Tumor Genomic Testing for >4,000 Men with Metastatic Castration-resistant Prostate Cancer in the Phase III Trial PROfound (Olaparib)

Metastatic castration-resistant prostate cancer (mCRPC) is a molecularly heterogeneous disease, with 20% to 30% of patients harboring deleterious alterations in DNA damage repair genes, including those with direct or indirect roles in homologous recombination repair (HRR; refs. 1–4). The PROfound study (NCT02987543) is the first phase III, randomized, multicenter trial to show that a PARP inhibitor, olaparib, improves radiographic progression-free survival and overall survival compared with control [physician's choice of next-generation hormonal agent (NHA) enzalutamide or abiraterone] in patients with mCRPC with alterations in genes involved directly or indirectly in HRR and disease progression on prior NHA (1, 5). It is the largest PARP inhibitor study to date (6) to conduct central, prospective tissue next-generation sequencing (NGS) to screen patients to determine eligibility for enrollment. Patients were required to have a qualifying alteration in one or more of the 15 prespecified genes for enrollment into one of two trial cohorts (cohort A: BRCA1, BRCA2, ATM; cohort B: BRIP1, BARD1, CDK12, CHEK1, CHEK2, FANCL, PALB2, PPP2R2A, RAD51B, RAD51C, RAD51D, RAD54L). On the basis of the findings of the PROfound study, the FDA approved olaparib for men with deleterious or suspected deleterious germline or somatic HRR gene-altered mCRPC (PPP2R2A not included) and disease progression following prior NHA. In addition, other agencies, including the European Medicines Agency (EMA), the Japanese Pharmaceutical and Medical Device Agency, and the Australian Therapeutic Goods Agency, have approved olaparib for patients with BRCA1 or BRCA2 alterations, and Health Canada has approved olaparib for patients with BRCA1, BRCA2, or ATM alterations (7, 8).

Identifying patients with mCRPC who may benefit from olaparib through molecular diagnostics can improve outcomes for patients with a poor prognosis. Guidelines from the National Comprehensive Cancer Network (NCCN), European Society for Medical Oncology (ESMO), and American Urological Association/American Society for Radiation Oncology/Society of Urologic Oncology (AUA/ASTRO/SUO), as well as other societies, have recently been updated to include recommendations of genomic tumor testing for HRR gene alterations in patients with prostate cancer (9–11). However, although tumor testing is the gold standard for molecular testing, ensuring it is routinely used remains a challenge. Identifying factors that may increase the likelihood of an NGS result being obtained is critical in encouraging routine uptake of tumor tissue testing in clinical practice.

Here, we report the results of molecular diagnostic testing of tumor tissue samples provided by more than 4,000 patients with mCRPC who were screened for the PROfound study. We discuss the real-world challenges to testing and aim to provide insights into how to maximize generation of NGS results based on our findings.

The PROfound study (NCT02987543) is a phase III, prospective, randomized, open-label, multicenter, global trial. The study comprises two cohorts: cohort A includes patients with alterations in BRCA1, BRCA2, or ATM (assigned regardless of any co-occurring alteration in other genes), and cohort B includes patients with alterations in 12 other HRR genes. Additional details of the study design can be found in the primary publication (1).

This trial was performed in accordance with the principles of the Declaration of Helsinki, Good Clinical Practice guidelines, and the AstraZeneca policies on bioethics and was approved by an Institutional Review Board or ethics committee in the investigational sites. All patients provided written informed consent.

Clinical study sites were requested to submit tumor tissue to Foundation Medicine, Inc. (FMI) that was formalin fixed and paraffin embedded (FFPE) as a block or multiple 4 to 5 μm (n = 20) unstained, unbaked slides. An accompanying original hematoxylin and eosin (H&E)–stained tissue section, where available, was also requested. If unavailable, H&E staining was performed at FMI. Archival tumor specimens were defined as those previously collected during routine diagnosis and care, and newly collected (referred to as recently obtained) samples were defined as tumor specimens collected solely for the purpose of the study. Furthermore, primary samples were those taken from prostate tissue, while all other non-prostate samples were considered as metastatic. For recently obtained samples, sites were instructed to use standard FFPE fixation (10% neutral-buffered formalin for 6–72 hours) and processing to preserve nucleic acid integrity. Strong acid decalcification was to be avoided for recently obtained bone biopsy and bone marrow trephine samples; instead, calcium extraction with ethylenediaminetetraacetic acid after formalin fixation was recommended. Details of the tumor sample (e.g., organ, collection method, and collection date) were recorded on the specimen collection module in the PROfound study electronic case report form as completed by participating sites.

