Retinoic Acid Receptor Activation Reduces Metastatic Prostate Cancer Bone Lesions by Blocking the Endothelial-to-Osteoblast Transition

Metastatic progression in bone signifies the lethal progression of advanced prostate cancer. Currently approved therapeutic agents for prostate cancer bone metastasis only modestly affect patient survival (1). Thus, there is an urgent need to develop strategies that improve therapy outcomes. Prostate cancer bone metastasis is frequently associated with osteoblastic bone-forming lesions (2), indicative of cross-talk between metastatic tumor cells and the stromal component within the tumor microenvironment. We and others have shown that BMPs secreted from prostate cancer cells are involved in the formation of bone lesions in prostate cancer bone metastasis. First, in human-derived osteoblastic patient-derived xenograft (PDX) MDA PCa-118b, we identified BMP4 by cytokine array as a factor that promotes ectopic bone formation of MDA PCa-118b (3). Second, in human prostate cancer specimens, we found that BMP4 expression was low in 15 of 15 primary prostate cancer specimens, but significantly higher in tumor cells in 10 of 12 needle biopsies of bone metastasis (4). Third, Nordstrand and colleagues (5) showed that BMP4 is expressed in human prostate cancer bone metastasis specimens and BMP4 expression correlates with increased bone formation. Fourth, Dai and colleagues (6) showed that BMP6 contributes to the bone-forming phenotype of prostate cancer bone metastasis. Moreover, Morrissey and colleagues showed that BMP7 (7) is highly expressed in bone metastasis. At present, the mechanism that leads to the upregulation of BMPs in prostate cancer bone metastasis is not clear. Dai and colleagues (8) reported that Wnt3a and Wnt5a induce BMP4 and BMP6 expression in prostate cancer cells. Morrissey and colleagues (7) showed that BMP7 levels are increased in castrate-resistant prostate cancer. Nordstrand and colleagues (5) showed that high BMP signaling inversely correlates with low tumor cell androgen receptor activity in human prostate cancer bone metastasis. Whether androgen depletion therapy leads to BMP upregulation in prostate cancer cells requires further investigation.

A role of osteoblasts (OSB) in prostate cancer progression in bone was demonstrated by Lin and colleagues (4), which showed that reduction of tumor-induced bone formation decreased osteogenic tumor growth in a genetically modified mouse model (4). In addition, inhibition of tumor-induced bone formation, by the BMP receptor inhibitor LDN-193189 (3) or an anti-BMP6 antibody (6), also reduced tumor growth in mouse models. Consistent with this notion, bone-targeting radionuclides, including Sr89 and Samarium-153 EDTMP, were developed for prostate cancer bone metastasis, but significant bone marrow suppression limited their clinical use (9). Radium-223, a bone-targeting alpha emitter that remodels bone by inducing cell death in bone microenvironment, increases survival, albeit modest, of patients with bone metastasis (10). As prostate cancer–induced bone formation also increases therapy resistance (11), inhibition of tumor-induced bone formation may improve therapy outcome.

The mechanism of prostate cancer–induced aberrant bone formation was thought to result from the expansion of existing OSBs. However, we recently elucidated an alternative mechanism whereby prostate cancer–induced aberrant bone formation occurs via endothelial-to-osteoblast (EC-to-OSB) transition mediated by tumor-secreted BMP4 (4). We showed that the tumor-associated OSBs in prostate cancer bone metastasis specimens and PDXs coexpress endothelial marker Tie-2 and OSB marker osteocalcin (4); BMP4 promoted EC-to-OSB transition of 2H11 ECs (4); BMP4-overexpression in nonosteogenic C4-2b prostate cancer cells led to ectopic bone formation under subcutaneous implantation (4); and tumor-induced bone was reduced in trigenic mice (Tie2^cre/Osx^f/f/SCID) with EC-specific deletion of OSB cell-fate determinant OSX when compared with that in bigenic mice (Osx^f/f/SCID) (4). Targeting this unique EC-to-OSB transition may reduce prostate cancer–induced bone formation.

This new understanding of EC-to-OSB transition in prostate cancer–induced bone formation has revealed similarities in the mechanism(s) of prostate cancer–induced bone and ectopic bone arising from the genetic disease fibrodysplasia ossificans progressiva (FOP), also known as stoneman disease (12, 13). FOP is a rare genetic disorder in which muscle and connective tissues are gradually replaced by bone, forming bone outside the skeleton (heterotopic ossification). FOP is caused by a hereditary gain-of-function genetic mutation (R206H) in the BMP type I receptor ALK2 (14, 15). Mechanistically, the cell source for ectopic bone formation is shown to be ECs that undergo endothelial-to-mesenchymal transition to become mesenchymal stem cells, which further differentiate into bone tissues (12). Thus, in both diseases, the aberrant bone originates from ECs with overactivated BMP signaling. We posit that therapies to prevent heterotopic ossification in FOP could be repurposed for inhibiting prostate cancer–induced aberrant bone formation.

Various strategies have been developed for the treatment of FOP, with most targeting the ALK2 mutation (R206H; ref. 16). Recently, Shimono and colleagues (17) showed that compounds that activate the nuclear receptor, retinoic acid receptor gamma (RARγ), are effective in inhibiting chondrogenesis and preventing heterotopic ossification in FOP. As cartilage development normally requires an unliganded RARγ repressor function, the presence of ligands (RARγ agonists) may interfere with transcriptional programming for chondrogenesis and osteogenesis (18). Currently, palovarotene (Palo) is being evaluated as therapy for patients with FOP in a phase III clinical trial (NCT03312634). Palo may be repurposed to inhibit prostate cancer–induced aberrant bone formation.

