PSMA⁺ Extracellular Vesicles Are a Biomarker for SABR in Oligorecurrent Prostate Cancer: Analysis from the STOMP-like and ORIOLE Trial Cohorts

The concept of oligometastatic disease was first suggested in 1995 by Hellman and Weichselbaum as a unique biological state between localized and widespread metastatic disease (1). This concept has changed the dogma that all metastatic disease was beyond cure, now proposing that patients with early metastatic disease may still be cured with a combination of local and metastasis-directed therapies (MDT). SABR-COMET was the first randomized trial to evaluate stereotactic ablative radiotherapy (SABR) in various oligometastatic cancers (2). This trial was also the first to demonstrate a survival benefit with SABR; however, prostate cancer only represented 16% of the treatment group. In prostate cancer, two randomized clinical trials have since shown an oncologic benefit of SABR in the oligometastatic setting (STOMP and ORIOLE; refs. 3, 4).

SABR is rapidly emerging as a safe curative-intent treatment option in oligometastatic castration-sensitive prostate cancer (omCSPC), helping prolong disease-free survival. It has the potential to delay or avoid the use of noncurative systemic therapy, thereby delaying the negative impact of androgen deprivation on the quality of life and the risk of developing castration-refractory disease. Both the STOMP and ORIOLE trials reported long-term disease-free survival in ∼20% of patients; however, most patients progressed and ultimately received systemic therapy (3, 4). This divergence in outcomes suggests that selected patients with omCSPC can benefit from SABR alone, whereas others may require the addition of systemic therapy. This highlights an unmet need to develop predictive biomarkers that can help select patients who may attain a durable response from SABR alone without systemic therapy. Unfortunately, to date, no such effective liquid biomarkers exist.

Tumor-derived extracellular vesicles (EV) are microscopic particles released by tumors into the bloodstream and are emerging as noninvasive biomarkers for cancer detection and prognosis (5). An important benefit of circulating tumor-derived EV is that, unlike circulating tumor cells and ctDNA, they are detectable across the entire spectrum of the disease from localized prostate cancer to widely spread metastatic castration-resistant prostate cancer (CRPC; refs. 6, 7). We recently demonstrated the positive association of circulating levels of PSMA-positive (PSMA⁺) EV with radiographic tumor burden in prostate cancer (6). We also reported that pre-SABR PSMA⁺EV levels can predict the risk of disease recurrence in patients with oligometastatic CRPC treated with SABR. Herein, we conducted a biomarker correlative study evaluating the prognostic and predictive value of PSMA⁺EV in a multi-institutional cohort of patients with omCSPC.

Three cohorts of patients with stored plasma available were obtained from Ghent University [principal investigator (PI): Piet Ost, Belgium; N = 80], the Iridium Network (PI: Carole Mercier, Belgium; N = 47), and the ORIOLE trial (PI: Phuoc Tran; N = 46; ref. 3). The Ghent University cohort enrolled patients between 2015 and 2020 under the existing STOMP trial protocol (PI: Piet Ost; ref. 4). In the Iridium Network cohort, patients were treated for omCSPC with SABR between 2018 and 2020. Detailed characteristics of the ORIOLE trial have been previously reported (3). Studies were conducted in accordance with the ethical principles outlined in the Declaration of Helsinki, with written informed consent obtained from all patients. The studies were approved by the Mayo Clinic Institutional Review Board (No. 21-004451).

Platelet-poor plasma samples were incubated with fluorescent PSMA antibodies (J591 clone), and concentrations of PSMA⁺EV were measured by nanoscale flow cytometry as previously described (6, 8). Based on calibration, PSMA⁺EV with a minimum diameter of 188 nm and a minimum fluorescence intensity of 460 molecules of equivalent soluble fluorochrome were counted. All measurements were performed blinded from clinical data. Each sample was run in three technical replicates, and the average was used for data analysis. Following standardized reporting of EV flow cytometry experiments, a detailed description of the methodology can be found in the MIFlowCyt-EV report (Supplementary Materials).

