Therapeutic biomarkers in metastatic castration-resistant prostate cancer: does the state matter?

Abstract

The treatment of metastatic castration-resistant prostate cancer (mCRPC) has been fundamentally transformed by our greater understanding of its complex biological mechanisms and its entrance into the era of precision oncology. A broad aim is to use the extreme heterogeneity of mCRPC by matching already approved or new targeted therapies to the correct tumor genotype. To achieve this, tumor DNA must be obtained, sequenced, and correctly interpreted, with individual aberrations explored for their druggability, taking into account the hierarchy of driving molecular pathways. Although tumor tissue sequencing is the gold standard, tumor tissue can be challenging to obtain, and a biopsy from one metastatic site or primary tumor may not provide an accurate representation of the current genetic underpinning. Sequencing of circulating tumor DNA (ctDNA) might catalyze precision oncology in mCRPC, as it enables real-time observation of genomic changes in tumors and allows for monitoring of treatment response and identification of resistance mechanisms. Moreover, ctDNA can be used to identify mutations that may not be detected in solitary metastatic lesions and can provide a more in-depth understanding of inter- and intra-tumor heterogeneity. Finally, ctDNA abundance can serve as a prognostic biomarker in patients with mCRPC.

The androgen receptor (AR)-axis is a well-established therapeutical target for prostate cancer, and through ctDNA sequencing, insights have been obtained in (temporal) resistance mechanisms that develop through castration resistance. New third-generation AR-axis inhibitors are being developed to overcome some of these resistance mechanisms. The druggability of defects in the DNA damage repair machinery has impacted the treatment landscape of mCRPC in recent years. For patients with deleterious gene aberrations in genes linked to homologous recombination, particularly BRCA1 or BRCA2, PARP inhibitors have shown efficacy compared to the standard of care armamentarium, but platinum-based chemotherapy may be equally effective. A hierarchy exists in genes associated with homologous recombination, where, besides the canonical genes in this pathway, not every other gene aberration predicts the same likelihood of response. Moreover, evidence is emerging on cross-resistance between therapies such as PARP inhibitors, platinum-based chemotherapy and even radioligand therapy that target this genotype. Mismatch repair-deficient patients can experience a beneficial response to immune checkpoint inhibitors. Activation of other cellular signaling pathways such as PI3K, cell cycle, and MAPK have shown limited success with monotherapy, but there is potential in co-targeting these pathways with combination therapy, either already witnessed or anticipated. This review outlines precision medicine in mCRPC, zooming in on the role of ctDNA, to identify genomic biomarkers that may be used to tailor molecularly targeted therapies. The most common druggable pathways and outcomes of therapies matched to these pathways are discussed.

Keywords:

• Metastatic castration resistant prostate cancer
• precision oncology
• targeted therapy
• circulating tumor DNA
• DNA damage repair

Introduction

Prostate cancer (PCa) is the most common non-cutaneous solid cancer in men and accounts for the second most deaths in men after lung cancer [1]. Most patients with lethal PCa succumb to metastatic disease. After the development of metastases, the cornerstone of the management of PCa is androgen deprivation therapy (ADT), as it has been since 1941 [2]. Almost inevitably, patients progress and develop metastatic castration-resistant prostate cancer (mCRPC).

Treatment options for mCRPC patients were limited for many years. In 1996, the type II topoisomerase inhibitor, mitoxantrone plus prednisone, proved to be more beneficial than prednisone alone [3]. In 2004, the TAX-327 trial showed superior results of 3-weekly docetaxel plus prednisone over mitoxantrone plus prednisone, and, in 2010, cabazitaxel showed improved overall survival (OS) over mitoxantrone after progression on docetaxel [4,5]. In the last decade, next generation androgen receptor (AR) signaling inhibitors (ARSI) proved to be effective in the mCRPC setting, with the COU-301 and COU-302 trials adding abiraterone to the armamentarium and the AFFIRM and PREVAIL trials adding enzalutamide [6–9]. Sipuleucel-T and radium-223 are two non-chemotherapeutic, non-hormonal agents for mCRPC, of which the latter is registered exclusively for patients with bone-only disease [10,11]. With these new therapies, the median OS increased but remained <3 years after diagnosis of mCRPC and <14 months in case of visceral metastasis [12,13].

In recent years, there has been a shift toward using therapies approved for mCRPC upfront in the castration-sensitive setting. The CHAARTED and STAMPEDE trials proved that 6 cycles of docetaxel upfront in addition to ADT increased the OS while the GETUG-15 trial did not show an OS benefit for 9 cycles of docetaxel upfront [14–16]. The LATITUDE and STAMPEDE trials showed an OS benefit for adding abiraterone to ADT, and the ARCHES, ENZAMET, and TITAN trails showed positive results for enzalutamide or apalutamide with ADT [17–21]. The PEACE-1 trial even showed an OS benefit for the triplet therapy of ADT, docetaxel, and abiraterone in de novo metastatic castration-sensitive PCa patients [22]. It should be noted that these studies varied in their inclusion criteria, particularly with regard to risk classification.

The management of metastatic PCa has now entered the era of precision oncology, with the most important breakthroughs being in the mCRPC setting. The aim of precision oncology is to act therapeutically on the extreme heterogeneity of mCRPC by classifying patients in groups based on biomarkers that predict response to a given treatment [23]. Current biomarker-driven efforts focus on matching therapies to their driver mutations. For genetically matched therapy (GMT), it is essential to sequence tumor DNA derived from liquid and solid biopsies and to interpret results correctly. This review on therapeutic biomarkers in mCRPC discusses data on both liquid and solid biopsies to allow allocation toward GMT, as well as the pros, cons, and pitfalls of identifying genetic targets through these approaches.

The state of matter

The genetic analysis of PCa has been traditionally performed through the sequencing of archival primary prostate tissue, and, less commonly, of biopsied metastatic tumor tissue. While the use of diagnostic tissue has shown to be effective in molecular profiling, obtaining metastatic tissue can be challenging because of inaccessible lesions [24,25]; up to 90% of PCa patients have bone metastases, and biopsy of these lesions can be arduous and has a lower success rate than that of soft tissue [26,27]. Also, the acquisition of fresh tumor tissue can result in patient morbidity and high costs, and the lead time can cause deterioration of the patient’s performance status and delay initiation of precision medicine; thus, alternative methods such as archival primary tissue profiling is a more practical approach.

