CXCL12γ Promotes Metastatic Castration-Resistant Prostate Cancer by Inducing Cancer Stem Cell and Neuroendocrine Phenotypes Free

Bone is the most frequent site of prostate cancer metastasis, and skeletal lesions are a common and disabling consequence of progressive disease (1, 2). First-line therapy for skeletal metastasis includes androgen ablation and chemotherapy. Yet, in the vast majority of cases, metastatic castration-resistant prostate cancer (m-CRPC) emerges and poses significant challenges for treatment (1–7). Unfortunately, many of the molecular mechanisms that lead to the development of m-CRPC remain unclear.

Growing evidence suggests that cancer stem cells (CSC) play a crucial role in cancer chemoresistance and recurrence in the development of m-CRPC (8–11). For example, after second-generation antiandrogens and chemotherapy, neuroendocrine carcinoma, potentially derived from CSCs, can accelerate metastatic castration-resistant progression, resulting in an aggressive clinical course (12–15). Thus, understanding the mechanisms by which CSCs and neuroendocrine differentiation impact the development of m-CRPC may facilitate the identification of biomarkers and development of novel therapeutics.

Stromal-derived factor-1 (SDF-1 or CXCL12) and CXCR4 signaling is known to play a central role in promoting tumor progression in many cancers and prostate cancer bone metastasis (16–23). The CXCL12–CXCR4 axis also directly promotes tumor cell motility based on chemokine gradients of CXCL12 and CXCR4 expression on tumor cells (21, 23). Recently, we demonstrated that enhanced levels of CXCL12 by cancer-associated fibroblasts (CAF) promote an epithelial–mesenchymal transition (EMT), supporting prostate cancer bone metastasis (18, 20). Moreover, CXCL12 signaling promotes activation of protein kinase C (PKC)/NFκB signaling, which contributes to CSC activities (24). These data provide direct evidence that CXCL12 plays a central role in the development of prostate cancer progression. To date, most studies have focused on the secreted isoforms of CXCL12 (CXCL12α, CXCL12β) as they promote metastasis and the survival of tumor cells (16–23). The role of the alternatively transcribed CXCL12γ isoform, an intracellular chemokine, in tumor progression remains unknown (25–30).

Here, the role of CXCL12γ expression in induction of CSC and neuroendocrine phenotypes in the development of m-CRPC was explored. These investigations demonstrate a unique and significant role of CXCL12γ in the reprogramming of CSCs and neuroendocrine cells, leading to the development of m-CRPC and reveal a potential central molecular pathway in the progression of aggressive disease.

Five- to 7-week-old male and female SCID mice (CB.17. SCID; Charles River Laboratories) were used as transplant recipients. All animal procedures were performed in compliance with the institutional ethical requirements and approved by the University of Michigan Institutional Committee for the Use and Care of Animals.

Human prostate cancer cell lines (PC3, DU145, and LNCaP) were obtained from the ATCC. The metastatic subclone of LNCaP, C4-2B, was originally isolated from a lymph node of a prostate cancer patient with disseminated bony and lymph node involvement. GFP-expressing prostate cancer cell lines (PC3^GFP and DU145^GFP cells) were established by lentiviral transduction. A castration-resistant subline of the LNCaP cells, LNCaP95, was kindly provided by Dr. Jun Luo (Johns Hopkins University, Baltimore, MD). All prostate cancer cell lines were routinely grown in RPMI1640 (Life Technologies) supplemented with 10% FBS (GEMINI Bio-Products), 1% penicillin/streptomycin (Life Technologies), and maintained at 37°C, 5% CO₂, and 100% humidity. In some cases, prostate cancer cells were cultured in phenol red-free RPMI1640/DMEM (HyClone) supplemented with 10% charcoal/dextran-treated FBS (HyClone). Normal human prostate epithelial RWPE-1 cells (ATCC, CRL-11609) were cultured in Keratinocyte-SFM (Life Technologies) with supplements (17005-042, Life Technologies). Neuroendocrine prostate cancer cells (NCI-H660; ATCC, CRL 5813) were obtained from ATCC and were grown in ATCC-formulated RPMI1640 Medium (cat. 30-2001) with supplements. In some cases, the prostate cancer cells were cultured in RPMI with 1% FBS supplemented with 200 ng/mL of human recombinant CXCL12 (cat. 350-NS, R&D Systems). The human breast cancer cell line, MCF7 cells were kindly provided by Dr. Max Wicha (University of Michigan, Ann Arbor, MI). MCF7 cell line was cultured in DMEM with supplements (Invitrogen). All cell lines were routinely monitored for mycoplasma contamination using the Mycoplasma PCR Detection Kit (cat. MP0035, Sigma), and all cells were authenticated by STR genotyping using the Promega 16 High Sensitivity STR Kit (cat. DC2100, Promega). The STR profiles were compared with databases (DSMZ/ATCC/JCRB/RIKEN).

A CXCL12γ overexpression plasmid vector, pLV-CXCL12γ and control vector, pLV were kindly provided by Dr. Ramirez (Viral Vector Facility, Technical Unit of Gene Targeting, Fundacion CNIC, National Centre for Cardiovascular Research, Madrid, Spain; ref. 29). pLV-CXCL12γ and pLV were packaged with lentivirus at University of Michigan. Lentiviral pLV-CXCL12γ or pLV were infected into prostate cancer cells (PC3, DU145, LNCaP, C42B) and breast cancer cells (MCF7). Infected cells were selected for 7 days in media containing 1 μg/mL puromycin and analyzed by real-time qPCR or immunofluorescence staining.