Biochemical progression-free survival (bPFS) and radiographic PFS (rPFS) were used as clinical endpoints to determine the association of PSMA⁺EV levels, serum PSA, and PSA doubling time (PSA DT) at 3 months with oncologic outcomes. Biochemical progression was defined as any of the following: (i) for patients who have undergone a radical prostatectomy, a PSA increase to ≥0.2 ng/mL from nadir after MDT; (ii) for patients who did not experience a PSA nadir below 0.2 ng/mL after MDT, the first increase in PSA after reaching nadir; (iii) for patients treated with primary radiation to the prostate, PSA nadir +2 ng/mL after MDT; or (iv) initiation of systemic therapy for local recurrence or distant recurrence before reaching the numerical definition of PSA as above. Radiographic progression was defined as new nodal lesions, intrapelvic or distant, bone lesions, or visceral lesions on conventional (bone scintigraphy) or molecular (C-11 choline PET/CT or PSMA PET/CT) imaging with the application of RECIST version 1.1. Kaplan–Meier estimates were used to estimate survival curves. For each Kaplan–Meier plot, P values were derived from the log-rank test for difference between groups. For the association of PSMA⁺EV and PSA with oncologic outcomes, concentrations were converted to categorical variables and patients were classified as high and low. The optimal cutoff values of PSMA⁺EV (2.1 × 10⁶ EV per mL for the STOMP-like cohort and 1.6 × 10⁶ EV for the ORIOLE cohort) and PSA (1.0 ng/mL for the STOMP-like cohort and 7.0 ng/mL for the ORIOLE cohort) were defined as the value with the most significant log-rank test split in a univariate Cox proportional hazards model using rPFS as clinical endpoint (9). The same cutoff values were used for all statistical analyses throughout the study. The HR of each biomarker was calculated with each clinical outcome (bPFS and rPFS). We tested the effect of PSMA⁺EV on bPFS and rPFS, controlling for the effect of PSA levels, number of lesions, and lesion location in multivariate Cox proportional hazards models. PSA DT was also treated as a categorical variable (≤ or >3 months). For predictive biomarker assessment, we tested whether the PSMA⁺EV level was predictive by including PSMA⁺EV, treatment group, and treatment-by-biomarker interaction term in a Cox proportional hazards model. HR with 95% confidence intervals (CI) were calculated. Statistical software (SAS, version 9.4, SAS Institute Inc.) was used for the multivariate Cox models. Prism v9.0.1 (GraphPad Software) was used for all other statistical analyses.

Deidentified data related to the results reported in this article are publically available upon request. Data are not published publically because of patient privacy concerns. The data will be available to researchers immediately and for 6 years following publication. Data requests and proposals should be directed toward the corresponding author Dr. Fabrice Lucien-Matteoni ([email protected]).

A total of 173 patients with plasma samples available were included (Ghent University n = 80, Iridium Network n = 47, and ORIOLE Trial n = 46). The diagram for the study cohorts is included in Supplementary Fig. S1. Sixteen patients with CRPC disease or local recurrence/nonmetastatic disease were excluded from the study. For the ORIOLE cohort, 30 patients were treated with SABR and 16 patients were part of the observation arm. Patient characteristics for the three cohorts are presented in Table 1. For all patients treated with SABR (n = 157), no active systemic therapies were received concomitantly with SABR. Per the classification developed by the European Society for Radiotherapy and Oncology (ESTRO) and European Organisation for Research and Treatment of Cancer (10), 92% (145/157) of patients were diagnosed with metachronous oligorecurrent prostate cancer. Oligometastatic disease was diagnosed with advanced PET imaging (80%) and conventional imaging (20%; Table 1). All patients were diagnosed with C-11 choline PET in the Ghent University cohort, whereas most patients (94%) received PSMA PET in the Iridium Network. The median follow-up was 45.7 months (range, 41.2–51.7). The median bPFS was 21.5 months in the Ghent University cohort, 17.5 months in the Iridium cohort, and 11.1 months in the ORIOLE SABR arm (Supplementary Fig. S2A). The median rPFS was 29.0 months in the Ghent University cohort, 32.1 months in the Iridium cohort, and 25.3 months in the ORIOLE SABR arm (Supplementary Fig. S2B).

