Prostate cancer–induced endothelial-cell-to-osteoblast transition drives immunosuppression in the bone–tumor microenvironment through Wnt pathway–induced M2 macrophage polarization

Significance

Immune checkpoint therapy has limited efficacy for patients with bone-metastatic castration-resistant prostate cancer (bmCRPC). Prostate cancer (Pca) in bone frequently induces aberrant bone overgrowth through bone morphogenetic protein 4 (BMP4)-mediated endothelial-to-osteoblast (EC-to-OSB) transition. We showed that EC-to-OSB transition produced paracrine factors, which induced M2 polarization and recruited M2-like tumor-associated macrophages (TAMs) to the bone–tumor microenvironment (bone-TME). These TAMs suppressed CD8⁺ T cells’ proliferation and cytolytic activity, and these effects were partially reversed by Wnt inhibitors. Genetic or pharmacological inhibition of Pca–induced EC-to-OSB transition reduced the levels of M2-like macrophages in osteogenic tumors. Our study demonstrates that Pca–induced EC-to-OSB transition drives immunosuppression in the bone-TME, suggesting that therapies that reduce Pca–induced bone formation may improve immunotherapeutic outcomes for bmCRPC.

Abstract

Immune checkpoint therapy has limited efficacy for patients with bone-metastatic castration-resistant prostate cancer (bmCRPC). To improve immunotherapy for bmCRPC, we aimed to identify the mechanism of bmCRPC-induced changes in the immune microenvironment. Among bmCRPC patients, higher levels of a 32-gene M2-like macrophage signature in bone metastasis samples correlated with shorter overall survival. Immunohistochemistry showed that CD206-positive (CD206⁺) macrophages were enriched in bmCRPC bone biopsy specimens compared with primary tumors or lymph node metastases. In preclinical osteogenic prostate cancer (Pca) xenograft models, CD206⁺ macrophages were recruited to areas with tumor-induced bone. RNA sequencing (RNAseq) analysis showed higher expression of an M2-like gene signature, with activated canonical and noncanonical Wnt pathways, in tumor-associated macrophages isolated from osteogenic tumors (bone-TAMs) than in TAMs isolated from nonosteogenic tumors (ctrl-TAMs). Mechanistic studies showed that endothelial cells (ECs) that had undergone EC-to-osteoblast (EC-to-OSB) transition, the precursors of tumor-induced OSBs, produced paracrine factors, including Wnts, CXCL14, and lysyl oxidase, which induced M2 polarization and recruited M2-like TAMs to the bone–tumor microenvironment (bone-TME). Bone-TAMs suppressed CD8⁺ T cells’ proliferation and cytolytic activity, and these effects were partially reversed by treating bone-TAMs with Wnt inhibitors. Genetic or pharmacological inhibition of Pca–induced EC-to-OSB transition reduced the levels of M2-like macrophages in osteogenic tumors. Our study demonstrates that Pca–induced EC-to-OSB transition drives immunosuppression in the bone-TME, suggesting that therapies that reduce Pca–induced bone formation may improve immunotherapeutic outcomes for bmCRPC.

Bone is the dominant site of prostate cancer (PCa) metastases, and bone metastasis occurs in more than 70% of patients with castration-resistant disease (1). Despite the development of cytotoxic agents, second-generation androgen-receptor-targeting therapies, and radioligand therapies to treat metastatic PCa, metastasis of PCa to bone signifies lethal disease (2). Although immune checkpoint therapy has significantly improved overall survival in a number of solid tumors, bone-metastatic castration-resistant PCa is relatively unresponsive to immune checkpoint inhibitors (3, 4). Several mechanisms may contribute to relative resistance of bone-metastatic castration-resistant PCa to immune checkpoint therapy. For example, Jiao et al. (3) showed that immune checkpoint therapy led to an increase in Th17 rather than Th1 subsets in bone marrow mediated by release of TGFβ during osteoclast-mediated bone resorption. Kfoury et al. (5) performed single-cell RNAseq analysis of bone from patients undergoing laminectomy and found that M2-polarized macrophages were enriched in PCa bone metastases. These observations suggest that M2-like macrophages in bone metastases may play a role in promoting PCa progression in bone. However, mechanisms for enrichment of M2-like macrophages in PCa bone metastases remain unknown.

A unique feature of PCa bone metastasis is the induction of aberrant bone overgrowth leading to osteoblastic lesions (6). This characteristic of PCa distinguishes it from most other solid tumors, including renal, lung, and breast carcinomas, which induce osteolytic bone lesions. PCa-induced aberrant bone formation supports tumor progression and resistance to therapies (7, 8). Bone morphogenetic proteins (BMPs), including BMP4 (9, 10), BMP6 (11), and BMP7 (12), secreted from PCa cells are involved in bone-forming lesions, consistent with their roles in promoting OSB differentiation. PCa-induced aberrant bone formation was thought to result from existing OSBs expansion (6). However, studies by us (13) demonstrated that tumor cells secreted BMP4 induces endothelial cell-to-osteoblast (EC-to-OSB) transition as a mechanism of PCa-mediated aberrant bone formation. In this study, we showed that tumor-induced EC-to-OSB transition contributes to the immunosuppressive bone–tumor microenvironment (bone-TME).

Results

High-Level M2-Like Macrophages in Bone Metastases Correlate with Shorter Survival in Patients with Metastatic Castration-Resistant PCa.

To examine the involvement of M2-like macrophages in PCa progression, we used a 32-gene M2-like macrophage gene signature derived from Cassetta et al. (14) to examine the castration-resistant PCa bone metastases in a metastatic castration-resistant PCa RNAseq dataset (15). We found that bone metastasis samples could be classified into M2-high and M2-low subsets using hierarchical clustering (Fig. 1A). In addition, immunosuppressive genes, including CD274 (encodes PD-L1), PDCD1LG2 (encodes PD-L2), CSF1, and TGFB1, which are associated with M2-like macrophages, had higher expressions in M2-high than M2-low bone metastasis samples, although the P value for TGFB1 (P = 0.06) did not reach statistical significance (Fig. 1B). We also observed a trend toward shorter survival in patients with the M2-high signature than with the M2-low signature (log-rank P = 0.0664) (Fig. 1C). Of note, since our studies sought to elucidate how bone-forming metastases modulate the bone-TME, we excluded tumors with neuroendocrine features as these tumors give rise to osteolytic bone metastases and represent a distinct subtype of PCa. Similar hierarchical clustering analyses were performed in lymph node metastases from the aforementioned metastatic castration-resistant PCa RNAseq dataset (15) and treatment-naïve primary PCa from the TCGA dataset (16). There was no significant difference in survival between the M2-high and M2-low populations in patients with lymph node metastases (log-rank P = 0.7749) (Fig. 1C) or in patients with primary PCa (log-rank P = 0.6810) (Fig. 1C). These observations suggest that M2-like macrophages play a more critical role in PCa bone metastases than in lymph node metastases or primary cancer.

Fig. 1.

High levels of M2-like macrophages in bone-metastatic PCa. (A) Analysis of RNAseq dataset of bone-metastatic Pca using M2-like macrophage gene signature. Samples were classified into M2-high and M2-low using hierarchical clustering. (B) Higher expression of several immunosuppressive genes in M2-high than M2-low bone metastasis samples. (C) Patients with M2-high signature in the bone metastasis (bone met) group, but not in lymph node metastasis (LN) or primary PCa groups, had shorter survival. (D) IHC of CD68⁺ and CD206⁺ cells in PCa specimens. CD68 or CD206 staining was quantified using ImageJ. B, bone; T, tumor. Fields, number of regions under microscope (×200) selected for quantification. (Scale bar, 50 µm.) **P < 0.01 and ***P < 0.001.

CD206⁺ Macrophages Are Enriched in Human PCa Bone-Metastasis Specimens.