RARγ acts as a transcriptional repressor by recruiting corepressors to target promoters in the absence of ligands. When ligands are present, RARγ interacts with coactivators to activate gene transcription (19, 20). Retinoic acid signaling plays a critical role in modulating skeletal development (21–23). Overexpression of RARα in transgenic animals leads to appendicular skeletal defects (24). A loss of retinoid receptor–mediated signaling is necessary and sufficient for expression of bone phenotype (25). Thus, retinoic acid signaling inhibits skeletal progenitor differentiation (23).

In this study, we examined whether Palo may be repurposed to inhibit prostate cancer–induced aberrant bone formation. We used MDA-PCa-118b, which is a human-derived PDX endogenously expresses BMP4, as well as prostate cancer cells engineered to overexpress BMP4 in our study. We found that RAR agonists, including Palo and all-trans-retinoic acid (ATRA) inhibit BMP4-induced EC-to-OSB transition and bone formation. In addition, RARs regulate BMP4-induced EC-to-OSB transition through a noncanonical RAR mechanism that is independent of RAR's transcription function. Furthermore, we found that plasma levels of Tenascin C (TNC), a protein secreted by EC-OSB cells, can be used for monitoring treatment response. ATRA treatment also does not lead to unwanted bone loss in healthy nontumor-bearing bones. Thus, Palo and ATRA represent new agents for the treatment of osteoblastic bone metastasis of prostate cancer.

Cell lines, antibodies, and reagents are listed in Supplementary Table S1. Cells were tested for Mycoplasma using MycoAlert Mycoplasma Detection Kit (LONZA, LT07-418) and authenticated using short tandem repeat.

2H11 cells treated with BMP4 and/or Palo were switched into adipocyte differentiation medium (RPMI Medium with 10% charcoal-stripped serum, 2 μg/mL insulin, 1 μmol/L dexamethasone, and 2 μmol/L rosiglitazone) for 10 days. For chondrogenic differentiation, treated 2H11 cells were cultured in chondrogenic differentiation medium (Lonza) for 14 days. Mesenchymal stem cell line C3H10T1/2 was used as a positive control. Oil Red O staining for adipocytes was performed for 30 minutes on cells grown in adipocyte differentiation medium. Alcian blue staining for chondrocytes was performed on cells cultured in chondrogenic medium.

Total RNA was extracted using RNeasy Mini Kit (Qiagen). cDNA was synthesized from 2 μg of total RNA using TaqMan Reverse Transcription Reagents (Life Technologies). qRT-PCR was performed using SYBR green (Applied Biosystems) fluorescence signals under the following PCR conditions: 40 cycles of 15-second denaturation at 95°C and 1-minute amplification at 60°C. All reactions were run in duplicates and normalized using Gapdh. The mouse-specific primer sequences are listed in Supplementary Table S2.

Cell lysates (20 μg) were resolved on SDS-PAGE, and protein bands were analyzed by immunoblotting with indicated antibody. Protein levels were quantified against loading control using ImageJ. For immunofluorescence analysis, cells were fixed and stained with primary antibodies (1:50–1:500) overnight, followed by Alexa-Fluor conjugated secondary antibodies. Images were acquired on a Nikon TE2000 widefield microscope system (26). For immunoprecipitation, nuclear fractions were isolated using lysis buffer as described previously (27). Lysates were incubated overnight with antibodies and the protein-antibody complexes were analyzed by Western blotting.

2H11 cells (murine lymphoid endothelial cells) were cultured in serum-free DMEM overnight followed by treatment with BMP4 (100 ng/mL), Palo (1 μmol/L), or both in serum-free medium for 48 hours, then switched into OSB differentiation medium (α-MEM containing 10% FBS, 100 μg/mL ascorbic acid, and 5 mmol/L β-glycerophosphate) for 12–14 days with fresh medium changes every 3 days as described previously (27). MS1 (murine pancreatic microvascular endothelial cells) and SVR (MS1 cells transfected with H-ras) were treated in 0.5% FBS DMEM plus BMP4, Palo, or both overnight, then switched into OSB differentiation medium for 12 to 14 days. Human umbilical vein endothelial cells (HUVEC) primary cells (Lonza) from passage 1–4 were used for study. HUVEC cells were cultured in EGM Endothelial Cell Growth Medium. Confluent HUVEC cells were cultured in EGM serum-free medium overnight, and switched to HMSC Osteogenic BulletKit medium (Lonza) with BMP4 and/or Palo for 14 to 21 days. Cells were fixed with formalin and incubated with Alizarin Red S solution or AgNO₃ solution (von Kossa staining) as described previously (4). The mineralized nodules were quantified using Image J software.

Mouse lung was cut into very small pieces and incubated in 0.2% collagenase DPBS with Ca²⁺/Mg (Gibco, 17100-017). The tissues were digested at 37°C for 45 minutes in an incubator with constant rotation. DMEM containing 10% FBS was added to the suspension to stop the digestion. The suspension was filtered through a 70 μm cell strainer and centrifuged at 400 × g at 4°C for 8 minutes. The pellet represents mouse lung endothelial cells (MLEC). MLECs were cultured in endothelial cell growth medium (high-glucose DMED with 10% FBS, 100 μmol/L nonessential amino acids, 100 μmol/L sodium pyruvate, 0.1 unit/mL heparin, and 100 μg/mL endothelial cell growth factor). For MLEC differentiation, MLECs were cultured in serum-free DMEM overnight followed by treatment with BMP4, Palo, or both in serum-free medium for 48 hours, then switched into Stempro OSB differentiation medium for 14 to 20 days.

