Longitudinally Monitoring Prostate Cancer Bone Metastasis and Treatment Response via Circulating Extracellular Vesicles

ABSTRACT

Lethal prostate cancer (PCa) is dominated by treatment‐refractory progression in bone. In this study, we used plasma‐derived extracellular vesicles (EVs) for monitoring the tumour and therapy‐induced changes in the bone tumour microenvironment (bone‐TME) in men with bone metastatic PCa (M1b PCa). Transcriptome analysis of differentially expressed genes in plasma EVs from patients with M1b and localized (M0) PCa, identified a ‘bone metastasis signature’ (BMS) that reflects tumour‐induced bone formation. In preclinical studies, the BMS was enriched in plasma EVs from mice inoculated with bone‐metastatic MDA PCa‐118b but not with adrenal‐metastatic MDA PCa‐250 tumours. Longitudinal monitoring of the BMS for up to 50 days showed that the levels of several genes increased prior to detectable growth of bone‐forming tumours. Real‐time monitoring of the BMS from tumour‐bearing mice treated with the bone‐targeting agent Ra‐223 showed reduction of their expression that correlated with reduction in bone formation and tumour growth. Development of resistance to Ra‐223 correlated with the induction of the BMS genes. In a cohort of PCa patients with clinically localized PCa with pathologic lymph node involvement, increases in BMS genes following hormonal therapy correlated with the subsequent development of metastasis during long term follow up. The BMS may enable longitudinal monitoring of the initiation, progression and treatment response of M1b PCa.

1. Introduction

Bone is the major site of prostate cancer (PCa) progression. The development of bone metastasis signifies lethal progression of PCa (Davila et al. 2015, Lee et al. 2011, 2015). In contrast to most solid tumours that produce lytic lesions, PCa bone metastases are distinctively osteoblastic due to PCa‐induced aberrant bone formation. Bone‐metastatic PCa (M1b PCa) is a microenvironment‐dependent disease in which tumour–bone stromal cell interactions create a bone tumour microenvironment (bone‐TME) that promotes lethal disease progression and resistance to therapy (Logothetis et al. 2013). The critical role of tumour–bone interaction in castrate‐resistant M1b PCa was revealed by Radium‐223 (Ra‐223), an α‐emitting radiopharmaceutical that targets the bone–tumour interface and prolongs overall survival (Dondossola et al. 2019; Parker et al. 2013). Notably, the benefit of Ra‐223 is not associated with expected proportional reduction in prostate‐specific antigen (PSA), a marker for PCa epithelial cells, but is associated with reductions in total alkaline phosphatase (ALP), a marker for osteoblasts. However, changes in ALP are neither sensitive nor specific enough to monitor the complex tumour microenvironment interactions that drive disease progression in the bone and resistance to therapy.

Bone tumour biopsies are currently the only method to monitor the complex biology of the bone‐TME. However, the invasive nature of the biopsies limits serial sampling for longitudinal studies. In addition, bone biopsy samples frequently contain only small amounts of tissue that may not contain any tumour cells, and sample decalcification interferes with many molecular‐pathologic analytical techniques. While circulating DNA and circulating tumour cells have been developed as complementary approaches for monitoring tumour progression and treatment response, these focus on epithelial cells and lack information on stromal components in the bone‐TME. Thus, there is a clinical need to identify biomarkers to longitudinally monitor the bone‐TME during the development of bone metastasis and in response to treatment. The development of methods that monitor components of the bone‐TME and their interplay across disease states is required for the development of biology‐based treatment strategies.

To develop novel methods to monitor the bone‐TME, our group has focused on the analysis of extracellular vesicles (EVs). EVs are intrinsically programmed secreted vesicles composed of proteins, DNA, RNA and lipids derived from the endosome (e.g., exosomes) or the ectosome (e.g., microvesicles) of cells (Fonseca et al. 2016; Kharaziha et al. 2012). Cancer cell‐derived EVs have been found to play important roles in promoting systemic inter‐cellular communication, modulating immune responses and setting up the premetastatic niches (Kharaziha et al. 2012). The contents of EVs reflect their cellular origin, including those in a bone‐TME, and can reveal changes in both tumour cells and TME during disease development and in response to therapy. Due to the stability and abundance in the plasma (especially in cancer patients), EVs are a promising source of circulating biomarkers.

Patients with PCa have higher number of circulating EVs in plasma, and the concentration of these EVs reached 89% sensitivity in distinguishing PCa from healthy individuals (Logozzi et al. 2021). Transcriptomic profiling of EV RNA derived from cell lines or preclinical models of M1b PCa showed an enrichment of transcripts associated with PCa (Casanova‐Salas et al. 2024). Wang et al. 2023 developed an EV‐based assay, and an 11‐gene metastasis score to distinguish between localized and M1b PCa but was not specific to bone metastasis. Importantly, our prior studies have demonstrated that EVs reflect molecular‐pathologic changes in the bone‐ and prostate‐TMEs (Gentile et al. 2023; Vardaki et al. 2021). Thus, profiling plasma circulating EVs represents a liquid biopsy approach to longitudinally monitor the bone‐TME to attain molecular insights during disease progression and treatment response of M1b.

In this study, we sought to identify genes associated with bone metastasis in plasma EVs in patients with M1b PCa by comparing their plasma EV transcriptome in those with only localized disease and identified a set of seven genes that constitute a ‘bone metastasis signature (BMS)’. Using osteogenic tumours in mouse models, the bone specificity, response to bone‐targeted therapy and the time course of detecting the BMS in the plasma were determined.

2. Materials and Methods

2.1. Patient Samples and Clinical Trials

Three different clinical trials were used in this project for discovering and applying the BMS. Our first analysis compared screening plasma samples of patients with high‐risk localized PCa (M0, n = 25) treated for 6 months preoperatively with apalutamide with or without abiraterone (Apalutamide and Gonadotropin‐Releasing Hormone Analog With or Without Abiraterone Acetate in Treating Participants With PCa, ClinicalTrials.gov ID NCT03279250) with screening plasma samples of patients with M1b (M1b, n = 9) receiving best systemic therapy with or without definitive treatment to the primary tumour (radiation or surgery) (Best Systemic Therapy or Best Systemic Therapy (BST) Plus Definitive Treatment (Radiation or Surgery), ClinicalTrials.gov ID NCT01751438). The bone related markers identified in the comparison between the M0 and M1b trials were examined in patients (Ra‐223, n = 25) with bmCRPC treated with the bone‐targeting agent Ra‐223 (Radium 223 in Castrate Resistant PCa Bone Metastases (ClinicalTrials.gov ID NCT02135484, Vardaki et al. 2021). In these clinical trials, the status of the primary tumour and lymph nodes for staging purposes was determined by conventional CT and/or MRI. The presence or absence of bone metastases was determined by Technetium‐99m (Tc‐99m) bone scans.

