Background

Prostate cancer is the most commonly detected male cancer in Europe and the third major cause of cancer-related death among men [1]. Although the majority of metastatic hormone-naïve prostate cancers (mHNPCs) demonstrate a reliable response to initial androgen deprivation therapy which targets AR signaling, progression to a castration-resistant state is inevitable. However, the treatment landscape for metastatic castration-resistant prostate cancer (mCRPC) is evolving with the recent approval of several new drugs translating to an increased overall survival [2,3,4,5,6]. Multiple additional avenues exist as genomic profiling of metastatic tissue revealed that the majority of mCRPC patients harbor clinically relevant alterations beyond the AR signaling pathway [7].

The most promising non-approved treatment avenue in metastatic prostate cancers (mPCs) exploits synthetic lethality in treating homologous recombination-deficient cancers with poly (ADP-ribose) polymerase (PARP) inhibitors [8]. Approximately one fifth of mCRPC carry mutations in DNA repair genes [7]. However, the mutational signatures of biallelic inactivation are heterogeneous between different DNA repair genes [9], and future studies are therefore needed to determine which genes are associated with a response to PARP inhibition. Approximately 3% of mPC are driven by microsatellite instability (MSI) [7, 10]. Pembrolizumab recently became the first drug to be approved by the U.S. Food and Drug Administration based on the MSI phenotype, irrespective of tumor type [11]. Although checkpoint blockade did not confer a survival advantage as compared with placebo for chemotherapy-relapsed mCRPC [12], anecdotal cases have been reported to display partial or complete responses [10, 13,14,15].

The emergence of additional drugs, both towards common and rare mPC phenotypes such as PTEN-deficient [16, 17] and neuroendocrine cancers [18], raises questions of how to efficiently translate the multitude of treatment options to improved patient outcomes. The genomic heterogeneity of mCRPC [7] and, in turn, the low response rates of currently approved drugs [2,3,4,5, 19, 20] argue for the urgent need of predictive biomarkers. Ineffective trial-and-error decisions inevitably lead to unnecessary side effects and unsustainable costs [21]. The AR splice variant 7 (AR-V7) [22] demonstrated promising results as a negative response biomarker for androgen receptor signaling inhibitors. However, follow-up studies have been unable to validate the initial clear-cut finding [23, 24], and although AR-V7 is clearly prognostic, clinical implementation remains debated due to (1) lack of treatment options for AR-V7-positive patients and (2) lack of data from a prospective clinical trial demonstrating the predictive power of AR-V7 for treatment selection [25]. Recently, Scher and colleagues demonstrated increased survival for AR-V7-positive patients receiving taxanes in a blinded retrospective multicenter study [26]. However, AR-V7 assay positivity and taxane therapy were both correlated to tumor burden, obscuring interpretation and further highlighting the need for a prospective randomized clinical trial to determine if AR-V7 can be applied as a predictive biomarker [27].

The lack of predictive biomarkers is in part due to the difficulty of obtaining temporally matched metastatic tissue as the majority of mPCs metastasize to the bone. Multiple studies on the acquisition of tumor tissue with or without direct image guidance report a range of success rates [28,29,30,31]. A recent effort, focusing on methodological improvements, obtained > 20% cell content in the majority of bone biopsies [32]. Circulating tumor DNA is a viable alternative to metastatic tissue with demonstrated high fractions of ctDNA [33,34,35,36] enabling sensitive detection of somatic variation, and direct comparisons to metastatic tissue have revealed high concordance [33, 37, 38]. Circulating tumor DNA has several advantages as sampling through simple blood draws is fast, cost efficient, and without side effects and allows for longitudinal monitoring and the detection of multiple resistance alleles during therapy [38, 39].

Although ctDNA has the potential to accelerate biomarker-driven trials in mPC, several questions remain unanswered, e.g. if it is possible to detect MSI directly from liquid biopsies and how ctDNA fractions correlate to line of therapy. The ctDNA fraction determines the sensitivity to detect somatic variation which in turn has consequences for the design of prospective biomarker studies relying on liquid biopsies. Here, we present a retrospective analysis of 217 cases and 364 blood samples covering the entire spectrum of mPC. The purpose of this study was to gather information relevant for future liquid biopsy-driven biomarker studies with a focus on (1) how ctDNA fractions vary from mHNPC to end stage castration-resistant disease; (2) a rationale for how to treat samples with low ctDNA fraction; (3) the relative impact of different types of somatic variation, affecting the sequencing strategy; (4) the detection of potentially predictive biomarkers; (5) and finally, how clonal expansions in the hematopoietic stem cells [66]. A large fraction of the mutations was detected in the WBC DNA, corroborating our conclusions that to avoid false positive variant calls, sequencing of WBC DNA should be undertaken to the same or higher depth as the cell-free DNA. We therefore discourage the use of commercial assays that only analyze cfDNA from plasma.

Although multiple potentially predictive biomarkers have been reported for metastatic prostate cancer, no level 1 evidence currently exists from prospective randomized clinical trials. Based on our experience to date, we consider ctDNA profiling to be at Technology Readiness Level 7 according to the definition applied in the Horizon 2020 calls [67]. We have therefore initiated a prospective outcome-adaptive, multi-arm, open-label, multiple-assignment randomized biomarker-driven trial in patients with mCRPC where ctDNA profiling will be applied to identify somatic alterations (ProBio, EudraCT Number 2018-002350-78). The goal of the trial is to determine whether treatment choice based on a biomarker signature can improve progression-free survival compared to standard of care in patients with mCRPC and to evaluate the predictive capability of the investigated biomarker signatures.

Conclusions

This study strengthens the accumulating evidence that ctDNA profiling mirrors the somatic alteration landscape from metastatic tissue by demonstrating, for the first time, that the MSI phenotype may be detected directly from cell-free DNA. To enable acceleration of clinical trials through ctDNA analysis, intronic sequencing of tumor suppressors in combination with synchronous profiling of white blood cells must be applied to prevent inaccurate somatic variant calls, which in turn may reduce the power to identify predictive biomarkers.