PC3 or DU145 cells at 60% confluence were seeded onto 6-well culture plates. After 24 hours, negative control siRNA (cat. 4390843, Ambion) or CXCR4 siRNA (cat. 4390824, Ambion) with OPTI-MEM (cat. 31985-062, Life Technologies) were transfected into prostate cancer cells using Lipofectamine RNAiMAX (cat. 56532, Life Technologies) according to the manufacturer's instructions. Transfected cells were incubated at 37°C for 72 hours, and the cells were used to various cell assays. Silencing was verified by Western blot analysis.

For analysis of a CSC phenotype (CD133⁺/CD44⁺), overexpression of CXCL12γ in prostate cancer cells or control cells (PC3, DU145; 1 × 10⁵) were seeded onto 12-well culture plates and were cultured for 4 days. The cells were incubated with PE-anti-CD133 antibody (cat. 130-080-901, Miltenyi Biotec) and APC-anti-CD44 antibody (cat. 559942, BD Biosciences) for 20 minutes at 4°C. For CXCR4⁺ cell analysis, the cells were incubated with PE-anti-human CD184 (CXCR4) antibody (cat. 306506, BioLegend) or mouse IgG-PE (cat. 130-092-212, Miltenyi Biotec) for 20 minutes at 4°C. The CD133⁺/CD44⁺ or CXCR4⁺ fractions were analyzed with a FACS Aria High-Speed Cell Sorter (BD Biosciences). Apoptosis was measured by flow cytometry (FACSAria dual laser flow cytometer, Becton Dickinson) using PE Annexin V Apoptosis Detection Kit I (cat. 559763, BD Biosciences). The prostate cancer cells were pretreated AMD3100 (5 μg/mL) or siCXCR4 and treated with of docetaxel (Taxotere; 0.5–1 μg/mL, Hospira). In some cases, the prostate cancer cells were treated with XTANDI (enzalutamide; 0.5 μg/mL; Selleck Chemicals).

Prostate cancer cells that overexpress CXCL12γ or control (PC3, DU145) were dissociated to single cells by standard trypsinization and washed three times with PBS. The cells were plated in stem cell culture medium (DMEM:F12 plus 10 ng/mL bFGF, 20 ng/mL EGF, 5 mg/mL insulin, and 0.4% BSA) supplemented with 1% KO serum replacement (cat. 10828-028, Invitrogen/Gibco) at a density of 1,000 cells/mL in low attachment 6-well culture plates (31). Seven-day-old spheres are enumerated as size >50 cells.

Total RNA was extracted from cells using the RNeasy Mini or Micro Kit (Qiagen) and converted into cDNA using a First-Strand Synthesis Kit (Invitrogen). Quantitative PCR was performed on an ABI 7700 sequence detector (Applied Biosystems) using TaqMan Universal PCR Master Mix Kit (Applied Biosystems) according to the directions of the manufacturer. TaqMan MGB probes (Applied Biosystems) were as follows: CXCL12γ (hsSDF-1YJC), CXCR4 (Hs00237052_m1), AR-V7 (Hs04260217_m1), E-cadherin (CDH1; Hs010113953_m1), N-cadherin (CDH2; Hs00169953_m1), vimentin (Hs00185584_m1), α-SMA (Hs00426835_g1), CD44 (Hs01075861_m1), SNAIL1 (Hs00195591_m1), SNAIL2 (Hs0095344_m1), ZEB1 (Hs00232783_m1), ZEB2 (Hs00207691_m1), TGFBR2 (Hs00234253_m1), androgen receptor (AR; Hs00171172_m1), N-MYC (NYCN; 00232074_m1), NDRG1 (Hs00608387_m1), ENO2 (Hs00157360_m1), and synaptophysin (Hs00300531_m1). β-Actin (Hs99999903_m1) was used as internal controls for the normalization of target gene expression.

For the analysis of gene expression for circulating tumor cells (CTC), the healthy normal subjects (n = 2) and patients with m-CRPC (n = 26) were enrolled in the companion of the protocol “Evaluation of blood and bone marrow elements that contribute to genitourinary cancer metastasis” at the University of Michigan Comprehensive Cancer Center after obtaining a separate written informed consent approved by the University of Michigan Institutional Review Board in accordance with the Declaration of Helsinki. The whole blood (5 mL) was drawn from the health normal subjects and prostate cancer patients (IRB: 00052405, University of Michigan). CTCs were isolated from blood using anti-EpCAM conjugated Dynabeads (Thermo Fisher Scientific; refs. 32–34). After 5 times wash with PBSF (PBS plus 0.1% FBS), CTCs were lysed on beads with the lysis buffer and stored at −80°C. mRNA from CTCs was captured using Oligo(dT)25 mRNA Dynabeads (Thermo Fisher Scientific) and reverse transcribed into cDNA. Target gene amplified library was generated with multiplex PCR with a pooled primer set including 150 genes. Real-time PCR was then used to determine individual gene expression level. Blood from healthy normal controls (no history of cancer) were used to establish baseline gene expression related to WBC contamination.

CXCL12γ expression was also analyzed in normal human tissues and C57BL/6 mouse tissues (n = 3 mice). Total human RNA was purchased from Life Technologies; brain (cat. AM 7962), heart (cat. AM7966), lung (cat. AM7968), liver (cat. AM7960), spleen (cat. AM7960), kidney (cat. AM7976), prostate (cat. AM7988), testis (cat. AM7972), and skeletal muscle (cat. AM7982). The human or mouse-specific Cxcl12γ primers were designed through Life Technologies (human CXCL12γ primer, cat. SDF-1YJC; mouse Cxcl12γ primer, cat. CXCL12C). The human CXCL12γ TaqMan probes are F–5′-CCA AGA GTA CCT GGA GAA AGC TTT AA-3′, R–5′-CGG TCC ATC GGC AGG AA-3′, TaqMan probe–5′-FAM-TTC TCT TCT TCT GTC GCT TCT-TAMRA-3′ (Applied Biosystems) and the mouse Cxcl12γ TaqMan probes are F–5′-CCA AGA GTA CCT GGA GAA AGC TTT AA-3′, R–5′-CAC GGA TGT CAG CCT TCC T-3′, TaqMan probe–5′-FAM- AAG TAA GCA CAA CAG CCC-TAMRA-3′ (Applied Biosystems).