Prospective, central sample analysis was performed using an investigational clinical trial assay (CTA), based on the FoundationOneCDx NGS test, developed by FMI. An H&E-stained slide for each specimen underwent a specimen adequacy review by a trained pathologist to assess tumor nuclei percentage and available tissue volume. A minimum tumor content (ratio of tumor nuclei to all nuclei) of 20% and a minimum tissue volume of 0.2 mm³ was required to proceed to DNA extraction for the CTA used on the PROfound trial. The cutoffs used on the PROfound trial were based on validated cutoffs for the clinical trial assay as defined by FMI. The majority of samples (∼96%) assessed were, however, >0.6 mm³ (note: 0.6 mm³ is the minimum tissue volume required for FoundationOneCDx NGS, which was FDA approved after the PROfound trial started; ref. 12). For specimens with <20% overall tumor nuclei percentage, macrodissection to enrich tumor nuclei percentage above the required threshold was attempted if feasible based on tissue size and tumor distribution. Overall, a minimum of 50 ng of DNA postextraction was required for samples to proceed to library construction (rare exceptions of samples yielding <50 ng of DNA that proceeded to testing were included). For each tumor specimen that passed tissue input adequacy requirements, subsequent library construction, hybrid capture, and sequencing quality controls, a CTA report was generated specifying the presence or absence of qualifying gene alterations. In the PROfound study, a reported failed result referred to samples that did not meet preanalytic tissue input adequacy requirements and/or did not pass analytic quality control metrics during processing at FMI, with reasons ranging from pathology review to computational analysis of sequencing data. For the purpose of analysis performed in this article, generation of an NGS result was defined as samples that completed the FMI testing process according to FMI standards/test requirements and yielded an NGS result [NGS results included samples for which all sequencing quality control metrics were met regardless of biomarker mutation status (“pass”) or samples where one or more postsequencing quality metrics (e.g., coverage, computational tumor purity) did not meet pass criteria; however, NGS biomarker result was still deemed reportable but with potential for reduced sensitivity (“qualified” report)].

Because of the higher-than-anticipated rates of not obtaining an NGS result observed in the early stages of the PROfound study, a global WebEx session for study sites participating in the PROfound study was held in September 2017 to furnish insights into the current status of the samples with a view to providing advice on how to improve testing results. During the WebEx, the study sites were given advice on sample collection and processing before submission to FMI; they also had the opportunity to ask questions. Afterward, they were provided with updated educational material in the form of an updated education manual and leaflets on best practices and learnings from PROfound up to that point. Education focused on advice for pathologists selecting the samples for testing, recommendations on sample collection and preparation for recently obtained samples, and best practices for slide sectioning to avoid cross-contamination. This included advice on collecting multiple core needle biopsies (CNBs), creating blocks for cytology specimens using enrichment methods to maximize yield, minimizing tissue depletion of the blocks for initial diagnosis, guidance on using an 18-gauge needle and CT for lymph node and soft tissue, guidance on using an 11- to 14-gauge bone-cutting needle (with or without a coaxial trocar) for bone samples, optimal sample preservation methods, and preventing cross-contamination between samples. To evaluate whether this focus on education led to an improvement in generation of NGS results, data were analyzed for samples received for testing at FMI prior to and after October 1, 2017, at both sample and patient level.

Statistical analyses were performed using SAS version 9.4. For data on rates of NGS result generation, 95% confidence intervals (CIs) were obtained using the Clopper–Pearson method. For individual factors for which P values are reported, χ² and Wilcoxon and Kruskal–Wallis rank-sum tests were applied as appropriate.

To evaluate the predictive power of the sample characteristics on obtaining an NGS result, a multivariate logistic regression model was generated. For the modeling analysis, the characteristics considered were sample type, organ site, collection method, sample age, total tissue volume, tumor content (nuclei), and DNA yield. The model was built, using data from the PROfound study, through selection of any significant two-way interactions of sample characteristics, as well as purposeful selection of single characteristics using available information from the tissue samples. As the model relied on complete information, any missing values were either excluded or imputed (to demonstrate the impact of exclusion on the overall model), and continuous variables such as DNA yield, tumor content, total tissue volume, and sample age were log transformed to remove skew. Any samples with variables that were recorded as zero were set to near non-zero values to enable log transformation (130 sample records had percentage tumor nuclei imputed from 0 to 1, and 21 records had tissue volume imputed from 0 to 0.01; 548 records with missing tumor DNA were excluded). The model was evaluated on the fit, taking into account a number of factors, including the Akaike information criterion for the whole (overall) model, the significance of the model factors (parameter estimates of effect likelihood ratio test/Wald tests), and the predictive power using ROC AUC assessment. A lack-of-fit test (χ² test) was also evaluated. Leave-one-out cross-validation was performed on the model, and outputs were checked against the study descriptive results for two-way interactions (tumor DNA, total tissue volume, and tumor nuclei), such that the model can be used to predict outcomes when no real information is available.

Data underlying the findings described in this article may be obtained in accordance with AstraZeneca's data-sharing policy described at https://astrazenecagrouptrials.pharmacm.com/ST/Submission/Disclosure. To protect patient privacy, the raw data generated in this study are not publicly available. Individual requests will be considered, and data will be made available once the risk to patient identification has been assessed.

In total, 4,425 patients were screened for the PROfound study at 206 sites in 20 countries. Samples from 4,069 (92%) patients were received at FMI; samples were not available for 356 (8%) screened patients. Of the 4,069 patients with available samples, 4,047 (99.5%) were eligible for testing (i.e., proceeded to the first step in the test process, pathology assessment); the other 22 (0.5%) patients were ineligible for testing (e.g., incorrect sample submitted, such as fresh tissue or blood, or sample was received after the testing period was closed). From the 4,047 patients with samples eligible for assessment, some patients submitted more than one sample (archival and more recently collected or multiple archival or multiple recently collected) resulting in a total of 4,858 unique tumor samples evaluated at FMI with results (an NGS result or failed result) reported to the FMI web portal. Generation of an NGS result was defined as samples that completed the FMI testing process according to FMI standards/test requirements and yielded an NGS readout (regardless of biomarker mutation status). An NGS result was reported for 57.5% (2,792/4,858) of samples, equating to 69% (2,792/4,047) of patients overall. Of these, 316/4,047 (7.8%) patients obtained an NGS result with a subsequent sample that was submitted after a failure of an initial sample test submitted. Additional analyses of the proportion of patients generating an NGS result by country are shown in Supplementary Table S1; however, due to small sample sizes from some countries, the data should be interpreted with extreme caution.