We next performed immunohistochemistry (IHC) for CD68, a general macrophage marker, and CD206, a marker for M2-like macrophages, on human bone metastasis, lymph node metastasis, and primary PCa specimens. We found that bone metastasis specimens, including areas of woven bone and tumor, had higher levels of CD68⁺ macrophage infiltration (macrophages accounted for 2.62 ± 0.47% of the tumor area) than lymph node metastasis specimens (1.10 ± 0.46%, P = 0.011) or primary PCa specimens (0.14 ± 0.05%, P = 0.005) as shown by quantification of CD68⁺ cells using ImageJ (Fig. 1D and SI Appendix, Fig. S1A). IHC staining of the specimens with anti-CD206 antibody also showed a higher level of CD206⁺ macrophage infiltration in bone metastasis specimens (3.32 ± 0.67%) than in lymph node metastasis specimens (1.50 ± 0.63%, P = 0.023) or primary PCa specimens (0.78 ± 0.32, P = 0.005). There was no significant difference in the level of CD206⁺ macrophages between lymph node metastasis and primary PCa specimens (P = 0.306) (Fig. 1D and SI Appendix, Fig. S1A). These results suggest that M2-like macrophages are uniquely enriched in the bone-TME.

Tumor-Induced Bone Increases F4/80⁺ and CD206⁺ Macrophage Infiltration.

To examine whether tumor-induced bone contributes to the increased density of CD206⁺ cells in the bone-TME, we performed IHC staining on MDA PCa-118b xenograft tumors, osteogenic tumors derived from a bone biopsy specimen of a patient with castration-resistant PCa (9). F4/80 was used as a general mouse macrophage marker. We observed high densities of F4/80⁺ and CD206⁺ macrophages in the areas with tumor-induced bone (Fig. 2A and SI Appendix, Fig. S1B), consistent with densities in human bone metastasis specimens (Fig. 1D). We showed that MDA PCa-118b expresses BMP4, leading to ectopic bone formation (7). To examine whether tumor-induced bone plays a role in the increased M2-like macrophage infiltration observed in bone metastases (Fig. 1D), we generated an osteogenic PCa model C4-2b-BMP4 by ectopically expressing BMP4 in the nonosteogenic C4-2b PCa cells. As we reported (13), BMP4 overexpression in C4-2b cells led to woven bone formation in subcutaneous tumors (Fig. 2B). When tumors were immunostained for F4/80 and CD206, we observed densities of F4/80⁺ and CD206⁺ macrophages were significantly higher in the areas adjacent to tumor-induced bone of C4-2b-BMP4 tumors (Fig. 2B and SI Appendix, Fig. S1C). We also generated osteogenic MycCaP-BMP4 tumors by transfecting mouse BMP4 cDNA into MycCaP cells. When tumors were immunostained for F4/80 and CD206, we observed higher densities of F4/80⁺ and CD206⁺ macrophages in the MycCaP tumors than in the C4-2b-vec tumors (Fig. 2C vs. Fig. 2B and SI Appendix, Fig. S1D vs. SI Appendix, Fig. S1C), suggesting intrinsic differences in the levels of macrophage infiltration in tumors with different genetic backgrounds. We observed higher densities of F4/80⁺ and CD206⁺ macrophages in the MycCaP-BMP4 tumors than in the MycCaP tumors (Fig. 2C and SI Appendix, Fig. S1D). However, macrophage densities did not differ between the tumor-induced bone and tumor areas in MycCaP-BMP4 tumors (Fig. 2C), possibly due to high levels of macrophage infiltration in this genetic background.

Fig. 2.

Osteogenic tumors have higher levels of TAMs than nonosteogenic tumors. (A) IHC of CD206 in MDA PCa-118b tumors. Macrophages were enriched in areas with tumor-induced bone. (B) IHC of CD206 in C4-2b-vec (C4-2b) and C4-2b-BMP4 tumors. Macrophage levels were similar in C4-2b-vec and C4-2b-BMP4 tumors, but macrophages were enriched in the areas adjacent to tumor-induced bone of C4-2b-BMP4 tumors. (C) IHC of CD206 in MycCaP and MycCaP-BMP4 tumors. Macrophage levels were higher in MycCaP-BMP4 tumors and areas adjacent to tumor-induced bone of MycCaP-BMP4 tumors than in MycCaP tumors. (D and E) FACS analysis of macrophages in (D) C4-2b-vec (C4-2b) and C4-2b-BMP4 tumors and (E) MycCaP and MycCaP-BMP4 tumors. Equal weights of tumors were digested, and the cells were subjected to FACS. Average of two biological replicates from 1 experiment is shown. The study was repeated three times with similar results. vec, cells transfected with empty lentiviral vector. B, tumor-induced bone. Arrows, CD206⁺ macrophages. (Scale bars, 50 µm.)

To further confirm the IHC results, we performed fluorescence-activated cell sorting (FACS) to characterize myeloid subsets using CD45⁺ (pan-immune marker), CD11⁺ (pan-myeloid marker), GR1⁻ (granulocyte marker), F4/80⁺ (general mouse macrophage marker), and CD206⁺ (M2-like macrophage marker) in C4-2b-vec and C4-2b-BMP4 tumors and in MycCaP and MycCaP-BMP4 tumors. We found that the levels of CD45⁺CD11b⁺Gr1⁻F4/80⁺ and CD45⁺CD11b⁺Gr1⁻F4/80⁺CD206⁺ cells were higher in C4-2b-BMP4 tumors than in C4-2b-vec tumors (Fig. 2D) and were higher in MycCaP-BMP4 tumors than in MycCaP tumors (Fig. 2E). The P value for the difference in the CD206⁺ population between the C4-2b-vec and C4-2b-BMP4 tumors was 0.08, because the number of CD206⁺ cells in the C4-2b-vec tumors was very low, resulting in high variations in cell counts. Taken together, these results suggest that osteogenic tumors have higher M2-like macrophage infiltration than nonosteogenic tumors.

Conditioned Medium (CM) from EC-OSB Hybrid Cells Enhances the Recruitment of M2-Like Macrophages.

We showed that PCa-induced bone is produced through EC-to-OSB transition by PCa-secreted BMP4 (13). In this process, BMP4 induces ECs to transition into EC-OSB hybrid cells that then undergo osteoblastic differentiation leading to mineralization (17). To examine the role of EC-OSB cells in macrophage recruitment, 2H11 ECs were treated with 100 ng/mL BMP4 for 48 h to generate EC-OSB cells (Fig. 3A). CM from 2H11 and EC-OSB cells was assessed for its chemotactic effect on immortalized bone marrow–derived macrophages (iBMDM), RAW264.7 cells, and mouse BMDM. iBMDM and RAW264.7 cells were polarized to M2-like macrophages by incubation with IL-4 for 24 h as shown by upregulation of M2 macrophage markers Arg1 (arginase 1), Mrc1 (CD206), and Clec10a and to M1-like macrophages by incubation with lipopolysaccharide as shown by upregulation of M1 macrophage markers Tnf, Nos2, and Ccl2 (SI Appendix, Fig. S2). Similarly, BMDM were polarized to M2-like macrophages by incubation with IL-4 for 72 h as shown by upregulation of M2 macrophage markers (SI Appendix, Fig. S2). The Boyden chamber migration assay was used to examine the effect of CM on macrophage recruitment. CM from EC-OSB hybrid cells (EC-OSB CM), but not control 2H11 cell-CM or BMP4, which was used to induce EC-to-OSB transition, dramatically promoted the migration of M2-polarized, but not M1-polarized, iBMDM (Fig. 3B). Similar results were observed with RAW264.7 cells (Fig. 3C). These effects were further confirmed using BMDM that were polarized to M2 phenotype by IL-4 (Fig. 3D). These results suggest that factors secreted from EC-OSB cells recruit M2-like macrophages.

Fig. 3.

