Molecular reflex testing in patients with early metastatic castration-resistant prostate cancer within the PROMPT-study

Abstract

Background

Precision oncology using genotype-matched treatments (GMT) offers potential to improve survival in metastatic castration-resistant prostate cancer (mCRPC).

Patient and methods

In the PROMPT study (NCT04746300), reflex tumour testing through next-generation sequencing was performed in treatment-naïve or first-line mCRPC patients. All patients received a molecular tumour board (MTB) recommendation for GMT based on predefined druggable targets (DT). The main objective was to identify clinicopathological variables associated with DT.

Results

Analysis included 340 tissue samples from 307 patients, 51% of samples were newly biopsied. Valid results were generated in 84% (76% new, 92% archived; P < 0.01). DT were identified in 39% of the patients, with PI3K-AKT (26%) and Homologous Recombination (HR; 21%) pathways most frequently affected. Metastatic tissue, especially from mCRPC setting (P < 0.01), yielded higher GMT recommendations than primary tissue (P = 0.03). HR-associated genes were linked with shorter ADT-to-CRPC time (OR 3.77, 95%CI 1.62–10.32, P < 0.01). PI3K-AKT alterations were associated with metachronous metastatic disease (OR 0.48, 95%CI 0.26–0.89, P = 0.021) and longer time to CRPC (OR 0.47, 95%CI 0.25–0.87, P = 0.017). No distinct variables predicted DT genotypes.

Conclusion

Molecular tumour testing should preferably be done on metastatic mCRPC tissue. No combination of features could robustly identify druggable genotypes; therefore, reflex molecular characterisation should be routine for every mCRPC patient.

Introduction

Over the past decade, four life-prolonging therapeutic agents have been introduced into the routine care for patients with metastatic castration-resistant prostate cancer (mCRPC) in the European Union [1–4]. Despite these advancements, the median overall survival for mCRPC remains limited to 31 months [5]. Precision oncology offers the potential to improve survival outcomes for this molecular heterogeneous disease by introducing genotype-matched treatments (GMT) into routine care and within trials [6–8].

Current GMT options in the European Union include the poly ADP-ribose polymerase inhibitor (PARPi) olaparib, approved by the European Medicine Agency for patients with BRCA alterations [3]. Additionally, in some countries, programmed death-(ligand) (PD(L))-1 inhibitors are available for patients with microsatellite instability (MSI) and/or a high tumour mutational burden (TMB) [9]. The potential of precision medicine is further emphasised by basket, umbrella and platform trials [10–13]. By exploring novel uses for existing drugs, basket trials such as the DRUP or MSK-IMPACT expand treatment options for patients when standard therapies are no longer viable. While challenges remain—such as the limited efficacy of targeting a single pathway in a cancer-agnostic setting and difficulties in enroling patients with rare alterations—these trials have nonetheless achieved notable successes [14, 15]. For instance, the DRUP study in the Netherlands has a clinical benefit rate of 33% and led to the pan-cancer approval of the PD-1 inhibitor nivolumab, representing a significant breakthrough in linking novel trial designs to reimbursement decisions in precision medicine [10, 15]. Other successful DRUP cohorts include targeting of HER2, BRCA1/2, and targeting of BRAF in rare cancers [16–18].

Genomic and genetic testing is recommended by the European Society for Medical Oncology and the National Comprehensive Cancer Network, to determine eligibility for GMT based on individual genomic alterations [19, 20]. Additionally, such testing plays a crucial role in identifying families at risk for cancer predisposition and guiding appropriate screening recommendations for carriers. Nevertheless, large-scale implementation of reflex (routine) molecular testing in clinical practice faces several barriers [21–24]. These include the availability and quality of archival tumour tissue and logistical challenges, including obtaining new biopsies. Biopsies are particularly challenging to obtain successfully in patients with bone-predominant disease following routine imaging modalities. A PSMA-PET/CT guided biopsy may be helpful in selecting the optimal biopsy site, and in achieving higher tumour fractions [25], but is not yet integrated in routine care. Beside challenges in acquiring material for genomic testing, regional access to a molecular tumour board (MTB) is often lacking. However, both are essential to provide optimal advice on GMTs and maximise access to dedicated clinical trials and precision medicine trials including basket trials. Additionally, financial coverage for broad next-generation sequencing (NGS) remains challenging. In the Netherlands, BRCA testing has been made eligible for reimbursement since early 2022, but it remains inaccessible to some patients. Centralised expertise in academic or comprehensive cancer centres with an MTB can support regional networks, facilitate patient referrals, and provide knowledge and advice.

Current guidelines or expert opinion do not provide consensus on the optimal timing for tumour NGS in metastatic prostate cancer (PCa) [26]. Previous research at our tertiary centre indicated that timely NGS and early identification of druggable driver mutations could lead to higher utilisation rates of precision medicine and improved survival outcomes [27]. This large retrospective study indicated that 47% of patients received a GMT advise, while only 62% of these patients initiated the recommended treatment [27]. The main reason for not following recommendations was deterioration in performance status, with many patients being already heavily pre-treated before advice was received.