The majority of tested samples were obtained from archival primary prostate tissue (Fig 1A–D) and were collected by a variety of methods; CNB was the most common (Fig. 1E). Prostate CNB samples generally consisted of one to several residual cores of material remaining after primary diagnostic assessment, measuring < 1 mm in diameter and with lengths varying from 1 to 20 mm. There were a few anomalies with the reported collection methods, such that a small number of non–prostate-derived samples were recorded as having been collected during transurethral resection of the prostate (TURP), and non-bone samples were recorded as trephine biopsies; these were assumed to be incidental sampling during the procedure in question. The majority of prostate samples were archival CNBs collected as part of the initial patient diagnosis (see Supplementary Fig S1A and S1B for the distribution of collection methods used for each organ type and the organ type for each collection method, respectively). The age of samples ranged from < 1 to > 10 years (Fig. 1F); additional analyses of sample age by site characteristics are shown in Supplementary Table S2. For example, of the non-prostate samples, bone samples had the lowest median age, whereas lung samples had the highest median age (P < 0.0001). Of the collection methods, the highest median age was seen for radical prostatectomy samples, and trephine biopsies had the lowest (P < 0.0001). Figure 2 shows a selection of histological images of samples collected in the study that did and did not generate NGS results.

Of the 4,858 FMI tested samples that had an outcome reported to the FMI web portal in the PROfound study, 2,792 (57.5%) generated an NGS result. At sample level, the proportion of samples to give an NGS result was marginally lower for archival biopsy samples (56.9%) than for recently obtained samples (63.9%; Fig. 3A), and for primary samples (56.2%) than for metastatic (63.9%; Fig. 3B). There was no difference in obtaining an NGS result based on submitted sample format: 57.1% (95% CI, 55.2–59.0) for blocks (n = 2,679) and 57.7% (95% CI, 55.6–59.8) for slides (n = 2,167). Of the specimen tissue sites, samples collected from lymph nodes had the highest rates of NGS results obtained (74.7%), and bone samples had the lowest (42.6%; Fig. 3B). The sample collection method with the lowest rates of NGS results obtained was prostate CNB (52.4%), while a higher rate of obtaining an NGS result were reported for samples derived from radical prostatectomy (74.0%) and TURP specimens (69.8%; Fig. 3C). Although trephine biopsies had the highest rate of generation of an NGS result of all collection methods (86.7%), these findings are based on only 15 samples, five of which were recorded as being derived from non-bone sites. Furthermore, many bone samples were recorded in the CNB category; therefore, this finding is not representative of all bone biopsies collected in the study. As expected, and in agreement with findings from the recently obtained versus archival samples, a higher proportion of recently collected samples generated an NGS result, with a gradual decline with increasing sample age (Fig. 3D). The association of older sample age with greater chance of no NGS result being generated was shown via χ² test for association (P < 0.001); for example, 68.1% (95% CI, 65.0–71.1) of samples that were < 1 year old generated an NGS result compared with samples that were >10 years old at 47.3% (95% CI, 43.0–51.5). There was a similar age-related decline for CNBs when assessed separately that amplified the low tissue volume effect and further reduced the rate of generation of an NGS result in the CNB cohort from 65.5% (95% CI, 61.7–69.1) for samples ages < 1 year to 38.9% (95% CI, 33.2–44.9) for samples ages > 10 years (Supplementary Fig. S2A). In comparison, the rate of obtaining an NGS result was 74.7% (95% CI, 69.0–79.9) for samples < 1 year old and 56.6% (95% CI, 50.4–62.7) for samples > 10 years old for other collection methods (Supplementary Fig. S2B). Additional country level analyses of the proportion of samples generating an NGS result by sample type, collection method, and organ type are shown in Supplementary Tables S3 to S5; however, due to small sample sizes from some countries, the data should be interpreted with extreme caution.