CM from EC-OSB hybrid cells enhances recruitment of M2-like macrophages. (A) EC-OSB cells were generated by treating 2H11 ECs with 100 ng/mL BMP4 for 48 h. The CM from 2H11 cells and EC-OSB cells were collected for migration assays. (B–D) Boyden chamber migration assay for (B) M1- or M2-polarized iBMDM, (C) M1- or M2-polarized RAW264.7, and (D) M2-polarized BMDM. EC-OSB CM promoted migrations of M2-polarized, but not M1-polarized, macrophages. (E) RNAseq analysis of EC-OSB cell–secreted factors from 2H11 cells treated with or without BMP4. The top 20 genes with P < 10⁻⁴ are shown. The values are expressed as fold changes between EC-OSB cells and 2H11 cells. GREM1, gremlin 1; GREM2, gremlin 2; NOG, Noggin. (F and G) Confirmation of (F) LOX and (G) CXCL14 expression in EC-OSB cells by qRT-PCR, western blot, and/or ELISA, and IHC of MDA PCa-118b tumor. B, bone. T, tumor. Arrows, EC-OSB cells. (Scale bar, 50 µm.) **P < 0.01 and ***P < 0.001.

Lysyl Oxidase (LOX) and CXCL14 in EC-OSB CM Are Candidate Factors for M2 Macrophage Recruitment.

To identify secreted factors in EC-OSB CM that led to M2 macrophage recruitment, RNAseq analysis of RNA from 2H11 and EC-OSB cells (GSE168321) was performed, and results were analyzed for differentially expressed secreted factors (Fig. 3E). The most abundantly secreted factors were BMP inhibitors, including gremlin 1, gremlin 2, and Noggin, indicating a negative feedback mechanism following BMP4 treatment. Among other secreted factors up-regulated in EC-OSB CM were LOX and CXCL14, which have been shown to recruit macrophages (18, 19). Upregulation of LOX and CXCL14 in the EC-OSB cells was confirmed by qRT-PCR (Fig. 3 F and G). Western blot and enzyme-linked immunosorbent assay (ELISA) further showed that both LOX and CXCL14 proteins were secreted into the EC-OSB CM (Fig. 3 F and G). The differences between 2H11 and EC-OSB cells in LOX protein levels were greater than the differences between 2H11 and EC-OSB cells in Lox mRNA levels (Fig. 3F). This may be due to LOX is a secreted protein, which accumulates in CM. IHC staining of MDA PCa-118b tumors localized the expression of LOX and CXCL14 to EC-OSB cells rimming the tumor-induced bone (Fig. 3 F and G, arrows). These findings suggest that LOX and CXCL14 secreted from EC-OSB cells during tumor-induced bone formation are involved in M2-like macrophage recruitment.

LOX and CXCL14 Secreted by EC-OSB Hybrid Cells Promote Macrophage Migration.

To examine the function of LOX as a macrophage chemoattractant in vitro, medium supplemented with recombinant LOX protein (200 ng/mL) was used in a transwell assay. LOX increased M2-polarized, but not M1-polarized, immortalized bone marrow-derived macrophage (iBMDM), RAW264.7, and BMDM migration compared with control medium (Fig. 4A). Knockdown of Lox using shRNA resulted in a significant decrease in the Lox mRNA expression by qRT-PCR and protein in EC-OSB cell lysates by western blot (SI Appendix, Fig. S3A). Using CM from shLox-transfected or control shvector–transfected 2H11 cells in a transwell assay showed that knockdown of Lox led to a decrease in the migration of M2-polarized iBMDM and RAW264.7 cells (Fig. 4B). Knockdown of Lox did not have significant effects on BMP4-mediated EC-to-OSB transition as measured by Bglap (osteocalcin) mRNA expression (SI Appendix, Fig. S3B) or on mineralization (SI Appendix, Fig. S3C), suggesting that Lox is not essential for EC-to-OSB transition. A similar approach was used to examine the function of CXCL14 as a macrophage chemoattractant in vitro. Recombinant CXCL14 protein (100 ng/mL) increased migration of M2-polarized macrophages compared with control medium (Fig. 4C), and shRNA knockdown of Cxcl14 in 2H11 cells led to a decrease in the migration of M2-polarized iBMDM and RAW264.7 cells (Fig. 4D and SI Appendix, Fig. S3D). Knockdown of Cxcl14 did not have significant effects on BMP4-mediated EC-to-OSB transition (SI Appendix, Fig. S3 C and E). Together, these results suggest that LOX and CXCL14 secreted during EC-to-OSB transition recruit M2-like macrophages into the bone-TME (Fig. 4E).

Fig. 4.

LOX and CXCL14 secreted by EC-OSB hybrid cells promote macrophage migration. (A) Recombinant LOX protein (200 ng/mL) increased the migration of M2-polarized, but not M1-polarized, iBMDM, RAW264.7 cells, and BMDM compared with control medium. n = 3 to 4 biological replicates. (B) CM from EC-OSB cells expressing Lox shRNA (shLox) reduced the migration of M2-polarized iBMDM and RAW264.7 cells compared with shvector EC-OSB CM. (C) Recombinant CXCL14 protein (100 ng/mL) increased the migration of M2-polarized, but not M1-polarized, iBMDM, RAW264.7 cells, and BMDM compared with control medium. n = 3 to 4 biological replicates. (D) CM from EC-OSB cells expressing CXCL14 shRNA (shCxcl14) reduced the migration of M2-polarized iBMDM and RAW264.7 cells compared with shvector EC-OSB CM. (E) LOX and CXCL14 secreted from EC-OSB cells recruit M2 macrophages. Error bars indicate SD. *P < 0.05, **P < 0.01, and ***P < 0.001, Student’s t test.

Inhibition of EC-to-OSB Transition by Palovarotene or LDN193189 Decreases CD206⁺ Macrophage Infiltration in Tumors.

We next examined whether inhibiting PCa-induced bone formation would reduce CD206⁺ macrophage infiltration in tumors. Our studies showed that treatment of mice with MDA PCa-118b tumors with LDN193189, a BMP receptor kinase inhibitor, reduced tumor-induced bone formation (7). We also showed that RARγ agonists, including palovarotene and all-trans retinoic acid (ATRA), inhibited BMP4-mediated EC-to-OSB transition through promoting pSmad1 degradation (20) (Fig. 5A). To examine whether inhibition of PCa-induced EC-to-OSB transition reduces M2-like macrophage infiltration, mice with MDA PCa-118b tumors were treated with LDN193189 or ATRA. IHC analysis showed that LDN193189 or ATRA treatment significantly reduced CD206⁺ macrophages in MDA PCa-118b tumors (Fig. 5B). Similarly, LDN193189 or palovarotene treatment significantly reduced CD206⁺ macrophages in C4-2b-BMP4 tumors (Fig. 5C and SI Appendix, Fig. S4). Palovarotene and LDN193189 had minimal effects on macrophage proliferation in vitro as measured by the MTS assay (SI Appendix, Fig. S5). These results suggest that inhibition of tumor-induced bone formation decreases CD206⁺ macrophage infiltration in osteogenic tumors.

Fig. 5.

Inhibition of tumor-induced bone formation in C4-2b-BMP4 or MDA PCa-118b tumors reduces CD206⁺ macrophage infiltration. (A) LDN193189 (LDN), palovarotene (Palo), and ATRA inhibit components of the BMP4 signaling pathway. (B) Inhibition of tumor-induced bone by LDN193189 (3 mg/kg/twice a day, intraperitoneally) or ATRA (10 mg/kg/d, orally) in MDA PCa-118b tumors reduces CD206⁺ macrophages. (Scale bar, 50 µm.) (C) IHC of CD206 in C4-2b-vec (C4-2b) or C4-2b-BMP4 tumors from mice treated with or without (control) palovarotene (Palo) (2 mg/kg/d, orally) or LDN193189 (3 mg/kg/d, intraperitoneally). (Scale bar, 50 µm.) (D) IHC of CD206 in MDA PCa-118b and C4-2b-BMP4 tumors treated with or without (control) Ra223. (Scale bar, 50 µm.) (E) Levels of CD206⁺ macrophages in C4-2b-BMP4 tumors are lower in tumors from trigenic mice than from bigenic mice. Levels of CD206⁺ macrophages were quantified by ImageJ. B, tumor-induced bone. T, tumor. N, number of tumors examined. Multiple random fields were selected for quantification using ImageJ. (Scale bar, 50 µm.)

Radium 223 Treatment Decreases F4/80⁺ and CD206⁺ Macrophage Infiltration in MDA PCa-118b and C4-2b-BMP4 Tumors in SCID Mice.