Comparative analysis between baseline (pre-SABR) PSA levels demonstrated no significant difference between the Ghent University and Iridium Network cohorts (1.9 and 2.0 ng/mL, P > 0.99; Supplementary Fig. S3A). There was a significant difference in baseline PSA when comparing the Ghent University and Iridium Network cohorts (P < 0.0001) with the ORIOLE cohort (6.8 ng/mL), which could be explained by the lower sensitivity of conventional imaging (ORIOLE) over PET imaging (both Belgian centers) for detecting radiographic recurrence. Comparative analysis between baseline PSMA⁺EV demonstrated no significant difference between the Ghent University and Iridium Network cohorts (5.91 × 10⁶ and 4.98 × 10⁶ EV/mL, P = 0.74). Similar to baseline PSA, there was a significant difference in baseline levels of PSMA⁺EV in the ORIOLE cohort (median, 2.12 × 10⁶ EV/mL) when compared with those in the Ghent University (P < 0.0001) and Iridium Network (P = 0.0004) cohorts (Supplementary Fig. S3B). No correlation was found between baseline levels of PSA and PSMA⁺EV in the three cohorts (Supplementary Fig. S3C). Given the similarities in patient characteristics, treatment plan, and oncologic outcomes of the Ghent University and Iridium Network cohorts with the STOMP trial (4), we combined both cohorts and referred to them as the “STOMP-like” cohort hereafter. Oncologic outcomes of patients stratified by the number of metastatic lesions can be found in Supplementary Fig. S4.

Following the stratification of patients into low and high baseline level groups based on the PSMA⁺EV levels in the pooled cohort of the ORIOLE trial and STOMP-like cohorts, the median bPFS were 26.1 and 15.0 months (P = 0.005; Fig. 1A), and the median rPFS were 36.0 and 25.0 months for patients with low and high PSMA⁺EV levels, respectively (P = 0.003; Fig. 1B). Low baseline levels of PSMA⁺EV were associated with a lower risk of bPFS (HR, 0.59, 95% CI, 0.40–0.85; P = 0.005) and rPFS (HR, 0.55, 95% CI, 0.34–0.81; P = 0.003). Patients stratified based on PSA and PSA DT demonstrated similar differences in PFS outcomes, but baseline PSA was a superior predictor of outcome compared with PSA DT (Supplementary Figs. S5–S7).

PSMA⁺EV remained an independent predictor of the risk of both bPFS and rPFS when controlling for the effect of PSA levels, number of lesions, and lesion location in multivariate Cox proportional hazards models. In the pooled cohort, low PSMA⁺EV was also an independent predictor of bPFS (HR, 0.59, 95% CI, 0.40–0.85; P = 0.005) and radiographic progression (HR, 0.55, 95% CI, 0.34–0.81; P = 0.003; Supplementary Table S1).

Although patient stratification based on PSMA⁺EV levels alone produced significant differences in PFS, combining PSMA⁺EV and PSA resulted in identification of long-term responders to SABR (Table 2). Patients were stratified into PSA-low/PSMA⁺EV-low, PSA-high/PSMA⁺EV-high, PSA-low/PSMA⁺EV-high, and PSA-high/PSMA⁺EV-low groups. In the pooled cohort, 15% (19/127) of patients presented with PSA-low/PSMA⁺EV-low levels. The median bPFS and rPFS were not reached, and PSA-low/PSMA⁺EV-low level was a superior prognostic marker of bPFS and rPFS (Fig. 2A and B). The combination of baseline levels of PSMA⁺EV and PSA was associated with a lower risk of bPFS (HR, 0.34, 95% CI, 0.18–0.58; P = 0.0002) and rPFS (HR, 0.22, 95% CI, 0.09–0.44; P = 0.0001). PFS stratified by PSMA⁺EV and the combination of PSMA⁺EV and PSA for the individual cohorts is given in Supplementary Figs. S8 and S9.