Analysis of NGS result generation by tumor content (nuclei), total tissue volume, and DNA yield indicated an increase in the proportion of NGS results obtained with increased tumor content and DNA yield (Fig. 3E and G). Despite one tissue volume grouping (1–5 mm³) showing a slight decrease of NGS result obtained compared with the 0.6–1.0 mm³ group, there was a visible overall trend toward increased NGS result generation with increased total tissue volume as an individual factor (Fig. 3F). Further evaluation of NGS result generation by tumor content showed a corresponding increase in DNA yield with an increase in tumor content (Supplementary Table S6). It is worth noting that for DNA yield, there was a maximum input that would be analyzed by NGS (i.e., 50–1,000 ng of DNA was required for library construction, with 1,000 ng being the maximum input), so increased DNA yield would not necessarily always reflect increased input for NGS (for samples with DNA yield > 1,000 ng) but may reflect a sample yielding higher-quality DNA. Mean and median percentage tumor content, total tissue volumes, and DNA yields were evaluated for different sample categories (Supplementary Table S7). Higher mean and median DNA yields were observed for more recently obtained and non-prostate (metastatic) samples than with archival and primary (prostate) samples. Of the metastatic samples, those taken from lymph node samples had the highest mean/median tumor content and DNA yield, whereas bone samples had the lowest tumor content and DNA yield. Specifics of the decalcification methods for archival bone specimens were not well documented; however, histologic appearance suggested use of strong acids in many cases, a likely contributor to the low DNA yields observed for bone samples. Notably, samples with total tissue volume > 0.2 mm³ were eligible for evaluation in this study, whereas the recommended minimum total tissue volume for the FoundationOneCDx NGS test is 0.6 mm³. The majority (96.1%; 4,669/4,858) of samples collected in our study had a cutoff total tissue volume of > 0.6 mm³. The overall proportion of samples generating an NGS for the cohort of samples with tissue volume ≥ 0.6 mm³ was higher (58.9%; 95% CI, 57.4–60.3) than for the 0.2–< 0.6 mm³ cohort (32.1%; 95% CI, 24.4–40.6), as was DNA yield (median 357.8 vs. 88.3 ng, respectively; Supplementary Table S8).

Of the 4,858 unique samples that had a result reported in the PROfound study, testing failed to deliver a result in 2,066 (42.5%) cases. Figure 4A shows a flow chart of the processing stage at which samples failed and the proportion of samples failing at each stage. Overall, the main reasons for samples failing to meet FMI's standard requirements for the test process and consequently not obtaining an NGS test result were “not meeting specimen adequacy requirements” (pathology review—a minimum tumor content of 20% and a minimum tissue volume of 0.2 mm³) and failed DNA extraction (Fig. 4B). Archival samples were most likely to fail at the DNA extraction stage, whereas recently obtained samples were more likely to fail at the pathology adequacy review stage. The reasons for failure at pathology review stage were primarily insufficient tumor content/purity and insufficient tissue size; however, low cellularity or the sample being presumably exposed to high temperature during slide preparation (baked) were also factors. Reasons for failing to meet FMI's standard requirements for the test process varied across sample collection methods (Fig. 4C). For example, larger samples, such as from radical prostatectomy, were more likely to meet FMI's requirements at pathology review because of abundant tissue and typically larger foci of tumor, that could be enriched by macrodissection, rather than other collection methods; however, they were more likely to fail computational biology review because they were more likely to be archival and have lower DNA quality. Further analysis found that samples that generated an NGS result had a lower mean and median age than those that did not, and samples that failed to meet FMI requirements later in the testing process were older than those that failed at the beginning of the test process (Supplementary Table S9). For example, samples that failed at pathology review were typically younger in age (mean ± SD 4.4 ± 4.6 years) than those that failed to meet quality control metrics required during computational review of sequencing data (6.7 ± 4.7 years; Supplementary Table S9). Additional analyses of the reasons for failure by country are shown in Supplementary Table S10; however, due to small sample sizes from some countries, the data should be interpreted with extreme caution.

To determine the factors affecting the generation of an NGS result of samples that passed pathology adequacy review, a logistic regression model (with log transformation of continuous variables) found that the combination of all sample variables had a substantial contribution to whether an NGS result was obtained (AUC = 0.9172; Fig. 5). For this model, DNA yield was the main variable affecting whether an NGS result could be obtained (AUC = 0.9061), with other variables such as sample age (AUC = 0.6292) and percentage tumor content (AUC = 0.5965) also having a potential effect but to a lesser extent (Supplementary Fig. S3). A model without log transformation of continuous variables was not as predictive of whether an NGS result was obtained (AUC = 0.8378); however, the ranking of individual factors did not change relative to the main model (Supplementary Fig. S3). The main model generated was based on complete records, such that any samples from which DNA was not extracted (e.g., failed pathology review) were not included. The impact of removal of incomplete sample data, imputing missing data and using DNA yield as a surrogate for NGS result generation is further described in the Supplementary Appendix (Supplementary Figs. S4 and S5). Despite the observations of changes in predictive power, the main model with no imputation of missing DNA was retained and any missing records excluded to ensure that the model was built on known records.

This model was used to evaluate various sample characteristic combinations to predict outcome in specific scenarios, even where data were not available (Table 1). The selection of cutoffs was based on selecting a range of minimum requirements required (50 ng DNA input; 0.2 mm³ and 20% tumor content) and expected optimal conditions (200-ng input; 1-mm³ tissue volume), as well as a range of tumor content spanning various potential observations in a real-world setting. As a supplement to the observed outcomes in PROfound, the model enabled us to evaluate various scenarios of sample characteristics that may be observed in a real-world setting. The two-way interaction model comprises purposefully selected key sample characteristics such as DNA yield, percent tumor nuclei, tissue volume, and age, regardless of their significance in the model. Within the characteristics evaluated in Table 1, age < 5 years was shown to predict increased likelihood of acquiring an NGS result for the categories outlined in Table 1, including those with the lowest total tissue volume and percentage tumor content. Our model showed that selecting an increased tissue volume when combined with the other characteristics shown in Table 1 is not predicted to affect the likelihood of obtaining an NGS result if considered in combination with the factors: collection method, DNA yield category, age category, and log-transformed percent tumor nuclei. The influence of tissue volume as an individual factor is, however, visible in the analysis of the actual data (Fig. 3F). Furthermore, even for samples ages < 5 years, an increase in the proportion of NGS results obtained was predicted as other categories, such as tumor content and DNA yield, were increased. Interestingly, however, for samples > 5 years, the model predicted minimal impact of different percentage tumor content when the DNA input was between 50 and 200 ng but a predicted increase in likelihood of obtaining an NGS result when the DNA was > 200 ng and increased tumor content. Additional modeling that included factors such as tumor grade and Gleason score was also performed (see details below and Supplementary Appendix).