Cells and tumor sections were used for immunostaining. Cells were fixed and permeabilized with Perm/Wash Buffer (cat. 554723, BD Biosciences). Tumor sections were blocked with Image-iT FX signal enhancer for 30 minutes and incubated for 2 hours at room temperature with primary antibodies combined with reagents of Zenon Alexa Fluor 488 (green; cat. Z25002, Invitrogen) or 555 (red; cat. Z25305, Invitrogen) labeling kit. For the CXCL12γ antibody, a rabbit polyclonal antiserum (anti-CXCL12γ) was raised against the peptide KVGKKEKIGKKKRQ, mapping to the specific C-terminal region of CXCL12γ. The N-terminal cysteine enables direct conjugation of the peptide to the protein carrier and is not present in the native sequence. Peptide synthesis, coupling, immunization, ELISA titration, and affinity purification were done by BioGenes GmbH (29). Cxcl12α (mouse, cat. ab25117, Abcam), CD133 (cat. 130-090-423, Miltenyi Biotec), CD44 (cat. ab6124, Abcam), CXCR4 (cat. ab181020, Abcam), CD24 (cat. MS-1279-P, Thermo Fisher Scientific), ENO2 (cat. NB110-58870, Novus), synaptophysin (cat. ab23754, Abcam), pan-cytokeratin (cat. ab86734, Abcam), and α-tubulin (cat. ab52866, Abcam) antibodies were used as primary antibody. After washing with PBS, tumor sections were mounted with ProLong Gold antifade reagent with DAPI (cat. P36931, Invitrogen). Images were taken with Olympus FV-500 confocal microscope. In some cases, hematoxylin and eosin (H&E) stain, ENO2, synaptophysin, pan-cytokeratin antibodies were used for IHC staining in subcutaneous tumors. Images were taken with Olympus 51A microscope. Human prostate tissue microarrays (TMA) were obtained from The Tissue Core of the University of Michigan Comprehensive Cancer Center and US Biomax, Inc. TMAs consist of normal prostate tissues (n = 8), benign (n = 6), Gleason 6–7 prostate cancer tissues (n = 12), Gleason 9-1-0 prostate cancer tissues (n = 25). Tumors were graded using stage progressing system. Human prostate metastatic tissues [lymph node (n = 15), lung (n = 10), liver (n = 12), and bone (n = 19)] also obtained from The Tissue Core of the University of Michigan Comprehensive Cancer Center. Staining intensity was ranked on a scale from 1 to 4 (1, negative; 2, weak; 3, moderate; and 4, strong).

Whole-cell lysates were separated on 4% to 20% Tris-Glycine gel and transferred to PVDF membranes. The membranes were incubated with 5% milk for 1 hour and incubated with primary antibodies overnight at 4°C. Primary antibodies used included anti-MARCKS (1:1,000 dilution, cat. 5607, Cell Signaling Technology), anti-p-MARCKS (Ser159/163; 1:1,000 dilution, cat. 11992, Cell Signaling Technology), anti-PKCα (1:1,000 dilution, cat. 2056, Cell Signaling Technology), anti–p-PKCα (1:1,000 dilution, cat. 9373, Cell Signaling Technology), monoclonal anti-NFκBp65, 1:1,000 dilution, cat. 8242, Cell Signaling Technology), monoclonal anti-p-NFκBp65, 1:1,000 dilution, cat. 3033, Cell Signaling Technology), anti-CXCR4 (1:1,000 dilution, cat. ab181020, Abcam), polyclonal anti-N-MYC (1:1,000 dilution, cat. 9405, Cell Signaling Technology), monoclonal anti-NDRG1 (1:1,000 dilution, cat. 9485, Cell Signaling Technology), polyclonal anti-ENO2 (1:1,000 dilution, cat. 2056, Cell Signaling Technology), and polyclonal anti-synaptophysin (1:1,000 dilution, cat. 4329, Cell Signaling Technology). Blots were incubated with peroxidase-coupled anti-mouse IgG secondary antibody (cat. 7076, 1:2,000 dilution, Cell Signaling Technology) or peroxidase-coupled anti-rabbit IgG secondary antibody (cat. 7074, 1:2,000 dilution, Cell Signaling Technology) for 1 hour, and protein expression was detected with SuperSignal West Dura Chemiluminescent Substrate (cat. Prod 34075, Thermo Fisher Scientific). Membranes were reprobed with monoclonal anti–β-actin antibody (1:1,000 dilution; cat. 4970, Cell Signaling Technology) to control for equal loading.

PC3 or DU145 cells (1 × 10⁵) were seeded onto 12-well culture plates and cultured for 4 days. Whole-cell lysates were prepared to evaluate PKC kinase activity by following the directions of the manufacturer (cat. ADI-EKS-420A, Enzo Life Sciences). The levels of PKC kinase activity were normalized to total cell numbers.