We analyzed the baseline levels of PSMA⁺EV in both the SABR and observation arms of the ORIOLE cohort to evaluate their predictive value. Patients who presented with low levels of PSMA⁺EV showed benefit from SABR compared with patients in the observation arm (Fig. 3A). The median bPFS was 24.3 and 5.8 months for SABR and observation arms, respectively (P = 0.003). Patients treated with SABR had a significantly lower risk of biochemical progression (HR, 0.19, 95% CI, 0.065–0.644; P = 0.004). In contrast, SABR did not show any benefit for patients with high baseline levels of PSMA⁺EV compared with the observation arm (5.9 and 7.1 months, P = 0.95; Fig. 3B). The risk of biochemical progression was not statistically different between both groups (HR, 1.17, 95% CI, 0.498–2.766; P = 0.715). Baseline levels of PSMA⁺EV did not significantly affect the risk of radiographic progression (Supplementary Fig. S10). There was no significant treatment effect within the PSMA-high group (HR, 1.39, 95% CI, 0.50–3.84, P = 0.512) or within the PSMA-low group (HR, 0.57, 95% CI, 0.14–2.22; P = 0.419). In patients with low baseline PSA levels, bPFS benefit was observed from SABR compared with observation, but the dichotomy (P = 0.03) was not as significant as that with PSMA⁺EV (P = 0.003; Supplementary Fig. S11).

Systemic therapy with androgen-deprivation therapy in combination with next-generation androgen receptor–targeted therapy has level one evidence for overall survival benefit in patients with low-volume metastatic CSPC. The benefit of systemic therapy intensification with chemotherapy remains very modest in patients presenting with metachronous low-volume disease (≤3 nonvisceral metastases; ref. 11). The concept of MDT with SABR or surgery in oligorecurrent prostate cancer is gaining traction following the positive results from the STOMP and ORIOLE randomized trials (3, 4). MDT represents a safe and effective treatment option to eliminate visible metastases on imaging without using systemic therapy. Of note, ∼10% to 15% of patients achieve prolonged disease control with no sign of recurrence on imaging for at least 4 years after treatment (12–16). Although local control remains excellent (∼90% at 2-year follow-up) after SABR treatment, oligoprogression remains the major cause of clinical failure after SABR, and it suggests that patients may present at diagnosis with micrometastatic disease below the threshold of detection of both conventional and advanced imaging (17). To date, treatment with MDT and/or systemic-intensified hormonal therapy for patients with oligorecurrent CSPC is informed by clinical presentation such as the volume and number of metastases, metastasis location, patient comorbidities, and physician experience. The integration of tumor biomarkers with imaging variables better supports treatment decisions and estimation of response to SABR (12). Our current work demonstrates the clinical value of a pretreatment PSMA⁺EV liquid biopsy as a noninvasive biomarker to improve stratification of oligometastatic disease state and better select patients for SABR.

In this study, we assessed three potential prognostic biomarkers to SABR; PSMA⁺EV, PSA, and PSA DT in a large multi-institutional dataset of 157 patients with omCSPC. We found that high baseline concentrations of PSMA⁺EV were a prognostic biomarker of both biochemical and radiographic progression following SABR treatment, which is in line with our original proof-of-concept study in oligometastatic CRPC (6). The combination of baseline PSA and PSMA⁺EV was superior to all three individual biomarkers. On assessment with multivariate Cox proportional hazards models, controlling for the effect of PSA levels, number of lesions, and lesion location, PSMA⁺EV level remains an independent prognostic biomarker. In addition, a total of 29/157 (18.5%) patients were identified with low PSA and PSMA⁺EV levels, and this combination was associated with long PFS. Similar to STOMP and ORIOLE studies, 30/157 patients (19.1%) did not experience either biochemical or radiographic progression at a median follow-up of 40.3 months (95% CI, 36.2–48.5). Among those patients, 16 patients (53.3%) were classified as pre-SABR PSA-low and PSMA⁺EV-low. Conversely, 109/157 patients (69.4%) showed evidence of radiographic progression at a median follow-up of 20.4 months (95% CI, 17.0–25.30), and only seven (6.4%) had both PSA-low and PSMA⁺EV-low levels. Our work demonstrates that pre-SABR blood levels of PSA and PSMA⁺EV are blood-based biomarkers that can identify patients with truly omCSPC from those who likely harbor micrometastatic disease at diagnosis and may be predictive of response to SABR alone.

