Early On-treatment Changes in Circulating Tumor DNA Fraction and Response to Enzalutamide or Abiraterone in Metastatic Castration-Resistant Prostate Cancer

Metastatic castration-resistant prostate cancer (mCRPC) is fatal, but widespread availability of androgen receptor pathway inhibitors (ARPI) have resulted in clinically-meaningful improvements in long-term patient outcomes (1, 2). Despite recent level 1 evidence supporting use of combination treatment with ARPI and androgen deprivation therapy in metastatic hormone-sensitive prostate cancer (mHSPC; refs. 3–6), physician choices or barriers to access are reflected in real-world treatment patterns where most patients do not receive ARPI until first-line therapy for mCRPC (7–9). However, 20% to 30% of mCRPC will exhibit primary or rapidly-acquired resistance to first-line ARPI and progress within 6 months of starting treatment (1, 2). Current biochemical or radiographic assessments of treatment response are not considered reliable before 12 weeks due to early fluctuations in serum prostate-specific antigen (PSA) and flares in imaging (10). Therefore, to improve mCRPC management in an era with multiple life-prolonging treatment options including chemotherapy, PARP inhibitors, and radionuclide therapy, we require earlier indicators of ARPI response.

Longitudinal testing of plasma circulating tumor DNA (ctDNA) is an emerging approach to monitor disease (11). On-treatment changes in ctDNA fraction (the proportion of total plasma cell-free DNA that is tumor-derived) are associated with clinical outcomes in some advanced cancers (12, 13). However, prior prostate cancer studies were hampered by inconsistent plasma sampling, heterogeneous study populations across different treatment lines, and ctDNA detection methods with limited sensitivity or inability to filter variants related to clonal hematopoiesis (14–17). Furthermore, because ctDNA fraction prior to treatment is a strong prognostic indicator in mCRPC (18–20), it is currently unknown if on-treatment ctDNA measurements will provide additional clinical utility. Here, we leverage a prospective multicenter study to test whether baseline and 4-week on-treatment ctDNA fraction measurements can identify patients with mCRPC that have nondurable responses to first-line ARPI.

This study represents a pooled analysis of patients prospectively enrolled in two observational multicenter studies conducted in the Netherlands between July 2015 and October 2018: OPTIMUM (ClinicalTrials.gov identifier: NCT02426333) and ILUMINATE (NCT02471469). In both studies, patients with biochemically/radiographically-progressive mCRPC and asymptomatic or minimally-symptomatic disease were eligible if they had a life expectancy of ≥6 months. Exclusion criteria included prior systemic therapy (ARPI or chemotherapy) for mCRPC, although docetaxel for mHSPC was permitted in accordance with established standard-of-care (21, 22). ARPI was not locally indicated for mHSPC, where these studies were conducted. Consequently, no patient received ARPI in the mHSPC setting. Patients were excluded if they stopped treatment before the second study visit at 4 weeks. Eligible patients received either abiraterone acetate 1000 mg daily plus prednisolone 10 mg daily (OPTIMUM), or enzalutamide 160 mg daily (ILUMINATE). The studies and correlative ctDNA analysis were approved by the medical ethics committee at the Radboud University Medical Center and the University of British Columbia Clinical Research Ethics Board. The studies were conducted in accordance with Good Clinical Practice guidelines and the Declaration of Helsinki. All participants provided written informed consent.

A key prespecified objective of both studies was to define blood-based biomarkers associated with durable treatment response. Therefore, peripheral blood (9 mL) for plasma ctDNA analysis was collected in a single Streck Cell-Free DNA BCT tube at two set timepoints: (i) study baseline prior to treatment initiation (not exceeding 2 weeks before commencing ARPI), and (ii) 4 weeks after commencing study treatment. The protocol-specified 4-week timepoint was selected based on practical reasons since it represents the first hospital visit in standard clinical practice. In addition, at 4 weeks the steady-state concentration of abiraterone and enzalutamide in the blood is reached (23). Furthermore, 4 weeks is within the period where radiographic reimaging (i.e., for response evaluation) may be obfuscated by false positives due to flare phenomenon. An additional whole blood EDTA sample (6 mL) was collected for resolving germline variants from putative somatic mutations and to exclude somatic mutations present in white blood cell DNA (including from clonal hematopoiesis). Plasma cell-free DNA and white blood cell DNA was subjected to deep targeted sequencing with a laboratory-developed panel covering coding regions and select introns of 73 prostate cancer-related genes and a genome-wide grid capturing common heterozygous germline SNPs. CtDNA fraction (ctDNA%) at baseline and 4-week timepoints were estimated using allele fractions of autosomal somatic mutations in non-amplified regions, or genome-wide copy number and heterozygous SNP allele imbalance in cases with no eligible somatic mutations. Methodology for plasma processing, targeted sequencing, bioinformatic ctDNA% estimation and justification of ctDNA% detection threshold is provided in Supplementary Materials and Methods.