Archival primary tissue profiling has the advantage of rapidly providing comprehensive molecular profiles from biopsies obtained for diagnostic purposes [28]. However, the genetics of primary PCa tissue and mCRPC tissue differ, with PCa being one of the tumor types with the largest differences in genomic features between primary and metastatic disease [29]. Mateo et al. characterized molecularly the matched, same-patient diagnostic and mCRPC tissues of 61 patients and found increased prevalence of TP53, RB1, and PI3K pathway aberrations in mCRPC [30]; they discussed that changes in the DNA damage repair (DDR) pathway were likely early events and could already be found in primary tissue, although this was supported only by a small sample size of nine patients with aberrations in genes associated with DDR. Robinson et al. found mutations in TP53, AR, and BRCA2 (among others) to be enriched in mCRPC, and van Dessel et al. found aberrations in AR, MYC, TP53, and PTEN (among others) to be enriched in mCRPC [24]. It is worth noting that the latter studies were not same-patient populations; therefore the presence of enriched BRCA2 aberrations could be explained by these patients having an aggressive phenotype that more commonly leads to mCRPC.

Even profiling of a freshly biopsied metastatic lesion may not fully reflect the possible intra-tumor and inter-tumor heterogeneity of PCa [31,32]. Moreover, for close disease monitoring, repeated biopsies, which would bring harm to the patient and incur extra costs would be necessary.

Given these limitations, the use of liquid biomarkers obtained from minimally invasive procedures is becoming increasingly important in the field of cancer research. Circulating tumor DNA (ctDNA) has emerged as an alternative to tissue biopsy that provides real-time insight into the genomic changes in the tumor and enables monitoring of treatment response and detection of resistance mechanisms.

Circulating tumor DNA

The presence of cell-free DNA (cfDNA) was first described in 1948. It comprises DNA fragments circulating freely in the human blood that originate from benign and malignant sources [33,34]. Typically, individuals with cancer have a greater amount of cfDNA compared to healthy individuals. If a tumor is present, a proportion of the cfDNA can be tumor-derived and is referred to as the ctDNA fraction. It can represent genetic aberrations present in the tumor. The fragment sizes of ctDNA molecules can deviate from non-tumoral cfDNA [35–37]. The fraction of ctDNA in cfDNA can also vary, depending on the tumor burden, with increasing levels of ctDNA occurring in more advanced disease stages [38–40]. The threshold of current ctDNA assays limits its detection in a proportion of patients, thus at present limiting its full potential in patients harboring “undetectable” ctDNA levels [38–41]. However, the variable abundance of ctDNA in cfDNA is strongly prognostic for mCRPC outcomes, with undetectable or low ctDNA levels being associated with a longer progression-free survival (PFS) and OS for both ARSI and taxane chemotherapy [42–45].

The concordance between ctDNA and matched tissue biopsies is high, making ctDNA a valuable source of tumor DNA for precision oncology in mCRPC. Wyatt et al. found a 94% concordance between genetic mutations present in metastatic PCa tissue and mutations detected using ctDNA from the same patient [46]. An evaluation of copy number variations in samples with a ctDNA fraction >35% and solid tumor biopsies demonstrated a correlation with an R-squared value of 0.76 (range 0.28−0.94). Interestingly, the same group reported a lower concordance (80%) between ctDNA and primary diagnostic tissue for 53 newly diagnosed de novo metastatic PCa patients [40]. The feasibility of genomic profiling of ctDNA has been demonstrated by Tukachinsky et al. who successfully sequenced the plasma of 3334 mCRPC patients, with 94% of the patients having detectable ctDNA [47].

Sequencing of ctDNA presents several advantages over tissue sequencing. First, liquid biopsies are less invasive compared to tumor biopsies, that is, they can be considered minimally invasive. Moreover, blood draws can be performed outside the hospital and blood can be transported by ambient temperatures to academic hospitals or specialized laboratories for ctDNA analysis, thus reducing patient travel and costs. Second, the half-life of cfDNA/ctDNA is short (16 min to 2 h), enabling a real-time snapshot of the tumor genomics [48]. Third, while frequent tissue biopsies are not possible, liquid biopsies can be obtained during treatment and can enable early detection of resistance mechanisms. Finally, ctDNA is a better reflection of tumor heterogeneity compared to a single tissue biopsy [49].

In addition to these advantages, ctDNA also has limitations, including the need for deep-sequencing, predominantly because of low tumor fractions, which can result in higher costs. Additionally, there are cases where the ctDNA percentage is below the detection limit, rendering the test incapable of producing a result. Another drawback of the use of ctDNA is the occurrence of clonal hematopoiesis, which may lead to false positive results if it is not carefully corrected for (discussed in detail later). Furthermore, the determination of copy number variations and certain structural variants may be less accurate because of the limited tumor content and short fragment size of ctDNA molecules.

Precision oncology in mCRPC

The integration of ctDNA sequencing in mCRPC care has the potential to gain momentum for precision oncology, especially because there are ample druggable targets in mCRPC. According to Robinson et al., 89% of mCRPC patients had potentially clinically actionable genetic aberrations, particularly in the AR-axis, the PI3K signaling pathway, and genes associated with DDR [24]. Our institute explored the implementation of GMTs in the care of mCRPC patients [50] and found that 46% of the mCRPC patients harbored a druggable genetic target, as per the ESMO Precision Medicine Working Group recommendations for NGS in advanced PCa [51]. GMTs were recommended to 102 patients (47%), with 63 patients (62%) starting the treatment. Patients who initiated GMTs had promising outcomes, including a PFS of ≥6 months in 41%, a ≥ 50% PSA decline (PSA50) in 44%, and a radiographic response in 39% of cases. Patients with aberrations in genes associated with DDR were the most likely to receive a recommendation and to initiate GMT, and the agents targeting the DDR pathway also showed the best responses.

DNA damage repair

Impaired DDR machinery may be the Achilles’ heel of mCRPC. Deleterious genetic aberrations in genes associated with mismatch repair (MMR) or homologous recombination repair (HRR) confer a highly druggable genotype that results in clinically relevant therapeutic responses. The maintenance of genomic integrity by means of effective DDR is a fundamental aspect of cellular physiology. Conversely, defects in DNA repair can lead to genomic instability, a hallmark of cancer [52]. The prevalence of DDR aberrations in metastatic PCa adds up to almost 30%, with approximately 16% representing somatic and 12% germline aberrations [53,54].

PCa patients with HRR aberrations, especially germline variants, present with higher-grade tumors and an increased risk of nodal involvement and distant metastasis at diagnosis [55,56]. The impact of HRR aberrations on mCRPC patients’ response to ARSI or taxanes has not reached consensus because of inconsistent study results, possibly because of the distinct effects of germline and somatic aberrations [57–60]. A recent meta-analysis compared the efficacy of abiraterone, enzalutamide, and docetaxel in patients with tumors harboring BRCA1 or BRCA2 aberrations (BRCAm) in terms of PSA50, PFS and OS [61]. In total, 348 patients from 16 studies were included. All agents showed comparable response rates in the genetic subgroups. By pooling first- and second-line ARSIs, a PFS benefit of enzalutamide over abiraterone was observed (hazard ratio [HR]: 0.47, 95% confidence interval [CI]: 0.26–0.83, p = 0.010). The response of patients with MMR-deficient (MMRd) PCa to ARSI or taxanes has been explored only in small retrospective studies, but aberrations in MMR genes do not appear to negatively impact the response to standard therapies [62–65]. It is challenging to determine the impact of DDR gene defects on prognosis, given the ongoing treatment revolution in which certain gene alterations such as BRCA2 loss-of-function alterations can potentially change from being an unfavorable to a favorable prognostic factor when they become selection criteria for targeted therapies.