For the analysis of serum enolase 2, chromogenin A, and PSA levels for CTCs from prostate cancer patients, serum samples were collected from whole blood, which were drawn from prostate cancer patients (IRB: 00052405, University of Michigan). Serum enolase 2 and chromogenin A were evaluated following the manufacturer's directions (enolase2/NSP, cat. DENL20, R&D System; chromogenin A, cat. DY9098, R&D Systems). PSA data were provided by University of Michigan Cancer Center (IRB: 00052405, University of Michigan).

Dual-luciferase reporter assays (cat. E1910, Dual-Lucuferase Reportor Assay System, Promega) were performed according to the manufacturer's instructions. Cells plated in 24-well plates were transiently transfected with NFκB promoter-firefly luciferase plasmid and internal control plasmid pRL-TK (Promega), which encodes Renilla luciferase and is used to normalize transfection efficacy. The cells were incubated for 48 hours. All firefly luciferase activity was normalized to Renilla luciferase activity or total DNA (cat. P11496, picoQuant-iT PicoGreen dsDNA Assay Kit, Molecular Probes), and data were presented as relative luciferase activity where control values were set to 1 for each cell line.

Tumors were established by injecting prostate cancer cells (CXCL12γ-overexpressing prostate cancer cells or controls) or breast cancer cells (CXCL12γ-overexpressing breast cancer cells or controls; 1 × 10⁶) in growth factor–reduced Collagen Type I gel (cat. 354236, BD Biosciences) subcutaneously into 5- to 7-week-old male or female SCID mice. The animals were monitored daily. Tumor volumes were calculated using the formula V = (the shortest diameter) × (the longest diameter) × height. After 4 weeks, the animals were sacrificed. The tumors were measured and prepared for histology. For in vivo metastasis assays, IHC for prostate cancer cells in the marrow was also used. The numbers of prostate cancer cells were quantified on the endosteal region defined as 10 cell diameters from the bone surfaces.

The prostate cancer cells overexpressing CXCL12γ or control cells (1 × 10⁶ cells) were injected into male 5- to 7-week-old male SCID mice by intracardiac injection. Groups were PBS (n = 2), PC3-control cells (n = 7), and PC3- CXCL12γ-overexpressing cells (n = 8). After 72 hours, brain, lung, liver, and bone (spine, pelvis, femur, and tibia) tissues were collected. The cells from animal tissues were incubated first with a RBC lysis buffer (Qiagen, cat. 8305333), and then were incubated with a PE/Cy7 anti-mouse H-2Kd (MHC) antibody (cat. 116622, BioLegend) and APC/Cy7 anti-human HLA-ABC antibody (cat. 311426, BioLegend) for 30 minutes at 4°C. The disseminated prostate tumor cells (prostate cancer DTC) were analyzed with a FACSAria II Cell Sorter by gating on mouse MHC-negative and HLA-ABC–positive cells. Data are presented as mean ± SD (Student t test).

Results are presented as mean ± SD. Significance of the difference between two measurements was determined by unpaired Student t test. Correlation was analyzed with simple linear regression to calculate Pearson correlation coefficient (r) and P value. Analyses were conducted in with GraphPad Prism version 7 software. Values of P < 0.05 were considered statistically significant.

We previously reported that elevated expression of CXCL12 and its receptor CXCR4 are found in prostate cancer tissues and cell lines and are associated with increasing malignant potential (22). Here, the role of the alternatively transcribed CXCL12 isoform CXCL12γ, an intracellular chemokine, was explored in relation to prostate tumor progression. Using human prostate tumor and metastatic TMAs, CXCL12γ expression was examined using a polyclonal antibody raised against a peptide specific to the C-terminal region of CXCL12γ. Little or no expression of CXCL12γ was detected in normal prostate and benign prostate cancer tissues, and relatively low levels of CXCL12γ expression were observed in low-grade prostate cancer tissues and significant expression of CXCL12γ was detected in high-grade prostate cancer tissues dominated by small-cell carcinoma (Fig. 1A and B). In the metastatic tissues from m-CRPC patients, significant CXCL12γ-expressing small-cell carcinoma was detected in lymph node, lung, liver, and bone tissues (Fig. 1C and D). In contrast, CXCL12γ expression was not detected in regions dominated by luminal type of adenocarcinoma cells from the high-grade prostate cancer tissues (Fig. 1A; Supplementary Fig. S1A and S1B). CXCL12γ expression in bone metastases from m-CRPC patients mirrored that of the primary lesions; CXCL12γ expression was observed in epithelial regions (cytokeratin-positive) dominated by a small-cell phenotype, whereas it was not observed in regions dominated by luminal type of adenocarcinoma cells (Fig. 1E; Supplementary Fig. S1C and S1D).

The intriguing observation that CXCL12γ expression is associated with progressive disease and with cellular regions dominated by a small-cell phenotype raises the possibility that CXCL12γ may serve as a marker of CSCs in m-CRPC. To address these possibilities, cells enriched for CSC-like activities were isolated from prostate cancer cell lines based upon their expression of CD133 and CD44 (14) and examined for CXCL12γ expression. CXCL12γ expression in both hormone-sensitive and castration-resistant CSCs was significantly enhanced relative to non-CSCs (Fig. 1F). Importantly, even among the CSC populations, CXCL12γ expression was significantly enhanced in castration-resistant cell line LNCaP95 compared with the hormone-sensitive cell line LNCaP (Fig. 1F). CXCL12γ expression was also significantly enhanced in CSCs relative to non-CSCs from the metastatic cell lines PC3 and DU145 (Fig. 1G). These data suggest that CXCL12γ expression is associated with the CSC phenotype and m-CRPC.