An additional model assessed imputation of missing DNA data at results less than the required minimum of 50 ng. By imputing the DNA yield to 0.01 or 49 ng, while the general trend of individual factors influencing NGS result generation was similar to the main model, there was a shift in ranking of sample age and percentage tumor nuclei specifically compared with the main model (Supplementary Fig. S4A and S4B, respectively). With no imputation for DNA (i.e., the main model), the predictive power for percentage tumor nuclei was AUC = 0.5965, which increased to AUC = 0.6763 with imputation of tumor DNA to 0.01 ng. This increase is in line with that observed for the overall main model prediction (from AUC = 0.9172 to AUC = 0.9400). By removing DNA yield completely from the model, the predictive power of the model decreased (AUC = 0.7604), but there was a shift in ranking of sample age and percentage tumor nuclei compared with the main model (Supplementary Fig. S4C).

Given that DNA yield had the greatest predictive power of NGS result generation, an additional model was evaluated using DNA yield as a surrogate for NGS result generation. Using a DNA yield of >50 ng as the cut-off value, the predictive power of the overall model decreased from AUC = 0.9172 to AUC = 0.7749, suggesting that it is not just sufficient DNA that is required for an NGS result to be obtained and that other factors should be considered (Supplementary Fig. S5A). In this model, the predictive power of tumor nuclei increased (AUC = 0.6785) relative to the main model (AUC = 0.5965). Conversely, evaluation of higher DNA yield cut-off values (>100 and >200 ng) found that the predictive power of the overall model decreased further, as did the predictive power of tumor nuclei (Supplementary Fig. S5B and S5C, respectively).

To check the reliability of the model, we evaluated the predicted model outcomes against data observed in the study, and high concordance was observed (Supplementary Table S11).

Although DNA yield and percentage tumor nuclei were higher for samples that had an NGS result, when evaluated at organ level, it was noted that some metastatic samples (e.g., liver and lung) had less overlap in the range of percentage tumor nuclei and DNA yields for samples yielding an NGS result over samples where an NGS result was not obtained (data not shown), suggesting that other clinical factors (e.g., tumor grade and Gleason score) may potentially influence generation of a test result. To investigate further, another model evaluated available data including primary tumor grade information and Gleason score as additional factors (Supplementary Fig. S6). These data were primarily available for patients randomized in the PROfound study; therefore, the dataset was biased to a population of patients with samples yielding NGS results (with a select few who did not have an initial result) and should thus be interpreted with caution.

Evaluation of rates of NGS result generation for all samples found an 8% improvement from 51.9% (95% CI, 49.2–54.6) for the 1,380 samples received at FMI on or before October 1, 2017 to 59.7% (95% CI, 58.0–61.3) for the 3,478 samples analyzed after October 1, 2017, following the educational sessions in September 2017 (Supplementary Table S12). A similar improvement was observed at patient level, whereby the proportion of samples where an NGS result was obtained increased from 61.1% (95% CI, 58.2–63.9) for the 1,172 patients whose samples were received before October 1, 2017, to 69.9% (95% CI, 68.3–71.6) for the 2,968 patients whose samples were received on or after October 1, 2017. Evaluation of the change in the percentage of samples not obtaining an NGS result over time showed a gradual decline from 48.1% at September 30, 2017 (1,380 samples) to 42.5% by December 31, 2018 (4,858 samples) at sample level (Supplementary Fig. S7A) and from 38.9% at September 30, 2017 (1,172 patients) to 31.1% by December 31, 2018 (4,047 patients) at patient level (Supplementary Fig. S7B). These results suggest that educating pathologists and laboratory staff on the preparation of samples is a key aspect of improving generation of an NGS result.

The PROfound study is the largest study to date to screen patients prospectively for a qualifying alteration in a panel of genes with a direct or indirect role in HRR by central tumor tissue NGS testing. The PROfound study has shown the feasibility of molecularly targeted therapy in men with mCRPC (when tumor tissue is available), with an NGS result generation rate of 69% when calculated at patient level (1), comparable with that observed in several other prostate cancer studies using similar standardized operating procedures (13–15). We sought to evaluate the observations from the PROfound study, as successful implementation of genomic testing in clinical practice is critical to identify men with mCRPC for whom there is a significant unmet clinical need and who might be eligible for precision medicine with PARP inhibitors, immunotherapy, and targeted agents.