Homologous recombination

HRR is a DDR mechanism by which cells repair double-stranded breaks (DSBs) in DNA. The process involves the use of a homologous template such as a sister chromatid or a homologous chromosome to repair the broken DNA.

DSBs occur with or without prior single-stranded breaks, for instance, through exposure to ionizing radiation or chemotherapy or after unsuccessful single-stranded break repair. The majority of DSBs are physiologically generated through topoisomerase II activity during DNA repair or during meiosis [66]. DSBs can be repaired through various mechanisms that include HRR and the more error-prone non-homologous end joining.

Following the occurrence of a DSB, MRN complexes can bind near the free ends of the DNA at the break [67]. The MRN complex is composed of the MRE11, RAD50, and NBN/NBS1 proteins. After trapping the DNA, the complex recruits and activates ATM kinases to the site of the DSB [68]. Once activated, ATM phosphorylates other transducers such as histone H2AX, creating a cascade that recruits a plethora of DDR-related proteins. This process leads to an increase in the expression of genes that are important for DNA repair, cell cycle arrest, and even cell death (apoptosis), including CHK2 (encoded by CHEK2), p53 (encoded by TP53) and BRCA1 [67, 69]. While the involvement of BRCA1 in HRR is broadly accepted, its exact role is yet to be elucidated [67, 70,71].

To trigger HRR, CtIP, a damage repair protein, binds to the MRN complex and allows the complex to create a 3′ overhang at each site of the DSB, after which the MRN complex detaches. The 3′ overhang is then occupied by multiple replication protein A proteins [67]. BRCA2, together with its partner and localizer PALB2, facilitates the replacement of replication protein A with RAD51, creating a single strand overhang occupied with RAD51 proteins [72].

The RAD51 proteins catalyze the binding of the 3′ overhang of the broken DNA strand to the complementary strand of a sister chromatid, after which the RAD51 proteins are removed. It has been proposed that BRCA1 plays a role in the timely removal of RAD51 proteins [71]. The 3’ overhangs are then extended complementary to the strand on the opposite side of the break, resulting in error-free restored double-stranded DNA [73].

In DDR, genes associated with HRR are most frequently altered. Deleterious aberrations, including pathogenic mutations or copy number losses of genes associated with HRR (HRRm), can lead to a homologous recombination-deficient phenotype (HRD). In the phase 3 PROfound study, 28% of the 2792 successfully screened mCRPC patients were defined as HRRm, predominantly because of BRCA2 aberrations (9.7% of successfully screened patients) [74]. The mutations occur with roughly equal frequency in germline and somatic variants [75]. In the PROfound trial, ctDNA analysis of 491 patients showed high agreement (81% positive and 92% negative agreement) with tissue analysis for BRCA and ATM status [76]. A separate study found 93% concordance of BRCA aberrations in 837 matched ctDNA and tissue samples [47]. Because HRRm mCRPC patients are more sensitive to certain therapies, this is a prime genotype for precision oncology. The most prominent treatment approach in this context is the use of PARP inhibitors (PARPi).

PARP inhibitors for mCRPC

PARPi trap the enzymes, PARP1, and to lesser extent PARP2, onto the DNA [77]. These enzymes are involved in the resealing of single-stranded breaks in the DNA. It has been proposed that the mechanism of action behind PARPi results in a surplus of single-stranded breaks that cannot be restored correctly in HRD patients, leading to apoptosis and programmed cell death. However, the underlying mechanism of PARP inhibition is likely more complex and multifaceted. Several models have been proposed, including base excision repair impairment and the trapping of PARP enzymes on the DNA [77,78]. In 2009, the first evidence of PARPi efficacy in mCRPC was reported [79]. Olaparib was used to treat 60 patients with advanced solid cancers, including three with mCRPC. One mCRPC patient who harbored a BRCA2 aberration experienced a PSA50. This sparked interest in PARP inhibition for mCRPC. As monotherapy, four PARPi, olaparib, rucaparib, niraparib and talazoparib, have shown antitumor activity in mCRPC.

Olaparib

After the positive phase 2 trials, TOPARP-A for genetically unselected and TOPARP-B for genetically selected mCRPC patients, the phase 3 PROfound trial was initiated in 2017 [74, 80,81]. This included HRRm patients with deleterious aberrations in BRCA1, BRCA2, or ATM in cohort A, and patients with deleterious aberrations in other genes linked to HRR in cohort B. All patients had previously received at least one ARSI and were allowed to have received prior taxane chemotherapy. Patients were randomized in a 2:1 ratio to receive olaparib (300 mg tablets twice daily) or a second ARSI as control.

Of the 4425 patients enrolled for screening, 387 were HRRm and met the inclusion criteria. The median PFS was significantly longer with olaparib (5.8 months) than with a second ARSI (3.5 months) (p < 0.001). Cohort A showed the largest benefit in PFS, 7.4 versus 3.6 months, with BRCA1 and BRCA2 aberrations driving the superior outcomes (HR 0.41 and 0.21, respectively). Overall, no benefit was seen for those with ATM aberrations (HR 1.04). Responses in cohort B were not reported in the main article, but data reconstruction by Stopsack et al. revealed that there was no evidence that olaparib prolonged PFS in cohort B (HR 0.88, 95% CI 0.58–1.34) [82]. Aberrations in PPP2R2A were found to have a worse outcome with olaparib compared to an ARSI (HR 6.61). The PROfound study led to Food and Drug Administration (FDA) and European Medicines Agency (EMA) approval for the use of olaparib in genetically selected mCRPC patients who progressed on at least one line of ARSI.

Rucaparib

The phase 2 TRITON2 trial clearly identified that patients with aberrations in genes beyond BRCA showed considerably lower response rates to PARPi treatment when compared to BRCA-deficient patients [83,84]. The response rates observed in patients with deleterious BRCA aberrations in this trial were so promising that, in May 2020, the FDA granted accelerated approval for the use of rucaparib for mCRPC patients with deleterious BRCA aberrations who progressed on at least one ARSI and a taxane [85].