Under normal culture conditions, little or no expression of CXCL12γ was detected in human prostate cancer cell lines (Supplementary Fig. S2A). To further explore the role that CXCL12γ plays in tumor progression, CXCL12γ was overexpressed in prostate cancer (PC3 and DU145). Overexpression of CXCL12γ was confirmed at the RNA and protein levels (Fig. 2A and B). As higher levels of CXCL12γ expression were observed in the CSC population in castration-resistant prostate cancer cells in Fig. 1F, studies were performed to evaluate the impact of CXCL12γ on the CSC phenotype in vitro. Overexpression of CXCL12γ led to a significant increase in the expression of the CSC phenotype (Fig. 2C) and facilitated the development of spheres in culture (Fig. 2D). To validate that CXCL12γ plays an active role in tumor growth, prostate cancer cells were implanted subcutaneously into SCID mice and evaluated for tumor growth. Intriguingly, tumors generated from the CXCL12γ-overexpressing prostate cancer cells grew significantly larger than tumors generated from the control cells (Fig. 2E–G). We further confirmed that significant numbers of CSCs (CD133⁺/CD44⁺ cells) in prostate cancer tumors were detected in tumors generated with CXCL12γ-overexpressing cancer cells compared with control cells (Fig. 2H and I). Consistent with more aggressive and larger tumor growth observed following subcutaneous injection, enhanced numbers of disseminated tumor cells (DTC) were observed in bone tissues of these animals following injection of CXCL12γ overexpressing prostate cancer cells compared with control cells (Fig. 2J and K). The enhanced number of DTCs found in the bone marrows of these animals was likely made possible by the induction of an EMT in cells overexpressing CXCL12γ (Supplementary Fig. S2B). To further validate that CXCL12γ plays an active role in tumor metastasis, the prostate cancer cells overexpressing CXCL12γ or control cells were injected intracardially into SCID mice and evaluated for DTCs. Significant numbers of prostate cancer DTCs were observed from PC3 CXCL12γ-overexpressing cells compared with PC3-control cells in brain, lung, liver, and bone tissues (Fig. 2L; Supplementary Fig. S2C). In addition, we also observed that high levels of CXCL12γ expression are present in normal brain, lung, liver, and heart, adrenal gland, and bone tissues (Supplementary Fig. S2D and S2E). Together, these data suggest that CXCL12γ induces a CSC phenotype, which promotes prostate cancer tumor outgrowth and ultimately enhanced levels of DTCs in the soft and bone tissues. To further validate that CXCL12γ plays an active role in tumor metastasis, the prostate cancer cells overexpressing CXCL12γ or control cells were injected intracardially into SCID mice and evaluated for DTCs. Significant numbers of prostate cancer DTCs were observed from PC3- CXCL12γ-overexpressing cells compared with PC3-control cells in brain, lung, liver, and bone tissues (Fig. 2L; Supplementary Fig. S2C). In addition, we also observed that high levels of CXCL12γ expression are present in normal brain, lung, liver, heart, adrenal gland, and bone tissues (Supplementary Fig. S2D and S2E).

To further explore the role that CXCL12γ plays in tumor progression, CXCL12γ was overexpressed in breast cancer cell line (MCF7). Under normal culture conditions, little or no expression of CXCL12γ was detected in human breast cancer cell line MCF7 (Supplementary Fig. S3A). Overexpression of CXCL12γ was confirmed at the RNA and protein levels (Supplementary Fig. S3A and S3B). Overexpression of CXCL12γ led to the development of spheres in culture (Supplementary Fig. S3C and S3D). To validate that CXCL12γ plays the role in tumor growth, breast cancer cells were implanted subcutaneously into SCID mice and evaluated for tumor growth. The tumors generated from the CXCL12γ-overexpressing breast cancer cells grew significantly larger than tumors generated from the control cells (Supplementary Fig. S3E–S3G). We further confirmed that significant numbers of CSCs (CD24⁻/CD44⁺ cells) in breast cancer tumors were detected in tumors generated with CXCL12γ-overexpressing cancer cells compared with control cells (Supplementary Fig. S3H and S3I). We also found that breast cancer cells overexpressing CXCL12γ increase the induction of EMT markers, suggesting the implication of metastatic potentials (Supplementary Fig. S3J).

Together, these data suggest that CXCL12γ induces a CSC phenotype that promotes tumor outgrowth and ultimately enhanced levels of DTCs in the soft and bone tissues.

Neuroendocrine prostate cancer disease is commonly observed in relapsing individuals particularly after treatment with second-generation antiandrogens (e.g., abiraterone and enzalutamide; refs. 10, 12, 14, 35–38). Emerging evidence indicates that the neuroendocrine cell differentiation and more aggressive disease are associated with enhanced expression of CSC and EMT phenotypes (14, 35, 36). To explore the extent to which CXCL12γ expression is associated with neuroendocrine differentiation, we examined the expression of a neuroendocrine phenotype in CXCL12γ-overexpressing prostate cancer cells. In vitro, neuroendocrine markers were significantly increased in CXCL12γ overexpressing prostate cancer cells compared with control cells (Fig. 3A and B). To validate the in vitro studies, enolase-2 (ENO2) and synaptophysin staining were performed on the tumors generated with CXCL12γ-overexpressing prostate cancer cells compared with control cells. A higher density of neuroendocrine cells with extensive vascularization (Fig. 3C and D, H&E stain), ENO2, and synaptophysin expression were observed in tumors generated from the cells overexpressing CXCL12γ compared with control cells (Fig. 3C and D). Pan-cytokeratin expression was used to confirm that these neuroendocrine prostate cancer cells were of human origin (Fig. 3C and D). To further explore the extent to which CXCL12γ expression is associated with neuroendocrine differentiation in breast cancer cell line (MCF7), we examined the expression of a neuroendocrine phenotype in CXCL12γ-overexpressing breast cancer cells or control cells. In vitro, neuroendocrine markers were significantly increased in CXCL12γ-overexpressing breast cancer cells compared with control cells (Supplementary Fig. S4A). We also examined the expression of neuroendocrine markers in breast cancer tumors. A higher density of neuroendocrine cells (Supplementary Fig. S4B, H&E stain) with ENO2 and synaptophysin expression were observed in tumors generated from the cells overexpressing CXCL12γ compared with control cells (Supplementary Fig. S4B). Pan-cytokeratin expression was used to confirm that these neuroendocrine cells were of human origin (Supplementary Fig. S4B).