In addition to predefined sampling timepoints, all outcome measurements were prespecified in the study protocol. Prior treatment history, Eastern Cooperative Oncology Group (ECOG) performance status and baseline laboratory tests (including hemoglobin, alkaline phosphatase (ALP), and lactate dehydrogenase (LDH)) were obtained at study entry. Patients underwent clinical assessment prior to the start of therapy (defined as “baseline”) and at 4 weeks, 3 months, and 6 months after commencing protocol treatment. PSA measurements were taken at each scheduled review, and radiographic tumor assessment (CT or whole body MRI combined with bone scan) was performed at baseline, 3 months and 6 months after commencing ARPI therapy. Radiographic progression was evaluated by local investigator assessment, with subsequent independent blinded central review using Prostate Cancer Working Group 2 (PCWG2) guidelines for bone disease and RECIST v1.1 for soft-tissue disease (24, 25). For 3 patients the independent blinded central review was not performed. Local investigator assessment was used for calculating radiographic progression-free survival (PFS), with clinical decisions taken on local imaging results. After 6 months of protocol treatment, the interval between clinical assessments, PSA testing, and radiographic imaging was not mandated, but instead left to the discretion of site investigators as per local institutional practices.

Because of the exploratory nature of biomarker discovery in both studies, no formal sample size calculation was performed for this pooled analysis. Baseline and 4-week plasma samples were classified on the basis of presence of ctDNA (detected vs. undetected; see Supplementary Materials and Methods for limits of detection). On-treatment ctDNA response was assessed by comparing baseline and 4-week ctDNA%, and categorizing into four scenarios: (i) detected to detected, (ii) detected to undetected, (iii) undetected to undetected, and (iv) undetected to detected. Kaplan–Meier survival estimates were used to assess the association between ctDNA% and clinical outcomes: (i) PFS (prospectively assessed and defined as the time from baseline to prostate cancer-related clinical progression according to the treating physician; radiographic progression according to local assessment; or death due to prostate cancer), and (ii) overall survival (OS), defined as the time from baseline to death from any cause. Exploratory outcome measurements included: (i) time on treatment (baseline to therapy discontinuation); (ii) biochemical PFS (bPFS; time from baseline to PSA increase ≥25% beyond 12 weeks and confirmed with a second PSA measurement); and (iii) combined PFS, defined as time from baseline to first of prostate-related clinical progression, radiographic progression, PSA progression, or death due to prostate cancer. The independent utility of ctDNA% change to predict PFS and OS was assessed using multivariable Cox proportional hazard models [covariates: ctDNA% at baseline, PSA > cohort median, LDH > upper limit of normal (ULN), and ALP > ULN; refs. 18, 26, 27]. For comparison, an additional multivariable analysis was performed with LDH, ALP, and log-transformed PSA as continuous variables and an exploratory multivariable analysis was performed with ctDNA% change at 4 weeks (detected vs. undetected at baseline and 4 weeks), PSA change at 4 weeks (PSA decline of 30% achieved vs. not achieved), and LDH normalization at 4 weeks (LDH below vs. above ULN). For PFS analyses, patients were censored at the therapy stop date in cases where no PFS was observed, or at last follow-up if patients remained on treatment. For OS analyses, patients were censored at the last follow-up date. Follow-up was calculated from the date of baseline blood draw to last patient contact.

The potential for ctDNA% changes to differentiate patients with and without durable response was evaluated using positive predictive value (PPV) and negative predictive value (NPV). Durable treatment response was defined as PFS >6 months according to the local assessment. The 6-month threshold was selected on the basis of the registration trial for enzalutamide in chemotherapy-naïve mCRPC, in which the median radiographic PFS of the placebo arm was 5.4 months (28). Patients that ceased therapy before 6 months without PFS were excluded from the PPV/NPV analysis. The PPV and NPV of ctDNA change was compared with the PPV and NPV of a PSA decrease of 30% from baseline to 4 weeks (PSA30), which has been described as an early marker for treatment response (29, 30). Statistical significance was defined as P < 0.05 and all statistical tests were two-sided. Statistical analyses were carried out using R v.3.6.2 using the survival and survminer packages.