The recently published phase 3 trial, TRITON 3, randomized mCRPC patients with a deleterious aberration in BRCA1, BRCA2, or ATM between rucaparib (600 mg orally twice daily) or a physician’s choice of docetaxel or ARSI [86]. Eligible patients had progressed on one ARSI but did not receive a taxane in the castration-resistant setting (while 23% of patients had received docetaxel in the castration-sensitive setting). Again, the ATM subgroup did not show a significant benefit from treatment with a PARPi; the HR for progression or death was 0.95 (95% CI 0.59–1.52) for rucaparib versus control. In contrast, the BRCA subgroup had a significantly longer PFS of 11.2 months with rucaparib compared to 6.4 months with control (HR 0.50, p < 0.001). The radiologic response rate in the BRCA subgroup was 45% to rucaparib compared to 17% in the control group. Notably, the TRITON 3 trial differed from the PROfound trial in that docetaxel could be given in the control arm. Rucaparib still resulted in a reduced risk of progression or death in the BRCA subgroup compared to docetaxel (HR 0.53, 95% CI 0.37–0.77). OS data remained immature at the time of publication.

Niraparib and talazoparib

The more recent phase 2 GALAHAD and TALAPRO-1 trials investigated the use of niraparib and talazoparib, respectively, as monotherapy in patients with mCRPC [87,88]. The results of these trials again illustrated that HRRm patients with genetic aberrations in genes beyond BRCA showed lower response rates to treatment with PARPi when compared to those with BRCA aberrations. The overall outcomes were similar to those observed in the PROfound and TRITON2 trials, with comparable PFS and response rates for patients with BRCA mutations. Supportive of the role of ctDNA sequencing in mCRPC, the GALAHAD trial included 4043 patients who submitted plasma samples, and only 0.8% (n = 31) of results were unavailable because of technical failure [87]. A recent review analyzed data on the response to PARPi and ranked the potency of each PARPi from highest to lowest [89]; from in vitro data, the study found that talazoparib was the most potent PARPi, followed by rucaparib, olaparib, and lastly, niraparib. However, these putative differences in efficacy do not translate clearly to differences in objective response rates in the previously mentioned clinical studies.

The impact of gene alterations on PARP inhibitor response

Across these four trials, a pattern emerges that aberrations in non-BRCA genes associated with HRR confer less sensitivity to PARP inhibition than aberrations in BRCA1 or BRCA2. When combining multiple published studies, the responses to the four PARPi monotherapies can be used to categorize patients as BRCA-deficient and those with HRRm beyond BRCA (Table 1). When the results from the four PARPi studies were pooled, 54% (269/498) of the BRCA-deficient patients had a PSA50 and 42% (116/276) had a radiologic response, while, of the non-BRCA HRRm patients, only 9% (33/353) had a PSA50 and 9% (19/217) had a radiologic response. A recent study demonstrated superior response rates to PARPi in BRCA2-deficient mCRPC patients when compared to BRCA1-deficient patients, but potential confounding factors such as zygosity and concurrent TP53 mutations were identified [91,92]. Interestingly, in a biomarker analysis from the TOPARP-B trial, germline and somatic BRCA2 variants were found to be associated with similar benefit from olaparib, but homozygous BRCA2 loss yielded the greatest benefit [93]. These findings were confirmed by biomarker analyses from the TRITON2 and TALAPRO-1 trials [94,95]. Immunohistochemical loss of ATM was associated with a better outcome than genetically identified ATM alterations, probably because immunohistochemistry selected only homozygous losses [93, 96].

Table 1. PARP inhibitor monotherapies.

		Olaparib		Rucaparib		Niraparib		Talazoparib
Approval for mCRPC		EMA: BRCAm after ARSI FDA: HRRm^1 after ARSI		EMA: Not approved FDA: BRCAm after ARSI & taxane		EMA: Not approved FDA: Not approved		EMA: Not approved FDA: Not approved
Trial (Phase) [Ref]		PROfound (3) [74]		TRITON3 (3) [86]	TRITON2 (2) [84]	GALAHAD (2) [87]		TALAPRO-1 (2) [88]
Prior treatment lines		ARSI: ≥ 1^A^,^B,^C Taxane: allowed^A^,^B,^C		ARSI: 1^C Taxane: allowed^A	ARSI: 1 or 2^A^,^B,^C Taxane: 1^C	ARSI: ≥ 1^C Taxane: ≥ 1^C		ARSI: ≥ 1^C Taxane: ≥ 1^A^,^C
Main endpoint		rPFS		rPFS	ORR	ORR		ORR
Biomarker group		BRCAm^α	Non-BRCAm^2	BRCAm^δ	Non-BRCAm^3,^ζ	BRCAm	Non-BRCAm^4	BRCAm	Non-BRCAm^5^,^θ
Zygosity		Mono- and biallelic		Mono- and biallelic		Biallelic or germline		Mono- and biallelic
PSA50	n/total	58/94	15/151^β	111/201^ε	9/78	61/142	4/81	39/61	5/43
	Fraction	61.7%	9.9%	55.0%	11.5%	43.0%	4.9%	63.9%	11.6%
ORR	n/total	25/57	5/84^γ	37/82	6/43	26/76	5/47	28/61	3/43
	Fraction	43.9%	6.0%	45.1%	14.0%	34.2%	10.6%	45.9%	7.0%
rPFS	n	102	156	201	78	142	81	61	43
	Months	9.8	Min. cohort B: 4.8 Max. ATM: 5.4	11.2	Min. CDK12: 3.7 Max. other: 11.6^η	8.08	3.71	11.2	Min. other: 1.8 Max. PALB2: 5.6

Based on Slootbeek PHJ, Overbeek JK, Ligtenberg MJL, et al. PARPing up the right tree; an overview of PARP inhibitors for metastatic castration-resistant prostate cancer. Cancer Lett. 2023 2023/09/07/:216367. https://doi.org/10.1016/j.canlet.2023.216367.

Prior treatment lines:

^A For mCSPC.

^B For nmCRPC.

^C For mCRPC.

Gene panels (genes in bold are genes in which aberrations were identified):

¹ ATM, BARD1, BRCA1, BRCA2, BRIP1, CDK12, CHEK1, CHEK2, FANCL, PALB2, RAD51B, RAD51C, RAD51D, RAD54L.

² ATM, BARD1, BRIP1, CDK12, CHEK1, CHEK2, FANCL, PALB2, PPP2R2A, RAD51B, RAD51C, RAD51D, RAD54L.

³ ATM, BARD1, BRIP1, CDK12, CHEK2, FANCA, NBN, PALB2, RAD51, RAD51B, RAD51C, RAD51D, RAD54L.

⁴ ATM, BRIP1, CHEK2, FANCA, HDAC2, PALB2.

⁵ ATM, ATR, CHEK2, FANCA, MLH1, MRE11A, NBN, PALB2, RAD51C.