Radium 223 (Ra223), a high-energy α-emitting calcium mimetic that preferentially localizes to hydroxyapatite in newly formed matrix within osteogenic metastases, induces double-strand DNA breaks in adjacent cells (21). Ra223 is approved for treating PCa patients with bone metastasis based on improvement in overall survival (22). We next examined whether Ra223 reduced M2-like tumor-associated macrophages (TAMs) in osteogenic tumors. MDA PCa-118b cells were injected subcutaneously into SCID mice. Tumor-bearing mice were treated with or without one dose (300 kBq/kg) of Ra223. IHC analysis of MDA PCa-118b tumors showed that Ra223 treatment reduced CD206⁺ TAMs (Fig. 5D). Similar results were observed in osteogenic C4-2b-BMP4 tumors (Fig. 5D). These results suggest that treatment with Ra223 in bone-forming tumors decreases levels of CD206⁺ macrophages within the bone-TME.

Reduction of Tumor-Induced Bone in Tumors Implanted in an EC-Specific Osterix Deletion Model (Tie2^cre;Osx^f/f;Scid^+/+) Decreases CD206⁺ Macrophage Infiltration.

We next examined whether reducing PCa-induced bone formation had an effect on CD206⁺ macrophage infiltration in osteogenic tumors generated in a genetically modified mouse model (Tie2^cre;Osx^f/f) with EC-specific deletion of Osx (osterix; also known as Sp7), a cell fate determinant gene that controls the development of the OSB lineage (23). We used this model to examine the role of EC-to-OSB transition in PCa-induced osteogenesis (Fig. 5E). Tumor-induced bone in C4-2b-BMP4 tumors was significantly reduced in tumors implanted in Tie2-Cre;Osx^f/f;Scid^+/+ trigenic mice, in which Osx in ECs was specifically deleted, compared with tumors implanted in control Osx^f/f;Scid^+/+ bigenic mice (13). IHC analysis of C4-2b-BMP4 tumors showed significantly lower levels of CD206⁺ macrophages in tumors from trigenic mice compared with tumors from control bigenic mice (Fig. 5E). These results suggest that reducing tumor-induced bone formation through genetic deletion of Osx in ECs reduces CD206⁺ macrophage infiltration in the bone-TME.

An M2-like Macrophage Signature Is Enriched in MycCaP-BMP4 Osteogenic Tumors Compared with MycCaP Tumors in FVB Mice.

To define molecular differences between MycCaP and MycCaP-BMP4 tumors, we performed bulk RNAseq on total RNA isolated from these tumors (GSE241343). MycCaP or MycCaP-BMP4 cells were implanted in immunocompetent Friend Virus B (FVB) mice, the mouse strain from which MycCaP tumors were generated (24, 25). Because MycCaP-BMP4 cells with a selection marker (Green-fluorescence protein (GFP), luciferase, or Neo) could not grow in FVB mice, due to an immune response against the selection marker, marker-negative MycCaP-BMP4 cells were generated using dilutional cloning. MycCaP-BMP4 tumors in FVB mice showed tumor-induced bone formation and recruitment of F4/80⁺ and CD206⁺ macrophages to the area with tumor-induced bone (SI Appendix, Fig. S6), similar to what was observed in SCID mice (Fig. 2 C and E).

Using the M2 gene signature shown in Fig. 1A, gene set enrichment analysis (GSEA) of GSE241343 RNAseq data demonstrated enrichment of an M2-like macrophage signature in MycCaP-BMP4 tumors compared with MycCaP tumors (SI Appendix, Fig. S7A). Using CIBERSORT deconvolution algorithms (26) to analyze for M0, M1, and M2 populations revealed that M2-like macrophages were enriched in MycCaP-BMP4 tumors compared with MycCaP tumors, while M0 and M1 macrophages were enriched in MycCaP tumors compared with MycCaP-BMP4 tumors (SI Appendix, Fig. S7B). Differential gene expression analysis comparing MycCaP-BMP4 tumors with MycCaP tumors showed that M2 markers Mrc1, Clec10a, Il10, Cxcl13 (27), Il1r2 (28), Vegfa, and Vegfd were up-regulated while M1 markers Tnf, Ccl2, Ccr2, Il16 (29), and Ifnlr1 (30) were down-regulated (SI Appendix, Fig. S7C). qRT-PCR confirmed upregulation of M2 macrophage markers Mrc1 (CD206), Il10, and Clec10a, in MycCaP-BMP4 tumors (SI Appendix, Fig. S7D) and upregulation of M1 macrophage markers Ccr2, Nos2, Ccl2, and Il1b in MycCaP tumors (SI Appendix, Fig. S7D). These results suggest that MycCaP-BMP4 osteogenic tumors are enriched with M2-like macrophages compared with MycCaP tumors.

An IFN-Primed TAM (IFN-TAM) Gene Signature Is Reduced, and Canonical and Noncanonical Wnt Pathways Are Activated in MycCaP-BMP4 Tumors Compared with MycCaP Tumors.

To characterize the molecular features of macrophages recruited to the TME of MycCaP and MycCaP-BMP4 tumors, CD45⁺CD11b⁺Gr1⁻F4/80⁺ macrophages were isolated by FACS from MycCaP/FVB (ctrl-TAMs) and MycCaP-BMP4/FVB subcutaneous tumors (bone-TAMs) (Fig. 6A), and RNAseq analysis was performed (GSE241344). GSEA showed suppression of interferon (IFN)-gamma and IFN-alpha pathways in bone-TAMs compared with ctrl-TAMs (Fig. 6B). Consistently, IFN-regulated genes, including Cxcl9, Cxcl10, Cxcl11, Isg15, and Oas3, were down-regulated in bone-TAMs compared with ctrl-TAMs (Fig. 6C). Furthermore, several MHC-II genes, including H2-Oa, H2-Ob, and H2-DMa, were down-regulated in bone-TAMs compared with ctrl-TAMs (Fig. 6C). TAMs are classified into functionally distinct subsets, including IFN-TAMs, immune-regulatory TAMs, inflammatory cytokine-enriched TAMs, lipid-associated TAMs, proangiogenic TAMs, RTM-like TAMs, and proliferating TAMs (31). IFN-TAM gene signature was dramatically down-regulated in bone-TAMs compared with ctrl-TAMs (SI Appendix, Fig. S7E). Importantly, IFN-TAM-associated genes (Cxcl9, Cxcl10, Cxcl11, H2-Oa, and H2-Ob) were down-regulated in bone-TAMs compared with ctrl-TAMs as confirmed by qRT-PCR (SI Appendix, Fig. S7 F and G). These results suggest that osteogenic tumors have reduced IFN-TAMs in the bone-TME compared with nonosteogenic tumors.

Fig. 6.

M2 macrophage–related genes are up-regulated and M1 macrophage–related genes are down-regulated in MycCaP-BMP4 tumors compared with MycCaP tumors. (A) Schema for isolating TAMs from MycCaP and MycCaP-BMP4 tumors. (B) GSEA showed that IFN-gamma and -alpha signaling were suppressed in bone-TAMs compared with ctrl-TAMs. (C) Decreases in the expression levels of IFN-regulated genes and MHC-II genes in bone-TAMs compared with ctrl-TAMs. (D) GSEA showed that Wnt signaling was activated in bone-TAMs compared with ctrl-TAMs. (E) Immunofluorescence showed an increase of active β-catenin in the nuclei of bone-TAMs compared with ctrl-TAMs. (Insets) higher magnification of a ctrl-TAM or bone-TAM. (Scale bar, 30 µm.) (F) Western blot of nuclear fractions of ctrl-TAMs and bone-TAMs. The levels of active β-catenin and pGSK3β were higher in bone-TAMs than in ctrl-TAMs. (G) GSEA of genes involved in noncanonical Wnt signaling showed upregulation in bone-TAMs compared with ctrl-TAMs. (H) Fold changes of genes in noncanonical Wnt signaling. (I) Western blot confirmed increased levels of ROR1 and its downstream target pCAMK2 in bone-TAMs compared with ctrl-TAMs.