Detection of CTCs in blood is a potential prognostic marker in m-CRPC patients and has also been used as a tool to assess gene expression of metastatic tumors (4–6, 12). For example, AR-V7–based CTC detection has been shown to be an accurate predictive marker of response to enzalutamide and abiraterone in patients with m-CRPC (4–6). Here, we measured expression of CXCL12γ and the neuroendocrine marker ENO2 in CTCs of m-CRPC patients to assess the association of CXCL12γ with neuroendocrine disease (Fig. 3E). Critically, we observed that CXCL12γ expression was significantly higher in men with increased ENO2 expression (P = 0.018), suggesting that CXCL12γ is associated with a neuroendocrine state (Fig. 3F; ref. 10). In addition, we observed a trend of positive linear relationship in ENO2 mRNA expression in CTCs and serum enolase and chromogranin A (P = 0.129 and P = 0.131, respectively) and an inverse linear relationship (P = 0.250) in m-CRPC patients (Supplementary Fig. S5A–S5C). Together, these data suggest that CXCL12γ induces an aggressive neuroendocrine phenotype, which contributes to the development of m-CRPC.

To determine the impact of the major CXCL12 receptor, CXCR4, in CSC activities and resistance, CSCs were examined from hormone-sensitive cell line or castration-resistant cell line and CXCR4 expression was also examined in CSCs or non-CSCs from hormone-sensitive cell line or castration-resistant cell line. CSCs were significantly elevated in the castration-resistant LNCaP95 cells compared with the parental hormone-sensitive LNCaP cells (Fig. 4A) and neuroendocrine prostate cancer cells (NCI-H660; Supplementary Fig. S6A and S6B). CXCR4 expression in castration-resistant CSCs was significantly enhanced relative to hormone-sensitive CSCs (Fig. 4B). Total CXCR4 protein levels were enhanced in castration-resistant LNCaP95 cells compared with the parental hormone-sensitive LNCaP cells (Fig. 4C). These data suggest that CXCL12γ signaling is linked to induction of a CSC phenotype possibly through its receptor, CXCR4. We further verified that the expression levels of CXCL12γ and CXCR4 were dramatically enhanced in cells with a neuroendocrine phenotype (NCI-H660) compared with PC3 and DU145 cells (Fig. 4D and E). Critically, colocalization of CXCL12γ with CXCR4 near the nuclear membrane was observed in NCI-H660 cells (Fig. 4F). We next probed the cells for expression of CXCR4, which is strongly linked to metastasis in solid tumors, including prostate cancer (23). Overexpression of CXCL12γ significantly increased the expression of CXCR4 in prostate cancer cells (Fig. 4G). Critically, intracellular CXCR4 colocalized with CXCL12γ near the nucleus of CXCL12γ-expressing prostate cancer cells in a similar pattern as in neuroendocrine prostate cancer cells (NCI-H660; Fig. 4H; Supplementary Fig. S6C). To further examine CXCR4 involvement in CXCL12γ-mediated CSC induction, the prostatosphere formation assays were performed in the presence or absence of the small-molecule inhibitor of CXCR4, AMD3100. AMD3100 inhibited the formation of the prostatospheres in control cells, but had minimal impact on the prostatosphere formation in cells overexpressing CXCL12γ, further suggesting that intracellular CXCR4 is more involved in CSC activities than the membrane-binding CXCR4 (Fig. 4I and J). Consistent with these observations, blocking CXCR4 signaling with AMD3100 sensitizes prostate cancer cells to docetaxel chemotherapy. However, when cells were induced to overexpress CXCL12γ, AMD3100 did not sensitize prostate cancer cells to docetaxel chemotherapy (Fig. 4K). We also examined the effect of CXCL12γ/CXCR4 signaling on docetaxel-induced microtubule stabilization. CXCL12γ/CXCR4 signaling significantly reduced docetaxel-induced microtubule stabilization in CXCL12γ-overexpressing prostate cancer cells compared with control prostate cancer cells (Fig. 4L).

Furthermore, to determine the role of CXCL12γ in AR-dependent prostate cancer cells, CXCL12γ was overexpressed in LNCaP and C42B cells (Supplementary Fig. S6D and S6E). The AR mRNA was enhanced when CXCL12γ was overexpressed in LNCaP and C42B cells (Supplementary Fig. S6F). We also observed that less apoptotic cells in CXCL12γ-overexpressing LNCaP cells compared with control cells following enzalutamide treatment and in CXCL12γ-overexpressing C42B cells compared with control cells following docetaxel treatment (Supplementary Fig. S6G and S6H). We further found that neuroendocrine prostate cancer cells (NCI-H660) are highly resistant to docetaxel treatment (Supplementary Fig. S6I). We also found a significant enhancement of CXCR4 expression in breast cancer cell overexpression of CXCL12γ compared with control cells (Supplementary Fig. S7A). Together, these data suggest that the intracellular CXCL12γ and CXCR4 interactions associate with the induction of a CSC phenotype and play a significant role in the regulation of resistance to chemotherapy.