Manner of data collection:

^α Data from exploratory analysis from the abstract by De Bono et al. [90].

^β Calculated by subtracting Cohort A from the overall population of de Bono et al. [74] and adding the ATM mutated population from the abstract by De Bono et al. [90].

^γ Calculated by adding ATM mutated population from the abstract by De Bono et al. [90] to Cohort B, calculated by Stopsack et al. [82].

^δ ATM subgroup from TRITON3 had a PSA50 rate of 2/69 (3%, not separately presented in the article), ORR rate of 0/24 (0%) and rPFS of 8 months.

^ε A PSA50 response rate of 55% was reported by Fizazi et al. This could either imply 110 or 111 patients with a PSA50.

^ζ Counted from figure 2 from Abida et al. [84].

^η Not presented by Abida et al. [84], data retrieved from Study Results section of NCT page (NCT02952534). Median rPFS: ATM 5.3 months; CDK12 3.7 months; CHEK2 9.4 months; Other genes 11.6 months.

^θ Counted from table 2 from de Bono et al. [88].

ARSI, androgen receptor signaling inhibitor; BRCAm, BRCA mutated; EMA, European Medicines Agency; FDA, Food and Drug Administration; HRRm, homologous recombination repair mutated; mCSPC, metastatic castration-sensitive prostate cancer; mCRPC, metastatic castration-resistant prostate cancer; n, number of patients; nmCRPC, non-metastatic CRPC; ORR, objective radiologic response; PSA50, ≥50% prostate specific antigen decline; rPFS, radiologic progression-free survival.

Drawing definite conclusions on PARPi responsiveness is challenging when considering alterations in the many rarely mutated HRR genes. Risdon et al. bundled responses on PARPi for 11 genes associated with HRR [97]. They noted potential benefit in patients with a PALB2 aberration, but the data were insufficient to reach a definitive conclusion. Aberrations in CDK12 and NBN were considered to have a modest effect on PARPi susceptibility. For BRIP1, CHEK2, FANCA, RAD51B, and RAD51D, the data were too limited to draw conclusions. The list of genes that might also show PARPi sensitivity upon loss-of-function alterations, based on their physiological functions, is extensive. It includes but is not limited to BARD1, CHD1, HDAC2, MRE11, RAD50, RAD54L, and various Fanconi anemia genes. This shows that the role of each individual HRR gene in PARPi sensitivity is not yet fully understood and further research on predictive biomarkers is warranted.

Combination therapy

Co-inhibition of the AR and PARP can provide a synergistic anti-tumor effect and may offer a greater therapeutic advantage than either approach alone. The underlying biology of this combination has revealed a reciprocal relationship between AR signaling and the HRR pathway. AR signaling plays a vital role in promoting MRN foci formation, leading to ATM activation and efficient accumulation of activated H2AX and RAD51 foci, all of which promote HRR [98]. Conversely, blocking AR signaling results in the downregulation of many HRR-related genes, including BRCA1, RAD51C, and RAD54L, as well as less efficiently-activated H2AX foci [99,100]. These observations suggest that ADT impairs HRR, resulting in features resembling a “BRCAness” phenotype [99]. The downregulation of the HRR pathway necessitates the repair of double-stranded DNA damage through alternative DDR mechanisms, all of which begin with PARP. There is indeed an upregulation of PARP signaling in response to the blockade of the AR-axis; thereby, ADT has a synergistic effect, and is possibly synthetically lethal, when combined with PARP inhibition [98].

In addition to its role in DNA damage repair, PARP1 also influences the expression of tumor suppressors and oncogenes by binding to RNA polymerase II and modifying the transcription of hormone-dependent genes controlled by estrogen, progesterone, and AR [101,102]. PARP1's ability to increase AR binding to chromatin can result in the enhanced ability of AR to promote transcription and gene expression [103]. As a result, the reciprocal relationship between AR and PARP creates a synergistic effect when both are inhibited, with each treatment amplifying the effect of the other.

Currently, phase 3 trials are investigating the four PARPi that have been shown to be beneficial as monotherapies in combination with an ARSI in mCRPC patients (Table 2). The final results of the PROpel and MAGNITUDE trials investigating olaparib and niraparib, respectively, in combination with abiraterone and prednisone have both been published [106,107]. Only preliminary data on the combination of talazoparib and enzalutamide in the TALAPRO-2 study are currently available, while the first results of the CASPAR trial, which is examining the combination of rucaparib with enzalutamide, are yet to be reported [108,109]. The allowed prior treatment lines and main outcomes of these studies, stratified by HRR and BRCA mutational status, are shown in Table 2.

Table 2. PARP and AR co-inhibition.

Trial name		PROPEL [104]			MAGNITUDE			TALAPRO-2			CASPAR
PARPi, ARSI		Olaparib, Abiraterone			Niraparib, Abiraterone			Talazoparib, Enzalutamide			Rucaparib, Enzalutamide
NCT		NCT03732820			NCT03748641			NCT03395197			NCT04455750
Prior treatment lines		Abiraterone: Not allowed Enzalutamide: Allowed^A*^,^B Taxane: Allowed^A			Abiraterone: Allowed^B* Enzalutamide: Allowed^A,^B Taxane: Allowed^A			Abiraterone: Allowed^A Enzalutamide: Not allowed Taxane: Allowed^A			Abiraterone: Allowed^A,^B Enzalutamide: Not allowed Taxane: Allowed^A, B
Main endpoint		rPFS by investigator			rPFS by BICR			rPFS by BICR			rPFS by BICR & OS
Genetic subgroups		HRRm: 226/796 (28%)^1; BRCAm: 85/226 (38%)			HRRm: 423/423 (100%)^2; BRCAm: 225/423 (53%)			HRRm: 169/805 (21%)^3; BRCAm: 60/169 (36%)			^4
Treatment arm		PARPi + ARSI	ARSI	HR (95% CI) / OR [95% CI]	PARPi + ARSI	ARSI	HR (95% CI) / p-value	PARPi + ARSI	ARSI	HR (95% CI) / p-value	Results are not yet available
HRRm		111/399 (2%)	115/397 (29%)		212/212 (100%)	211/211 (100%)		85/402 (21%)	84/403 (21%)
ORR	Unselected	94/161 (58%)	77/160 (48%)	1.60 [1.02–2.53]	All patients were HRRm			74/120 (62%)	58/132 (44%)	p = 0.005
	HRRm	Not reported			55/92 (60%)	23/82 (28%)	p < 0.001	Not reported
	BRCAm	Not reported			29/56 (52%)	15/48 (31%)	p = 0.04	Not reported
rPFS ^α	Unselected	24.8 months	16.6 months	0.66 (0.54–0.81)	All patients were HRRm			Not reached	21.9 months	0.63 (0.51–0.78)
	HRRm	Not reached	13.9 months	0.50 (0.34–0.73)	16.5 months	13.7 months	0.73 (0.56–0.96)	27.9 months	16.4 months	0.46 (0.30–0.70)
	BRCAm	Not reached ^β	8.4 months ^β	0.23 (0.12–0.43)	16.6 months	10.9 months	0.53 (0.36–0.79)	Not reported
	HRRp	24.1 months	19.0 months	0.76 (0.60–0.97)	Closed due to futility: HR 1.09 (0.75–1.57)			Not reached	22.1 months	0.66 (0.49–0.91)

Based on Slootbeek PHJ, Overbeek JK, Ligtenberg MJL, et al. PARPing up the right tree; an overview of PARP inhibitors for metastatic castration-resistant prostate cancer. Cancer Lett. 2023 2023/09/07/:216367. https://doi.org/10.1016/j.canlet.2023.216367.