To prospectively assess the impact of routine molecular characterisation in early mCRPC patients, the PROMPT-study was initiated. PROMPT conceptualises the regionalisation of molecular characterisation for mCRPC in (South)eastern Netherlands, utilising an established network of regional community/general hospitals. The study aims to identify druggable targets (DT) early in the mCRPC state to maximise the provision of GMT and improve outcomes for patients harbouring poor prognostic but predictive genomic alterations. Here, we report the findings of the real-world observational PROMPT study from the first patient cohort.

Patients and methods

Patient cohort

Eligible patients had confirmed mCRPC and were either therapy-naïve or had received a maximum of one line of systemic treatment in the mCRPC setting. Previous therapy up to six cycles of docetaxel or androgen receptor signalling inhibitors (ARSI) in the hormone-sensitive setting was allowed. Clinical trials combining routine treatment with an experimental drug or placebo were considered one line of treatment. All patients demonstrated adequate organ and bone marrow function and had an estimated life expectancy of at least one year. Full eligibility criteria are provided in the supplementary data.

Study design

PROMPT (NCT04746300) is an ongoing observational, non-investigational medical product study (Supplementary Fig. 1) [28]. The study is conducted at Radboudumc, Nijmegen, a tertiary referral hospital harbouring a multidisciplinary MTB specialising in PCa genomics and genetics. The Radboudumc provides expertise and accepts referrals predominantly from hospitals in the (South-)Eastern Netherlands [27]. Patients provided written informed consent. Standardised clinical and demographic data were collected at screening, pseudonymized and entered into the cloud-based system CASTOR (www.castoredc.com). PROMPT aims to recruit 600 patients.

NGS was performed using the TruSightOncology500 (Illumina) panel; prior to receiving the next line of therapy. A virtual PCa-specific diagnostic filter was applied, limiting analysis to 43 genes (Supplementary Table 1), which included common PCa prognostic genes and all current druggable genes, MSI-status and TMB. This filter was selected to focus on clinically actionable alterations relevant to advanced PCa and GMT implications. Valid molecular results met all pre- and post-analytical criteria, and required a minimal pathological tumour content (ratio of neoplastic cell nuclei to all nuclei, TC) or inferred TC assessed after NGS of 30%. This threshold is necessary to accurately assess biallelic gene inactivation. Specific methodology for TC estimation and TMB scoring can be found in the supplementary methods. In case of unsuitable material or an invalid result, a second sample was obtained. MTB discussions determined GMT recommendation based on DT defined by Precision Medicine Working group criteria and/or by Drug Rediscovery Protocol criteria at time of discussion [10, 20]; Supplementary Table 1. For all tumour suppressor genes, biallelic inactivation was required to define druggability, with exceptions for BRCA1/2; also mono-allelic mutations were considered eligible for PARPi, given the absence of a distinct relationship between zygosity status and/or outcome [29–31].

Formalin-fixed paraffin-embedded (FFPE) material was used from prostate or metastatic tissue biopsies, preferably from newly obtained samples. A predefined logistic workflow (depicted in Fig. 1) was implemented to standardise and optimise the generation of valid sequencing results. Bone biopsy site selection was focused on PSMA-positive lesions with substrate on (low-dose) CT scan, and favouring pelvis over the vertebrae and other axial sites [32].

Fig. 1. Overview of the PROMPT logistic workflow used to increase valid sequencing results.

After inclusion in PROMPT, depending on physician assessment, either archived material was analysed, or patients were discussed in a multidisciplinary team meeting to select the most appropriate biopsy location depending on the availability of radiological images. Archived material was more often chosen in case of aggressive disease, a time between the start of androgen deprivation therapy and castration-resistant disease below 24 months or no suitable biopsy location. In case of invalid NGS results, a second attempt to acquire fresh or archive material could be made.

Patients were categorised as ‘DT’ or ‘non-DT’ based on the potential druggability of the identified aberrations. The DT cohort was subdivided into categories based on a pre-defined hierarchy of efficacy: (1) mismatch repair deficient (MMRd)/MSI or high TMB (hTMB) with a non-synonymous (ns)TMB of ≥15 mut/Mb, (2) alterations in homologous recombination repair-associated genes (HRR-genes), (3) alterations in the PI3K-AKT pathway, or (4) other (the remaining) DT. Within the timeframe of this study, AR pathway alterations were not druggable with novel AR degraders or third-generation ARSIs, and therefore not considered a DT within these analyses. All patients with increased risk of cancer predisposition were referred for genetic counselling according to the Dutch Consensus criteria [33]. At inclusion in the PROMPT study, patients were not offered routine germline analysis.

Objectives

The primary objective of this current analysis is to identify correlates associated with DT in MMRd/MSI/hTMB, HRR-genes and PI3K-AKT pathway. Secondary objectives include i. describing the genomic landscape of first- or second-line mCRPC patients in a real-world setting, ii. assessing the relevance of tissue source used for NGS regarding hormone-sensitive or refractory status, iii. evaluating success rates of a regionalised, standardised molecular characterisation workflow.