2.2. Total EV Isolation and RNA Extraction From Human Plasma

EVs were isolated from Human plasma as previously described (Gentile et al. 2023). Briefly, 0.1 mL of the Exosome isolation kit (Thermo Fisher Scientific (RRID: SCR_008452), Cat# 4484450) was added to 0.5 mL of plasma and homogenized by vortexing. After 30 min incubation, the samples were centrifuged for 10 min at 10,000 ×g. The EV pellets were resuspended in 0.1 mL of PBS, and total RNA was extracted from the isolated EVs using a miRNeasy Micro Kit (QIAGEN (RRID: SCR_008539), Cat#: 217084) according to the manufacturer's instructions. RNA purity and quantification analysis were performed by an RNA 6000 Pico kit on the Agilent 2100 Bioanalyzer Instrument (RRID: SCR_018043).

2.3. Transcriptome Analysis: Affymetrix‐Clariom D Assay

RNA profiling of EVs was performed by University of Texas MD Anderson Advanced Technology Genomics Core Facility (RRID: SCR_026261). Affymetrix Clariom^TM D Assay, Human (Applied Biosystems (RRID: SCR_005039), Cat# 902924‐902622) provided a detailed view of the transcriptome with all known coding and noncoding splice variants. Affymetrix Transcriptome Analysis Console Software (RRID: SCR_018718), was used to analyse and visualize global expression patterns of genes, exons, pathways and alternative splicing events.

2.4. Ultra Low Input RNA Sequencing

Low Input mRNA Libraries were prepared using the Smart‐Seq V4 Ultra Low Input RNA Kit (Takara Bio Inc. (RRID: SCR_021372), Cat# 634442) and KAPA HyperPlus library preparation kit (Roche (RRID: SCR_001326), Cat# 07962401001). Briefly, full length, double‐stranded cDNA was generated and amplified by 11–13 cycles of LD‐PCR, then purified using AMPure beads Agencourt (New England Biolabs (RRID:SCR_013517), Cat# E6260). After evaluation using the Agilent Bioanalyzer High Sensitivity DNA Kit (Agilent 2100 Bioanalyzer Instrument, RRID: SCR_018043) and the Qubit dsDNA HS Assay Kit (Thermo Fisher Qubit 2.0 Fluorometer Qubit 2.0, RRID: SCR_020553), the cDNA was enzymatically fragmented, and 5 ng of the fragmented cDNA was used to generate Illumina‐compatible libraries. The libraries were quantified using the Qubit dsDNA HS Assay (Thermo Fisher Qubit 2.0 Fluorometer Qubit 2.0, RRID: SCR_020553), then multiplexed and sequenced on the Illumina NovaSeq 6000 Sequencing System (RRID: SCR_016387) using the 100 nt PE format.

2.5. Animal Studies

All animal studies were conducted using male NOD/SCID mice from the Jackson Laboratory, USA. The current regulations and standards of the U.S. Department of Agriculture, the U.S. Department of Health and Human Services and the National Institutes of Health were followed and approved by The University of Texas MD Anderson Cancer Center Institutional Animal Care and Use Committee.

2.6. Patient‐Derived Xenografts Studies

In this study, two different PCa patient–derived xenografts (MDA PCa) models were used: MDA PCa 250 (negative control) and MDA PCa 118b (osteogenic bone model) (Anselmino et al. 2024). The models were developed in the laboratory of Dr. Navone at the PCa Patient–Derived Xenografts Program, Department of Genitourinary Medical Oncology, MD Anderson Cancer Center. The tumours were implanted in the right flank of 6‐ to 8‐week‐old male NOD/SCID, and tumour growth was recorded weekly. For selected experiments, Radium‐223 (Ra‐223) (Xofigo, Bayer, Leverkusen, Germany), a radioactive alpha‐emitting isotope, used to treat bone metastases in M1b, was given to the osteogenic bone metastasis tumour, MDA PCa 118b at a dose of 300 kBq/kg intravenously. To achieve longitudinal monitoring of the BMS, blood was collected through in vivo intracardiac puncture at different time points (t0, t14, t28, t42 and t56 days) and the isolated plasma was stored at −80°C. At the end of experiment, the mice were euthanized by asphyxiation with carbon dioxide, and tumour and blood were collected for future analysis.

2.7. Micro‐Computed Tomography

Bone samples were fixed in 70% ethyl alcohol in a 15 mL tube. They were scanned on a Bruker microCT Skyscan 1276 (Bruker, Kontich, Belgium, University of Texas MD Anderson Cancer Center Small Animal Imaging Facility, RRID:SCR_027643) at 55 kV, 200 µA with a 0.25 mm Al filter, with resolution 2016 × 1344 px, pixel size 13 µm, 435 ms exposure time, and step and shoot mode with rotation step 0.400 degrees. Backward projection data sets of all the mice were reconstructed using InstaRecon software (Bruker, Kontich, Belgium, University of Texas MD Anderson Cancer Center Small Animal Imaging Facility, RRID: SCR_027643) with windowing 0–0.08 intensity, ring artefact reduction 5, 23% beam hardening and automatic post‐alignment correction. Analysis was performed using a CT software (Bruker, Kontich, Belgium, University of Texas MD Anderson Cancer Center Small Animal Imaging Facility, RRID: SCR_027643). An in‐house script was used to segment the bone samples, and then a threshold of 120 was applied. The tissue volume (mm³), bone volume (mm³) and percent bone volume (%) was calculated used the 3D analysis tool (Bruker, Kontich, Belgium, University of Texas MD Anderson Cancer Center Small Animal Imaging Facility, RRID: SCR_027643).

2.8. EVs Isolation From Mouse Plasma and RNA Extraction

Longitudinal monitoring was used during all in vivo experiments performed in this study. Blood was collected at different time points, and plasma was separated from the blood samples and stored at−80°C. EVs were isolated from mouse plasma samples using ultracentrifugation as previously described (Gentile et al. 2023). The ultracentrifuge Sorvall WX80 (Thermo Fisher Scientific (RRID: SCR_008452), Cat# 7500080) was run at 35,000 rpm 4°C for 2 hours. Pellets were resuspended in 0.700 mL of QIAzol Lysis Reagent, and RNA was isolated from EVs using miRNAeasy Micro kit (QIAGEN (RRID: SCR_008539), Cat#: 217084), according to manufacturer's protocols.

2.9. Gene Expression Analysis

Gene expression analysis was performed as previously described (Gentile et al. 2023). 100 ng of total RNA (concentration 10 ng/µL) from isolated EVs was used to synthesize cDNA with SuperScript IV VILO Master Mix (Thermo Fisher Scientific (RRID:SCR_0084520), Cat# 11756050) and 0.010 mL of cDNA were mixed with PowerUp SYBR Green Master Mix (Thermo Fisher Scientific (RRID:SCR_008452, Cat# A25741) and specific target primers at a concentration of 450 nM (Sigma‐Aldrich (RRID:SCR_008988), KiCqStart SYBR Green Primers) for gene expression analysis in real‐time PCR instruments (Life Technologies QuantStudio 3 Real Time PCR System (RRID:SCR_020238). Standard cycling mode was running for 40 cycles according to PowerUp SYBR Green Master Mix manufacturer's protocols. SYBR Green Primer sequences used in this project are:

Human, BMPR2, Forward Primer Sequence ‐TATCTGTATACTGCTGCCATC‐ reverse Primer Sequence ‐CCACCTCATTCATTTAACCG‐;