Prior treatment lines:

^A* For (m)CSPC but only without progression during treatment and stopped 12 months before randomization.

^A For (m)CSPC.

^B* For mCRPC with maximum treatment duration of ≤4 months prior to randomization.

^B For nmCRPC.

^C For mCRPC.

Gene panels:

¹ ATM, BRCA1, BRCA2, BARD1, BRIP1, CDK12, CHEK1, CHEK2, FANCL, PALB2, RAD51B, RAD51C, RAD51D, and RAD54L. Including 18 patients with unknown HRRm status.

² ATM, BRCA1, BRCA2, BRIP1, CDK12, CHEK2, FANCA, HDAC2, and PALB2.

³ ATM, ATR, BRCA1, BRCA2, CHEK2, FANCA, PALB2, RAD51C, NBN, MLH1, MRE11A, and CDK12. Including patients with unknown HRRm status.

⁴ A subgroup analysis will be performed on patients with pathogenic alterations in BRCA1, BRCA2, and PALB2.

Manner of data collection:.

^α rPFS reported according to the main endpoint of the study.

^β Not in main article. Data from the presentation by Dr. Saad at ESMO 2022 [105].

ARSI, Androgen receptor signaling inhibitor; BICR, blinded independent central review; BRCAm, BRCA mutated; CI, confidence interval; HR, hazard ratio; HRRm, homologous recombination repair mutated; HRRm, homologous recombination repair proficient; NCT, ClinicalTrials.gov identifier; OR odds ratio; PARPi, PARP inhibitor; rPFS, radiographic progression-free survival.

The outcomes from the three trials with reported results clearly indicate a hierarchy in the response of genetic subgroups to the combination of PARPi and an ARSI. The best responses were observed in BRCAm patients, followed by HRRm patients, and then the molecularly unselected mCRPC patients. Notably, where the non-HRRm arm of the MAGNITUDE was closed for futility, the molecularly unselected as well as the HRR-proficient patients in the PROpel trial did show a significant PFS benefit for the combination olaparib and abiraterone over abiraterone alone (Table 2). Final OS results are available only for the PROpel trial [110]. The OS benefit of the combination for the BRCAm subgroup was impressive, with an HR of 0.29 (95% CI 0.14–0.56), while the PFS benefit of the combination for molecularly unselected patients did not translate into a significant OS benefit (HR 0.81, p = 0.054). The first results for the combination rucaparib-enzalutamide and the final results for talazoparib-enzalutamide are awaited. For the latter combination, a second cohort comprising only HRRm patients, consisting of 399 patients, will be reported; this includes 169 HRRm patients from the initial 805 all-comers cohort and an additional 230 HRRm patients [108].

The combination of olaparib plus pembrolizumab was investigated in the phase 3 KEYLYNK-010 trial [111]. The included patients had progressed on one ARSI and docetaxel and were randomized between a different ARSI than was received before or olaparib plus pembrolizumab. The dual primary endpoints, radiographic PFS and OS, did not differ between both arms, with an HR for PFS of 1.02 (95% CI 0.82–1.25) and an HR for OS of 0.94 (95% CI 0.77–1.14). The combination of a PARPi plus radium-223 has not yet shown beneficial results in larger trials [112,113].

Platinum-based chemotherapy for mCRPC

While PARPi are FDA- and EMA-approved and are becoming readily available for HRRm mCRPC patients, platinum-based chemotherapy is still not standard of care in this molecularly selected subgroup of mCRPC patients, despite accumulating evidence of their effectiveness in the BRCA-altered group [104, 114–117]. Platinum compounds exert their anti-tumor activity through formation of covalent inter- and intra-strand adducts with cellular DNA, resulting in DNA damage [118]. The repair mechanism after platinum exposure relies on HRR, as indicated by an increase of RAD51 foci [119]. Therefore, the efficacy of platinum agents may be comparable with the efficacy of PARPi in HRRm mCRPC patients.

Platinum compounds have been tested alone or in combination with taxanes in unselected mCRPC patients. A meta-analysis of these trials showed significant benefits in the objective radiologic response rate and PSA50 rate, but no differences in PFS or OS [120]. In a study by Corn et al. adding carboplatin to cabazitaxel improved the median PFS from 4.5 to 7.3 months (HR 0.69, p = 0.018), but no OS benefit was observed [121].

Multiple case reports and series have suggested an association between HRRm status and clinically meaningful responses to platinum compounds in mCRPC patients [116, 122,123]. The largest cohort study describing responses to platinum-based chemotherapy in molecularly selected mCRPC patients was described by Schmid et al. [117]. In this analysis of 508 patients, 80 were identified as DDR- (including HRR) deficient, 98 were identified as DDR-proficient, and 330 were not tested for DDR status. A platinum-agent was prescribed as monotherapy in 20% of patients, while the remaining patients received a combination therapy, with 83% receiving docetaxel, paclitaxel, or etoposide. The use of platinum-based chemotherapy also varied by treatment line, with 33% administered in the first line, 29% in the second line, and 13% in the third line. The DDR-deficient patients had a PSA50 rate of 47% and radiologic response rate of 48%, compared to 36% (p = 0.200) and 31% (p = 0.070), respectively, for the DDR-proficient cohort. The proportion of patients with a PSA50 varied per aberration: 64% (23/44) for BRCA2, 0% (0/3) for BRCA1, 36% (4/11) for ATM, and 29% (6/21) for other aberrations. Moreover, 50% of BRCA2-deficient patients had a radiologic response.

Three retrospective cohort studies reported similar response rates (Table 3) [104, 114,115]. Two studies compared patients with HRRm to HRR-proficient patients and observed significant differences that favored the HRRm subgroup, with the best responses from BRCA2-deficient mCRPC patients. Pomerantz et al. compared only BRCA2-deficient patients to BRCA2 wildtype patients and demonstrated a marked difference in response rates.