Statistical analysis

Descriptive statistics were used to characterise the study population. Subgroups were compared using Chi-square test for categorical variables and the nonparametric Mann-Whitney U or Kruskal Wallis, for continuous variables. To evaluate the utility of clinical correlates in predicting the presence of DT, we implemented multivariable analysis (MVA) using backward selection. Variables with a P-value < 0.20 were considered predictive in this analysis. A P-value < 0.05 was considered significant in all other analyses. Statistical tests were performed, and data were visualised using RStudio (2023.12.1 + 402). Code availability: The R code used is available upon request.

Results

Patient characteristics

From February 2020 until December 2021, a total of 321 patients were included. Fourteen patients were subsequently excluded as illustrated in the trial CONSORT diagram (Supplementary Fig. 1). Baseline characteristics for the remaining 307 patients are detailed in Table 1. At initial diagnosis, 64% of patients had a minimum Grade Group of 4, and 61% (n = 187) presented with de novo metastatic (synchronous) disease, with 62% of these cases presenting with high-volume disease. In metastatic hormone-sensitive PCa (mHSPC) setting, 46% of patients (n = 140) received intensification of androgen-deprivation therapy; 89% received additional docetaxel and 11% an ARSI, including three patients who received enzalutamide ± pembrolizumab (NCT04191096). The median age was 70 years (interquartile range [IQR]: 65–74). In the CRPC setting, 55% of the patients were treatment-naive, whereas 45% had received one prior line of therapy, predominantly an ARSI.

Table 1.

Baseline characteristics.

		Total 1		DT		Non-DT
	n	Median [IQR] No. of pts (valid%)	n	Median [IQR] No. of pts (valid%)	n	Median [IQR] No. of pts (valid%)
At initial diagnosis
Age, yr.	307	65 [59–70]	115	64 [59–70]	192	65 [60–69]
iPSA, ng/mL	299	59 [14–276]	110	61 [12–231]	179	59 [16–300]
PGG
PGG ≤ 3	296	107 (36)	35	35 (31)	184	72 (38)
PGG 4-5	296	189 [64]	77	77 [69]	184	112 [58]
T-stage
<T3	219	52 (24)	87	17 (15)	132	35 (27)
T3-T4	219	167 [76]	87	70 [61]	132	97 [73]
N-stage
N0	243	96 (40)	98	42 (43)	145	54 (37)
N1	243	147 [60]	98	56 [57]	145	91 [63]
M-stage
M0	307	120 (39)	115	49 (43)	192	71 (37)
M1a	307	19 (6.2)	115	6 (5.2)	192	13 (6.8)
M1b	307	87 (28)	115	34 (30)	192	53 (28)
M1c	307	81 (26)	115	26 (23)	192	55 (29)
Volume status2
Low	151	57 (38)	55	29 (53)	96	32 (33)
High	151	94 (62)	55	26 (47)	96	64 (67)
At inclusion
Age, yr.	307	70 [65–74]	115	70 [65–75]	192	71 [65–74]
PSA, ng/mL	285	25 [8.0–77]	105	32 [6.4–100]	180	24 [9.3–51]
Time from ADT to CRPC
Median, mo.	307	20 [12–35]	115	21 [12–34]	192	20 [12–36]
<24 mo.	307	781 (59)	115	71 (62)	192	110 (57)
Previous CRPC-treatment
None	307	169 (55)	115	53 (46)	192	116 (60)
ARSI	307	107 (35)	115	45 (39)	192	62 (32)
Docetaxel	307	17 (5.5)	115	7 (6.1)	192	10 (5.2)
Cabazitaxel	307	2 (0.7)	115	1 (0.9)	192	1 (0.5)
Other	307	12 (3.9)	115	9 (7.8)	192	3 (1.5)
Bone disease
With bone involvement	307	247 (81)	115	92 (80)	192	155 (81)
Bone only disease	307	118 [38]	115	36 [31]	192	82 [43]
Visceral disease
With liver involvement	307	3 (1.0)	115	2 (1.7)	192	1 (0.5)
Non-liver visceral involvement	307	27 (8.8)	115	10 (8.7)	192	17 (8.9)
No visceral disease	307	277 (90)	115	103 (90)	192	174 (90)
Family history3
Familial prostate cancer	305	13 (4.3)	115	4 (3.5)	190	9 (4.7)
Breast cancer	305	31 [10]	115	10 (8.7)	190	21 [11]
Ovarian cancer	305	5 (1.6)	115	3 (2.6)	190	2 (1.1)
Pancreatic cancer	305	18 (5.9)	115	9 (7.8)	190	9 (4.7)

1: The table includes all included participants in which NGS results were generated independent of tumour content, this includes three patients in the GMT group and thirteen in the non-GMT group without a valid NGS. The five patients where NGS analysis was not possible are included in the non-GMT group, since no GMT advice could have been given.

2: Volume status for M1b and M1c assessed according to CHAARTED criteria: ≥ 4 bone metastasis, including ≥ 1 outside vertebral column or pelvis and/or visceral metastasis [51].

3: According to the Dutch Consensus criteria [52].