BMP6, Forward Primer Sequence ‐CTGTATGTGAGTTTCCAAGAC‐ reverse Primer Sequence ‐ATAAGGTGAACCAAGGTCTG‐;

SMAD4, forward Primer Sequence ‐GCAGAATCCATTTACTGGTC‐ reverse Primer Sequence ‐GAGGCAAATTAGGTGTGTATG‐;

SPARC, Forward Primer Sequence ‐CATTGCAGCTTCAACTTTTC‐ reverse Primer Sequence ‐AAACCATCCTTGACAACATC‐;

SPP1, Forward Primer Sequence—GTACACAAGCAGACACTTTC‐ reverse Primer Sequence ‐GATACTGTTCATCAGAAACCTG‐;

OSTM1, Forward Primer Sequence ‐ATGTGGAAGATGCAATGAAC‐ reverse Primer Sequence ‐CTCTGAGTGAAGAAAGCTAC‐;

OSCAR, forward primer sequence ‐CAGACATCACTCCGTCTG‐ reverse primer sequence‐AGGCTTGAAAAAGTCCAAATC‐.

Mouse: Bmpr2, Forward Primer Sequence ‐GATAATGCGGCTATAAGTGAG‐ reverse Primer Sequence ‐GTCTGTACACCTCATAAACAC‐;

Bmp6, Forward Primer Sequence ‐GTGAGTTAAAAACAGCTTGC‐ reverse Primer Sequence ‐TACTCGGGATTCATAAGGTG‐;

Smad4, Forward Primer Sequence ‐CCAGCTACTTACCATCATAAC‐ reverse Primer Sequence ‐GAGCAGGATGATTGGAAATG‐;

Sparc, Forward Primer Sequence ‐TATGCAGCAATGACAACAAG‐ reverse Primer Sequence ‐GTGACCAGGACATTTTTGAG‐;

Spp1, Forward Primer Sequence ‐GGATGAATCTGACGAATCTC‐ reverse Primer Sequence ‐GCATCAGGATACTGTTCATC‐;

Ostm1, Forward Primer Sequence ‐TCAAATGCAGAAACTGAACG‐ reverse Primer Sequence ‐TAATGTTCATTGCATCCTCC‐;

Oscar, forward primer sequence—CACCTACTGTTGCTATTACC‐ reverse primer sequence ‐GAACCTTCGAAACTGATGAC‐.

Housekeeping control used were:

Human 18s, Forward Primer Sequence ‐CAGTTATGGTTCCTTTGGT‐reverse Primer Sequence TTATCTAGAGTCACCAAGCC‐;

Mouse 18s, forward primer sequence CAGTTATGGTTCCTTTGGTC‐ reverse primer sequence ‐TTATCTAGAGTCACCAAGCC‐.

2.10. Immunohistochemistry

Immunohistochemistry (IHC) analyses of protein expression were performed on mouse tissue specimens obtained from in vivo experiments. Tumours were collected at the end of the experiment and were fixed in formalin overnight. The next day the tumours were stored in 70% ethanol at 4°C. Paraffin blocks were then made, and a microtome was used to cut thin slices of the tissues and mounted on slides. Tissues sections were then stained with the following antibodies detecting both human and mouse proteins: BMPR2 dilution 1:800 (BMPR‐II, Novus Cat# NBP2‐37625, RRID: AB_3296451), SPP1 (Anti‐Osteopontin, Abcam Cat# ab283669, RRID: AB_3739926) dilution 1:1500, SPARC (Thermo Fisher Scientific Cat# MA1‐43027, RRID: AB_2195095) dilution 1:800. The stained slides were digitalized in a Leica Aperio AT2 scanner (RRID: SCR_021256) Leica Biosystems) and quantified using ImageJ (RRID: SCR_003070)

2.11. Statistical Analysis

Statistical analysis was performed by GraphPad Prism (RRID: SCR_002798). Two‐tailed t tests were used to compare the mean differences between the groups of samples, and the differences were considered statistically significant when P < 0.05. Ingenuity Pathway Analysis system (RRID: SCR_008653) was used for bioinformatics analysis. A z‐score algorithm was used to predict gene expression changes.

3. RESULTS

3.1. Bone‐Related Transcripts Are Enriched in EV From Patients With M1b PCa Compared With Those Without Bone Metastasis (M0 PCa)

To identify circulating molecular markers of PCa bone metastasis, we first compared the plasma EV transcriptome of patients with bone metastatic disease (M1b, n = 9, Figure 1A, Table S1) to those with clinically localized disease (M0, n = 25) (Figure 1A, Table S2). We found an enrichment of bone related genes in the EVs derived from M1b patients compared to M0 (Figure 1B, C). Ingenuity Pathway Analysis showed that several of the top 15 pathways enriched in EVs from M1b patients are involved with the bone‐TME and immune suppression (e.g., WNT‐, IL‐8‐, NET‐, CXCR4‐ and HGF signalling) (Figure S1A). We focused on pathways and genes associated with the bone‐TME. We found that the EVs from M1b patients were significantly enriched in transcripts from bone‐related pathways, such as RANK, TGFβ, EGF, BMP and IGF signalling pathways, compared with the M0 patients (Figure S1B), consistent with an increase of bone‐forming activity in M1b patients.

FIGURE 1.

Plasma EV transcriptome identifies a bone metastasis‐associated gene signature in patients with M1b PCa. (A) Schema of patients’ plasma used for the gene expression comparison. (B) Bone related genes enriched in the EVs from M1b patients. (C) Identified genes enriched in the EVs from the M1b PCa patients compared to M0 patients. The identified M1b genes are highlighted in red. (D) The effect of Ra‐223 on the expression of 25 genes in patients who had a favourable and unfavourable response to Ra‐223. (E–G) Bioinformatic analysis of publicly available PCa data set showed that transcripts of SPP1 (E), SMAD4 (F) and OSCAR (G) are enriched in patients with bone metastasis compared to other indicated sites of metastasis. P values: *, <0.05; **, <0.01; ***, <0.001; ****, <0.0001. (H) The seven genes identified are from osteoblasts or osteoclasts or are involved in the signalling pathway of bone formation.

To identify markers enriched in bone metastasis, the top 25 bone‐related genes enriched in M1b patients were selected (Figure 1C). To further refine the genes modulated by PCa‐induced bone formation, we applied two independent criteria. The first was to identify genes that are modulated by bone‐targeting agent Ra‐223 in patients with castrate‐resistant M1b PCa (Figure 1C, Table S3). We reasoned that genes modulated by Ra‐223 would be most relevant for monitoring bone metastases clinically. In a prior study of Ra‐223 (NCT02135484; Vardaki et al. 2021) in patients with castrate‐resistant M1b PCa, patients were stratified into two groups based on overall survival, with a favourable group evidencing an overall survival of > 2 years and unfavourable group evidencing an overall survival of < 2 years (Vardaki et al. 2021). Among the 25 bone‐related genes, we found that OSTM1, MEMO1, RASSF2, BMPR2, TGFB1, SP1, SPARC, OSTF1, BMP6 and BMP2K were decreased in patients in the favourable group compared to the unfavourable group (Figure 1D).