C4-2b-BMP4-LT cell line, expressing BMP4, luciferase reporter, and red fluorescence protein Tomato, was generated as described previously (4). C4-2b cells that expressed both BMP4 and Luc were selected with G418 followed with FACS. TRAMP-BMP4 and MycCaP-BMP4 cell lines were generated by using a bicistronic retroviral vector that contains mouse BMP4 cDNA. The luciferase reporters were introduced into TRAMP-BMP4 and MycCaP-BMP4 cells using bicistronic retroviral vector pBMN-Luc-IRES-Tomato and a lentivirus vector FUW-Luc-mCh-Puro, respectively. The TRAMP-BMP4-LT or MycCaP-BMP4-LC cells that coexpressed both BMP4 and Luc were selected with puromycin (for BMP4) and FACS (for Tomato or mCherry).

Tumor cells were injected subcutaneously in male SCID mice. Mice were treated by oral gavage with Palo (2 mg/kg/day), ATRA (6–12 mg/kg/day), or corn oil (as control).

TRAMP-BMP4-LT (1 × 10⁶) or MycCaP-BMP4-LC (0.5 × 10⁶) cells were injected into femurs of SCID mice. Palo (2 mg/kg/day) or ATRA (6–12 mg/kg/day) was administered by daily oral gavage 3 days before tumor inoculation and continued throughout the treatment period. Tumor growth was monitored weekly by bioluminescent imaging (BLI). Femurs were fixed in 10% paraformaldehyde for μCT analysis using Bruker SkyScan 1276 (Bruker). CT Analyzer (Bruker) software (version 1.17.7.2) was used for data analyses.

Tumor sections were stained for von Kossa (silver staining) or Goldner trichrome to visualize mineralization.

Cell proliferation was determined by viable cell counting by CellTiter 96 AQueous Non-Radioactive Cell Proliferation Assay (MTS) Kit (Promega).

2H11 cells cultured in serum-free DMEM were treated with BMP4 (100 ng/mL) for 24 hours, fixed and coincubated overnight with pairs of primary antibodies (sc, Santa Cruz Biotechnology; cs, Cell Signaling Technology) as follows: anti-Smad1 (mouse sc7965, 1:50) and anti-RARγ (rabbit cs8965, 1:100); anti-Smurf1 (Smad ubiquitination regulatory factor1; mouse sc-100616, 1:50) and anti-pSmad1/5 (rabbit cs9516, 1:100); anti-Smurf1 (mouse sc-100616, 1:50) and anti-RARγ (rabbit cs8965, 1:100). Proximity ligation assays (PLA) were performed using DUOLink In Situ Red Starter Kit (DUO92101; Sigma) with anti-Mouse MINUS (DUO92004) and anti-Rabbit PLUS (DUO92002) PLA probes as described previously (26). Incubations with a single antibody or no antibody were used as negative PLA controls. PLA images were acquired on a Nikon TE-2000 confocal microscope and analyzed using NIS-Elements 5.21.02 software (Nikon; ref. 26).

2H11 cells were treated with BMP4 (100 ng/mL) with or without Palo (1 μmol/L) in serum-free medium for 48 hours. Total RNA was extracted using RNeasy Mini Kit (Qiagen) and prepared for RNA sequencing (RNA-seq) analysis at Arraystar Inc. Transcriptome data have been deposited at GSE168157.

RARα, RARβ, RARγ, or Smurf1 in 2H11 cells were knocked down using MISSION pLKO.1 lentiviral short hairpin RNA (shRNA; MilliPORE Sigma). shRNA clones were selected with 5 μg/mL puromycin. Cells infected with empty pLKO.1 lentiviral vector were used as controls. Primers for shRNA are listed (Supplementary Table S2).

MDA PCa-118b PDX was maintained by serial passage in SCID mice as described previously (3). MDA PCa-118b (1 × 10⁶) were injected into SCID mice subcutaneously. After 3 weeks, mice bearing MDA PCa-118b tumors were grouped into two groups based on similar tumor size. ATRA (6–12 mg/kg/day) was administered through oral gavage daily 6 days per week. Control mice were administered with corn oil. Tumor sizes were measured by caliper.

TNC protein level was determined using mouse TNC ELISA Kit (NBP2-78767; Novus Biological Inc). Serum samples from mice bearing MycCaP-BMP4 tumor treated with or without ATRA or Palo were diluted in PBS and used in ELISA.

Bone histomorphometry analysis on nondecalcified mouse femurs was performed by M.D. Anderson Cancer Center Bone Histomorphometry Core Laboratory as described previously (28). Femurs were sectioned at 5 μm thickness in a longitudinal plane. The measurement area excluded the 150 μm area adjacent to the cortical bone. For each sample, measurements for OSBs and osteoclasts were performed on two separate 5-μm-thick Toluidine Blue–stained and TRAP-stained sections, respectively. Blood vessel quantification was performed on one Toluidine Blue–stained 5-μm-thick bone section.

To isolate mouse adipocyte stem cells (mADSC), adipose tissues were collected from 8-week-old mice and cut into small pieces. Collagenase (0.2%) was added and adipose tissues were incubated at 37°C for 1 hour. DMEM containing 10% FBS was added to the suspension and centrifuged at 1,200 rpm for 5 minutes. mADSCs at the bottom of the tube were isolated and cultured in adipocyte differentiation medium or OSB differentiation medium.

Basement Membrane Matrix (356260, Corning) was added to a 48-well plate (100 μL/well) and incubated for 12 hours at 37°C for solidification. 2H11 cells (2 × 10⁴) were added to the Matrigel with or without various concentrations of Palo. Cells treated with 0.1 μmol/L cabozantinib was used as a positive control. Tube formation was observed at 0 or 20 hours following transfer to Matrigel and quantified by detecting the nodes in four randomly chosen fields.