ADT: androgen deprivation treatment; GMT: genetically matched treatment; iPSA: initial PSA at diagnosis; mo.: months; ARSI: androgen receptor signalling inhibitor, i.e., abiraterone acetate or enzalutamide; other CRPC treatment: either radium-223, combination therapy or therapy within clinical trials; PGG: prognostic grade group; yr.: years.

Percentages below 10% are specified with one decimal place.

Based on the reported family history (FH), 4.6% (n = 14) of patients met the Dutch consensus criteria for familial PCa. Due to presence of specific cancers and/or (early) onset in a first- or second-degree relative, referrals for assessment were warranted for hereditary breast and ovarian cancer in 16% (n = 48) of cases, due to the occurrence of breast cancer in 9.8% (n = 30), ovarian cancer in 1.6% (n = 5), and pancreatic cancer in 5.9% (n = 18) in relatives. No significant differences in FH were observed between the ‘DT’ and ‘non-DT’ groups. Based on aberrant cancer predisposition gene alterations detected in tumour tissue, 22 patients were referred for genetic counselling [34].

Regionalisation within the PROMPT-network

PROMPT was initiated to facilitate molecular testing in (South-)Eastern Netherlands, though referrals were accepted from all Dutch hospitals. Ultimately, 240 patients were referred from 54 distinct hospitals, including two academic hospitals and one comprehensive cancer centre (Supplementary Fig. 3).

Tumour tissue testing

A total of 340 tumour tissue samples from 307 patients were acquired for analysis through the predefined logistic workflow, of which 172 samples (51%) were new biopsies in a castration-resistant setting. The samples were predominantly from prostate (n = 182, 54%), followed by bone (n = 80, 24%) and lymph nodes (n = 69, 20%; Supplementary Fig. 4a, b). Visceral metastases were uncommon in this early mCRPC setting and were infrequently biopsied, with only six samples collected. Archived biopsies were primarily obtained during the hormone-sensitive stage (n = 129, 77%). Dominant biopsy site for archival material was the prostate, whereas newly obtained samples were predominantly from bone.

The median TC was 60% (range: 20–95%; Supplementary Fig. 4c) for the 305 samples subjected to NGS. Among the three most common sites of tumour tissue (each with >50 samples), the highest median TC was observed in lymph nodes (70%, IQR: 45–80, n = 63, P = 0.007), followed by prostate (60%, IQR: 40–70, n = 172), and bone (55%, IQR: 40–70, n = 62). Overall, the median TC did not differ between archived and newly obtained tissue (P = 0.5, MWU). No significant differences in TC within tissue sites were observed between archived and newly obtained samples, or when comparing localised, mHSPC or mCRPC settings (Supplementary Fig. 5a, b).

Towards valid molecular characterisation

Tumour NGS yielded valid results for 285 patients (93%), adhering to the TC threshold and quality control criteria defined per study criteria. Figure 2 illustrates the proportion of samples failing quality control across the different analytical steps. The proportion of samples meeting pathology criteria and proceeding to DNA isolation was higher for archival tissue (n = 162, 96%) compared to newly obtained (n = 146, 85%) samples, P < 0.01. Despite several attempts, no suitable tissue meeting pathology standards was obtained for five patients (visualised in Supplementary Fig. 2 as NGS not possible). For 305 samples, library preparation and sequencing were initiated. With hindsight, NGS data for twenty tissue samples were excluded from this evaluation because of an (inferred) TC lower than 30%. For 92% (n = 155) archival and 76% (n = 130, P < 0.01) newly obtained material, NGS data were used for further evaluation.

Fig. 2. Flowchart representing the process from collecting tissue samples to NGS results.

Upper row represents the newly obtained tissue samples collected through new biopsies. The bottom row represents the archived tissue samples formerly obtained during the disease. Tumour tissue contains all the samples received for assessment. Formalin-fixed paraffin-embedded (FFPE) specimen insufficient includes those samples in which tumour tissue is not present at macroscopic evaluation. At the pathological assessment, material was judged microscopically by a specialised pathologist. No tumour present contains all samples without (previous) evidence of containing vital tumour cells. Next DNA was isolated, before next-generation sequencing was performed. The last numbers represent the tissue samples with a minimal tumour content of 30% as required and were considered valid.

In 85% (261/307) of the patients, the first tissue assessed led to valid results, 74% (23/31) of replacement samples yielded valid results, and 50% (1/2) of third attempts. Tissue samples collected from visceral metastases had the highest proportion of valid NGS results (5/5, 100%), while tissue from bone metastasis had the lowest (54/80, 67.5%; Supplementary Fig. 6a). A TC < 30% was the main reason for invalid NGS results (20/31, 48,8%; Supplementary Fig. 6B). In more than 50% (24/41) of the cases that failed pre- or post-analytical quality control, the tissue source was bone.

Of the 69 newly performed bone biopsies, site selection was PSMA-PET/CT guided in 50 patients (72%). PSMA-PET-based site selection did not result in higher TC compared to non-PET-based selection of biopsy sites (P = 0.8), nor did it lead to significantly higher proportions of valid NGS results (66% [33/50] versus 53% [10/19] for PET-based versus non-PET-based, respectively, P = 0.5).