Unprocessed de-identified sequencing data are available at the European Genome-phenome Archive (accession code EGAS00001006856) under standard controlled release.

Of 97 patients enrolled, 81 patients were eligible for ctDNA analysis (51 received abiraterone acetate and 30 received enzalutamide; Fig. 1A). Baseline patient characteristics are presented in Table 1 (see also Supplementary Table S1). At data cut-off (26 February 2021), the median follow-up was 27.4 months [inter quartile range (IQR), 17.7–34.9] with 52 deaths (64%) and 60 patients (74%) experiencing radiographic and/or clinical progression. The median PFS was 10.2 months and median OS was 20.6 months. In total, 27 (33%) patients experienced nondurable response (PFS ≤6 months) and 50 (62%) patients experienced a durable response; 4 patients could not be assessed due to treatment cessation for toxicity within 6 months.

ctDNA was detected in 48/81 (59%) patients at baseline and 29/81 (36%) patients at 4 weeks (Fig. 1B; Supplementary Table S2). Patients with detected ctDNA at baseline were enriched for clinical prognostic markers of disease aggression, including higher baseline PSA (median 64 ng/mL vs. 29 ng/mL, Mann–Whitney U, P = 0.02), five or more bone metastases (73% vs. 36%, Fisher exact test, P = 0.001), elevated ALP (54% vs. 15%, Fisher exact test, P < 0.001), and elevated LDH (43% vs. 18%, Fisher exact test, P = 0.03). Nineteen patients converted from detected ctDNA at baseline to undetected at 4 weeks; no patients converted from undetected to detected. Among patients with detected ctDNA at both timepoints, the median ctDNA% (as a proportion of total cell-free DNA) was lower at 4 weeks than at baseline (14.5% vs. 5.0%; P = 0.017; Mann–Whitney U test; Fig. 1C). The majority of patients showed a reduction in both ctDNA and PSA at 4 weeks of treatment, whereas LDH was mostly stable and below the ULN (Supplementary Figs. S1 and S2). Genomic alterations identified in ctDNA were consistent with previous cohorts of clinical mCRPC assessed by tissue or liquid biopsy (Supplementary Figs. S3 and S4; refs. 18, 31–34).

Detected ctDNA at baseline was associated with shorter PFS and OS compared with undetected ctDNA [median PFS 5.77 vs. 20.23 months; HR, 2.48; 95% confidence interval (95% CI), 1.45–4.23; P = 0.001, univariate; Fig. 2A; median OS 22.70 months vs. not-reached (NR), HR, 3.56; 95% CI, 1.89–6.72; P < 0.001, univariate; Fig. 2B]. Consistent with our prior work, very high baseline ctDNA% (>30%) was linked to particularly poor outcomes in comparison to patients with undetected ctDNA [median OS 15.80 months vs. not reached (NR); HR, 8.12; 95% CI, 3.75–17.61; P < 0.001, univariate; Supplementary Fig. S5; refs. 18, 20].

A total of 29 of 48 (60%) patients with detected ctDNA at baseline also had ctDNA detected in their 4-week blood collection. Compared with patients with undetected ctDNA at both timepoints, these patients experienced the shortest PFS and OS (median PFS 4.82 months; HR, 4.79; 95% CI, 2.62–8.77; P < 0.001; and median OS 16.00 months; HR, 5.49; 95% CI, 2.76–10.91; P < 0.001; Fig. 2C and D). No significant PFS difference was observed between patients who converted from detected to undetected ctDNA by 4 weeks and patients with undetected ctDNA at baseline, although OS was shorter (Fig. 2C and D; Table 2). Similar relationships were observed for other outcome measures including bPFS in isolation, and a combined PFS measure incorporating bPFS with radiographic and/or clinical progression (Supplementary Fig. S6).

Clinical prognostic indicators of high tumor burden were enriched in patients with ctDNA detected at baseline and at 4 weeks (Supplementary Fig. S7). In addition, baseline ctDNA% (Supplementary Fig. S8; two-sided t test; P = 0.002) and correspondingly the prevalence of detected baseline AR gain and TP53 alterations was higher in this group compared with patients with undetected ctDNA at 4 weeks (Supplementary Fig. S7; chi-square; P = 0.021 and 0.039, respectively). However, detection of ctDNA at 4 weeks remained independently associated with PFS and OS in a multivariable model incorporating ctDNA detection at baseline and clinical prognostic factors (multivariable HR, 4.98 compared with patients with undetected ctDNA at both timepoints, P < 0.001 and HR, 3.69; P = 0.005 for PFS and OS, respectively (Table 2; Supplementary Table S3).