The second filter we applied was to confirm which of the 25 bone metastasis genes are enriched in tumours from patients with bone metastasis using publicly available data sets from the SU2C and Fred Hutchinson cohorts (Kumar et al. 2016; Robinson et al. 2015). We found that the EV‐enriched genes SPP1 (osteopontin) and SMAD4 were significantly enriched in the transcriptomes of bone metastasis samples compared to transcriptomes of tissues from localized disease or from visceral metastasis in the Fred Hutchinson data set (Figure 1E,F). Similarly, SPP1 and SPARC were enriched in the bone metastasis samples compared to lymph node or liver metastasis in the SU2C data set (Figure S1C–E). In the SU2C data set we also found that OSCAR that was enriched in M1b EVs (Figure 1C) was also enriched in bone metastatic tumours compared to lymph node (LN) and visceral (liver) (Figure 1G). Together, seven bone‐related genes, (BMP6, BMPR2, SMAD4, SPP1, SPARC, OSTM1 and OSCAR), were associated with changes in the bone‐TME of M1b and were used as the BMS genes (BMS) (Figure 1H).

3.2. EV‐Derived Bone‐Related Genes Are Enriched in the Plasma EVs From Mice Bearing Bone Metastasis–Derived but Not Adrenal Metastasis–Derived PDX Tumours

To confirm that the BMS genes identified in the plasma‐derived EVs from M1b patients are derived from bone metastasis tissues, we used two MDA PCa PDX models, the bone metastasis‐derived osteogenic MDA PCa 118b and the adrenal metastasis‐derived MDA PCa 250 (Figure 2A). In PDX tumours, we have previously shown that species‐specific primers can identify tumour‐ versus stromal‐ derived changes in the bone (Vardaki et al. 2021). At time 0 (t0), baseline blood was collected from six NOD SCID mice prior to implantation in the right flank with MDA PCa 250 (n = 6 mice) or MDA PCa 118b (n = 6 mice) tumours. The MDA PCa 118b tumour growth was followed for 49 (t49) days (Figure 2B), and the bone formation in the tumours was measured by micro‐CT (Figure 2C and S2A) and by histological analysis using hematoxylin and eosin (H&E) staining at the end of the experiment (Figure 2F). We found a positive correlation between tumour sizes and the bone volume (Figure S2B). Because MDA PCa 250 is a fast‐growing tumour, growth was followed for 21 (t21) days, and blood was collected at that time point (Figure 2D and Figure S2C). Using mouse‐specific PCR primers, we found that the seven BMS genes, BMPR2, BMP6, SMAD4, SPP1, SPARC, OSTM1 and OSCAR, were enriched in the plasma EVs from MDA PCa 118b tumour–bearing mice (Figure 2E). In contrast, none of the bone‐related genes were detected in the plasma EVs from MDA PCa 250 tumour–bearing mice or control mice with no tumours implanted (Figure 2E).

FIGURE 2.

BMS genes are selectively enriched in plasma EVs from mice bearing MDA PCa 118b osteogenic PCa tumours but not from mice bearing non bone‐forming MDA PCa 250 adrenal tumours. (A) Schema of NT, MDA PCa 250 and MDA PCa118b in vivo experiment. Blood was collected at t0 from mice with NO TUMOUR mice, at 21 days from mice with MDA PCa 250 tumours, and at 49 days from mice with MDA PCa 118b tumours. (B) Weekly monitoring of tumour volume for MDA PCa 118b mice (left panel). The box indicates the tumour from mouse #B642, which was used as representative for the IHC images shown in E. (C) Micro‐CT images of the tumours collected at day 49 (right panel). (D) Weekly monitoring of tumour volume for MDA PCa 250 for 21 days. (E) Comparison of gene expression using mouse‐specific primers of BMS genes in EVs isolated from: no tumour NOD SCID mice (NT), MDA PCa 250 (250) and MDA PCa 118b (118B). P values: *, < 0.05; **, < 0.01. (F) H&E and IHC images of MDA PCa 118b tumour tissue of mouse #B642. The whole tumour and 4× and 20× images were captured for H&E and IHC for BMPR2, SPARC and SPP1 (T: MDA‐PCa 118b tumour cells, B: Tumour‐induced bone).

Next, the expression of these EV genes at the protein level in the MDA PCa 118b tumours was examined by IHC (Figure 2F). BMPR2, SPP1 and SPARC were strongly expressed in the tumour‐induced osteoblasts of MDA PCa118b tumours. Interestingly, these proteins were also expressed in the tumour cells in MDA PCa 118b tumours (Figure 2F). Thus, we performed RT‐qPCR of the plasma EVs by using human‐specific primers to quantify the levels of transcripts that were derived from tumour cells in the MDA PCa 118b tumours. We found that the human‐specific bone metastasis genes were also enriched in the plasma EVs from MDA PCa 118b mice compared with MDA PCa 250 mice (Figure S2D). The expression of bone metastasis genes in MDA PCa 118b tumour cells is likely due to the osteo‐mimetic property that has been described for bone metastasis–derived tumour cells (Knerr et al. 2004; Rucci and Teti 2018). These results suggest that the transcripts may reflect tumour‐induced changes in both the bone‐TME and the PCa cells.

3.3. Expression of Bone‐Related Genes in Plasma EVs Correlates With Ra‐223–Mediated Tumour and Bone Growth Inhibition

Ra‐223 targets the bone‐TME and was shown to prolong overall survival in a subset of men with castrate‐resistant M1b (Parker et al. 2013). To determine if changes in the expression of the identified BMS genes would correlate with response to Ra‐223, we treated mice bearing MDA PCa 118b tumours with Ra‐223 or vehicle and measured tumour and bone volumes (Figure 3A). After implantation in the right flank, tumour growth was monitored in living mice (n = 5) as depicted in the schema. When tumours reached detectable size and had evidence of bone formation as determined by x‐ray (Figure 3C), Ra‐223 treatment started (t0). Blood was drawn again from the same mice 15 days after treatment (t15) (Figure 3A). X‐ray images were acquired in vivo at baseline and 4 days after Ra‐223 treatment, and tumour‐induced bone formation was detected in the tumours (Figure 3C). Micro‐CT analysis was performed ex vivo at the end of the experiment (Figure 3D,E). We found that both tumour growth and bone formation were inhibited by Ra‐223 in MDA PCa 118b tumours (Figure 3B,E). The expression of the BMS genes in EVs from mouse plasma was determined by RT‐qPCR using mouse primers (Figure 3F). We found that the levels of the microenvironment (mouse)–derived genes Bmpr2, Smad4, Sparc and Ostm1 were significantly reduced by Ra‐223 (Figure 3F), consistent with decreases of bone formation by Ra‐223 treatment (Figure 3E). Expression of the tumour (human)–derived genes was also significantly downregulated for BMP6, SMAD4, SPP1, SPARC and OSCAR in response to Ra‐223 (Figure S3A), corresponding to inhibition of tumour growth by Ra‐223 (Figure 3B). Downregulation of the BMS genes by Ra‐223 correlated with the corresponding protein levels in both tumours and tumour‐induced bone formation as determined by IHC (Figure 3G,H and S3B). These data suggest that the expression of the identified BMS genes correlates with the efficacy of Ra‐223 (Vardaki et al. 2021) and may be used as a pharmacodynamic measure of Ra‐223 activity

FIGURE 3.