Molecular landscape and precision medicine allocation

Of the 285 patients with valid NGS results, 219 (77%) had at least one (likely) pathogenic aberration in the 43 genes prospectively assessed. An overview oncoplot is visualised in Fig. 3, with further details found in Supplementary Figures 7+8. Mutations in TP53 (31%) were observed most frequently, followed by aberrations in AR (27%).

Fig. 3. Oncoplot showing an overview of the molecular landscape of somatic aberrations (n = 219).

The oncoplot is divided into three categories based on the stage in which the tested tissue was originally obtained. Rows show the affected genes sorted per pathway; columns represent affected individuals. The frequency of gene alterations is shown on the right side. The upper histogram shows non-synonymous mutations per mb (nsTMB), the colour of the bar represents the microsatellite status. The type of aberration is shown in the legend below. Mono-allelic losses, without a second hit, were omitted from the oncoplot except for BRCA1/2. Cases with more than one aberration in a gene are represented by multi-hit. In case of loss of the wild-type allele this is indicated with a dot. The bottom row shows information about the tissue samples and tumour stage wherein the aberration was found.

MTB-based GMT recommendations, either in routine care or within a dedicated clinical trial active at time of inclusion, were strictly defined (Supplementary Table. 1). Advice for GMT could be provided for 112 patients (39%). DT were more commonly detected in metastatic tissue compared to primary prostate tissue (47% versus 34%, P = 0.03), and mCRPC compared to mHSPC or localised setting (48% versus 28% and 38%, respectively; P < 0.01, Fig. 3). DT in all metastatic samples were distributed evenly by tissue origin: 46% (25/54) in bone, 48% (27/56) in lymph nodes, and 44% (4/9) in other tissue samples (P = 0.9). In contrast to tissue of metastatic origin, DT within primary prostate tissue did not significantly differ across stages: 45% (17/38) of the mCRPC compared to 29% (32/108) for mHSPC and 29% (5/17) for localised samples; P = 0.2.

Pathways most frequently affected were PI3K-AKT in 26% (n = 74) and homologous recombination (HR) in 21% (n = 60) of the patients. The PI3K-AKT pathway was more commonly affected in mCRPC compared to mHSPC tissue (33% versus 16%, P < 0.01). Similar results within the PI3K-AKT pathway were found when comparing NGS-results from tissue obtained from mCRPC or mHSPC setting, within the subgroup of patients presenting with synchronous metastatic disease at initial diagnosis (28% versus 8.9%, P = 0.03, Supplementary Fig. 7). Within samples obtained in mCRPC setting, no significant differences were found when comparing tissue from patients with synchronous disease to those with metachronous metastatic disease (28% versus 37%. P = 0.34, Supplementary Fig. 8). No significant differences in presence of HRR-gene pathogenic variants were seen between disease states (21% for both mHSPC and mCRPC tissue samples, P = 1). Additionally, the prevalence was similar for samples collected in the CRPC setting, regardless of whether the initial diagnosis was localised or synchronous metastatic cancer. Likewise, prevalence of HR aberrations was similar between samples tested from synchronous mHSPC with mCRPC (Supplementary Fig. 7). HRR-gene alterations were found in 19% of the metastatic tissue samples and 12% of prostate tissue samples (P = 0.1). Limiting the analysis to DT in BRCA1 and/or BRCA2 (n = 22) revealed similar distributions across tissue types (P = 0.8) and stages (P = 0.5). Bi-allelic inactivation of PTEN (n = 47; 16%), any alteration of BRCA2 (n = 21; 7.3%), and bi-allelic inactivation of ATM (n = 11; 3.6%) were the most frequently altered, druggable genes, leading to GMT recommendations.

Median nsTMB across all samples was 2.4 mut/Mb (IQR: 1.6–3.9). Stratified per stage, the median nsTMB was 1.6 (IQR: 0.8–2.6) for localised, 2.4 (IQR: 1.6–3.2) for mHSPC, and 2.4 (IQR: 1.6–3.9) for mCRPC samples, respectively. In fifteen cases (4.9%) the minimum nsTMB threshold of 15 mut/Mb was met, in 80% this co-occurred with MSI. Median nsTMB in the MMR/hTMB subgroup was 34 (IQR: 19–45). MSI was detected in 5/120 mHSPC samples and in 7/145 mCRPC samples (P = 0.8).

Clinicopathological predictors of GMT

Three predefined druggable pathways within the ‘DT’ group were assessed for potential predictors using clinicopathological data, with each pathway compared individually to the remaining study population. Baseline characteristics for each subgroup are detailed in Supplementary Table. 2.

MMRd/hTMB

Patients harbouring an MMRd/hTMB (n = 16) genotype exhibited a trend towards having a higher Prostate Grade Group (PGG) at initial diagnosis ( ≥ 4; OR 3.62, 95%CI 0.97–23.47, P = 0.10) and were less likely to have bone-predominant disease (OR 0.38, 95%CI 0.10–1.15, P = 0.11). The only variable that was significantly associated with a MMRd/hTMB genotype was a PSA level above the median ($24$ $\mu g/L$), with almost four-fold higher odds compared to the remaining study population (OR 3.90, 95%CI 1.21–17.44, P = 0.038), depicted in Fig. 4A. MVA displayed in Supplementary Table. 3A indicated that a combination of PGG ≥ 4, age at CRPC > 70 years, and PSA levels above the PROMPT population median were independent predictors of MMRd/hTMB.