To identify the molecular mechanisms that are functionally associated with CXCL12γ-mediated induction of CSC and neuroendocrine phenotypes, PKCα/NFκB signaling was explored. We found that PKCα and NFκB signaling were preferentially activated in prostate cancer cells overexpressing CXCL12γ compared with control cells (Fig. 5A–D). To prove that PKCα/NFκB activation is responsible for the enhancement of CSC activities, a pan-PKC inhibitor (Go6983) or a NFκB inhibitor (IKK2 inh) was employed. Using each of these inhibitors, significant reductions in CSC phenotype and sphere formation were observed (Fig. 5E–I). To further prove the role of CXCR4 in the activation of NFκB signaling, we examined the phosphorylation of the NFκB subunit p65 following AMD3100 treatment. AMD3100 inhibited CXCL12-mediated NFκB-p65 phosphorylation in parental prostate cancer cells, yet it had little to no impact on NFκB-p65 phosphorylation on the CXCL12γ-overexpressing prostate cancer cells (Fig. 5J and K). To further prove how CXCL12γ/CXCR4–mediated PKCα/NFκB signaling is functionally associated with the development of chemoresistance, myristoylated alanine-rich C-kinase substrate (MARCKS) signaling was explored. Activation of PKC signaling modulates microtubule cytoskeleton and cell morphology (39–41). MARCKS as a downstream target of PKC signaling pathway is known to be critical for regulating multiple pathophysiologic processes, including chemoresistance (40–42). Recent evidence also shows that docetaxel reduces PKC activity targeting of taxol-mediated suppression of NFκB (41, 42). Here, we found that phosphorylation of the MARCKS was preferentially increased in prostate cancer cells overexpressing CXCL12γ (Fig. 5L) and phosphorylation of the MARCKS was significantly reduced along with PKCα and NFκB-p65 phosphorylation in CXCR4 knockdown prostate cancer cells overexpressing CXCL12γ (Fig. 5M). Similarly, we found that significant levels of PKCα and NFκB signaling were activated in breast cancer cells overexpressing CXCL12γ compared with control cells (Supplementary Fig. S7B and S7C).

These data suggest that the intracellular CXCL12γ and CXCR4 interactions are associated with expression of a CSC phenotype and resistance to chemotherapy through PKCα/NFκB signaling pathways.

Our studies demonstrate that CXCL12γ expression in prostate cancer regulates induction of CSC and neuroendocrine phenotypes, promoting prostate cancer tumor outgrowth, metastasis, and chemoresistance, which support the significance impacts of CXCL12γ on the development of m-CRPC.

Despite many advances in therapy for m-CRPC, improvements in survival have generally been on the order of months rather than years (1–7). One possible explanation for this observation is the emergence of metastatic tumor clones or cells with tumor-initiating potential or CSCs, which are extraordinarily resistant to conventional therapy (8–11). This biologic process is mediated by therapy-initiated factors and pathways in the tumor. The second-generation antiandrogen therapies alter intrinsic factors and signaling pathways, including AR, PI3K/AKT/PTEN, HIF, IGF-1, WNT, IL6, SOX2, NANOG, and many others within tumor cells, which facilitate CSC expansion and subsequent m-CRPC development (11). Furthermore, CSCs are associated in the transition from androgen-sensitive to castration-resistant phenotype. For example, PKCα/NFκB signaling is implicated in activating stem-like prostate tumor-initiating cells and activation of NFκB signaling, which is thought to contribute to the development of m-CRPC in AR-independent cells (24, 43). These studies support our findings that CXCL12γ induces a CSC phenotype through CXCR4-mediated activation of PKCα/NFκB signaling, which promotes prostate cancer tumor outgrowth and metastasis (44–47).

In addition, chemotherapy with docetaxel alters protein targets involved in cell survival, normal physiologic functions, and oncogenesis (48–50). Docetaxel also increases circulating cytokines in castration-resistant prostate cancer patients (49). CXCL12–CXCR4 signaling is known to prevent docetaxel-induced microtubule stabilization via p21-activated kinase 4 (PAK4)-dependent activation of LIM domain kinase 1 in prostate cancer cells (48). Activation of PKC signaling modulates microtubule cytoskeleton and cell morphology (39–41) and MARCKS as a downstream target of PKC signaling pathway is known to be critical for regulating multiple pathophysiologic processes, including chemoresistance (40–42). Recent evidence also shows that docetaxel reduces PKC activity by targeting of taxol-mediated suppression of NFκB (41, 42). These studies support our data demonstrating that intracellular CXCL12γ and CXCR4 interactions associate with the induction of a CSC phenotype and resistance to chemotherapy through PKCα/NFκB signaling pathways. Thus, understanding the biology of CSC expansion by therapy and by what mechanisms CSCs drive resistance to the cytotoxic effects will implicate the development of potential novel therapeutics.

Neuroendocrine prostate cancer disease has an extremely aggressive clinical course with a high mortality rate and is commonly observed in the late stages of the prostate cancer coincident with metastatic spread and resistance to treatment (10, 12–15, 35–38). Neuroendocrine cells may arise by two possible processes, the differentiation of CSCs into neuroendocrine cells, or transdifferentiation of preexisting adenocarcinoma cells. In the first instance, preexisting CSCs differentiate into cancer cells with three distinct phenotypes (luminal, basal, and neuroendocrine) or a cellular reprograming of differentiated epithelial cells to stem-like states poised for differentiating to progenitors. Cells with a neuroendocrine phenotype highly express the stem cell surface markers PROM1/CD133 and CD44 (10), which suggests that neuroendocrine cells share features with CSCs (14, 35). Recent studies show that the association of a neuroendocrine phenotype with CSC and EMT phenotypes such that when injected into nude mice, a highly vascularized tumor with a high density of neuroendocrine cells expressing low levels of E-cadherin and β-catenin and high levels of vimentin are formed (14, 35, 36).