Ra‐223 inhibits tumour‐induced bone formation and reduces BMS gene expression. (A) Schema of MDA PCa 118b in vivo experiment. After assessment of initial tumour size by caliper and bone formation by x‐ray, Ra‐223 treatment was started at time zero (t0). Blood was collected at the end of experiment, 15 days after Ra‐223 treatment (t15). (B) Weekly monitoring of tumour volume after treatment as measured by calipers. (C) In vivo x‐ray images of vehicle‐ and Ra‐223‐treated mice 4 days after treatment. (D) Ex vivo micro‐CT analysis on selected mice. (E) Bone volume quantification. (F) RT‐qPCR of gene expression of the indicated mouse bone‐targeting genes after 14 days of treatment with Ra‐223. (G) H&E and IHC images of MDA PCa 118b tumour tissue from mice #932 V (vehicle treated) (blue column) and #955R (Ra‐223 treated) (red column); 20× images were captured for H&E and IHC for BMPR2, SPARC and SPP1. (H) Quantification of protein expression was done using ImageJ on the whole tumour for all mice in MDA PCa 118b groups. P values: *, < 0.05; **, < 0.01; ***, < 0.001.

3.4. The Expression of Bone Metastasis Genes in Plasma EVs Can Be Used to Longitudinally Monitor Bone Forming Tumour Growth

To determine whether the expression of the seven BMS genes tracked with tumour growth over time, we performed RT‐qPCR on five serially collected plasma samples at t0 (prior to inoculation) and at days 14, 28, 42 and 56 after tumour inoculation, from 6 mice (Figure 4A). Tumour growth measured at the indicated time points showed typical growth for the MDA PCa 118b tumours (Figure 4B). We measured the expression of each of the seven genes for the five time points and matched their levels to tumour growth for all the mice (Figure 4C–E, TME). The mean expression of all mice (n = 6) for each microenvironment‐derived mouse gene (TME) and each tumour‐derived human gene (Tumour) showed some differences (Figure 4C–E, TME, Tumour). Longitudinal analysis revealed that several BMS genes increased prior to detectable tumour growth by imaging. Specifically, increases in Bmpr2, Smad4, Sparc levels (TME) were observed as early as day 14 and reached maximal expression at day 28, preceding the onset of exponential tumour growth and radiologically detectable bone formation, which occurred at later time points (approximately day 42). This temporal dissociation indicates that EV‐derived BMS captures early molecular changes in the bone‐TME before structural alterations become apparent by imaging. Bmp6, Spp1, Oscar and Ostm1 (TME) exhibited delayed kinetics, with initial increases detected at day 28 and peak expression reached by day 42 (Figure 4B–E, TME). In contrast, tumour‐derived genes (BMP6, SPARC, SPP1, OSTM1, OSCAR and BMPR2, but not SMAD4) showed a continuous and gradual increase in circulating levels over time, closely paralleling tumour growth kinetics (Figure 4B,D,E, Tumour). These results suggest that the BMS genes may be associated with early bone formation, which has been shown to support tumour growth and disease progression (Lee et al. 2011, 2015; Lin et al. 2018). Furthermore, tumour‐derived genes in EV may enable longitudinal tracking of tumour growth.

FIGURE 4.

BMS expression dynamically reflects tumour‐induced bone formation and progression, with early bone‐TME activation preceding tumour growth. (A) Schema of MDA PCa118B in vivo experiment with the indicated blood collection time points at days 0, 14, 28, 42 and 56. (B) Tumour volume measurements at the indicated time points. Exponential growth of MDA PCa 118b tumour occurred 42 days after tumour inoculation. (C–E) Mean expression of the microenvironment‐derived bone metastasis mouse genes (TME) and of tumour‐expressed human genes (Tumour) measured at 0, 14, 28, 42 and 56 days, compared to tumour volume.

3.5. Expression of Bone Metastasis Genes Can Be Used to Monitor Response and Resistance of Osteogenic Tumours to Ra‐223

Next, we examined whether the bone metastasis genes may serve as a potential pharmacodynamic measure of response and resistance to Ra‐223 treatment. Tumours were inoculated and allowed to grow to measurable size before Ra‐223 treatments were started. Tumours were treated with Ra‐223 at 0 (t0) and 14 (t14) days. Tumour growth was monitored by calliper measurement, and blood was drawn from living mice (n = 6) at baseline (before Ra‐223 treatment) and 7, 14, 28 and 49 days after Ra‐223 treatment (t0) (Figure 5A). Following the initial decrease on t7 in tumour volume in response to Ra‐223 (response phase) and the maintenance of tumour size between t7 and t28, there was an exponential increase in tumour volume after 28 days of treatment, suggesting that tumours became refractory to Ra‐223 treatment (resistance phase) (Figure 5B). Ex vivo micro‐CT images captured at day 49 showed reduced bone growth in the Ra‐223‐treated tumours compared to untreated ones (Figure 5C and Figure S4A compared with Figure S2A). In contrast to untreated tumours, the correlation between tumour and bone volume diminished (Figure S4B compared with Figure S2B), suggesting bone‐independent growth in Ra‐223‐resistant tumours. Monitoring the microenvironment‐derived bone metastasis genes (TME) of all the mice (mean) over time showed a reduction in Bmp6, Bmpr2, Smad4, Spp1, Ostm1 and Oscar levels at t7 after Ra‐223 treatment (Figure 5D–F, TME). Bmpr2, Smad4, Sparc and Oscar expression levels remained decreased at the end of the experiment (t49). Bmp6 and Spp1 showed an increase in expression levels after t14 days, reaching maximal levels at the end of the experiment (t49). These data indicate that stromal expression of bone‐related genes was continuously inhibited during the resistance phases of Ra‐223 treatment. In contrast, when tumour growth resumed, the expression of tumour cell‐derived genes increased, indicative of bone‐independent proliferation of PCa epithelial cells.

FIGURE 5.

EV‐BMS gene expression reveals the acquisition of resistance to Ra‐223 in MDA PCa 118b osteogenic tumours. (A) Schema of MDA PCa118B in vivo experiment. Tumour sizes were measured by caliper. Blood was collected at baseline (BL) and (t) 7, 14, 28 and 49 days. Ra‐223 was injected at t0 and t14. (B) Tumour volume measurements after treatment with Ra‐223 at the indicated time points. (C) Ex vivo micro‐CT analysis on mice tumours at day 49. (D–F) Mean expression (RT‐qPCR) of the microenvironment‐derived bone metastasis mouse genes (TME) and mean expression of the tumour‐derived bone metastasis human genes (Tumour) measured at baseline (BL) and at (t) 7, 14, 28 and 49 days, compared to tumour volume (tv) measured at 0, 7, 14, 28 and 49 days of Ra‐223 treatment.