Fig. 4. Forest plot illustrating clinicopathological variables analysed per molecular subgroup against the remaining study population.

a Visualises the subgroup with mismatch repair deficiency (MMRd) or high tumour mutational burden; b Visualises the subgroup of patients with alterations in homologous recombination repair (HRR) associated genes; c Visualised the subgroup with an affected PI3K-AKT pathway. Events represents the percentage of patients in the relevant subgroup to meet the clinicopathological condition. Horizontal lines represent 95% confidence intervals (CIs) of individual clinicopathological conditions, while the black box gives representation of the corresponding odds ratio (OR). ALP Alkaline Phosphatase, CRPC castration-resistant prostate cancer, LDH lactate dehydrogenase, M + : metastatic disease present, mo months, NLR neutrophil-lymphocyte-ratio, NSE neuro-specific enolase, PGG prostate grading group, PSA prostate specific antigen.

HRR-genes

Patients with HRR-gene aberrations (n = 36) had a higher average presence of visceral disease (OR 2.49, 95%CI 0.92–6.11 P = 0.056) and a shorter time to CRPC development (start of ADT to CRPC < 24 months; OR 3.77, 95%CI 1.62–10.32, P < 0.01). No other clinicopathological predictors were identified. Figure 4B shows the results of the UVA. Only shorter time on ADT was significant and independently associated with HRR-genes in MVA.

PI3K-AKT

Patients harbouring an aberration in the PI3K-AKT (n = 49) pathway were twice as likely to have developed metachronous metastatic disease (absence of M+ de novo: OR 0.48, 95%CI 0.26–0.89, P = 0.021) and have a time from start of ADT to CRPC exceeding 24 months (OR 0.47, 95%CI 0.25–0.87, P = 0.017; Fig. 4C). Furthermore, non-elevated alkaline phosphatase (ALP) levels at inclusion were associated with higher odds for this subgroup (ALP > upper limit of normal: OR 0.42, 95% CI 0.20–0.82, P = 0.013). A multivariate model (Supplementary Table 3B) showed that a combination of time to CRPC > 24 months, age <69 years, absence of visceral disease, normal ALP levels, and an elevated neuro-specific enolase level were predictive for a DT within the PI3K-AKT pathway.

Discussion

In this initial report from the PROMPT-study, we describe our real-world cohort of 307 mCRPC patients receiving reflex molecular tumour testing at a tertiary cancer referral centre with an integrated MTB and streamlined logistic workflow for metastatic PCa. To our knowledge, this study is among the largest prospective studies reporting outcomes of routine tumour testing in mCRPC patients outside of precision medicine clinical trials.

Optimisation of molecular tumour testing and centralisation of biopsy procedures through a defined multidisciplinary discussed PROMPT logistic workflow led to valid NGS results in 84% of the tissue samples. Among the three most common biopsy sites, lymph node material had the highest median TC and the highest proportion of valid results. Bone tissue accounted for half of the failed procedures. Surprisingly, defining the optimal site of bone biopsy based on prior knowledge from a recent PSMA-PET/CT imaging, which was performed for 72% of bone biopsies in the logistic workflow, did not result in higher proportions of valid results in our centre [25, 32]. Adding extra parameters as maximum Standard Uptake Value or Hounsfield scales could possibly lead to higher results [32], and will be retrospectively assessed. Compared to large clinical trials that have reported their NGS success rates, the overall success rate of 84% in our study is relatively high, despite our strict criteria, including the requirement for a minimum TC of ≥30%. In the PROfound study, a successful NGS result was reported for 69% of the patients [35], with a minimum TC threshold set at 20%. Results were more frequently generated from newly obtained than from archived material (64% versus 57%) and from metastatic versus primary tissue (64% versus 56%). However, only 17% of the tested samples were from metastatic tissue, compared to about 50% in our study. Similar to our study, the lowest success rates were observed in bone tissue, with 43% in the PROfound study compared to 68% in ours. Both PROpel and TALAPRO-2 reported higher rates of successfully reporting a molecular result, 98% and 86% respectively [36, 37]. A key difference is that both studies utilised tumour tissue testing in conjunction with liquid biopsy to assess HRR-gene aberrations. Limiting results to tumour tissue, an HRR-gene status within PROpel could be established for 67% of patients. No additional details regarding the TC threshold, the success rate per tissue type, or disease stage have been disclosed.