Table 3. Platinum-based chemotherapy response by homologous recombination repair status.

		Mota et al. [114]			Slootbeek et al. [115]			Pomerantz et al. [104]
Prior therapies		Taxane naïve: none PARPi refractory: none			Taxane naïve: 2 out of 30 patients PARPi refractory: 3 out of 14 HRRm patients			Taxane naïve: 133 out of 144 patients PARPi refractory: none presumed^α
Type platinum agent and combination		Platinum agent		Combined with	Platinum agents		Combined with	Platinum agent	Combined with
		Carboplatin Cisplatin		Monotherapy: 33% Taxane: 64% Etoposide: 3%	Carboplatin: 100%		Monotherapy: 11% Docetaxel: 86% Cyclophosphamide: 3%	Carboplatin: 100%	Docetaxel: 100%
Genetic subgroup		BRCA2m	HRRm^1	HRRp	BRCA2m	HRRm^2	HRRp	BRCA2m^β	BRCA2p^β
PSA50	n/total	4/6	8/16	5/40	7/7	10/14	5/16	6/8	23/133
	Fraction	66.7%	50.0%	12.5%	100%	71.4%	31.3%	75.0%	17.3%
ORR	n/total	NR	NR	NR	7/7	7/12	3/14	NR	NR
	Fraction	NR	NR	NR	100%	58.3%	21.4%	NR	NR
OS	n	6	16	48	7	14	16	8	133
	Months	8.4	9.0	7.8	21.1	8.4	7.0	18.9	9.5

The genes in bold are found in patients enrolled in the corresponding trials.

¹ ATM, ATR, BRCA1, BRCA2, CDK12, CHEK2, FANCA, MRE11, NBN, PALB2, RAD50, RAD51 and RAD51C.

² ATM, ATR, BAP1, BARD1, BRCA1, BRCA2, BARD1, BRIP1, CDH1, CDK12, CHEK1, CHEK2, ERCC2, ERCC4, FANCA, FANCC, FANCD2, FANCE, FANCF, FANCG, FANCI, FANCL, FANCM, MRE11A, NBN, PALB2, PARP1, PARP2, PARP3, PPP2R2A, RAD51, RAD51B, RAD51C, RAD51D, RAD52, RAD54L and RPA1.

^α Due to inclusion period from 2001 to 2015. Prior treatment with PARP inhibitors not stated in the article.

^β Germline carriers.

BRCA2m, BRCA2 mutated (including homozygous loss); BRCA2p, BRCA2 proficient; HRRm, homologous recombination repair mutated; HRRp, homologous recombination repair proficient; n, number of patients; NR, not reported; ORR, objective radiologic response; OS, overall survival; PARPi, PARP inhibitor; PSA50, ≥50% prostate specific antigen decline.

Cross-resistance and reversion mutations

Cross-resistance between platinum-based chemotherapy and PARPi has been described in other tumor types and, in the wake of the approval of olaparib and rucaparib, information on cross-resistance is crucial in finding an optimal treatment sequence. A recent retrospective study reported same-patient responses of 28 mCRPC patients who received both PARPi and platinum-based chemotherapy [124]. In this cohort, PARPi induced more cross-resistance to platinum-based chemotherapy than platinum-based chemotherapy to PARPi. This difference was evident in terms of PFS, biochemical response and best radiographic response. Still, >40% of the cohort were responsive to either PARPi or platinum-based chemotherapy after receiving the alternative therapy. More information is needed on cross-resistance between these agents in mCRPC and on the optimal treatment sequence. The COBRA trial (NCT04038502) is comparing olaparib and carboplatin head-to-head in a cross-over design, and more information is anticipated from this study.

The topic on cross-resistance of HRD-targeting agents is being accelerated by the emergence of so called “reversion mutations” in mCRPC patients. These mutations restore the function of a protein mostly under selective pressure of a GMT. In mCRPC, it is often the function of BRCA2 that is restored during treatment with a PARPi or a platinum-based agent, as found in multiple case reports or series with small sample sizes [125–128]. Less commonly mutated genes such as PALB2 can also acquire reversion mutations. Reversion mutations are most commonly observed when the inactivating mutation is a frameshift mutation, where an additional insertion or deletion can restore the open reading frame; they are also seen after single-nucleotide variants, large rearrangements, or splice-site mutations that restore part of the function by exon-skipping [129–131].

The prevalence of these reversion mutations in mCRPC has only recently been elucidated in a ctDNA analysis from the TRITON2 trial [132]. In the final analysis, 100 patients who were BRCA-deficient and had progressed on rucaparib treatment were included. Plasma samples were collected at various timepoints, including prior to treatment with rucaparib and at progression. Of the 100 mCRPC patients, 23 had a homozygous BRCA loss, seemingly impossible to reverse. Post progression, 39 of the 100 patients had a BRCA reversion mutation, restoring the function of BRCA1 or BRCA2. Moreover, all reversions were, as expected, in the 77 patients with at least one (albeit mutated) allele present. This means that a reversion mutation occurred in 39 of the 77 patients (51%) with at least one BRCA1 or BRCA2 allele present. Preliminary data from the phase 2 GALAHAD trial (niraparib) showed an even higher occurrence of reversion mutations. Excluding homozygous deletions, reversion mutations were found in 28 of the 33 patients (85%) who donated an end-of-treatment ctDNA sample [133].

In the data from both the GALAHAD and TRITON2 trials, the presence of at least one reversion mutation at progression was associated with better response on the PARPi. As Vanderkerkhove proposes, these counterintuitive results indicate that if the mutation in BRCA1 or BRCA2 is truly a driver mutation, there will be response to a PARPi, with relatively slow progression, and the development of a reversion mutation [134].

HRRm and checkpoint inhibitors

In the last decade, the use of immune system-aiding agents, specifically checkpoint inhibitors, has advanced in the treatment of solid cancers. These inhibitors target proteins such as cytotoxic T-lymphocyte-associated protein 4 (CTLA-4), programmed cell death protein 1 (PD-1) and its ligand (PD-L1), that are utilized by tumors to evade immune surveillance [135]. Pembrolizumab, a PD-1 inhibitor, was the first cancer-agnostic therapy registered for MMRd patients [136].

Cancers with high somatic mutation frequencies or immune-friendly tumor microenvironments, such as melanoma, bladder cancer, and renal cell carcinoma, are often responsive to checkpoint inhibitors [137,138]. With mCRPC patients having a relatively low mutational burden and a generally immunosuppressive tumor microenvironment overall, the clinical trials with checkpoint inhibitors for unselected mCRPC cohorts unsurprisingly showed limited effect [139–143]. HRRm tumors tend to have more somatic gene aberrations, and the resulting neoantigens can be recognized by the immune system [135]. HRRm also activates the STimulator of INterferon Genes (STING) pathway, enhancing T cell responses against neoantigens and potentially increasing susceptibility to checkpoint inhibitors. [144–148].