Most tissue samples analyzed in the PROfound study were from archival tissue, and although the rate of NGS results generated was marginally higher in more newly obtained samples, generating an NGS result was possible with both. Furthermore, while most samples were taken from primary prostate tumor specimens, the rate of generation of an NGS result was higher for non-prostate metastatic samples. This may be a result of metastatic samples having higher tumor content and cellularity and incidentally higher DNA yield, as well as being more recently obtained and assessed. Despite lower rates of NGS result generation with primary and archival tissue, evaluation of HRR alterations in these samples is achievable for some patients and appropriate even after mCRPC progression, as the majority of HRR alterations in prostate cancer appear to occur early in the disease and before metastasis (16–18). We observed a decline in the rate of NGS results obtained with increased sample age, which may be a result of DNA degradation over time. The main reason for not obtaining an NGS result for archival samples was failure to extract sufficient DNA, possibly because of DNA degradation, with exacerbation of the effect by tissue collection techniques that yield smaller available tissue input volumes, such as CNB. Despite these findings, the generation of an NGS result was reported in almost half of samples ages >10 years, especially in samples of larger volume or high tumor content, suggesting that optimizing collection methods may help with obtaining NGS results from archival samples. The generation of an NGS result is also influenced by collection methods. For example, CNB is increasingly being used to obtain prostate tissue samples, as it is less invasive than many other collection methods, and it was the most common method in the PROfound study; however, rates of obtaining an NGS result were lower than for other collection methods. Several features specific to prostate CNB are likely to contribute to not obtaining an NGS result for samples, such as not meeting the requirements for adequate specimen volume and percentage tumor nuclei. The benign prostate gland comprises abundant fibromuscular stroma and interspersed secretory glands with large lumens, leading to an overall low density of nuclei compared with other anatomical sites. In contrast, areas with prostatitis and prostate adenocarcinoma, particularly those with higher Gleason grades, have higher density of nuclei. When a prostate CNB is obtained, depending on the path it traverses, there can be a large variance in the amount of tissue with low versus high nuclei density and, therefore less of a predictable relationship between tissue volume and DNA extraction yield. Thus, in relatively large cores with small foci of tumor, volume of tissue, and percentage tumor nuclei can be sufficient but total nuclei insufficient to yield the minimum DNA required. The observations from this study suggest that generation of an NGS result is achievable in CNB as long as this effect is accounted for and other factors are adhered to (e.g., selecting the most recently collected samples with highest tumor content where possible). Other considerations, such as embedding multiple CNBs into a single FFPE block, may increase the likelihood of generating an NGS result through increasing DNA yield (19). Macrodissection of the tumor area is also possible and is recommended to increase the tumor content of the sample if the overall volume and cellularity of sample is sufficient (20). Furthermore, with the increasing use of MRI-guided biopsies, higher tumor cellularity with likely higher tumor DNA yield will be found in CNBs, which will inevitably improve NGS result acquisition for this sample type (21).

In our study, surgical specimens, radical prostatectomies, and TURP were found to contain adequate tissue volume, thus often resulting in relatively high levels of NGS result acquisition; however, the percentage tumor content of these large specimens can be low, so macrodissection to enrich samples can sometimes be insufficient for obtaining an NGS result. Differences in rates of obtaining an NGS result across organ types were noted, with lymph nodes having the highest rates of NGS result acquisition and bone samples the lowest, similar to previous reports (13). The improved NGS result acquisition reported for lymph node samples may be the result of a combination of higher nuclear content per volume of tissue and metastatic prostate cancer deposits being localized in subcapsular, high-purity nodules, making them more easily visualized and accessible for biopsy (i.e., tumor sites are better localized with imaging for extraction with needle biopsies) and thereby having the highest tumor content and DNA yields compared with other organ sites. While the low rates of obtaining an NGS result observed for bone samples may be associated with difficulties in collection of sufficient tissue (22–24), leading to lower tumor content and tissue volume, the low DNA yields observed may be a result of other preanalytic factors unique to the collection of bone samples. Avoidance of strong acid decalcification in bone biopsies of tumor samples collected for genomic testing is critical for molecular testing (25). As a consequence, there was a recommendation for this study that decalcification should not be performed on recently obtained bone samples where possible; however, for archival bone samples, it was not possible to control for decalcification as information regarding processing of these samples was not routinely available. Country-level analyses of samples provided at screening found that the countries with samples that had an overall lower rate of NGS result generation had more failures after pathology review, suggesting that although the samples may have met the tissue sample requirements as assessed by the pathologist, other factors such as sample age and/or poor sample preservation may have resulted in failures later in the process.

Our modeling analysis confirmed the findings that multiple factors affect the generation of an NGS result, with some having a greater prediction value than others; for example, DNA yield and percentage tumor nuclei had a strong influence on generating an NGS result in the model. Furthermore, while the model predicted that generation of an NGS result would be higher with younger samples (<5 years), regardless of tumor content, total tissue volume or DNA yield, predicted rates of obtaining NGS results are still encouraging for older samples (≥5 years), particularly when other characteristics are considered. However, we did not evaluate other factors, such as biopsy procedure (i.e., MRI-guided or transrectal ultrasound–guided biopsy), imaging procedure (i.e., prostate-specific membrane antigen or CT guided), and experience of the radiologist.