We identified PSMA⁺EV as a predictive biomarker of bPFS in response to SABR from the ORIOLE trial. PSMA⁺EV was not found to be a predictive biomarker of rPFS likely because of the limited sample size and crossover that occurred in almost all patients in the observation arm, resulting in low differential PFS events. SABR demonstrated durable benefit in patients with low PSMA⁺EV levels, yet SABR did not provide any benefit in patients with high PSMA⁺EV levels compared with observation. Validation of PSMA⁺EV as a predictive biomarker of both biochemical and radiographic progression is currently ongoing in a prospective trial (DIVINE; NCT06378866) and could provide a promising blood-based predictive biomarker for MDT in omCSPC. Moving forward, the future management of omCSPC may be improved in which treatment decisions are better informed by advanced PET imaging, blood-based measurement of tumor burden, and molecular profiling. We propose that this management approach for oligometastatic prostate cancer should be evaluated in a prospective clinical trial that stratifies treatment based on PSMA⁺EV and PSA levels similar to the escalation/de-escalation design of the PREDICT-RT trial (NCT04513717).

Owing to a low tumor mutational burden and limited metastatic burden, ctDNA concentrations are very low in omCSPC compared with those in metastatic CRPC (18, 19). In the ORIOLE trial, 41% of patients had detectable ctDNA (mean, 1.3 mutations per participant), and ctDNA abundance was not associated with clinical outcomes (3). Although studies utilizing ultrasensitive sequencing technologies are needed to investigate the potential utility of ctDNA in omCSPC, the abundance and detection rate of PSMA⁺EV in this setting make it a suitable biomarker for risk stratification and treatment selection. As EV carry molecular cargo from donor cells and protect them from enzymatic degradation (20), characterizing the DNA content of PSMA⁺EV assay may provide further utility to EV as liquid biopsy.

This study is not without limitations. Although it had access to patient plasma samples from the ORIOLE trial, most of the patients in the STOMP trial had stored serum samples. Plasma samples are more suitable than serum for nanoscale flow-cytometry–based EV measurement (data not shown). Therefore, we utilized plasma samples from patients recruited outside the STOMP trial but under the same protocol. In line with this, we acknowledge the potential for bias and variability in collection from a nonrandomized cohort. We showed PSMA⁺EV as a potential predictive biomarker of response to SABR in the ORIOLE cohort. Although the data are promising, the sample size is small, and prospective studies in larger cohorts are warranted. Additionally, our study did not consider the tumor mutational profile of patients. Deek and colleagues (21) found several germline mutations (e.g., TP53, ATM, RB1, and BRCA1/2) in omCSPC that are associated with a higher risk of clinical failure following SABR (12). In the future, the definition of oligometastasis may not solely rely on the number of metastases but also on the intrinsic tumor biology, which our PSMA⁺EV assay does not currently assess. Further studies leveraging the somatic and germline mutational signatures combined with PSMA⁺EV assay may improve patient risk stratification and selection of appropriate treatment strategies. Finally, we found significantly lower median levels of PSMA⁺EV in the ORIOLE cohort compared with those in the STOMP-like cohort, which is counterintuitive considering that the ORIOLE trial enrolled patients diagnosed with conventional imaging. This was unexpected, and although we have no concrete explanations to explain this, we cannot exclude variability in blood collection, handling, and storage conditions which could have affected the PSMA⁺EV measurement retrospectively (22), including freeze-thaw cycles.