Formalin-fixed, paraffin-embedded C4-2b or MycCaP tumors were generated by injecting C4-2b or MycCaP cells subcutaneously into SCID mice treated with or without ATRA. The tumor sections were incubated with anti-CD31 antibody overnight at 4°C, followed with horseradish peroxidase–conjugated secondary antibodies. Images were scanned and analyzed by Aperio ImageScope software.

Animal studies have been approved by the Institutional Animal Care and Use Committee at M.D. Anderson Cancer Center.

Data were expressed as the mean ± SD. P value was computed by two-sided t test. P < 0.05 by Student t test was considered statistically significant.

Materials (plasmids) generated in this study will be made available upon reasonable request to the lead contact. Their delivery could require a material transfer agreement. The sequencing data generated from this study are publicly available in Gene Expression Omnibus (RRID:SCR_005012) at GSE168157 and GSE168321.

We employed 2H11 cells, a murine lymphoid EC line (29), for our EC-to-OSB transition studies. As 2H11 cells also express CD34 (29), a stem cell marker, we evaluated whether 2H11 cells possess multilineage differentiation potential beyond OSBs. We observed that 2H11 cells cannot be differentiated into chondrocytes or adipocytes (Supplementary Fig. S1A and S1B), when compared with the pluripotent stem cell line C3H10T1/2 that differentiated into chondrocytes or adipocytes under the same culturing conditions (Supplementary Fig. S1A and S1B). We found that 2H11 ECs can be induced to undergo EC-to-OSB transition by treating with BMP4 (100 ng/mL) for 48 hours, based on the expression of OSX, an OSB cell fate determination factor, and osteocalcin (Fig. 1A and B). Addition of Palo significantly inhibited BMP4-induced EC-to-OSB transition as shown by the inhibition of OSX and osteocalcin expression (Fig. 1A and B), and OSB mineralization, as indicated by Alizarin Red (Fig. 1C) and von Kossa staining (Fig. 1D), compared with BMP4 alone when cultured in OSB differentiation medium for 21 days. Without BMP4, Palo alone had no effect on EC-to-OSB transition. These results suggest that Palo inhibits BMP4-induced EC-to-OSB transition.

Upon BMP4 treatment, 2H11 cells first transition into OSB by day 2 (days 0–2, transition phase). EC-OSB hybrid cells then undergo OSB differentiation and express bone matrix proteins (days 3–14, differentiation phase), and finally mineralize by day 20 (mineralization phase; Fig. 1E). To examine whether Palo could inhibit BMP4-induced bone formation after EC-to-OSB transition has already been initiated, Palo was added to cells at different transition stages. Palo was most effective during early transition (D0–D2), reaching approximately 80% to 90% inhibition of BMP4-stimulated activities (Fig. 1G and I). Palo still inhibited OSX (Fig. 1F) and osteocalcin expression (Fig. 1H), albeit with decreasing efficacy when added to the cultures on D2, D5, D8, or D11 (Fig. 1F–I). Similarly, inhibition of mineralization was also most effective during early transition and less effective when Palo was added in later phases of EC-to-OSB transition (Fig. 1J and K). These observations suggest that Palo is most effective in inhibiting early stages of BMP4-induced EC-to-OSB transition.

To assess whether other ECs can also undergo transition into OSBs, we treated primary MLECs, MS-1 cells (a murine pancreatic microvascular EC line), SVR cells (an H-ras transfected MS-1 cell line; ref. 30), and primary HUVECs with BMP4 and cultured the cells in osteogenic medium. We found that all of these ECs can be induced to become OSBs by BMP4 as evidenced by an increase in Alizarin Red staining (Supplementary Fig. S1C–S1F). Addition of Palo significantly inhibited BMP4-induced EC-to-OSB transition as indicated by the inhibition of Alizarin Red staining (Supplementary Fig. S1C–S1F). These results suggest that BMP4 can induce EC-to-OSB transition in ECs derived from several organs and that Palo can inhibit this BMP4-induced EC-to-OSB transition.

We employed three osteogenic xenograft models to examine Palo effects on prostate cancer–induced bone formation. In the first model, C4-2b-BMP4 cells, which form osteogenic tumors (4), were injected subcutaneously into SCID mice (Fig. 2A). Mice were treated with or without Palo (2 mg/kg/day) by oral gavage. Palo significantly reduced tumor-induced bone formation compared with those in vehicle-treated control tumors, as determined by Goldner Trichrome staining (Fig. 2B). Palo also led to a moderate decrease in tumor sizes compared with controls, as determined by BLI (Fig. 2C) and tumor weight (Fig. 2D). However, Palo did not affect C4-2b-BMP4 cell proliferation in vitro (Supplementary Fig. S2A) or mouse body weight (Supplementary Fig. S2B). These observations suggest that Palo reduces tumor-induced bone formation and tumor size, consistent with previous observation that prostate cancer–induced bone supports prostate tumor growth (3, 6).

In a second model, we injected osteogenic TRAMP-BMP4 cells intrafemorally into SCID mice and treated these tumor-bearing mice with or without Palo (Fig. 2E). TRAMP-BMP4 cells, generated by BMP4-cDNA transfection into TRAMP-C2 cells (31), express high levels of BMP4 protein (Supplementary Fig. S2C). TRAMP-BMP4 increased the bone mineral density (BMD) of the tumor-bearing femurs (Fig. 2F, left), and induced bone formation in the bone marrow (Fig. 2F, middle) compared with control uninjected legs. Palo significantly reduced the BMD in the tumor-bearing femurs compared with those in nontreated mice (Fig. 2F, right), and moderately decreased prostate cancer growth in bone at week 4, although the decrease did not reach statistical significance (Fig. 2G). Palo did not affect TRAMP-BMP4 cell proliferation in vitro (Supplementary Fig. S2D). In this experiment, Palo led to a small decrease in mouse body weight compared with untreated mice (Supplementary Fig. S2E). These results suggest that Palo reduces TRAMP-BMP4–induced bone formation in mouse femurs.