As the second possible mechanism, the cells with a neuroendocrine phenotype arise from transdifferentiation of preexisting adenocarcinoma cells, which have accumulated multiple molecular alterations (14, 15, 51, 52). Growing evidence suggests that adenocarcinoma cells can undergo a transdifferentiation process to become neuroendocrine prostate cancer cells, which acquire a similar phenotype to normal neuroendocrine cells and express several neuroendocrine markers (53, 54). In this transdifferentiation process, cell-autonomous signaling pathways may facilitate acquisition of a neuroendocrine state, which in conjunction with microenvironmental clues promotes tumor outgrowth, survival, and therapeutic-resistant properties. Recent findings demonstrate that IL6, epinephrine, and forskolin induce neuroendocrine differentiation in prostate cancer cells through activation of intracellular cAMP and protein kinase A (PKA) suggests this possibility (55, 56). Together, these studies indicate that multiple signaling pathways are likely active in mechanisms associated with neuroendocrine differentiation. Furthermore, growing evidence also shows that deregulation of cellular signaling by genetic or epigenetic alterations are associated with neuroendocrine differentiation during therapies. Neuroendocrine prostate cancer is often associated with accumulation of new genetic alterations, such as loss of tumor suppressors (RB1, PTEN, and p53), TMPRSS2-ERG rearrangement, and expression of proto-oncogenes (AURKA, N-MYC, and BCL2; refs. 14, 29, 37, 38, 52, 57, 58). NDRG1 expression involves metastasis and neuroendocrine differentiation in prostate cancer cells by loss of RB in a hypoxia-dependent fashion, and these biologic events are associated with activation of the NFκB signaling pathway, which contributes to metastatic, castration-resistant, or neuroendocrine disease (58). In this investigation, our findings show that CXCL12γ induces N-MYC, NDRG1, and ENO2 expression in neuroendocrine phenotype of prostate cancer cells (37, 59), which contribute to aggressive disease.

Currently, there is a continued need to develop biomarkers that predict response to chemotherapy for m-CRPC patients. Although most studies have focused on the important role of AR during the transition to m-CRPC, these approaches do not explain the resistance process in conjunction with alternation of signaling pathways, suggesting that AR is not the only pathway involved in the formation of m-CRPC (7, 11). Moreover, CTC numbers have been shown to correlate with patient outcome with overall survival in m-CRPC. In fact, the AR splice variant 7 (AR-V7) is associated with resistance to both enzalutamide and abiraterone in prostate cancer cells and has been shown to be one potential marker for CTCs (4). Here, we show that CXCL12γ was significantly higher in men with increased ENO2 expression. Thus, CXCL12γ expression may serve as another potential marker of neuroendocrine prostate cancer disease and the development of m-CRPC.

Finally, our findings show for the first time that expression of the intracellular chemokine CXCL12γ facilitates the emergence of CSCs and neuroendocrine prostate cancer associated with metastatic progression, chemoresistance, and tumor recurrence. That the molecular signatures of CSCs have tumorigenic capability with a large proliferative potential suggests that CXCL12γ promotes a cellular reprograming of differentiated epithelial cancer cells to stem-like states poised for differentiation into alternative phenotypes.

No potential conflicts of interest were disclosed.

Conception and design: Y. Jung, F.C. Cackowski, J.K. Kim, K.J. Pienta, R.S. Taichman

Development of methodology: Y. Jung, F.C. Cackowski, A.M. Decker, Y. Wang

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Jung, F.C. Cackowski, K. Yumoto, A.M. Decker, J. Wang, E. Lee, Y. Wang, J.-S. Chung, A.M. Gursky, P.H. Krebsbach, T.M. Morgan, R.S. Taichman

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y. Jung, F.C. Cackowski, K. Yumoto, A.M. Decker, J. Wang, J.K. Kim, E. Lee, Y. Wang, J.-S. Chung, P.H. Krebsbach, T.M. Morgan, R.S. Taichman

Writing, review, and/or revision of the manuscript: Y. Jung, A.M. Decker, J.K. Kim, Y. Wang, P.H. Krebsbach, K.J. Pienta, T.M. Morgan, R.S. Taichman

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y. Jung, J. Wang, P.H. Krebsbach, R.S. Taichman

Study supervision: Y. Jung, P.H. Krebsbach, K.J. Pienta, R.S. Taichman

The authors wish to thank Taocong Jin and Sasha Meshinchi for technical and logistical support in Molecular Core and Microscopy & Image Analysis Laboratory (MIL), Comprehensive Cancer Center Microscopy Core at University of Michigan, Ann Arbor, MI. Dr. Juan C. Ramirez (National Centre for Cardiovascular Research, Spain) kindly provided pLV and pLV-CXCL12γ vectors and CXCL12γ antibody. This work is directly supported by the NCI (to R.S. Taichman; CA093900 and CA163124), the Department of Defense [to R.S. Taichman (W81XW-15-1-0413 and W81XWH-14-1-0403) and T.M. Morgan (W81XWH-14-1-0287)], and the Prostate Cancer Foundation Challenge Award (to R.S. Taichman; 16CHAL05). R.S. Taichman receives support as the Major McKinley Ash Collegiate Professor. F.C. Cackowski receives support from a Career Enhancement Award, Sub-Award (F048931) of NIH/NCI Prostate Cancer Specialized Program in Research Excellence (SPORE) to Arul Chinnaiyan at the University of Michigan (F036250).

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.