrophages (TAM) facilitate enzalutamide resistance in prostate cancer (21, 23) Accordingly, samples from ADT-treated OV6⁺ CSCs-derived orthotopic prostate cancers without or with ATG7 (or OCT4) inhibition still presented high OV6 expression and F4/80⁺ TAM infiltration (Supplementary Fig. S4A–S4C).

Thus, we examined the interaction between OV6⁺ CSCs and TAMs in prostate cancer. First, a positive correlation between OV6 and CD68 was observed in prostate cancer patients (Fig. 4F; Supplementary Fig. S4E). Furthermore, patients with high expression levels of both OV6 and CD68 exhibited the worst BCR and DFS (Fig. 4G and H; Supplementary Table S12). Second, freshly isolated healthy donor circulating monocytes (Supplementary Fig. S4F) and the human monocytic cell line THP-1 were exposed to conditioned medium (CM) from OV6⁻ or OV6⁺ cells, and their migration properties were tested. A higher number of monocytes migrating toward OV6⁺ CM than toward OV6⁻ CM was recorded (Fig. 4I). Fourth, in the presence of OV6⁺ CM, the monocytes expressed a higher percentage of M2 markers and decreased M1 phenotype genes than monocytes exposed to OV6⁻ CM (Fig. 4J). These results indicate that OV6⁺ cells can recruit monocytes and educate them toward TAMs.

To determine whether OV6⁺ cell-educated TAMs facilitated the stem-like properties of OV6⁺ CSCs, a coculture system was employed. After the coculture of TAMs and OV6⁺ CSCs, stem-associated gene expression and sphere-forming capacity were enhanced significantly (Fig. 4K–L). In addition, TAMs enhanced the enzalutamide resistance of OV6⁺ CSCs (Fig. 4M). Therefore, a reciprocal interaction between OV6⁺ CSCs and TAMs occurs in prostate cancer and influences ADT resistance.

To search for the inflammatory factors facilitating the interaction between OV6⁺ CSCs and TAMs, the cytokine profiles in the CM from monocytes, OV6⁺ CSCs alone, or cocultured monocytes/OV6⁺ CSCs were analyzed by a RayBio Human Cytokine Antibody Array (Fig. 5A–E; Supplementary Fig. S5A and S5B; Supplementary Table S13). Compared with the CM from monocytes or OV6⁺ CSCs cultured alone, the CM from cocultured monocytes/OV6⁺ CSCs presented upregulated cytokines (Fig. 5A and B; Supplementary Table S13). We then compared the upregulated proteins in both comparable groups and, by using a Venn diagram, found that 13 proteins were shared (Fig. 5C). To select the important upregulated proteins, we used metrics, combining both P value and fold change, to evaluate the protein relevance rank. The distribution of the rank score for each candidate upregulated protein is shown in Fig. 5D. In addition, we also used rank scores as filtering metrics to analyze the shared upregulated proteins. We used a rank score greater than the 75% quantile as the upregulation cutoff, and then the shared upregulated proteins were obtained by the intersection of the proteins via P value or rank scores (Fig. 5E). Finally, we obtained 8 shared upregulated proteins as candidates (Fig. 5E) and performed pathway analysis, which also indicated that IL6 signaling pathway was activated (Supplementary Fig. S5A and S5B). In conclusion, IL6 is identified as the crucial factor in the interaction between OV6⁺ CSCs and TAMs.