A total of 23 of 27 (85%) patients experiencing nondurable responses had ctDNA detected at both baseline and 4 weeks. Conversely, only 3 of 50 patients (6%) experiencing durable responses had ctDNA detected at both timepoints (Fig. 3A and B). PPV and NPV for predicting nondurable responses with baseline and 4-week ctDNA detection was (23/26) 88%, 95% CI, 72% to 96%, and (47/51) 92%, 95% CI, 82% to 97%, respectively. Fourteen patients experienced progression by 12 weeks on-treatment, suggestive of primary resistance to ARPI. A total of 13 of 14 patients had ctDNA detected at both timepoints.

Most patients converting from detected to undetected ctDNA at 4 weeks showed strong PSA responses at 4 and 12 weeks on-treatment (Supplementary Fig. S7), whereas many patients with persistent ctDNA showed no PSA decrease (Supplementary Fig. S9). However, the PPV and NPV of PSA30 at 4 weeks to identify nondurable responses was only (13/23) 57%, 95% CI, 40% to 72%; and (40/54) 74%, 95% CI, 66% to 81%, respectively (Fig. 3C), suggesting considerably stronger performance for 4-week ctDNA measurements compared with PSA responses. The per patient changes in PSA levels and ctDNA fractions following 4 weeks of treatment in relation to the durability of response are visualized in Supplementary Fig. S10. An exploratory multivariable analysis further emphasized that ctDNA measurements are independent of changes in conventional blood-based biomarkers at 4 weeks (PSA and LDH; Supplementary Table S3).

This study demonstrates that early on-treatment ctDNA% changes provide an accurate and independent indication of ARPI response in patients with mCRPC. Importantly, on-treatment blood draws for ctDNA analysis are practical to perform, and ctDNA% is increasingly reported by clinical test providers (35). ctDNA% is a variable that can be measured with a range of different assays and does not strictly require the bespoke prostate cancer panel that we applied in our study (36). Therefore, we believe that early ctDNA% measurements represent a pragmatic near-term strategy to improve patient management.

Current clinical tools for mCRPC do not accurately predict impending disease progression before radiographic assessments at 3 and/or 6 months (10). Although 4-week PSA30 responses have been associated with PFS in mCRPC receiving ARPI (29, 30), our data suggest suboptimal performance for identifying disease that will progress before 6 months of treatment. In line with our results, Petrylak and colleagues (37) reported that 4-week PSA30 failed the criteria for surrogacy to rapidly evaluate treatment response. In addition, the utility of PSA monitoring to reflect disease response is eroded with successive treatment exposure in mCRPC. It is estimated that up to 25% of patients with mCRPC experience progression on enzalutamide without rising PSA levels (38). Consequently, the biomarker utility of 4-week PSA30 should be treated with caution.

Prior studies have indicated that baseline ctDNA% is associated with PFS and OS in mCRPC (18–20), independent of clinical prognostic factors. Here, 4-week on-treatment ctDNA% changes provided additional prognostic resolution versus baseline ctDNA% measurements alone, and was markedly superior to 4-week PSA30 measurements and 12-week PSA30/PSA50 measurements for identifying patients that will experience rapid disease progression during ARPI treatment. This result should not diminish the emerging value of baseline ctDNA testing, where prognosis and genomic alterations predictive of targeted therapy vulnerability can be identified prior to treatment selection. Both on-treatment and baseline ctDNA% were independently associated with OS in our study. In fact, the majority of patients with persistent ctDNA detected at 4 weeks had 1% to 30% ctDNA at baseline (rather than >30%, which was the baseline ctDNA% threshold associated with the poorest outcomes). Although we considered other measures of on-treatment ctDNA% decline, we choose to focus on ctDNA conversion (i.e., detected to not detected) on the basis of prior studies (16–20, 39, 40), the lack of a validated threshold for a meaningful ctDNA% decline and the inherent level of uncertainty surrounding ctDNA% estimates. Importantly, 15 of 27 (56%) patients with a nondurable response exhibited more than a 50% decline in ctDNA at 4 weeks, suggesting that this metric (which is described in literature for other cancer types; refs. 41, 42) is less informative than ctDNA conversion for understanding duration of response in mCRPC.