In a third model, MycCaP-BMP4 cells were injected intrafemorally into SCID mice and treated with or without Palo (Fig. 2H). MycCaP-BMP4 cells, generated by BMP4-cDNA transfection into MycCaP cells that were derived from prostate tumor of Hi-Myc mice (32), express a high level of BMP4 protein (Supplementary Fig. S2F). X-Ray analysis showed that MycCaP-BMP4 in femurs induced an osteoblastic bone response in tumor-bearing femurs, and Palo significantly reduced this response (Fig. 2H). Palo also exhibited a trend toward reduced tumor growth (Fig. 2I). Palo did not affect MycCaP-BMP4 cell proliferation in vitro (Supplementary Fig. S2G) and mouse body weight (Supplementary Fig. S2H). Together, the combined results from these three osteogenic xenograft models suggest that Palo reduces prostate cancer–induced bone formation accompanied with a trend of decreased tumor growth.

Smad1 phosphorylation is one of the downstream BMP4 signaling pathways in 2H11 cells (27). pSmad1 levels were significantly increased by BMP4, and Palo reduced BMP4-stimulated pSmad1 levels (Fig. 3A). Total Smad1 was also increased by BMP4 and slightly decreased by Palo (Fig. 3A). BMP4 also increased nuclear localization of pSmad1, which was decreased by Palo (Fig. 3B). Similar patterns of distribution were observed with Smad1 (Fig. 3B). In contrast, total and nuclear RARγ levels were not affected by BMP4 or Palo (Fig. 3A and B). These observations suggest that Palo leads to a decrease in BMP4-stimulated nuclear pSmad1 levels.

Using nuclear extracts, we showed that immunoprecipitation of RARγ pulled down pSmad1, while the reciprocal immunoprecipitation of Smad1 pulled down RARγ (Fig. 3C), suggesting that pSmad1 forms a complex with RARγ in BMP4-treated 2H11 cells. Furthermore, in situ PLA for the potential physical association between pSmad1 and RARγ showed a significant increase in PLA signals, visualized as individual fluorescent spots, in the nucleus of BMP4-treated 2H11 cells but not in control cells (Fig. 3D). No PLA signals were observed in BMP4-treated cells incubated with a single antibody alone as controls (Fig. 3D). Palo was found to reduce BMP4-mediated nuclear pSmad1/RARγ complex formation by both co-immunoprecipitation (Fig. 3C) and PLA (Fig. 3D). These results suggest that Palo inhibits BMP4-stimulated pSmad1/RARγ nuclear complex formation.

We postulated that Palo inhibits BMP4-stimulated pSmad1/RARγ nuclear complex formation by enhancing its degradation via Smurf1 (33). Indeed, Palo led to the recruitment of E3-ubiquitin ligase Smurf1 to the nuclear pSmad1/RARγ complex (Fig. 3E). Furthermore, PLA showed a significant increase in Smurf1 association with pSmad1 (Fig. 3F) as well as Smurf1 association with RARγ (Fig. 3G), only in the presence of RAR activation by Palo. Together, these observations suggest that Palo recruits Smurf1 to the nuclear RARγ/pSmad1 complex.

We knocked down Smurf1 and found an increase in pSmad1 levels in 2H11-shSmurf1#1 and #2 clones compared with vector control (Fig. 3H), consistent with a role of Smurf1 in regulating pSmad1 levels. In 2H11-shSmurf1 clones, Palo-mediated decrease in pSmad1 was abrogated compared with vector controls (Fig. 3I), suggesting that Palo mediates pSmad1 degradation through Smurf1, likely through pSmad1 ubiquitination for proteasomal degradation. Together, our studies suggest that RAR activation by Palo causes BMP4-stimulated pSmad1 degradation by recruiting Smurf1 into the nuclear pSmad1/RARγ complex (Fig. 3J).

We characterized the Palo-regulated transcriptome by RNA-seq. RNA-seq analysis showed that mRNA levels of 591 genes were upregulated more than 1.5-fold by BMP4 when compared with 2H11 cells (2H11-BMP4 vs. 2H11; Fig. 4A; see GSE168321). By using fold enrichment to select for the top 215 pathways followed with lowest FDR, pathway analysis showed that regulation of OSB differentiation, bone morphogenesis and bone mineralization were the most significantly upregulated pathways by BMP4 (Fig. 4B), consistent with ECs transition into OSBs. Upon Palo treatment for 48 hours, mRNA levels of 756 genes were downregulated compared with BMP-treated sample [(BMP4+Palo) vs. BMP4] (Fig. 4A; see GSE168157). Pathway analysis showed that genes involved in regulation of ossification were most significantly downregulated by Palo (Fig. 4C), consistent with an effect of Palo on inhibiting bone formation.