On the basis of the ELISA results, IL6 secretion was elevated in the CM from cocultured TAMs/OV6⁺ CSCs compared with the CM from monocytes or OV6⁺ CSCs alone (Fig. 5F). In addition, the addition of a neutralizing antibody against IL6 to the CM of cocultured TAMs/OV6⁺ CSCs suppressed the recruitment and education of TAMs by OV6⁺ CSCs (Fig. 5G–J). Furthermore, treatment with IL6 antibody also inhibited the ability of TAMs to facilitate the stem-like properties of OV6⁺ cells, including stem-associated gene expression, self-renewal, and tumorigenicity (Fig. 5K–M). IL6 exerts its biological effects by binding to IL6 receptor (IL6R) and triggers STAT3 activation (32). As expected, the FDA-approved drug tocilizumab (an IL6R inhibitor) or STAT3 knockdown also alleviated the promotion of the stem-like properties of OV6⁺ C4-2B cells by TAMs (Fig. 5K–M; Supplementary Fig. S5C). In addition, tocilizumab (an IL6R inhibitor) abated the increased protein level of p-STAT3 and nuclear STAT3 expression in OV6⁺ C4-2B cells by TAMs (Supplementary Fig. S5D and S5E). Thus, IL6/STAT3 is required for the reciprocal interaction between OV6⁺ CSCs and TAMs in prostate cancer.

On the basis of above results, we examined whether a strategy targeting OV6⁺ CSCs and their network with TAMs reversed the ADT resistance of OV6⁺ CSCs in vivo. OV6⁺ C4-2B or C4-2 cells were injected into the prostates of mice, and all mice were randomly divided into five groups (Fig. 6A; Supplementary Fig. S6A). As shown in Fig. 6A–C and Supplementary Fig. S6A–S6C, enzalutamide-treated mice became resistant to ADT, and ATG7 silencing or IL6R inhibition alone achieved limited suppression of tumor growth and the enzalutamide resistance of OV6⁺ CSCs. However, jointly targeting OV6⁺ CSCs and their interaction with TAMs by ATG7 silencing and tocilizumab effectively inhibited tumor growth and improved the enzalutamide resistance of OV6⁺ CSCs in vivo (Fig. 6A–C; Supplementary Fig. S6A–S6C). In addition, the levels of OV6 and F4/80 were elevated in the enzalutamide-treated xenografts and targeting only ATG7 silencing or only IL6R inhibition partly attenuated the expression of OV6 and F4/80 (Fig. 6A; Supplementary Fig. S6D–S6H). However, the combination of the two treatments exerted better therapeutic effects on enzalutamide resistance of OV6⁺ CSCs in vivo (Fig. 6A; Supplementary Fig. S6A, S6D–S6E).

Tumor heterogeneity presents a great challenge in cancer treatment, as tumor cell populations evolve and adapt to the stress of the external environment (1, 2). CSCs, an important example of heterogeneity, are well known to survive many therapies including chemotherapy, radiotherapy and endocrine therapy (33, 34). Thus, the therapeutic eradication of existing CSCs may be a promising policy to overcome treatment resistance. Although many studies have examined the mechanisms driving the expansion and self-renewal of CSCs, the effect of scavenging CSC populations in vivo is not as ideal as in vitro. In our study, we found that an intracellular pathway, ATG7/β-catenin/OCT4, sustained the self-renewal of OV6⁺ CSCs and facilitated ADT resistance in prostate cancer. As expected, blocking the pathway by ATG7 or OCT4 inhibition impaired the stemness of CSCs and reversed the ADT resistance of OV6⁺ cells in vitro. However, experiments revealed that ATG7 or OCT4 knockdown treatment failed to inhibit ADT resistance in an orthotopic prostate cancer model.

In fact, the specific microenvironment or niche should receive equal consideration in eradicating CSCs and reversing the drug resistance of tumors (17). CSCs reside in the microenvironment and remodel their specific niche by recruiting mesenchymal stem cells, inflammatory cells, or immune cells (17). In addition, educated cells within the CSC niche produce factors that reciprocally facilitate the expansion and self-renewal of CSCs and promote tumor cell growth, invasion, and metastasis (35). Therefore, a better understanding of the biology and niche components of CSCs as well as their network signaling is beneficial for effectively eliminating CSCs and improving the therapeutic effect on tumors in clinical practice. Among the niche components, TAMs, which are derived from monocytes/macrophages, play a protumor role and correlate with poor prognosis in patients with various cancers (36, 37). Although the interaction between CSCs and TAMs has been reported to promote the growth, progression, and stem-like characteristics of tumors, understanding of the in-depth mechanisms and a combined CSC-TAM targeting policy in prostate cancer remain lacking and need further study.