Several recent studies in other solid cancers have also linked on-treatment ctDNA clearance with improved PFS and/or OS (43–45), advocating for ctDNA as a potential pan-cancer biomarker. In prostate cancer, our findings are supported by a prior study investigating plasma ctDNA alterations in patients with mCRPC after one cycle of abiraterone acetate treatment (17). In this published study, inability to detect baseline genomic alterations after one cycle of treatment was associated with good long-term outcomes. However, Jayaram and colleagues used a relatively small gene panel for mutation detection which may have hindered detection of ctDNA% between 1 and 10%. In our study, ctDNA% below 10% was still associated with poor treatment outcomes, suggesting that sensitivity for low ctDNA% will be important in any future clinical-grade tests. Furthermore, although we incorporated deep targeted sequencing of an established broad gene panel (18, 20) together with genome-wide copy number and allelic imbalance profiling, some patients likely harbored ctDNA% below our detection threshold. Use of personalized mutation panels and/or methylation-based cell-free DNA profiling might boost sensitivity for very low ctDNA% (<1%) and potentially refine outcome predictions (44, 46). However it is not always possible to profile relevant archival prostate cancer tissue in order to have ground truth for truncal patient-specific somatic mutations (47), and it is unknown whether very low on-treatment ctDNA levels are clinically-meaningful in patients with metastatic disease.

Although ctDNA% at baseline and 4-weeks was prognostic in our study, early ctDNA% measurements could also be used to stratify randomized clinical trial interventions such as treatment changes or intensification. mCRPC likely to progress rapidly on ARPI may still respond to other therapeutic options such as taxane-based chemotherapy or radionuclides. Interestingly, several patients with rapid resistance had germline and/or somatic DNA homologous recombination repair or mismatch repair gene alterations, which are associated with benefit from PARP and immune checkpoint inhibitors, respectively (48). Therefore, serial ctDNA profiling could serve a dual purpose of identifying inadequate disease control and indicating potential therapeutic vulnerabilities. Even if no further treatment options are available, early discontinuation of ineffective therapy on the basis of ctDNA measurements has potential to reduce patient exposure to unnecessary drug-related adverse effects and financial toxicity. Blood samples drawn at the end of cycle 1 would realistically provide ctDNA results within 2 weeks, meaning that decisions could be made before the end of cycle 2. Although ctDNA testing is not as cheap as PSA or other simple blood-based markers, ctDNA% may require a single on-treatment measurement, and more cost-effective techniques that could be implemented in molecular pathology laboratories are in development.

Our patient cohort was relatively small, and due to the exploratory design of our ctDNA analysis we did not perform formal sample size calculation. However, a key strength of our study above prior work was the mandated imaging at baseline, 3 and 6 months, together with the prospective assessment of radiographic and/or clinical progression. Although no imaging was mandated after 6 months, the strong relationships between ctDNA% and PFS were supported by OS data showing a 5.5 times shorter OS for patients with detected ctDNA at baseline and 4 weeks compared with patients with undetected ctDNA. Furthermore, the association of on-treatment ctDNA detection with both PFS and OS was independent of other prognostic clinical features (ctDNA% baseline, PSA, LDH, ALP), retaining high multivariable HR of 4.98 and 3.69, respectively. ctDNA measurements at 4 weeks also provided clear added value beyond PSA or LDH changes. In first-line mCRPC, most patients have LDH levels below the ULN, making LDH changes around this threshold less suited for monitoring treatment response. PSA is a valuable biomarker in HSPC and in mCRPC exhibits reasonable sensitivity as a response biomarker after several measurements and a longer time on treatment (>12 weeks). However, ctDNA changes at 4 weeks appear to be more suitable for very early identification of patients with mCRPC who are expected to exhibit a short response to ARPI.