GSEA showed that pathways related to Wingless/integrated (WNT) signaling were among the top 10 up-regulated pathways in bone-TAMs (based on the KEGG, Reactome, and WikiPathways collections) (Fig. 6D and SI Appendix, Fig. S8A and Table S1). In lung cancer, Sarode et al. (32) showed that macrophage-specific β-catenin-mediated Wnt signaling is centrally involved in the phenotypic transition of M1-like to M2-like TAMs. We examined whether the Wnt pathway is activated in bone-TAMs. Immunofluorescence showed increased β-catenin, a key signal transducer of the Wnt pathway, in the nucleus and cytoplasm of bone-TAMs compared with ctrl-TAMs (Fig. 6E). Furthermore, western blot showed increased nuclear β-catenin and pGSK3β, which regulates the stability of β-catenin, in bone-TAMs compared with ctrl-TAMs (Fig. 6F). qRT-PCR showed that mRNA levels of the downstream canonical Wnt pathway target genes Myc, Met, and Tnks1 (tankyrase 1), but not Tnks2 and Ccnd1, were up-regulated in bone-TAMs compared with ctrl-TAMs (SI Appendix, Fig. S8B). Besides the canonical Wnt pathway, GSEA showed upregulation of the noncanonical Wnt pathway in bone-TAMs compared with ctrl-TAMs (Fig. 6G). Genes in the noncanonical Wnt pathway that were up-regulated in bone-TAMs included Ror1, Wnt7b, Wnt5a, Fzd3, Dvl1, and Vangl2 (33–35) (Fig. 6H). Of note, several Wnt ligands were also up-regulated, including Wnt2, Wnt3, Wnt4, Wnt5a, Wnt6, and Wnt7b (Fig. 6H), suggesting autocrine activation of the Wnt pathways in bone-TAMs. qRT-PCR confirmed upregulation of genes involved in noncanonical Wnt signaling in bone-TAMs compared with ctrl-TAMs (SI Appendix, Fig. S8C). Western blot confirmed increased levels of ROR1 and its downstream target pCaMK2 in bone-TAMs compared with ctrl-TAMs (Fig. 6I). These results suggest that upregulation of both canonical and noncanonical Wnt pathways represents a key molecular feature of bone-TAMs.

EC-OSB Hybrid Cell–Mediated Wnt Pathway Activation Promotes M2 Polarization.

The polarization of TAMs is directly controlled by cytokines within the TME (36). We examined whether EC-OSB hybrid cells play a role in the M2-like polarization of bone-TAMs. Mouse BMDM treated with EC-OSB CM for 5 d promoted M2 polarization as shown by increased M2 markers (Fig. 7A) and increased CD206 expression by western blot (SI Appendix, Fig. S9A). Similar results were obtained when RAW264.7 cells were cultured with EC-OSB CM for 48 h (SI Appendix, Fig. S9B).

Fig. 7.

EC-OSB cell–secreted factors promote M2 polarization of macrophages. (A) Treatment of BMDM with EC-OSB CM up-regulated M2 macrophage markers detected by qRT-PCR. (B) Nuclear β-catenin levels were higher in bone-TAMs than in ctrl-TAMs. Treatment of bone-TAMs with Wnt pathway inhibitor XAV-939 or LGK-974 reduced nuclear β-catenin levels in bone-TAMs. (C) EC-OSB CM–mediated M2 polarization was suppressed by Wnt pathway inhibitor XAV-939 (XAV) as detected by qRT-PCR. (D) EC-OSB CM–mediated M2 polarization was suppressed by Wnt pathway inhibitor LGK-974 (LGK) as detected by qRT-PCR. (E) EC-OSB hybrid cells secreted several Wnt ligands as detected by qRT-PCR and western blot. (F) Treatment of BMDM or RAW264.7 cells with recombinant mouse Wnt5a (rmWnt5a) up-regulated M2 marker expression. (G) Graphical summary. EC-OSB cells secreted many factors, including Wnt ligands (Wnts), CXCL14, and LOX, that promoted M2 polarization of macrophages. Bone-TAMs also expressed Wnts that further promoted M2 polarization in an autocrine manner. Error bars, SD. *P < 0.05, **P < 0.01, and ***P < 0.001, Student’s t test.

To examine whether the Wnt pathway was involved in EC-OSB CM–mediated M2 polarization, we used XAV-939 and LGK-974, two small-molecule Wnt pathway inhibitors, to block Wnt/β-catenin signaling. XAV-939 is a tankyrase inhibitor that promotes the degradation of β-catenin (37, 38); LGK-974 targets Porcupine, a Wnt-specific acyltransferase (39). Treatment of bone-TAMs with XAV-939 or LGK-974 reduced levels of nuclear β-catenin in bone-TAMs (Fig. 7B). Pretreatment of BMDM with XAV-939 before incubation with EC-OSB CM reduced M2 markers, including Clec10a, Il10, and Mrc1 (Fig. 7C). Similarly, pretreatment of BMDM with LGK-974 reduced M2 markers (Fig. 7D). These results suggest that EC-OSB cell–mediated Wnt pathway activation is one mechanism that promotes M2 polarization of TAMs in the bone-TME.

EC-OSB Hybrid Cells Secrete WNTs, CXCL14, and LOX to Promote M2 Polarization.

Next, we investigated factors in EC-OSB CM that activate the Wnt pathway in bone-TAMs. RNAseq analysis of 2H11 versus EC-OSB hybrid cells revealed that Wnt4 and Wnt5a were up-regulated 1.79-fold (P = 0.0027) and 1.77-fold (P = 0.0015), respectively, in EC-OSB cells compared with 2H11 cells (GSE168321) (17). Wnt4 and Wnt5a were not included in Fig. 3E because only top 20 genes with P < 10⁻⁴ are shown. 2H11 ECs were treated with BMP4 for 48 h to examine the expression of Wnt ligands during EC-to-OSB transition. qRT-PCR confirmed that Wnt4 and Wnt5a were up-regulated in EC-OSB cells compared with 2H11 cells, and western blot showed upregulation of WNT5a in EC-OSB CM (Fig. 7E). qRT-PCR also showed that Wnt7, but not Wnt6, was up-regulated in EC-OSB cells (SI Appendix, Fig. S9C), suggesting that EC-OSB cells secreted multiple Wnt ligands. Of note, WTN5a is a ligand for noncanonical Wnt pathway activation (33, 35). Treatment with recombinant mouse WNT5a led to increased expression of several M2 markers in BMDM (Arg1 and Il10) and RAW264.7 cells (Arg1, Clec10a, Il10, and Tgfb1) (Fig. 7F), suggesting that WNT5a is involved in M2 polarization of macrophages. Interestingly, Mrc1 expression was down-regulated by WNT5a (Fig. 7F), suggesting that other EC-OSB cell–secreted factors besides Wnt ligands may also be involved in M2 polarization of bone-TAMs. We found that CXCL14 secreted from EC-OSB cells played a role in recruiting M2-like macrophages to the area with tumor-induced bone (Fig. 4 C and D). CXCL14 also functions as a chemokine for M2 polarization of macrophages (19). We found that knockdown of CXCL14 in EC-OSB cells reduced the EC-OSB cell–mediated upregulation of M2 macrophage genes (Arg1, Clec10a, Il10, and Mrc1) in BMDM (SI Appendix, Fig. S9D) and RAW264.7 cells (SI Appendix, Fig. S9E). Furthermore, addition of recombinant mouse CXCL14 protein to BMDM or RAW264.7 cells led to increased expression of M2 macrophage genes (SI Appendix, Fig. S9 F and G). Similarly, knockdown of shLox in EC-OSB cells inhibited M2 polarization while recombinant mouse LOX protein stimulated M2 polarization of BMDM (SI Appendix, Fig. S10 A and C) and RAW264.7 cells (SI Appendix, Fig. S10 B and D). These results suggest that several EC-OSB cell–secreted factors including WNTs, CXCL14, and LOX, play roles in EC-OSB cell–mediated M2 polarization (Fig. 7G). Furthermore, bone-TAMs also express Wnts (Fig. 6H and SI Appendix, Fig. S8C) to promote M2 polarization through in autocrine manner (Fig. 7G).