Androgen deprivation therapy (ADT) is a crucial treatment for advanced or metastatic prostate cancer. Despite the effectiveness of ADT in the initial stage, drug resistance is inevitable due to the induced heterogeneity and the microenvironment. A recent study indicates that targeting colony-stimulating factor 1 receptor (CSF1R) can reverse TAM-mediated resistance to ADT in prostate cancer (21). In addition, our study demonstrates that blockade of the association between ADT-induced transdifferentiation and TAMs inhibited enzalutamide resistance in an orthotopic prostate cancer model (23). In this study, CSCs and their interaction with TAMs in the niche were both considered. Combined targeting of OV6⁺ cells and the network signaling between OV6⁺ cells and TAMs ameliorated ADT resistance in an orthotopic prostate cancer model. In conclusion, our study indicates that targeting CSCs and their niche may prove to be a more powerful strategy than targeting the CSCs alone, which provides a rational approach to ameliorating ADT resistance in prostate cancer (Fig. 6D). Our future studies will further elucidate the interaction and network signaling between the heterogenicity of prostate cancer and its microenvironment, which will assist in searching for the critical target, overcoming ADT resistance and improving the prognosis of patients with advanced prostate cancer.

No potential conflicts of interest were disclosed.

Conception and design: C. Wang, X.-G. Cui, Y.-H. Sun

Development of methodology: H. Huang, C. Wang, F. Liu, G. Peng, Y.-Q. Cheng, D. Liu

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): H. Huang, C. Wang, F. Liu, H.-Z. Li, G. Peng, X. Gao, K.-Q. Dong, H.-R. Wang, D.-P. Kong, K.-J. Wang, Z. Zhou, X.-G. Cui, Y.-H. Sun

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): H. Huang, C. Wang, F. Liu, K.-Q. Dong, L.-H. Dai, Z. Zhou, J. Yang, Z.-Y. Yang

Writing, review, and/or revision of the manuscript: H. Huang, C. Wang

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C. Wang, F. Liu, M. Qu, Y.-Q. Cheng, Q.-Q. Tian

Study supervision: X. Gao, M. Qu, C.-L. Xu, D.-F. Xu, X.-G. Cui, Y.-H. Sun

The authors thank Dr. Leland Chung (Cedars-Sinai Medical Center, Los Angeles, CA) for providing prostate cancer cell line C4-2B and C4-2. They also thank Dr. Zhe-wei Wang at Translational Medicine Center, Changzheng Hospital, the Second Military Medical University (Shanghai, China) for flow cytometry analysis. The authors are grateful for technical support from the Department of Biophysics of Second Military Medical University (Shanghai, China) for electron microscopy. This work was supported by the National Natural Science Foundation of China (no. 81472397, 81773154, 81772747, 81772720, and 81301861), National Key Basic Research Program of China (973 Program, no. 2012CB518306), Research Program of Science and Technology Commission of Shanghai Municipality (no. 14411950100), the Program for Shanghai Municipal Health and Family Planning Commission Important Diseases Joint Research Project (no. 2013ZYJB0101), Shanghai Natural Science Foundation of China (no. 13ZR1450700), Innovation Program of Shanghai Municipal Education Commission (no. 2017-01-07-00-07-E00014), the Shanghai Medical Guidance (Chinese and Western Medicine) Science and Technology Support Project (no. 17411960200), Key Construction Project of Zhangjiang National Innovation Demonstration Zone the National New Drug Innovation Program (no. 2017ZX09304030), and Shanghai Clinical Medical Center for Urinary System Diseases (no. 2017ZZ01005).

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.