To improve on the rates of NGS result generation observed in the PROfound trial in the initial few months, a live education session and updated instruction materials were provided to participating sites, based on input from both FMI and medical experts (see Supplementary Appendix for more information). Results suggested that educating pathologists and laboratory staff on the preparation of samples is a key aspect of improving generation of an NGS result. A recent review detailing practical considerations for molecular diagnostic testing in prostate cancer highlights other aspects of sample collection and processing that should be considered for improving the rates at which NGS results are obtained in mCRPC, as well as guidance on DNA extraction and processing to optimize DNA yield and quality for molecular diagnosis (20). In addition, the importance of communication between clinicians and pathologists, as well as continued training and feedback between oncology, interventional radiology, and pathology teams, has also been emphasized (26).

Many institutions are establishing tumor gene panel testing and molecular characterization of prostate samples for HRR alteration status, and this is anticipated to increase given that guidelines, such as those from the NCCN, ESMO, and AUA/ASTRO/SUO, are now including recommendations for tumor testing for HRR gene alterations in patients with prostate cancer (10, 11, 27). In addition, recent advances in cell-free DNA–based tumor profiling based on blood samples (liquid biopsy) to identify alterations in circulating tumor DNA may also provide an additional test option for patients where an NGS result cannot be obtained from tumor samples or for whom no tumor tissue is available, although plasma DNA frequently has a tumor content below 30%, with most liquid biopsy NGS assays not being cleared for copy-number analyses (28–31). It is important to note that should cell-free DNA be negative, it does not rule out tissue positivity.

The majority of samples that progressed through testing and obtained an NGS result in the PROfound study met the validated quality control criteria of the molecular test used in the trial with very few exceptions. Although it is unclear how representative the testing experience and NGS result outcome metrics reported here will be for other clinical laboratories, which may have alternative pipelines, different thresholds for sample metrics and/or may choose to test samples in exception of quality control requirements (32), this study shows some key learnings and factors that may predict the generation of an NGS result and should be considered for collection of samples for testing, regardless of the test used. Sample age, tumor content, and DNA input were the three factors that had the biggest impact on obtaining an NGS result, both in our actual data and our modeled data. These three factors in particular should be considered when selecting samples for NGS testing. If multiple samples are available, consider selecting the younger samples if the tumor content is similar. For older samples >5 years; consider using samples that have the highest tumor content and from locations likely to produce high tumor content and subsequently high DNA yield (e.g., from lymph nodes). For newly collected samples, use of standard formalin fixation and avoidance of decalcification (for bone samples) may help to preserve nucleic acid integrity. Men should be counseled that this test pipeline only delivers an interpretable result approximately two-thirds of the time; that is, screening archival samples or submitting to new biopsies may not enable access to personalized targeted treatment. With FDA and EMA approval of olaparib for treatment of patients with the HRR gene (7) and BRCA alterations (8), respectively, genomic testing is a key element in treatment planning for patients with mCRPC, and so should be taken into consideration at sample collection.