CXCL14, LOX, and Wnt5a Are Expressed in EC-OSB Cells and Primary Mouse OSBs (PMOs).

While LOX, CXCL14, and Wnt5a are secreted by EC-OSB cells, it is unclear whether they are also part of normal bone formation. To address this question, we assessed the expressions of these proteins during the time course of BMP4-induced EC-to-OSB transition of 2H11 cells and PMOs isolated from newborn mouse calvariae. 2H11 ECs undergo transition into EC-OSB cells upon BMP4 treatment in serum-free medium (days 0 to 2) and undergo differentiation/mineralization upon incubation in serum-containing OSB differentiation medium (days 3 to 14) (20, 40) (SI Appendix, Fig. S11 A and B). The levels of Lox mRNA in BMP4-treated cells were highest on day 2 and returned to basal levels during days 5 to 14 (SI Appendix, Fig. S11 C, Left). LOX protein levels (by western blot) in the CM correlated with mRNA levels (SI Appendix, Fig. S11 C, Middle and Right). Similar to Lox, mRNA and protein levels of Wnt5a in BMP4-treated cells were highest at day 2 (SI Appendix, Fig. S11D). Cxcl14 mRNA levels were significantly up-regulated in BMP4-treated 2H11 cells at days 2 to 5 compared to untreated 2H11 cells, although the mRNA levels at day 2 were relatively low (SI Appendix, Fig. S11E). CXCL14 protein levels (by ELISA) in CM were higher in BMP4-treated cells than in untreated cells on days 2 to 8 but similar on days 11 to 14 (SI Appendix, Fig. S11 E, Right). The difference between mRNA and protein levels may be due to protein stability. These observations suggest that LOX, Wnt5a, and CXCL14 are expressed in a time-dependent manner during EC-OSB transition.

We have previously shown that the BMP4-induced EC-to-OSB transition in osteogenic tumor also occurred in physiological osteogenesis (41). To address whether expressions of CXCL14, LOX, and Wnt5a are part of normal physiological process of bone formation, we isolated PMOs from calvaria of newborn mice (SI Appendix, Fig. S12A). We found that CXCL14, LOX, and Wnt5a mRNAs and proteins were expressed in PMOs, similar to those observed in EC-OSB cells from 2H11 cells (SI Appendix, Fig. S11 C–E vs. SI Appendix, Fig. S12 C–E). Upon culturing PMOs in differentiation medium for 27 d, PMOs could mineralize as shown by Alizarin red staining (SI Appendix, Fig. S12B). In contrast to BMP4-induced EC-to-OSB transition, the levels of CXCL14, LOX, and Wnt5a mRNAs and proteins in mineralizing PMOs (day 27) were higher than those in the freshly isolated PMOs (SI Appendix, Fig. S12 C–E). These results suggest that CXCL14, LOX, and Wnt5a are produced in EC-OSB cells as well as in freshly isolated or in vitro mineralized PMOs. However, there are differences in the temporal expression of these genes during the mineralization process.

The expressions of LOX, Wnt5a, and CXCL14 in PMOs suggest that normal bone formation may be also associated with immunosuppression. To address this, we examine whether M2 macrophages are associated with calvaria. We isolated cells from newborn mouse calvariae and performed FACS analysis for total macrophages (CD45⁺/CD11b⁺/Gr⁻/F4-80⁺) and M2-like macrophages (CD45⁺/CD11b⁺/Gr⁻/F4-80⁺/CD206⁺). We found that newborn mouse calvaria were enriched with F4-80⁺ macrophages (3.06% of live single cells), in which 2.30% were CD206⁺ macrophages (SI Appendix, Fig. S12F). These results are consistent with a previous report that macrophages are associated with normal bone (42). The levels of M2-like macrophages in calvaria are similar to those observed in C4-2b-BMP4 tumors (Fig. 2D) and MycCaP-BMP4 tumors (Fig. 2E and SI Appendix, Fig. S6D). Thus, M2-like macrophages enrichment occurs in both normal bone development and pathological bone formation. We reason that the enrichment of M2-like macrophages in the tumor microenvironment of osteogenic tumors is due to aberrantly increased bone-forming activity by PCa-secreted BMP4.

Bone-TAMs Suppress CD8⁺ T Cell Proliferation.

To examine the effect of bone-TAMs on T cell function, we examined whether bone-TAMs could suppress T cell proliferation using carboxyfluorescein succinimidyl ester (CFSE)-labeled mouse splenic CD8⁺ T cells activated by anti-CD3 and anti-CD28 antibodies (CD3/CD28 Dynabeads). CD45⁺CD11b⁺Gr1⁻F4/80⁺ cells from C4-2b-BMP4 tumors significantly suppressed the proliferation of T cells compared with ctrl-TAMs from C4-2b-vec tumors (SI Appendix, Fig. S13A). Next, CD45⁺CD11b⁺Gr1⁻F4/80⁺CD206⁺ cells isolated from C4-2b-BMP4 tumors were found to inhibit T cell proliferation (SI Appendix, Fig. S13B), suggesting that reduction of T cell proliferation is due to CD206⁺ M2-like TAMs. Control C4-2b-vec tumor contained too few CD45⁺CD11b⁺Gr1^-F4/80⁺CD206⁺ cells to be included in the assay, consistent with low CD206⁺ in C4-2b-vec tumors (Fig. 2 B and D). Bone-TAMs from MycCaP-BMP4 tumors generated in SCID mice also inhibited T cell proliferation compared with ctrl-TAMs (Fig. 8A). Similar results were observed using ctrl-TAMs and bone-TAMs isolated from MycCaP and MycCaP-BMP4 tumors generated in FVB mice (SI Appendix, Fig. S13C). These results suggest that M2-like macrophages in osteogenic tumors inhibit T cell proliferation.

Fig. 8.

Bone-TAMs suppress proliferation and cytolytic activity of CD8⁺ T cells in part through the activated Wnt pathway. (A) CD45⁺CD11b⁺Gr1⁻F4/80⁺ macrophages from MycCaP-BMP4/SCID tumors showed higher inhibition of T cell proliferation than those from MycCaP/SCID tumors. (B) Treatment of bone-TAMs from MycCaP-BMP4 tumors in FVB mice with Wnt pathway inhibitors XAV-939 (XAV) or LGK-974 (LGK) attenuates effects of bone-TAMs on T cell inhibition. (C) FACS analysis of perforin-1⁺ CD8⁺ T cells in splenocytes cocultured with ctrl-TAMS or bone-TAMs. The number of perforin-1⁺ CD8⁺ T cells was reduced by bone-TAMs compared with ctrl-TAMs, and this effect was partially reversed by XAV-939 (XAV). (D) Effect of bone-TAMs on the cytolytic activity of CD8⁺ T cells was measured using ovalbumin-specific OT-1 cell-mediated killing of ovalbumin-expressing B16F10/OVA cells. After coculturing with OT-1 cells for 48 h, bone-TAMs significantly reduced the cytolytic activity of CD8⁺ cells compared with ctrl-TAMs, and this effect was partially reversed by XAV-939. Cytotoxicity was measured by LDH released in culture supernatants. Photographs of tissue culture at 48 h are shown. Red circles, clusters of proliferating CD8⁺ T cells that were more abundantly present in ctrl-TAM-treated than in bone-TAM-treated samples. Addition of XAV-939 partially reversed bone-TAM-mediated inhibition. (E) Time course of OT-1 T cell killing showed that maximum inhibition by bone-TAMs occurred around 48-72 h. XAV, XAV-939. (F) Graphical summary. EC-OSB cell–secreted factors promote M2 polarization of macrophages that inhibit T cell proliferation. Error bars, SD. *P < 0.05, **P < 0.01, and ***P < 0.001, Student’s t test.

Wnt Pathway Inhibition Attenuates Bone-TAMs’ Effect on T Cell Proliferation.