Our study was designed and accrued before several phase III clinical trials reported a survival benefit for use of ARPI upfront in mHSPC (3–6). Therefore, our cohort of ARPI-naïve mCRPC no longer represents all patients with newly-progressing mCRPC. However, real-world data demonstrate that although available only 3% to 46% of newly-diagnosed mHSPC patients in North America receive ADT with ARPIs despite reimbursement of therapy (8, 49–51), with even more limited use in other countries due to constraints from regulatory bodies, cost barriers, or disparities in access to care. As such, we believe that our results remain immediately relevant for a large subset of mCRPC. More importantly, it is plausible that early ctDNA% measurements can predict subsequent treatment response in the context of any line or class of systemic therapy for mCRPC. For example, Sumanasuriya and colleagues (40) reported that docetaxel or cabazitaxel responsive mCRPC had a median on-treatment ctDNA% of zero, whereas nonresponding disease was associated with detected levels of ctDNA (measured by low-pass whole-genome sequencing). Similarly, in a retrospective analysis of patients with mCRPC, Tan and colleagues (52) showed a link between undetected on-treatment ctDNA% (after three or four cycles of chemotherapy) and a favorable 3-month PSA response. Finally, Goodall and colleagues (53) showed that a strong reduction in cfDNA (used as an indirect measurement of ctDNA%) after 8 weeks of PARP inhibitor treatment was associated with improved radiographic PFS and OS. Potential differences in depth and dynamics of on-treatment ctDNA% changes in the context of distinct therapy types should be further explored in future studies. These studies should also include additional blood collections (e.g., at 2 months), to determine if there is utility in repeat on-treatment testing.

In conclusion, detection of ctDNA at baseline and 4 weeks after treatment initiation is strongly linked to a nondurable response to first-line ARPI and a shorter overall survival in patients with mCRPC. Early ctDNA% measurements may have utility for informing early therapy changes or intensification in patients unlikely to experience durable treatment responses.

E.M. Kwan reports personal fees from Astellas Pharma and Janssen and other support from Pfizer and Roche outside the submitted work. P. Hamberg reports personal fees from Astellas outside the submitted work. D.M. Somford reports advisory board/consultancy participation with Bayer, Astellas, Janssen, and MSD; research grants from Astellas and Besins, and sponsored research from Bayer, Astellas, Janssen, and Eli Lilly. I.M. van Oort reports grants and personal fees from Astellas and Janssen during the conduct of the study as well as grants and personal fees from Bayer and personal fees from MSD/AstraZeneca and AAA Novartis outside the submitted work. N. Mehra reports grants and personal fees from Astellas, AstraZeneca, MSD, and Pfizer; grants from Janssen-Cilag and BMS; and personal fees from Bayer outside the submitted work. N.P. van Erp reports grants from The Netherlands Organization for Health Research and Development (ZonMW) as part of the Goed Gebruik Geneesmiddelen (GGG) program, Janssen-Cilag BV, and Astellas during the conduct of the study as well as grants from Ipsen and Astellas outside the submitted work. A.W. Wyatt reports personal fees from Astellas, AstraZeneca, Bayer, EMD Serono, Janssen, Merck, and Pfizer and grants from ESSA Pharma outside the submitted work. No disclosures were reported by the other authors.

S.H. Tolmeijer: Conceptualization, data curation, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. E. Boerrigter: Conceptualization, data curation, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. T. Sumiyoshi: Data curation, formal analysis, investigation, methodology. E.M. Kwan: Data curation, investigation, visualization, writing–original draft, writing–review and editing. S. Ng: Data curation, software, formal analysis, visualization, methodology, writing–review and editing. M. Annala: Software, methodology, writing–review and editing. G. Donnellan: Investigation. C. Herberts: Data curation, software, writing–review and editing. G.E. Benoist: Resources, investigation, writing–review and editing. P. Hamberg: Investigation, project administration, writing–review and editing. D.M. Somford: Investigation, project administration, writing-review and editing. I.M. van Oort: Investigation, project administration, writing–review and editing. J.A. Schalken: Conceptualization, project administration, writing–review and editing. N. Mehra: Conceptualization, resources, supervision, methodology, project administration, writing–review and editing. N.P. van Erp: Conceptualization, resources, supervision, funding acquisition, methodology, project administration, writing–review and editing. A.W. Wyatt: Conceptualization, resources, supervision, funding acquisition, methodology, writing–original draft, project administration, writing–review and editing.

The observational studies and blood collections were part of the REFINE project (grant no. 836041013) that is funded by ZonMw, The Netherlands Organization for Health Research and Development, as part of the Goed Gebruik Geneesmiddelen (GGG) program, and supported by a research grant of Janssen Cilag BV and Astellas Pharma BV (to N.P. van Erp). The ctDNA sequencing and analysis was funded by a Canadian Cancer Society Challenge Grant (#707339 to A.W. Wyatt). E.M. Kwan was supported by a Prostate Cancer Foundation (PCF) Young Investigator Award and Killam Postdoctoral Fellowship. No funding sources were involved in the design or execution of the study. The authors are thankful to all contributing patients and their families.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.