In conclusion, PSMA⁺EV is a novel prognostic biomarker of tumor burden and micrometastatic disease in omCSPC and a predictive biomarker of biochemical progression for SABR in omCSPC. Oligometastatic prostate cancer represents a heterogeneous patient population, and the combination of PSMA⁺EV and PSA can help refine the selection of patients who can experience durable disease-free survival in response to SABR. This observational study provides the first clinical use of EV in a prospective trial of omCSPC and strengthens the clinical value of PSMA⁺EV for personalized radiotherapy. The notable divergence in survival curves stratified by PSMA⁺EV and PSA levels highlights the need for prospective validation of PSMA⁺EV as a predictive biomarker for SABR. Going forward, the testing of biomarker-directed risk stratification and personalization in prospective clinical trial protocols is needed.

J.R. Andrews reports personal fees from Bayer and Blue Earth Diagnostics outside the submitted work. R. Phillips reports nonfinancial support from Novartis AG and Veracyte, Inc outside the submitted work. D.S. Childs reports other support from Novartis, Janssen, and Abdera outside the submitted work. J.J. Orme reports grants from the Prostate Cancer Foundation, Department of Defense, and NCI during the conduct of the study. A.A. Chaudhuri reports nonfinancial support from Roche, personal fees and nonfinancial support from Illumina, personal fees from Myriad Genetics, Invitae, Daiichi Sankyo, AstraZeneca, Guardant Health, Caris, AlphaSights, DeciBio, Guidepoint, and Agilent; other support from Geneoscopy, Droplet Biosciences, and LiquidCell Dx; and grants and other support from Tempus outside the submitted work, as well as a patent for filings related to cancer biomarkers pending. P. Tran reports personal fees from Natsar Pharmaceuticals, RefleXion Medical, Bayer, Janssen, Regeneron, Lantheus, and Pfizer outside the submitted work, as well as a patent for “Compounds and methods of use in ablative radiotherapy; patent No. 9114158” licensed and with royalties paid from Natsar Pharmaceuticals. A. Kiess reports grants from Bayer during the conduct of the study, as well as grants from Novartis, Convergent, and Lantheus outside the submitted work. C. Mercier reports grants from RaySearch Laboratories outside the submitted work. P. Ost reports grants from Bayer and other support from Bayer, AstraZeneca, Novartis, and Janssen outside the submitted work. F. Lucien reports grants from NaNotics, personal fees from Mursla Bio, and grants and nonfinancial support from Early is Good outside the submitted work, as well as a patent for US 63-633200 issued. No disclosures were reported by the other authors.

J.R. Andrews: Conceptualization, resources, data curation, formal analysis, supervision, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. Y. Kim: Investigation, writing–review and editing. E. Horjeti: Validation, investigation, writing–review and editing. A. Arafa: Writing–review and editing. H. Gunn: Formal analysis, writing–review and editing. A. De Bruycker: Resources, writing–review and editing. R. Phillips: Resources, data curation, writing–review and editing. D. Song: Writing–review and editing. D.S. Childs: Writing–review and editing. O.A. Sartor: Supervision, writing–review and editing. J.J. Orme: Supervision, writing–review and editing. A.A. Chaudhuri: Resources, data curation, writing–review and editing. P. Tran: Resources, data curation, supervision, writing–review and editing. A. Kiess: Resources, data curation, supervision, writing–review and editing. P. Sutera: Resources, data curation, writing–review and editing. C. Mercier: Resources, data curation, supervision, writing–review and editing. P. Ost: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, project administration, writing–review and editing. S.S. Park: Resources, data curation, writing–review and editing. F. Lucien: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, project administration, writing–review and editing.

This work was approved by the Mayo Clinic Institutional Review Board with informed patient consent under IRB 21-004451.

This work was supported by the Erivan K. and Helga Haub Family Fund in Image-Guided Urology (F. Lucien).