We next tested whether Wnt pathway activation in bone-TAMs was involved in their ability to suppress T cell proliferation. Bone-TAMs isolated from MycCaP-BMP4 tumors generated in FVB mice were pretreated with Wnt pathway inhibitor XAV-939 or LGK-974 for 1 h before coculture with T cells. XAV-939 and LGK-974 attenuated the inhibitory effect of bone-TAMs on T cell proliferation (Fig. 8B and SI Appendix, Fig. S13D). Similar results were observed with bone-TAMs isolated from MycCaP-BMP4 tumors generated in SCID mice (SI Appendix, Fig. S13E). These results suggest that Wnt pathway activation in bone-TAMs is involved in inhibiting T cell proliferation. To test whether XAV-939 or LGK-974 directly affected T cell proliferation, XAV-939 (5 μM) or LGK-974 (2.5, 5, and 10 μM)-treated T cells were compared with vehicle-treated T cells. The inhibitor-treated and vehicle-treated cells had similar proliferation rates (SI Appendix, Fig. S13F), suggesting that XAV-939 and LGK-974 did not have a direct effect on T cell proliferation.

Bone-TAMs Suppress the Cytolytic Activity of CD8⁺ T Cells Partly through the Activated Wnt Pathway.

We examined the effects of TAMs on the cytolytic activity of T cells by coculturing TAMs with T cells. FACS analysis showed that the number of perforin-1⁺ CD8⁺ T cells was reduced by bone-TAMs compared with ctrl-TAMs (Fig. 8C and SI Appendix, Fig. S13G). This effect was partially reversed by XAV-939 (Fig. 8C and SI Appendix, Fig. S13G). Importantly, bone-TAMs significantly reduced ovalbumin-specific OT-1 cell-mediated killing of ovalbumin-expressing B16F10/OVA cells (Fig. 8D). This effect was partially reversed by XAV-939 (Fig. 8D). Time course of OT-1 T cell killing showed that the maximum inhibition by bone-TAMs occurred at approximately 48 to 72 h (Fig. 8E). These results show that PCa-induced EC-OSB cell–secreted factors lead to M2 polarization of macrophages that inhibit T cells’ proliferation and cytolytic activity, resulting in an immunosuppressive bone-TME (Fig. 8F).

Discussion

In this study, we found that osteoblastic bone metastases have an immunosuppressive bone-TME due to an enrichment of M2-like macrophages. We identified EC-to-OSB transition as an upstream event that drives the immunosuppression in the bone-TME (Fig. 8G). Our study unveils a mechanism that accounts for resistance of bone-metastatic PCa to immune checkpoint therapy. Our findings suggest that inhibition of PCa-induced EC-to-OSB transition may reverse immunosuppression to increase efficacies of immunotherapies for bone-metastatic PCa. We previously identified a novel therapy for targeting PCa-induced EC-to-OSB transition through a noncanonical retinoic acid receptor mechanism (20). Taken together, our previous study and current study suggest a strategy to improve immunotherapeutic responses.

M1-like and M2-like macrophages exist along a phenotypic continuum and are not easily characterized by single markers (31, 43). Although we measured the total population of M2-like macrophages in our study, we recognize that this population contains many subtypes. Due to their plasticity, TAMs can switch from one type to another depending on soluble and cellular factors present in the TME (44). It is also possible that M2-like macrophages in different osteogenic tumors, e.g., MDA PCa-118b, C4-2b-BMP4, and MycCaP-BMP4, have different subtypes due to the influence of tumor-secreted factors in addition to EC-OSB cells. Despite the heterogeneity, an enrichment of M2-like macrophages in the bone-TME may account for the resistance of bone-metastatic castration-resistant PCa to immune checkpoint therapy.

CXCL14 and LOX promote M2-like, but not M1, macrophage migration (Fig. 4), suggesting that receptors for these proteins were expressed in M2. CXCL14 receptor activation is complex and cell type–dependent (45, 46). CXCR4 (47, 48), GPR85 (49), and ACKR2 (50, 51) are implicated in CXCL14 function. Recently, Xu et al. (52) reported that integrin alpha11beta1 is an additional receptor for CXCL14. Chen et al. (18) reported that in glioblastoma, LOX recruits macrophages via the integrin beta1-PYK2 pathway. We examined the expression of receptors for LOX and CXCL14 in M1- or M2-polarized RAW264.7 and BMDM cells. Among the four candidate receptors, CXCR4 and integrin beta1, but not GPR85 and ACKR2, exhibited higher mRNA levels in M2 than in M1 macrophages in both RAW264.7 and BMDM cells (SI Appendix, Fig. S14 A–D). These results suggest that CXCR4 is involved in CXCL14 function and integrin beta1 is involved in LOX function on macrophages. One of the CXCR4 signaling pathways is ERK1/2 (53). Knockdown of CXCL14 in EC-OSB cells reduced the EC-OSB cell–mediated ERK1/2 phosphorylation in BMDM and RAW264.7 cells (SI Appendix, Fig. S14 E and G). Addition of recombinant mouse CXCL14 protein to BMDM or RAW264.7 cells stimulated ERK phosphorylation (SI Appendix, Fig. S14 F and H). Whether CXCL14-mediated polarization and recruitment of M2-like macrophages occur through the CXCR4-ERK1/2-pathway requires further investigation.

We showed that Ra223 treatment decreased CD206⁺ macrophage infiltration. Ra223 is known to induce apoptosis through DNA double-strand breaks. To examine whether Ra223 increased apoptosis of CD206⁺ macrophage, MDA PCa-118 tumors treated with or without Ra223 for 2 wk were examined by immunostaining for cleaved caspase 3 and CD206. In control MDA PCa-118b tumors, CD206⁺ cells were abundant and mostly localized at the vicinity of tumor-induced bone (SI Appendix, Fig. S15). Cells costained for CD206 and cleaved caspase 3 were very scarce (0.79 ± 0.1 cells per field). In Ra223-treated samples, more cells are costained for CD206 and cleaved caspase 3 (4.07 ± 0.51 cells per field) (SI Appendix, Fig. S15). The number of apoptotic cells in Ra223-treated samples may be underestimated because tumors were collected 2 wk after treatment, and Ra223 induced apoptosis in the first 4 d (54).

Macrophages are known to secrete various factors that regulate T cell activity. Studies by Kfoury et al. (5) showed that in human PCa bone metastases, macrophage-secreted CCL20 interacts with its receptor CCR6 on T cells to induce T cell exhaustion. We did not find upregulation of CCL20 in the RNAseq analysis of bone-TAMs. However, we identified other secreted factors, including cytokines (e.g., IL11) (SI Appendix, Table S2), that were overexpressed in bone-TAMs. IL11 is implicated in immune regulation in colon cancer (55). Our RNAseq analysis also found that bone-TAMs had increased expression of Tph1 and Ido1, 2 key tryptophan metabolism enzymes implicated in immunosuppression (56, 57). We are currently studying the mechanisms by which bone-TAMs regulate T cell activities.

The use of subcutaneously transplanted osteogenic tumor models allowed us to study the effects of tumor-induced bone on macrophages, but involvement of bone marrow components in the bone–immune cell interaction is not captured in these models. Bone metastasis models that deliver tumor cells to mouse femurs or tibia by intracardiac or caudal arterial (58) injection would be suitable for observing M2-like macrophage functions in bone marrow during EC-OSB transition. These metastasis models will be used in our future studies.

Given the significance of M2-like macrophages in the TME and their dominant role in tumor growth and progression (59), strategies targeting immunosuppressive macrophages are being developed for cancer treatment (43, 60, 61). However, the properties of TAMs may be continuously influenced by the recruitment and polarization signals from the tumor-induced bone. Thus, targeting M2-like macrophages may not be sufficient to reduce the immunosuppression in the bone-TME. Our data suggest that inhibiting PCa-induced bone would improve immunotherapeutic strategies for bone-metastatic PCa by targeting the biological process that drives the adaptive resistance. This hypothesis was supported by our preclinical data showing that inhibition of the EC-to-OSB transition via genetic approaches, small molecules, or Ra223 in osteogenic PCa tumors led to reduction of M2-like macrophages in the bone-TME. Clinical trials to test this hypothesis are in development.