M. Hussain reports grants from AstraZeneca during the conduct of the study, as well as grants from Genentech, Pfizer, Bayer, and Arvinas and personal fees from AstraZeneca, AZ/Merck, Janssen, BMS, Daiichi Sankyo, Novartis, AstraZeneca, Merck, and Astellas outside the submitted work. C. Corcoran reports other support from AstraZeneca and Mudskipper during the conduct of the study and other support from AstraZeneca outside the submitted work, and is an employee of AstraZeneca and holds shares with the company. K. Fizazi reports other support from AstraZeneca during the conduct of the study; other support from Astellas, Amgen, Bayer, Clovis, Janssen, Novartis, and Sanofi, and personal fees from Orion outside the submitted work. F. Saad reports grants, personal fees, and nonfinancial support from AstraZeneca and Merck during the conduct of the study and grants, personal fees, and nonfinancial support from Astellas, Janssen, Bayer, Novartis, and Sanofi outside the submitted work. N. Shore reports personal fees from AstraZeneca, Merck, Invitae, Clovis Oncology, Myriad, Astellas, Pfizer, Janssen, Foundation Medicine, Tempus, and Sema4 during the conduct of the study. S. Sandhu reports grants from Merck Sharp & Dohme, AstraZeneca, Amgen, Novartis, Genentech, and Pfizer and personal fees from Merck Sharp & Dohme and AstraZeneca outside the submitted work. J. Mateo reports nonfinancial support and other support from AstraZeneca during the conduct of the study; grants and personal fees from AstraZeneca and Pfizer Oncology; and personal fees from Roche, Guardant Health, and Merck Sharp & Dohme outside the submitted work. D. Olmos reports grants, personal fees, and other support from AstraZeneca, Bayer, and Janssen; other support from F. Hoffman La-Roche/Genentech; and personal fees from Merck Sharp & Dohme and Clovis outside the submitted work. N. Mehra reports grants and personal fees from AstraZeneca and Merck Sharp & Dohme during the conduct of the study; personal fees from Janssen and Bayer; grants from Astellas and Bristol Myers Squibb; and grants and personal fees from Pfizer outside the submitted work. M.P. Kolinsky reports other support from AstraZeneca during the conduct of the study and personal fees from AstraZeneca, Merck, Astellas, Bayer, Bristol Myers Squibb, Eisai, Ipsen, and Janssen outside the submitted work. G. Roubaud reports grants from Bayer; personal fees from Astellas, Sanofi, and AstraZeneca; and personal fees and nonfinancial support from Janssen outside the submitted work. N. Matsubara reports grants from AstraZeneca during the conduct of the study; grants and personal fees from Janssen and Sanofi; and grants from AstraZeneca, Bayer, Roche, Merck Sharp & Dohme, Taiho, Astellas, Amgen, Eisai, Eli Lilly, Takeda, Pfizer, and Chugai outside the submitted work. C. Gedye reports personal fees from Merck EMD Serono, Merck Sharp & Dohme, Bristol Myers Squibb, Astellas, Pfizer, and AstraZeneca outside the submitted work; and C. Gedye has participated in consulting roles and advisory boards with Merck EMD Serono, MSD, BMS, Astellas, Pfizer, AstraZeneca, and Astellas. All payments and fees from these activities are directed to a third-party not-for-profit; C. Gedye does not accept any payments personally: the third party is ANZUP Cancer Trials Group unless otherwise stated, and donations can be made to ANZUP at https://www.anzup.org.au/donate/donate.aspx to support their work to conduct clinical trials research to improve treatments of bladder, kidney, testicular, penile and prostate cancers. C. Gedye receives travel support (when we could travel!) from BMS, Astellas, MSD, was accepted in lieu of NSW Health employment contract allowances (pharma pays for conferences that would have been received via publicly funded TESL reimbursement). C. Gedye reports research funding direct to third-party (no salary or payments to individual) from Bristol Myers Squibb, Amgen, and Merck Sharp & Dohme; MRFF funding to COGNO/University of Sydney; consulting for Novotech-CRO PTY, to help them improve and enhance clinical trial concepts and proposals, aiming to make for more effective commercial cancer clinical trials; and scientific advisory committee membership with ANZUP Cancer Trials Group, COGNO, Mark Hughes Foundation, and International Kidney Cancer Consortium. A. Kohlmann reports other support from AstraZeneca during the conduct of the study and other support from AstraZeneca outside the submitted work. R. Huisden reports personal fees from AstraZeneca during the conduct of the study. J.A. Elvin reports other support from Foundation Medicine and F. Hoffmann La-Roche outside the submitted work. J. Kang reports personal fees from AstraZeneca outside the submitted work and is an employee of AstraZeneca and owns stocks of the company. C.A. Adelman reports employment and stock ownership with AstraZeneca. C. Poehlein reports other support from Merck & Company, outside the United States and Canada known as Merck Sharp & Dohme, outside the submitted work. J. de Bono reports grants and personal fees from AstraZeneca during the conduct of the study; grants and personal fees from Amgen, Astellas, Bayer, Bioxcel Therapeutics, Boehringer Ingelheim, Cellcentric, Daiichi, Eisai, Genentech/Roche, Genmab, GlaxoSmithKline, Harpoon, ImCheck Therapeutics, Janssen, Merck Serono, Merck Sharp & Dohme, Orion, Pfizer, Qiagen, Sanofi Aventis, Sierra Oncology, Taiho, Terumo, and Vertex Pharmaceuticals and grants from Menarini/Silicon Biosystems outside the submitted work; in addition, J. de Bono has a patent for DNA repair defects PARP (ICR owned patent) and does not receive royalties issued and a patent for Abiraterone acetate (ICR owned patent) but does not receive any royalties issued. No disclosures were reported by the other authors.

M. Hussain: Conceptualization, formal analysis, methodology, writing–original draft, writing–review and editing, patient accrual. C. Corcoran: Conceptualization, data curation, formal analysis, supervision, investigation, visualization, methodology, writing–review and editing, QC of biomarker data delivered for PROfound. C. Sibilla: Investigation, writing–review and editing, QC of biomarker data delivered for PROfound. K. Fizazi: Investigation, writing–review and editing. F. Saad: Supervision, investigation, writing–review and editing. N. Shore: Investigation, writing–review and editing. S. Sandhu: Data curation, investigation, writing–review and editing. J. Mateo: Investigation, writing–review and editing. D. Olmos: Investigation, writing–review and editing. N. Mehra: Investigation, writing–review and editing. M.P. Kolinsky: Investigation, writing–review and editing. G. Roubaud: Investigation, writing–review and editing. M. Özgüroǧlu: Investigation, writing–review and editing. N. Matsubara: Investigation, writing–review and editing. C. Gedye: Investigation, writing–review and editing. Y. Choi: Investigation, writing–review and editing. C. Padua: Investigation, writing–review and editing. A. Kohlmann: Conceptualization, supervision, investigation, methodology, writing–review and editing. R. Huisden: Formal analysis, validation, investigation, methodology, writing–review and editing. J.A. Elvin: Conceptualization, validation, visualization, writing–review and editing, pathology review at foundation medicine. J. Kang: Resources, formal analysis, supervision, investigation, methodology, writing–review and editing. C.A. Adelman: Investigation, writing–review and editing. A. Allen: Investigation, writing–review and editing. C. Poehlein: Investigation, writing–review and editing, analysis of data (clinical relevance). J. de Bono: Investigation, writing–review and editing.

This study was funded by AstraZeneca and is part of an alliance between AstraZeneca and Merck Sharp & Dohme Corp, a subsidiary of Merck & Co., Inc., Kenilworth, NJ. Medical writing assistance was provided by Juliet Fawcett, PhD for Mudskipper Business Ltd., funded by AstraZeneca and Merck Sharp & Dohme Corp., a subsidiary of Merck & Co., Inc., Kenilworth, NJ.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.