We identified 146 transcripts that were upregulated by BMP4 and downregulated by Palo (Fig. 4A). Pathway analysis showed that these 146 transcripts are enriched in genes involved in regulation of epithelial-to-mesenchymal transition and the function of OSBs, osteoclasts, and chondrocytes in rheumatoid arthritis (Fig. 4D), consistent with a role of Palo in inhibiting EC-to-OSB transition. Furthermore, we found that eight transcription factors that play critical roles in cell fate change, including Slug, Foxo1, and Id4, were downregulated by Palo (Fig. 4E). Indeed, mRNAs of these transcription factors were upregulated by BMP4 and downregulated by Palo (Fig. 4E). Slug, also known as Snai2, is a cell fate determinant (34). Foxo1 regulates OSB numbers, bone mass, and the differentiation of mesenchymal cells into OSBs (35). Id4 is a DNA-binding protein that is a downstream effector of BMP signaling and regulates various cellular processes (36). Because the expression of Id proteins (Id1, Id2, Id3, and Id4) has been reported to be regulated by BMPs (37, 38), we further examined the expression of all four Ids in 2H11 cells. Our RNA-seq analysis showed that only Id1 and Id4, but not Id2 and Id3, expression was downregulated by Palo (Supplementary Fig. S3A). qRT-PCR analysis showed that while all four Ids were upregulated by BMP4 during EC-to-OSB transition, only Id1 and Id4, but not Id2 and Id3, were downregulated by Palo (Supplementary Fig. S3B), consistent with the RNA-seq analysis (Supplementary Fig. S3A). The differential effects of Palo on BMP4-induced Id expression and whether Id1 and Id4 are involved in EC-to-OSB transition require further investigation. These results suggest that the binding of Palo to RARγ alters Smad1 signaling, leading to transcriptional changes of several cell fate–related transcription factors involved in osteoblastogenesis.

We also identified 39 secreted proteins upregulated by BMP4 and downregulated by Palo (Fig. 4F; Supplementary Table S3). These proteins, including growth factors, cytokines, proteases, and bone matrix proteins (Fig. 4G; Supplementary Table S3), collectively termed “osteocrines” (11), may be involved in stromal-prostate cancer communication critical for prostate cancer progression in bone. The mRNA levels of Fgf10, Tslp, Mgp, and Tnc were upregulated by BMP4 and significantly downregulated by Palo in 2H11 cells (Fig. 4H). Fgf10 is required for limb and lung development (39, 40). TNC, a bone matrix protein, plays a role in inducing stemness and increasing metastatic potential of tumor cells (41). Thymic stromal lymphopoietin, a Th2 cytokine, is involved in the induction and progression of various tumors (42). These “osteocrines” may be involved in Palo-mediated inhibition of tumor growth and can be used as serum biomarkers for monitoring response to Palo treatment.

RAR family proteins contain three isoforms, RARα, RARβ, and RARγ. Palo is a synthetic retinoic acid that has a high affinity for RARγ. Whether RARα and/or RARβ are involved in EC-to-OSB transition is not clear. We determined that 2H11 cells express all three isoforms (Fig. 5A–C). Knockdown of RARα, β, or γ (Fig. 5A–C) significantly reduced BMP4-induced osteocalcin expression in 2H11-shRARα, β and γ clones (Fig. 5A–C). Thus, all three RAR isoforms are involved in BMP4-induced EC-to-OSB transition. Knockdown of RAR isoforms also led to an inhibition of BMP4-induced pSmad1 levels (Fig. 5D), suggesting similar mechanisms of inhibition mediated by the three RAR isoforms.

We next examined whether retinoic acids with broad spectrum for RAR isoforms can also inhibit BMP4-induced EC-to-OSB transition. We found that tretinoin (ATRA) and isotretinoin (13-cis-RA), pan-RAR agonists used in the clinic (43), both inhibited EC-to-OSB transition in a dose-dependent manner (Fig. 5E–G), to a similar extent as that of Palo. ATRA and 13-cis-RA also inhibited BMP4-induced nuclear pSmad1 to levels similar as that by Palo (Fig. 5H). Thus, RAR agonists, including Palo, ATRA and 13-cis-RA, can inhibit BMP4-induced EC-to-OSB transition.

In an initial study, ATRA was found to reduce MycCaP-BMP4 tumor growth (Supplementary Fig. S4A), with a trend of reduced BMD, percent bone volume (BV/TV), and bone surface density (BS/TV) compared with control nontreated mice (Supplementary Fig. S4B). However, ATRA (12 mg/kg) led to a significant decrease in body weight (Supplementary Fig. S4C). Dose reduction to 6 mg/kg (Supplementary Fig. S4A, arrow) alleviated toxicity but also lessened the tumor response (Supplementary Fig. S4A). Thus, the study was repeated with modifications (Fig. 6A). Mice were inoculated subQ+intrafemur with MycCaP-BMP4 cells and one group received ATRA (10 mg/kg) and the other received vehicle. After 1 week, both groups were castrated (Fig. 6A, red arrow). Castration led to a temporary halt or decrease of tumor growth in both control and ATRA-treated mice, after which, tumor growth resumed in both groups, and tumor growth was reduced by ATRA (Fig. 6A). Subcutaneous MycCaP-BMP4 tumors exhibited ectopic bone formation (Fig. 6B), similar to those observed with C4-2b-BMP4 tumors (Fig. 2A). ATRA reduced MycCaP-BMP4–induced bone formation measured by μCT and von Kossa staining (Fig. 6B).

In the MycCaP-BMP4 tumor-bearing femurs, ATRA led to a trend of decreases in BMD, BV/TV, BS/TV, and trabecular bone density (Tb.N) in the tumor-bearing femurs compared with those in control mice (Fig. 6C), although some parameters did not reach statistical significance. ATRA did not affect MycCaP-BMP4 cell proliferation (Supplementary Fig. S4D), as also observed with Palo (Supplementary Fig. S2G). A small decrease in body weight was observed at day 28 (Supplementary Fig. S4E). Together, these results suggest that ATRA reduces MycCaP-BMP4–induced bone formation and tumor growth in mice undergoing castration.

MDA PCa-118b expresses BMP4 endogenously and generates bone-forming tumors when implanted subcutaneously (3). In MDA PCa-118b PDX model, due to significant variation in tumor sizes, ATRA treatment was initiated after 3 weeks, when tumors have reached measurable sizes. We found that ATRA also reduced bone formation and tumor growth of the MDA PCa-118b xenograft (Fig. 6D and E). In addition, we also repeated the MDA-PCa-118b studies using longer treatment times. After tumors were grown for 3 weeks, tumors that reached the size of 35 mm³ were randomized for ATRA treatment (Supplementary Fig. S5A). Tumors in the control mice grew fast and reached large tumor sizes after 18 days. However, tumors in the ATRA-treated mice grew slower and the treatment was continued for an additional 21 days (Supplementary Fig. S5B–S5D). Together, these results suggest that ATRA not only inhibits MDA PCa-118b bone formation but also reduces the growth of preformed prostate cancer tumor.