Materials and Methods (Details Are in SI Appendix, Materials and Methods)

Cell lines, antibodies, and reagents are listed in SI Appendix, Table S3.

RNAseq Database Analysis.

To examine gene expression with M1/M2 macrophage markers in human PCa, publicly available primary and metastatic castration-resistant PCa RNAseq datasets (15) were downloaded from cBioportal. An M2-like TAM gene signature (14) was used for hierarchical clustering of human bone-metastatic castration-resistant PCa samples. Patients with M2-high or M2-low gene signatures were compared for overall survival.

IHC of Human and Mouse PCa Specimens.

Formalin-fixed, paraffin-embedded human PCa specimens were obtained from the Prostate Cancer Tissue Bank of The University of Texas MD Anderson Cancer Center through an Institutional Review Board–approved protocol. IHC was performed using described procedures (17).

Generation of C4-2b-vec, C4-2b-BMP4, MycCaP, and MycCaP-BMP4 Tumors.

C4-2b-vec and C4-2b-BMP4 tumors were generated in male SCID mice subcutaneously as described (20). Similarly, MycCaP and MycCaP-BMP4 tumors were generated in male SCID or FVB mice.

Isolation of TAMs and Flow Cytometry.

Subcutaneous tumors were digested in dissociation solution using a Tumor Dissociation Kit (Miltenyi Biotec). Cells that passed through a 70-μm cell strainer (BD Falcon) were collected as “first-digested cells.” Bone-containing tissues that were left on the 70-μm cell strainer were further digested with 0.2% Collagenase II (Gibco). Cells that passed through strainer were collected as “second-digested cells.” The first- and second-digested cells were combined and stained with LIVE/DEAD viability dyes, CD45, CD11b, Gr-1, F4/80, and/or CD206 monoclonal antibodies. Macrophage populations were analyzed and sorted by CytoFLEX SRT Cell Sorter (Beckman Coulter Life Sciences).

Mouse Primary BMDM Isolation and M1 or M2 Macrophage Polarization.

Bone marrow samples were from the femur and tibia of 8- to 12-wk-old C57B6 mice. Isolated BMDM cells were treated with 15% L929 cell-CM to generate primary macrophages (M0 macrophages). M0 macrophages were treated with 60 ng/mL LPS or 40 ng/mL IL-4 for 72 h to differentiate the cells into M1 or M2 macrophages, respectively. A similar procedure was used to differentiate RAW264.7 cells or iBMDM cells to M1 or M2 macrophages, except the incubation time was 24 h.

Macrophage Migration Assay.

Cells were seeded into FluoroBlock Cell Culture inserts (BD Falcon). Serum-free medium with 100 ng/mL LOX, 50 or 100 ng/mL CXCL14, and 2H11 control-cell-CM or EC-OSB CM was placed in the lower chamber of a 24-well plate individually. After incubation for 14 h, cells migrated to the other side of the membrane were labeled with Calcein AM, and cells in five randomly chosen visual fields were quantified. When captured pictures had low background, the levels of migrated cells were quantified using ImageJ software and presented as relative intensities. When captured pictures had high background, the number of cells was counted, and results were presented as number of cells per field.

qRT-PCR.

Total RNA was isolated using a RNeasy Mini Kit (Qiagen). Complementary DNA (cDNA) was prepared using a cDNA Synthesis Kit (Applied Biosystems). qRT-PCR was performed as described (20) using mouse-specific primer sequences (SI Appendix, Table S4).

Immunoblotting and Immunofluorescence.

2H11 or EC-OSB cell lysates (20 μg) or 5 to 10 fold-concentrated CMs (20 μL) were analyzed by immunoblotting with indicated antibodies. For nuclear fractionation, TAMs were lysed in lysis buffer, and the pellets were lysed in Radioimmunoprecipitation Assay (RIPA) buffer to isolate nuclear fractions as described (17). For immunofluorescence analysis, cells were fixed with ice-cold 100% methanol and immunostained with the indicated antibodies as described (17, 20).

Knockdown of LOX or CXCL14 in 2H11 Cells.

MISSION pLKO.1 lentiviral shRNA (MilliporeSigma) was used to knock down LOX or CXCL14 in 2H11 cells. shRNA sequences are listed in SI Appendix, Table S4.

Mineralization Assays.

2H11 cells or 2H11 cells infected with shRNA were cultured in serum-free Dulbecco's Modified Eagle Medium (DMEM) overnight and then treated with or without BMP4 (100 ng/mL) for 48 h, switched to OSB differentiation medium for 14 d with fresh medium changes every 3 d as described (20).

Generation of MDA PCa-118b Tumors and Treatment with ATRA, Palovarotene, LDN193189, or Ra223.

MDA PCa-118b cells were injected into SCID mice subcutaneously. After 3 wk, mice bearing MDA PCa-118b tumors with similar tumor size were treated with ATRA, palovarotene, or LDN193189 as described (20). Treatment of C4-2b-BMP4 tumors with Ra223 was as described (62).

Generation of MycCaP-BMP4 Cell Line by Dilution Cloning.

cDNA for mouse BMP4 was first cloned into FUW-Luc-mCh-puro vector. Luciferase, mCherry, and Puro markers were excised from the plasmid to generate FUW-BMP4 vector, which was transfected into MycCaP cells, and cells were diluted into 96-well plates. The expression of BMP4 protein in the culture supernatants was detected by western blot. Clones that expressed higher levels of BMP4 were further diluted and screened. Clones that expressed relatively high levels of BMP4 were selected as MycCaP-BMP4 cell line.

RNAseq of Tumors and TAMs.

For RNAseq analysis of tumors, tumors were homogenized in RLT lysis buffer [RNAeasy Mini Kit (Qiagen)] using an economical benchtop homogenizer (PT1600E). For RNAseq analysis of TAMs, the digested cells from tumors were FACS-sorted for CD45⁺, CD11b⁺, GR1⁻, and F4/80⁺ cells. RNA was isolated using the RNAeasy Mini Kit. RNAseq analysis was performed at the UTHealth Houston Cancer Genomics Center (Houston, TX).

CD8⁺ T Cell Isolation and T Cell Suppression Assays.

Mouse spleens from 8- to 12-wk-old C57B6 mice were placed in C-Tube (Miltenyi Biotec). After mechanical disruption, the cells that passed through the 70-μm cell strainer were purified with the EasySep Mouse CD8⁺ T Cell Isolation Kit (STEMCELL Technologies Inc.). For T cell proliferation assays, T cells were labeled with CFSE, stimulated with CD3/CD28 T-Activator and mouse IL-2 (Gibco), and cocultured with F4/80⁺ macrophages at a 2:1 ratio (T cells:macrophages). Seventy-two hours later, cells were analyzed by flow cytometry. For Wnt pathway inhibition, bone-TAMs isolated from MycCaP-BMP4 tumors were pretreated with XAV-939 or LGK-974 for 1 h before being cocultured with T cells. For the cytotoxicity assay, splenocytes from OT-1 mice were treated with CD3/CD28 T-Activator and mouse IL-2 (Gibco) and ovalbumin peptide OVA257-264 (2 μg/mL) for 48 h. Ctrl-TAMs or bone-TAMs (2 × 10⁴/well) were then cocultured with splenocytes from OT-1 mice (2 × 10⁴/well). After overnight culture, B16F10/OVA cells (1 × 10⁴/well) were added to the cocultures. Lactate dehydrogenase (LDH) released in culture supernatants was measured using the CytoTox96 nonradioactive cytotoxicity assay (Promega) per manufacturer’s instructions.

Study Approval.

Animal studies were approved by the Institutional Animal Care and Use Committee at MD Anderson Cancer Center.

Statistical Analysis.

Data were expressed as mean ± SD. For pairwise comparisons, Student’s t test was used. For multiple group comparisons, pairwise comparisons between groups were performed. The Bonferroni correction method was used for adjusting P values. P < 0.05 was considered statistically significant.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

RNAseq data have been deposited in [GEO] (GSE168321) (63). All other data are included in the article and/or SI Appendix.