A potentially DT was identified in nearly 40% of patients with a valid NGS result, similar to previously reported druggable gene frequencies in the Netherlands [6], creating significant opportunities for precision medicine. In line with literature [9, 23], we observed that the MSI/hTMB genotype is relatively uncommon, affecting 4.9% of patients. The frequency of alterations in HRR-genes was 21% in our study, compared to 28% reported in the PROfound and PROpel studies [3, 38]. In contrast to our study, both studies defined a single allele alteration in non-BRCA HRR genes as a possible DT. More similar to our study, TALAPRO-2 reported HRR-gene alterations in 23% of patients and used comparable second-hit requirements [37]. No significant differences in tissue origin or stage were observed for the determinant of HRR-gene status, stressing the potential of using archival material to test for PARPi eligibility. The PI3K-AKT pathway was, with 26% the most affected pathway. We identified key differences in prevalence based on tissue source, with more alterations identified later in the disease course. Patients have been referred to the DRUP trial pan-cancer cohort with alpelisib, which is currently being analysed. Allocation to other precision medicine trials for these patients has been limited by the lack of access to pan-PI3K/mTOR and dual-selective PI3K inhibitors with acceptable toxicity profiles, as well as by the previously limited objective responses observed in monotherapy [39]. The Phase 3 IPATential150 trial investigating ipatasertib in combination with abiraterone did not demonstrate an overall survival benefit based on predefined biomarkers. In contrast, the Phase 2 PROCAID study evaluating capivasertib with docetaxel did show a survival benefit, which is currently being further investigated in the ongoing Phase 3 CAPItello-280 trial [40, 41]. For alterations in genes of the PI3K-AKT pathway, it has been established that PI3K-AKT alterations are associated with resistance to androgen deprivation and AR targeting agents [42, 43]. We anticipate that lessons from IPATential150—particularly the need for improved biomarker-driven patient selection—will inform future strategies. Strategies include combining AKT inhibitors, pan-PI3K/mTOR, or dual-selective PI3K inhibitors with ARSIs or docetaxel. Predictive biomarker identification will be key to overcoming previous challenges in effectively targeting the PI3K/AKT/mTOR pathway [44, 45].

We also aimed to assess whether clinicopathological variables, or baseline disease characteristics at castration-resistance, could help in distinguishing specific druggable genotypes. In patients with HRR-gene aberrations, we observed a higher-than-average prevalence of visceral disease and a shorter time to the development of CRPC, suggesting a more aggressive disease course. For patients with alterations in the PI3K-AKT pathway, less aggressive clinicopathological characteristics are associated its presence. Specifically, the development of metachronous metastatic disease and a longer time from ADT to CRPC were significantly associated with this genotype, which indicates distinct biological behaviour. However, the wide confidence intervals of clinicopathological variables in predicting certain genotypes, underscores considerable variability and affirms the molecular heterogeneity and differences in clinical presentation of PCa. Furthermore, the robustness may also be influenced by the small sample size per druggable genotype. Larger datasets are needed to validate these findings and observe more putative clinicopathological predictors. These results could be helpful in determining which patients should be tested.

PROMPT was initiated to prospectively assess the impact of routine molecular testing on treatment outcomes and GMT allocation. The study was preceded by our findings from retrospective research [27]. Comparing both studies showed that reflex testing of all patients in the early mCRPC setting, as opposed to selective testing of patients lacking standard of care treatment options, identified 39% versus 47% patients eligible for GMT allocation. Besides patient selection, the use of targeted sequencing in this study, compared to more inclusive whole-genome sequencing in the retrospective study, or increased precision medicine trial options in the retrospective study might explain in part the differences. Future publications will focus on the rate of GMT allocation for those with a DT characterised under the PROMPT protocol and should be compared to real-world data on nationwide allocation of precision medicine. We anticipate that reflex molecular characterisation at the onset of CRPC will lead to higher rates of precision medicine allocation, possibly resulting in improved overall survival. Particularly, for the 10% of patients with identified MMRd/hTMB or BRCA alterations, knowledge of the molecular status should lead to a more optimal treatment planning, earlier access to GMT, and a better baseline performance status for these more aggressive phenotypes [46]. One of the first impacts of this approach can be seen in the successful INSPIRE trial [47], for which 67 biomarker-positive PROMPT patients were screened. After receiving dual immunotherapy, an objective response rate of 75% was reached for the 33 patients with MSI, combined with a median progression-free survival of 32.7 months and a median overall survival not yet reached. Future research should look at precision medicine options beyond those researched within PROMPT. An overview of clinically relevant pathways with relevant clinical trials is stated in Supplementary Table. 4.

An important future consideration in the implementation of GMT in mCRPC is the optimal strategy for identifying DTs. Solid tumour biopsies remain in the Netherlands the current standard for genomic profiling; however, in mCRPC, outside of tertiary centres these are often logistically challenging to perform. Liquid biopsies, particularly circulating(ct-)DNA profiling, represent a promising alternative or complementary approach [48, 49]. With continued advances in cell-free (cf)DNA-based technologies, including broader genomic coverage and improved sensitivity for copy number variation and structural alterations, ctDNA may increasingly enable reliable DT detection in routine practice. Ongoing translational efforts aim to compare DT detection across solid tumour tissue, cfDNA, and combined hybrid approaches to better define concordance and clinical utility [50]. Such studies are particularly relevant for extending the benefits of GMT beyond academic centres, into broader community practice.