Tumour expression of the bone‐related human genes (Tumour) BMP6, SMAD4, SPARC, SPP1, OSTM1 and OSCAR showed reduced expression after 14 days of treatment (Figure 5D–F, Tumour). The levels of these genes increased at t14 (except SMAD4, which showed an increase at t28), reaching maximal levels at t49. The expression of these tumour‐derived genes preceded tumour sizes measured at t28. IHC staining for BMPR2, SPP1 and SPARC confirmed the presence of these proteins in the tumour tissue after 49 days of Ra‐223‐treatment (Figure S4C,D). These data indicate that while stromal expression of bone‐related genes was continuously inhibited during the resistance phases of Ra‐223 treatment, tumour expression of bone‐related genes was inhibited during the initial response to Ra‐223 but increased with the acquisition of Ra‐223 resistance. Thus, Ra‐223–resistant tumours are not caused by insufficient bone inhibition, but rather by development of bone‐independent proliferation of PCa epithelial cells in these tumours. These findings suggest a shift towards tumour cell‐driven growth of osteogenic tumours.

3.6. Increased BMS After ADT+ARPi Correlates With Subsequent Development of Bone Metastasis

We next wanted to test if the EV‐BMS could predict the development of bone metastases. To approach this, we identified a cohort of 8 patients from the neoadjuvant clinical trial that were discovered to have lymph node involvement (LN) following pathological examination of the radical prostatectomy specimen (NCT03279250, Efstathiou et al. 2020). Since lymph node involvement at the time of radical prostatectomy predicts for a higher risk of developing bone metastases, we tested for the BMS in blood samples in this cohort (Figure 6A) and correlated with the subsequent development of bone metastases (Moschini et al. 2016). All 8 patients showed response to therapy prior to surgery with 7 achieving a PSA of ≤0.1 and 1 with a PSA of 0.1 (Table 1). BAP levels at baseline and after hormonal treatment were within normal range, 38–126 IU/l for all the patients, consistent with none having bone metastases at the time of trial enrollment.

FIGURE 6.

BMS expression after therapy predicts subsequent development of bone metastasis in PCa patients treated with ADT+ARPi and surgery. (A) Timeline of the treatment of eight patients and subsequent follow‐ups. (B) BMS expression at baseline (BL) in patients (Pat) that developed bone metastasis compared to BMS expression in M1b patients, red bar (Pat 18: P value **, 0.0017; Pat 21: P value *, 0.0303; Pat 56: P value **, 0.0080). (C) BMS expression at BL in patients that did not develop bone metastasis compared to BMS expression in M1b (red bar) (Pat 14: P value **, 0.0148; Pat 59: P value **, 0.0063; Pat 40: P value *, 0.0118). (D) BMS average expression at BL and EoT in patients that developed bone metastasis compared to BMS expression in M1b (red bar), (BL: P value **, 0.0038; BL versus EoT: P value **, 0.0056).(E) BMS average expression at BL and EoT in patients that did not develop bone metastasis compared to BMS expression in M1b (red bar), (BL: P value **, 0.0020; EoT: P value **, 0.0054). (F) PSA, BAP and BMS expression at BL and EoT in individual patients that developed metastasis compared to BMS expression in M1b (red bar), (Pat 18: P values **, 0.0017, *, 0.0175; Pat 21: P values *, 0.0303, *, 0.0346; Pat 56: P values **, 0.0080, *, 0.0113). (G) PSA, BAP and BMS average expression at BL and EoT in M0 patients with lymph node metastasis that did not develop bone metastasis compared to BMS expression in M1b (red bar), (Pat 14: P values *, 0.0148, **, 0.0031, *, 0.0196; Pat 49: P value *, 0.0478; Pat 59: P values **, 0.0063, **,0.0062; Pat 40: P values *, 0.0118, **, 0.0083).

TABLE 1.

LN n = 8 patients.

ACC#	PSA at BL	Gleason	Baseline clinical TNM	PSA at EoT	BAP at BL	BAP at EoT	BMS at EoT	Pathologic TNM	LN mets at surgery	Time to bone metastasis	Bone mets
									(1 = Y, 0 = N)		(1 = Y, 0 = N)
18	60.6	9	cT3cN0cM0	0	39	52	*	ypT3b ypN1	1	>3 years	1
21	26	8	cT3cN0cM0	0	78	91	*	ypT3a ypN1	1	>3 years	1
56	20.4	9	cT3cN0cM0	0	49	49	*	ypT3b ypN1	1	>3 years	1
14	5.8	9	cT3cN0cM0	0	68	49	**	ypT3b ypN1	1	No bone metastasis	0
40	18	9	cT2cN0M0	0	72	79	**	ypT2 ypN1	1	No bone metastasis	0
49	57.8	10	cT3cN1cM0	0	60	55	**	ypT3b ypN1	1	No bone metastasis	0
59	10.8	7	cT3cN0cM0	0	61	72	**	ypT2 ypN1	1	No bone metastasis	0
22	15.9	7	cT3cN0cM0	0.1	84	98	—	ypT3b ypN1	1	No bone metastasis	0

With a median follow‐up of 6.5 years, three (patients: 18, 21 and 56) of these 8 patients developed bone metastases at 3.4, 4 and 3.4 years later, respectively. Analysis of baseline (BL) BMS levels before ADT+ARPi therapy (BL) in these 8 patients showed no clear correlation with the subsequent development of bone metastasis (Figure 6B,C). In contrast, in the three patients that developed bone metastasis, there was increased BMS expression at EoT compared with BL (Figure 6D). Furthermore, the 5 patients that did not develop bone metastasis had reduced average BMS expression at EoT (Figure 6E). As part of the precision oncology initiative, we performed patient‐specific analysis of the BMS expression individually for the 8 patients. We found that in the three patients that developed bone metastasis after the treatments, that is, patients 18, 21 and 56, BMS expression was increased at the EoT to similar levels to those identified in patients with M1b (Figure 6F). From the five patients that had LN metastasis after treatment but did not develop bone metastasis, four of them had decreased BMS levels, which were also lower than M1b at EoT (Figure 6G). Notably, patient 22 had already high BMS levels at BL, and BMS levels were increased after treatment. However, the increase in BMS was not statistically significant. These data indicate that monitoring dynamic changes in BMS following treatment with ADT+ARPi may inform the likelihood of developing bone metastasis in the future in patients with clinically localized PCa undergoing surgery.

4. Discussion

Bone is the most common site of PCa metastases and bone metastases represent a lethal disease state. Unlike most other solid tumours which are lytic, PCa bone metastases are characteristically blastic (osteogenic). For this reason, the detection of bone metastases has historically relied on Technetium‐99m (Tc‐99m) scan, a technique where the isotope is linked to a bisphosphonate carrier that naturally accumulates in areas of high bone turnover. Tc‐99m bone scans, however, have limited sensitivity and specificity to detect new metastases and in patients with established bone metastases, Tc‐99m bone scans are not used to assess response given the ‘flare’ phenomenon (Scher et al. 2016). While PET‐PSMA scans have now become routine in the management of patients with PCa, an important limitation in their role assessing bone metastases is the fact that PSMA expression, an epithelial marker, is heterogeneously expressed and absent in up to 25% of M1b CRPC (Sayar et al. 2023; Zhao and Ji 2022). For this reason, our group routinely employs bone scans in addition to PET‐PSMA scans to assess bone metastases. In addition, we incorporate measurements of alkaline phosphatase (total and/or bone specific) as a clinical tool to judge response in bone and as a prognostic marker, particularly with therapies that target bone stromal cells (e.g. osteoblasts, endothelial cells) more than PCa epithelial cells (Som et al. 2012; Varkaris et al. 2016; Wan et al. 2014).