In preparation for using Palo or ATRA to reduce tumor-induced bone formation in patients with castrate-resistant osteoblastic bone metastasis, we examined TNC levels, which was increased during EC-to-OSB transition (Fig. 4H; ref. 44), in the plasma of SCID mice with intrafemur MycCaP-BMP4 tumors treated with or without ATRA or Palo. Both ATRA and Palo significantly reduced plasma TNC levels compared with untreated controls, as quantified by ELISA (Fig. 6F), consistent with inhibition of EC-to-OSB transition by ATRA or Palo. These results suggest that plasma TNC levels could be used to monitor response to treatments.

We also examined whether ATRA leads to unwanted bone loss in healthy nontumor-bearing bones. Bone histomorphometry analyses of contralateral nontumor-containing femurs from MycCaP-BMP4 castration plus ATRA-treated animals (Fig. 6C) showed that there were no significant changes in several bone parameters compared with control castrated mice (Fig. 6G). Similarly, no consistent changes on parameters related to activities of OSBs and osteoclasts (Supplementary Fig. S6A and S6B) were observed. Our previous study showed that castration reduced bone volume and an increase in osteoclast density (45). These results suggest that ATRA does not further worsen the osteoporotic effects from castration.

Our studies identify a strategy to target prostate cancer–induced bone formation to interfere with the cross-talk between stromal components and metastatic prostate cancer tumor cells. We show that activation of RAR by agonists, including Palo and ATRA, inhibits stromal reprogramming through blocking BMP4-induced EC-to-OSB transition and reduces prostate cancer–induced aberrant bone formation and tumor growth in vivo (Fig. 6H). We also show that Palo and ATRA inhibit EC-to-OSB transition through a noncanonical RAR pathway (Fig. 6H). Because prostate cancer–induced aberrant bone formation enhances prostate cancer progression (3) and also resistance to therapy (11), our studies suggest that Palo or ATRA may be used to reduce tumor-induced bone formation to improve therapy outcomes for patients with prostate cancer bone metastasis. Presently, effective therapies directed at blocking mechanisms of prostate cancer bone metastasis are limited. Palo is currently being evaluated for its potential to reduce heterotopic (extraskeletal) ossification for the rare genetic disease FOP (46). ATRA and 13-cis-RA have previously been used for the treatment of prostate cancer with limited or no efficacy (47). However, the potential to “repurpose” these agents to target tumor-induced bone in the treatment of bone metastatic prostate cancer has not been explored. We also propose to use a new generation retinoid acid agonist, Palo, which has a more stable pharmacokinetic profile and does not appear to induce CYP26 enzymes, allowing more predictable dosing regimens [for review, see Hind and Stinchcombe (48)]. Palo was shown to be safe for long-term treatments (46), making its combination with other cancer-targeted therapies for prostate cancer bone metastasis promising. Our studies thus identify a novel therapy strategy that targets tumor-induced stromal reprogramming as treatment for prostate cancer bone metastasis.

Our previous discovery that osteoblastic bone lesions arise from EC-to-OSB transition (4) led us to hypothesize that inhibiting this transition would represent a novel therapy strategy for bone-metastatic prostate cancer. The current study credentials the EC-to-OSB transition as a valid therapy target. This occurred without significant effect on nontumor-involved healthy bone in castrated mice as measured by bone histomorphometry. The fact that plasma levels of TNC correlated with ATRA's effect on EC-to-OSB transition suggests their potential use as a blood-based biomarker to monitor response to ATRA treatment (Fig. 6H). Together, our studies provide a framework for the clinical development of ATRA or Palo for the treatment of bone metastatic prostate cancer.

Although RARs are nuclear transcription factors, our study suggests that activation of RARs by Palo or ATRA inhibits EC-to-OSB transition via a noncanonical RAR pathway. We found that RARs form a complex with pSmad1 during BMP4-stimulated EC-to-OSB transition. It is possible that the formation of RAR/pSmad1 complex prevents the degradation of pSmad1 by Smurf1. Upon retinoic acid treatment, Smurf1 is recruited to the pSmad1/RAR/ligand complex to degrade pSmad1 (Fig. 6H). Our studies likely reveal a novel mechanism for retinoic acids’ effects on modulating skeletal development (21–23). Future studies should determine whether RARs signal through other yet-to-be identified pathways in different cellular contexts.

While we showed that ECs are one source of the tumor-induced bone (4), several cell populations with osteogenic capacity have also been described in the bone marrow, including NG2+ pericytes (49), LepR+ cells (50, 51), and Gli1+ cells (52). Although lineage tracing studies allowed us to follow OSBs that are derived from endothelial precursors (4), it is not feasible in this model to establish the fraction of OSBs that are contributed from endothelial source in the total OSBs. Thus, the possibility that other progenitor cells may be recruited and contribute to aberrant bone formation in prostate cancer bone metastasis cannot be excluded. Strategies that can distinguish OSBs from different progenitor sources are needed to address this intriguing question.

Adipocyte stem cells have been reported to differentiate into OSBs when cultured in osteogenic medium (Supplementary Fig. S7A; refs. 53, 54). We examined whether Palo can inhibit adipocyte-to-OSB transition. In contrast to Palo's inhibitory effect on EC-to-OSB transition, we found that Palo alone can enhance adipocyte-to-OSB transition (Supplementary Fig. S7B), consistent with the report by others (54, 55), and Palo did not inhibit BMP4-induced adipocyte-to-OSB transition (Supplementary Fig. S7B). Thus, Palo-mediated decrease in EC-to-OSB transition and bone formation is unlikely to be due to an inhibition of adipocyte-to-OSB transition.