There are several limitations of our study that need to be considered. First, there was only a limited number of patients per genomic subgroup. In this heterogeneous patient population, this will lead to significant variability, making it challenging to identify distinct and clinically useful predictive characteristics. Additionally, GMT advice provided by the MTB depended on routine care, precision medicine options and dedicated clinical trials and availability of specific cohorts within the basket trials open for inclusion at time of discussion. This not only meant that the potential for druggability varied under the PROMPT protocol, but also that, based on the obtained knowledge, stricter selection criteria could be implemented for precision medicine allocation. This might involve adjusting tumour types or genes within a selected druggable pathway, leading to a more stringent identification process. While we believe it is crucial to optimise precision treatment strategies in mCRPC through the exploration of precision medicine allocation, for instance via basket trials, it must be acknowledged that lessons from the past have shown that exceptional responses are rare, and overall clinical benefit is witnessed only in one in three patients. Therefore, allocation to genotype-matched therapies (GMT) with experimental agents should be approached with caution and carefully balanced when effective standard treatments remain.

Conclusion

This study analysed data on molecular tumour diagnostics of the first patient cohort included in the PROMPT study. In our cohort, routine NGS and MTB-discussion led to GMT advice in almost 40% of the patients. Overall, our data suggests that molecular tumour testing should preferably be done on metastatic tissue in an mCRPC setting to optimise the chance of finding a DT. Although clinicopathological variables were enriched in specific druggable genotypes, no combination of variables could robustly predict presence versus absence of druggable genotypes; therefore, wherever possible, reflex molecular characterisation should be routine for every mCRPC patient.

Author contributions

Conception and design, Iris S.H. Kloots, Minke Smits, Winald R. Gerritsen, Niven Mehra. Acquisition of data Iris S.H. Kloots, Peter H. J. Slootbeek, Sofie H. Tolmeijer^,, Minke Smits, Sjoerd van Helvert, Leonie I. Kroeze, Jolique A. van Ipenburg, Katrien Grünberg, C. Marleen Kets, Marjolijn J.L. Ligtenberg, James Nagarajah, Joyce M. van Dodewaard, Theo van Voorthuizen, Frederiek Terheggen, Tineke Smilde, Inge M. van Oort, Haiko J. Bloemendal, Winald R. Gerritsen, Niven Mehra. Analysis and interpretation of data Iris S.H. Kloots, Peter H.J. Slootbeek, Sofie H. Tolmeijer, Sandra van Wilpe, Sjoerd van Helvert, Leonie I. Kroeze, Jolique A. van Ipenburg, Katrien Grünberg, C. Marleen Kets, Marjolijn J.L. Ligtenberg, Niven Mehra. Drafting of the manuscript Iris S.H. Kloots, Niven Mehra. Critical revision of the manuscript for important intellectual content. All authors. Statistical analysis. Iris S.H. Kloots. Obtaining funding Winald R. Gerritsen, Niven Mehra, Minke Smits, Iris S.H. Kloots. Administrative, technical, or material support. Peter H.J. Slootbeek, Sofie H. Tolmeijer, Sandra van Wilpe. Supervision Jack A. Schalken, Inge M. van Oort, Haiko Bloemendal, Winald R. Gerritsen, Niven Mehra. Other (specify) The authors read and approved the final manuscript.

Funding

The PROMPT-study is in part supported by a ‘Betaalbaar Beter’ grant, funded by a health-insurance company (VGZ) and supported by the Radboudumc. Additional funding was obtained through the Paul Speth Foundation (no Grant number) and the Radboud Oncologie Fonds; Grant number: ROF1907. The funding bodies had no role in the design of this study nor in the collection, analysis, and interpretation of the data, and writing of this manuscript.

Data availability

Data can be requested through the corresponding author.

Competing interests

MJL received consulting fees from GlaxoSmithKline B.V., Janssen-Cilag B.V., AstraZeneca; received honoraria for lectures: AstraZeneca, Janssen, Roche, MSD, Pfizer. All payments to Radboudumc. WRG is a member of the advisory boards of Astellas, Coretag holding AG, Merck Sharp and Dohme, and Novartis; received personal fees for coaching activities by Bayer; and is member of the board of directors of ORCA therapeutic B.V. NM received consultancy fees from Astellas Pharma, AstraZeneca, Bayer, Janssen-Cilag, Johnson & Johnson, Merck Sharp and Dohme, and Pfizer; and research funding (all paid to the institution) from Astellas Pharma, AstraZeneca/Merck, Bristol Myers Squibb Foundation, and Janssen-Cilag. All other authors declare that they have no competing interests.

Ethics approval and consent to participate

The trial was conducted in accordance with the principles of the Declaration of Helsinki and classified by the medical ethical committee of “Commissie Mensgebonden Onderzoek” (CMO) regio Arnhem-Nijmegen. No ethics approval was needed for this study. All patients provided written informed consent. The study is registered in ClinicalTrials.gov under the number NCT04746300.

Consent for publication

Not applicable.

Supplementary information

The online version contains supplementary material available at 10.1038/s41416-025-03243-7.