Towards a goal to develop better markers for monitoring bone, we used EVs as a liquid biopsy platform to identify molecular markers that reflect the biology of PCa bone metastases. We have identified a seven‐gene BMS that may enable longitudinally monitoring of bone metastasis development and response to therapy. Utilizing PDX tumours in mice, we showed that dynamic changes in levels of tumour‐ and stroma‐derived genes in circulation were consistent with those occurring in tumours and bone‐TME. We found that Ra‐223 treatment induced a reduction in the BMS. Importantly, our data suggest that resistance to Ra‐223 treatment may, in part, be mediated by tumour cells, which progress in a bone‐independent state.

A key finding of this study is that changes in EV‐derived BMS genes precede imaging‐detectable tumour growth, suggesting that the BMS reflects early remodelling events within the bone‐TME. This observation highlights a fundamental limitation of current radiologic approaches, which primarily detect structural changes rather than the underlying biological processes driving disease progression. Because the BMS captures signals derived from both tumour cells and bone stromal components, it provides a dynamic readout of tumour–bone interactions that emerge prior to overt lesion formation. This is consistent with the concept that PCa bone metastasis is initiated and sustained through early microenvironmental reprogramming, which may not yet be visible on imaging.

All of the genes comprising the signature (BMP6, BMPR2, SMAD4, SPP1, SPARC, OSTM1 and OSCAR) have been shown to play a role in bone formation as well as in creation of the bone metastatic niche to promote osteogenic tumour growth (Fang and Kang 2022; Huang et al. 2022; Li and Wang 2021; Yan et al. 2020). These genes and their corresponding proteins have been detected in EVs, but their presence has not been previously correlated specifically with M1b PCa. The bone morphogenic protein (BMP)‐driven pathway (BMP‐BMPR‐SMAD) has been previously shown to play a fundamental role in normal bone development and homeostasis as well as pathological settings such as osteogenic tumours (Dai et al. 2005; Lee et al. 2011, 2015; Lin et al. 2017; 2018; Yuen et al. 2008). BMP6 is a bone morphogenetic protein that induces bone and cartilage formation by binding to its receptor BMPR2 and transducing downstream signals in a SMAD4‐dependent manner. BMP6 has been shown to be a prognostic predictor of M1b PCa (Yuen et al. 2008). Osteopontin (SPP1) is a secreted glycoprotein in the bone extracellular matrix and has been identified as a component of signatures for PCa metastasis to the bone (Pang et al. 2019). The presence of SPP1 in EVs has been well characterized, and it is thought to play a role in bone regeneration (Faqeer et al. 2023). SPARC (Osteonectin) is a protein secreted primarily by osteoblasts and is involved in bone formation and remodelling. Osteonectin plays a role in castrate‐resistant bmPCa by acting as a chemoattractant for PCa cells to invade bone (Lopez‐Moncada et al. 2019). The two osteoclast markers that have been identified, OSCAR and OSTM1, play a role in osteoclast maturation and activation. Osteoclast‐activated receptor (OSCAR) provides co‐stimulatory signals needed for osteoclast‐genesis and activation (Nedeva et al. 2021). OSCAR has been shown to correlate with increased metastasis in immunosuppressive tumour microenvironments (Liao et al. 2021). In addition to its bone modelling functions, OSCAR has been associated with promoting an immunosuppressive TME (Liao et al. 2021). Osteopetrosis‐associated transmembrane protein 1 (OSTM1) is also required for osteoclast maturation and activation (Vacher 2022). The combination of these seven genes as a BMS reflects tumour‐induced changes in the bone‐TME, allowing detection of the micro‐environment–dependent stage of PCa progression.

Our preclinical studies demonstrated heterogeneity in the kinetics of detecting the presence of the bone metastasis genes in circulation. The bone‐related markers were detected earlier in tumour cells (human‐derived) than stromal cells (mouse‐derived), supporting the model that Oste‐mimicry of PCa cells initiates osteogenic PCa bone metastasis (Fang and Kang 2022). For some genes, expression in EVs was seen earlier than palpable tumour growth, while in others expression in EVs and tumour growth occurred in parallel. It is not clear whether this is due to the biology of tumour‐induced bone formation or tumour growth dynamics, and the question warrants further investigation. The ability to monitor the BMS genes in circulation early allows for earlier intervention and refinement of treatment strategies to address adaptive resistance.

Since Ra‐223 was approved for use in 2013, efforts to build on its initial success have been challenged by the absence of markers to predict response and/or resistance, which in turn has hindered efforts to identify novel therapy targets and develop novel therapy strategies. PSA levels reflect activity of PCa cells and do not correlate with benefit from Ra‐223. While changes in total ALP levels do correlate with benefit to Ra‐223 total ALP is not specific to osteoblasts and many patients with radiologically detected bone metastasis do not have elevated total ALP levels. In our studies using human and mouse primers to measure tumour‐ and bone–TME‐derived genes, respectively, we found that resistance to Ra‐223 treatment correlates with increased expression of the BMS in epithelial cells but not stromal cells. This data suggests that PCa cells upregulate Oste‐mimicry pathways as an adaptive resistance mechanism to Ra‐223. Indeed, Oste‐mimicry represents a pro‐survival mechanism for PCa cells within the bone‐TME in response to hormonal therapies (Bock et al. 2021). More recently, a randomized multicenter open label phase III trial comparing enzalutamide versus a combination of Ra‐223 and enzalutamide in patient bone mCRPC (EORTCGUCG 1333/PEACE‐3) demonstrated that the addition of Ra‐223 significantly improved radiographic progression free survival (rPFS) and OS (Tombal et al. 2025). However, responses were heterogeneous, suggesting adaptive resistance to Ra‐223. This tumour‐intrinsic mechanism of Ra‐223 resistance warrants further investigation and may provide strategies of rational combinations to overcome this resistance.

Furthermore, although Ra‐223 was used as a representative bone‐targeting therapy in this study, the relevance of the BMS extends beyond this specific agent. The BMS reflects fundamental features of bone‐TME biology, including dynamic tumour–bone–stromal interactions, which are modulated by a range of therapeutic approaches. With the emergence of newer radiopharmaceuticals such as Lu‐PSMA and other targeted therapies, there remains a critical need for biomarkers that capture biological responses within the bone microenvironment. Because the BMS integrates signals from both tumour and TME compartments, it has the potential to serve as a broadly applicable pharmacodynamic readout of therapy response.