Our finding that Palo or ATRA decreases tumor volumes raises the question of whether RAR agonists are exerting a direct effect on tumor cells or an indirect effect through inhibition of tumor-induced bone. Others have shown that RAR agonists are not very effective in inducing growth arrest in prostate cancer cell lines in vitro, as high concentrations (∼1–10 μmol/L) are generally needed (56–58), in agreement with our results that Palo or ATRA does not inhibit proliferation of C4-2b-BMP4, TRAMP-BMP4, and MycCaP-BMP4 cells (Supplementary Fig. S2A, S2D, and S2G). Given that ATRA has been reported to inhibit angiogenesis in acute promyelocytic leukemia (59) and our bone histomorphometry analysis showed ATRA has effects on several EC parameters (Supplementary Fig. S6C), we examined whether Palo has an effect on EC proliferation and tube formation in vitro. We found that Palo moderately inhibited the proliferation of 2H11, MS1, and SVR cells (Supplementary Fig. S8A). Palo also inhibited EC tube formation of HUVEC, 2H11, MS1, and SVR cells (Supplementary Fig. S8B). When tested in vivo on subcutaneously generated nonosteogenic MycCaP, C4-2b and TRAMP-C2 tumors, ATRA causes a small decrease of tumor volumes (Supplementary Fig. S8C). We also examined ATRA treatment on tumor microvessel density using CD31 staining (Supplementary Fig. S8D). We found that the microvessel density of MycCaP and C4-2b tumors was moderately inhibited by ATRA treatment. Together, these results suggest that ATRA has an effect on angiogenesis. Because our previous studies showed that reduction of bone formation resulted in a decrease of tumor-induced bone formation in a genetic mouse model and a xenograft model (3, 4), these observations suggest that ATRA or Palo's effects on inhibiting osteogenic tumors are likely due to both a reduction in tumor-induced bone formation and inhibition of tumor angiogenesis.

We note that ATRA or Palo's effects on bone-forming tumor growth are moderate. Thus, we posit that ATRA or Palo in combinations with epithelial-targeting therapies (e.g., docetaxel) will be more effective for tumor growth inhibition. As we previously showed that tumor-induced bone is one source of therapy resistance (11), it is possible that Palo or ATRA, by reducing tumor-induced bone formation, will reduce resistance to therapies. Such an application requires further investigation.

Although Palo has been reported to be an RARγ-specific agonist in the literature (17, 48) and several clinical trials were performed on the basis of such a property (NCT 02190747, NCT 03312634, NCT 02279095, NCT03442985), the binding affinities of Palo to the three RAR isoforms have not been published. However, the binding affinities of Compound 7, which shares a close structural identity with Palo (Supplementary Fig. S8E), as reported in U.S. patent no. 6777418 are 260, 140, 17 nmol/L for purified RARα, β, γ proteins, respectively, suggesting that Palo may be more potent for RARγ than RARα or RARβ isoforms by about 15- and 8-fold, respectively.

With the advancement of immune therapies, performing the experiments in immunocompetent mice would yield information on the impact of immune infiltrates in the context of increased bone-formation and retinoic acid signaling. Engblom and colleagues (60) reported that OSBs supply cancer-promoting neutrophils to modulate the immune system in lung cancer. Retinoic acid and RARs are modulators of the immune system (61). Thus, it is possible that both tumor-induced bone formation and Palo have an impact on immune responses. Establishing syngeneic mouse tumor models will allow us to determine whether the proposed therapy approach is also effective in immunocompetent mouse. Such an approach is currently being developed.

In conclusion, we found that activation of RARs by agonists, including Palo and ATRA, targets the molecular basis of prostate cancer–induced aberrant bone formation, and Palo and ATRA are promising therapeutic agents for the treatment of prostate cancer bone metastasis.

S.K. Agarwal reports other support from Biology of Inflammation Center during the conduct of the study. C.J. Logothetis reports personal fees from Merck, Sharp & Dohme, Bayer, Amgen, Exelixis and grants from Janssen, ORIC Pharmaceuticals, Novartis, Aragon Pharmaceuticals outside the submitted work. No disclosures were reported by the other authors.

G. Yu: Conceptualization, data curation, formal analysis, validation, investigation, writing–original draft. P.G. Corn: Conceptualization, formal analysis, writing–original draft. P. Shen: Data curation. J.H. Song: Data curation. Y.-C. Lee: Data curation. S.-C. Lin: Data curation. J. Pan: Data curation. S.K. Agarwal: Resources, formal analysis. T. Panaretakis: Formal analysis. M. Pacifici: Formal analysis. C.J. Logothetis: Conceptualization, funding acquisition, writing–original draft. L.-Y. Yu-Lee: Conceptualization, formal analysis, supervision, funding acquisition, validation, writing–original draft, project administration, writing–review and editing. S.-H. Lin: Conceptualization, formal analysis, supervision, funding acquisition, validation, writing–original draft, project administration, writing–review and editing.

This work was supported by grants from the NIH RO1 CA174798 (to S.-H. Lin and L.-Y. Yu-Lee), P50 CA140388 (to C.J. Logothetis, S.-.H Lin), P30 CA016672, Cancer Prevention and Research Institute of Texas CPRIT RP150179 and RP190252 (to S.-H. Lin and L.-Y. Yu-Lee), and the Biology of Inflammation Center Arthritis Research Endowment AR00108 (to L.-Y. Yu-Lee).

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.