Wang et al. developed a metastasis signature of 11 genes (RLN1, SLC45A3, ACP3, FOLH1, HOXB13, MSMB, KLK4, TMPRSS2, KLK2, KLK3 and STEAP2) and a met score that was shown to be significantly higher in patients with M1b PCa than in patients with localized PCa (Thomsen et al. 2020; Wang et al. 2023). The Wang signature was developed to distinguish localized from M1b PCa and is composed predominantly of tumour‐derived epithelial transcripts that reflect metastatic status and disease burden. In contrast, the BMS was specifically derived from plasma EVs enriched in patients with bone metastases and captures molecular changes associated with the bone‐TME, including osteoblast, osteoclast and bone formation‐related pathways. Consistent with this, the BMS was enriched in bone metastases but not in other metastatic sites and tracked dynamic changes in tumour‐induced bone formation, response to Ra‐223 and acquisition of treatment resistance. These findings suggest that the BMS reflects tumour–bone–stromal interactions that are not captured by tumour cell–intrinsic metastatic signatures alone. Thus, rather than representing overlapping biomarkers, the Wang signature and BMS provide complementary information, with the former informing overall metastatic disease status and the latter providing a molecular readout of bone‐TME activity. Importantly, the ability of the BMS to detect changes preceding radiologically detectable structural alterations suggests that EV‐based monitoring may complement conventional imaging approaches for the longitudinal assessment of bone metastatic progression and treatment response.

We have previously shown that BMP4 secreted from PCa tumour cells induces the transition of endothelial cells to osteoblasts to produce aberrant bone. Thus, the development of bone‐forming phenotype of M1b is initiated by tumour cells (Lin et al. 2017). As a result, targeting the tumour cells with ADT+ARPI decreases PCa‐induced bone formation. In a cohort of patients with pathologic evidence of pelvic lymph node involvement after 6 months neoadjuvant treatment with ADT+ARPi, an increase in BMS genes at the completion of hormonal treatment was associated with the subsequent development of bone metastases up to 4 years later. In contrast, the complete PSA response seen in all 8 patients did not correlate with the subsequent development of bone metastases. These observations suggest that dynamic changes in the BMS after a finite duration of hormonal therapy may help identify men at risk to develop bone metastases years before they occur. In patients with localized disease, a failure of ADT+ARPi to suppress osteomimicry pathways might suggest secondary prevention strategies with bone‐targeting agents (e.g. RANK ligand inhibitors or Ra‐223) to reduce the risk of developing bone metastases after local treatment.

The retrospective analysis of the high‐risk cohort is limited by the small sample size and by the fact that these patients were treated before the widespread adoption of PSMA PET imaging and genomic risk stratification tools such as Decipher. It is possible that a subset of patients may have bone metastatic disease at baseline that was not detectable using conventional staging approaches at the time. The observed increase in BMS following ADT+ARPi may reflect early activation of the bone‐TME rather than overt metastatic burden detectable by imaging. This is consistent with the concept that bone metastasis development is preceded by microenvironmental reprogramming that may not yet be visible using radiologic methods.

While contemporary approaches such as PSMA imaging and genomic classifiers have significantly improved disease detection and risk stratification, they primarily capture anatomical disease burden or tumour‐intrinsic features. In contrast, the EV‐derived BMS provides a dynamic and non‐invasive readout of tumour‐bone interactions, offering complementary biological information on the state of the bone‐TME. In this regard, BMS‐based monitoring is not intended to replace current diagnostic modalities but rather to complement them by providing additional molecular insight into bone‐TME biology. Dynamic changes in BMS following therapy may identify patients with early TME activation who are at increased risk of developing clinically detectable metastases, even in the presence of favourable PSA responses or negative imaging. Thus, integration of EV‐based biomarkers with advanced imaging and genomic tools in future prospective studies will be essential to define their additive clinical value in contemporary treatment paradigms.

There are several limitations associated with this study. First, the temporal sequence of bone development may not be fully reflected in the EV‐BMS since it was derived using patients with established bone metastases. The bone lesions in these patients may represent the later stages of bone developmental programs activated in bone M1b PCa. It is likely that the biologic significance of the BMS would be enhanced by including genes involved in earlier steps of the bone development program (e.g. b‐catenin, runx2, osx, notch3 and gli1/2). Second, the BMS needs to be validated in a larger cohort of patients with localized PCa and correlated with the development of bone metastasis. This will allow for the development of a bone metastasis ‘score’ that can be used in prospective studies to identify men at highest risk for developing bone metastases. Third, we examined the therapy effects on the EV‐BMS using (i) specific and relevant models that may not, though, fully represent the diversity of PCa and (ii) only two classes of life‐prolonging treatments, ARPis and Ra‐223, while there are now multiple approved agents with different mechanism of action. Despite the advent of novel radioligands like Lutetium‐177 (^{1 7 7}Lu‐PSMA), Ra‐223 remains an integral life prolonging agent in the sequential treatment of patients with M1b PCa and is being studied in combination with ^{1 7 7}Lu‐PSMA (Kostos et al. 2025; Mehr and Hauke 2025). Modulation of the BMS by these agents is currently being studied by our group.

This study highlights the potential clinical utility of the BMS to provide molecular information of the bone‐TME that complements current clinical tools (alkaline phosphatase, imaging). Through serial liquid biopsies, the BMS can provide insights into disease progression and therapy resistance. Measurement of the BMS complements the current approach of monitoring the emergence of M1b using PSA and imaging by adding information about the bone‐TME, the predominant site of treatment‐refractory disease. The results are a step towards achieving the aspirational goal of developing individualized treatment strategies using combinations or sequences that account for disease state and heterogeneity, including rational combinations of drugs, to overcome adaptive resistance.

Author Contributions

Emanuela Gentile: investigation, methodology, formal analysis, data curation, visualization, validation, Writing – review and editing, Writing – original draft. Paul G. Corn: investigation, conceptualization, formal analysis, data curation, writing – review and editing, writing – original draft, resources. Martha Zavridou: investigation, validation, writing – review and editing. James P Long: investigation, formal analysis. Jian H Song: investigation, validation, formal analysis, data curation. Anh Hoang: investigation, validation, methodology, formal analysis. Andrew W Hahn: investigation, writing – review and editing, Validation, data curation. Eleonora Dondossola: investigation, methodology. William Norton: investigation, methodology, validation. Sue‐Hwa Lin: investigation, writing – original draft, writing – review and editing, formal analysis, conceptualization. Christopher J. Logothetis: conceptualization, writing – original draft, funding acquisition, writing – review and editing. Theocharis Panaretakis: conceptualization, investigation, funding acquisition, writing – original draft, validation, visualization, writing – review and editing, formal analysis, project administration, data curation, supervision, resources.

Conflicts of Interest

Andrew W Hahn, Consultant/Advisory board: Janssen, Intellisphere, AVEO, Exelixis, Eisai, Pfizer and Tolmar. Honoraria: Medscape, Binaytara Foundation, Projects in Knowledge, Curio Science, Dava Oncology and Mashup Media. Institutional research funding from Bayer, Bristol Myers Squibb, Eisai and Halda Therapeutics

Supporting information

Supporting Information: jev270364‐sup‐0001‐SuppMat.docx

Data Availability Statement

All data generated or analysed during this study are included in this article and its supplementary information files.