Indeed, in three clinical trials, KEYNOTE-199, CheckMate 650, and STARVE-PC, HRRm mCRPC patients demonstrated increased response rates, as shown in a pooled analysis of their responses (Table 4) [142,143, 149]. Although the number of patients with a PFS of ≥6 months was limited because of the absence of this data in KEYNOTE-199, the proportion of patients in the HRRm group was still significantly higher than in the HRR-proficient patients. Furthermore, the HRRm subgroup showed a significantly higher rate of patients with a PSA50 compared to the HRR-proficient subgroup. A borderline significant difference that favored HRRm patients was observed in the radiographic response. It is noteworthy that none of these studies excluded MMRd cancers that lead to MSI. However, in the KEYNOTE study, none of the responders were MSI/MMRd, while in the STARVE-PC trial, none of the patients who were tested for MSI/MMRd status were MSI/MMRd. The MSI/MMRd status of patients in the CheckMate 650 trial was not reported.

Table 4. Response om immune checkpoint inhibitor by homologous recombination repair status.

	n, total	HRR proficient		HRR mutated		p-value
		n	Fraction	n	Fraction
PSA50	205	11/163	6.7%	9/42	21.4%	0.008
ORR	178	11/143	7.7%	7/35	20.0%	0.054
PFS6	59	12/46	26.1%	8/13	61.5%	0.022

Genes associated with HRR per included trial. The genes in bold are found in patients enrolled in the corresponding trials.

KEYNOTE-199^a [141]: ATM, BARD1, BRCA1, BRCA2, BRIP1, CDK12, CHEK1, CHEK2, FANCL, PALB2, PPP2R2A, RAD51B, RAD51C, RAD51D and RAD54L.

CheckMate 650 [142]: ATM, BARD1, BRCA1, BRCA2, BRIP1, CDK12, CHEK2, FANCA, NBN, PALB2, RAD51, RAD51B, RAD51C, RAD51D and RAD54L.

STARVE-PC^b [143]: all DDR related genes according to Mateo et al. [80]. Genes identified: ATM, BRCA2, ERCC4, FANCM.

^a The KEYNOTE-199 study did not provide detailed information on which genes were altered.

^b The limitation of this study is the range of sequencing panels used in which several HRR-related genes are missing: CancerSELECT-R^TM 88 (n = 3), CancerSELECT^TM 125 (n = 1), CancerSELECT^TM 203 (n = 6), PlasmaSELECT^TM 64 (n = 4), Foundation® Panel (n = 1).

HRR, homologous recombination repair; n, number of patients; ORR, objective radiologic response; PFS6, progression-free survival of ≥6 months; PSA50, ≥50% prostate specific antigen decline.

CDK12 stands out among the HRR-related genes because of its distinct characteristics. CDK12 aberrations are associated with poor prognosis and show only limited response to PARPi [150]. At a genomic level, CDK12-altered tumors do not resemble typical HRD. Instead, they display a focal tandem duplicator phenotype and a high gene fusion burden that result in abundant fusion-induced neoantigens. The predicted total neoantigen burden in CDK12-deficient tumors approaches that of MSI PCa. CDK12 aberrations increase immune infiltration and expand T-cell clones, implying that there is a higher susceptibility to checkpoint inhibitors. However, this has been demonstrated only in small retrospective studies [151,152].

Checkpoint inhibitors for HRRm or CDK12-eficient mCRPC, either as monotherapy or as combined therapy, are being investigated in several trials (NCT04104893, NCT03506997, NCT03040791, NCT04717154, NCT03570619, NCT03061539, and NCT03554317).

HRRm and other inhibitors

Within the complex double-stranded DDR pathway, which includes HRR, loss-of-function alterations in many genes do not always lead to aberrant pathway activity because of dispensable or redundant function of individual genes and alternative pathways of double-stranded DNA repair. Therefore, these gene aberrations are not all targeted effectively by PARPi. In particular, the DDR pathway through ATM, with CHK2 (encoded by CHEK2) serving as an effector, as well as the pathway through ATR, with CHK1 (encoded by CHEK1) as effector, may best be targeted with different agents. Several studies, both in vitro and in vivo, have shown that mCRPC patients with pathogenic aberrations in ATM have inferior responses to PARP inhibition compared to those with BRCA mutations [153–155]. While ATM inhibitors are being investigated, these studies are only in early phases and have not yet included PCa patients [156]. A first-in-human study of an ATR inhibitor, conducted in 2021, included four mCRPC patients [157]. Two phase 1/2 trials that are evaluating the use of ATR inhibitors in PCa patients are now underway (NCT03682289, NCT03787680) [158].

The downstream effectors, CHK1 and CHK2, can also be targeted. Currently, the assessment of CHK2 inhibitors is limited to in vitro studies. A phase 2 clinical trial examining a dual CHK1/CHK2 inhibitor was prematurely terminated (NCT02203513). The results of the phase 1/2 trial evaluating the selective CHK1 inhibitor, SRA737, have not been published but are available on ClinicalTrials.gov. It is important to exercise caution when interpreting these preliminary results: 13 out of 16 mCRPC patients were available for radiographic response. None of these patients witnessed a complete or partial response, according to RECIST 1.1 criteria, and 8 (62%) had stable disease (NCT02797964) [159].

Radioligand therapies

Radium-223 is a therapeutic option for mCRPC patients with symptomatic bone-only disease [10]. Although radium-223 is not a GMT, it has been found to be effective in HRRm mCRPC patients, probably because of its ability to induce DSB, which is similar that of platinum-based chemotherapy [160,161]. In a study with 93 mCRPC patients who underwent molecular profiling, 30% had a pathogenic aberration, most commonly in BRCA1/2 (30%), ATM (8.6%), and CDK12 (4.3%) [161]. There was a significant OS benefit for HRRm patients (median 36.3 versus 17.0 months; HR 2.29; p = 0.011) with a particularly strong benefit seen in those with BRCA1/2 aberrations (median 36.8 months versus 20.5 months; HR 2.73; p = 0.038). There was no significant difference in either alkaline phosphatase or PSA response. Conversely, the largest cohort of 127 genetically profiled and mCRPC patients treated with radium-223 showed that 15% had a pathogenic aberration in BRCA1/2, 12% in ATM, and 2% in CDK12 [162]. Notably, no aberrations were found in other HRR-related genes. Again, none of the molecular subgroups demonstrated a significant difference in biochemical response, but in this study, PFS and OS also did not differ significantly between subgroups.

Prostate-specific membrane antigen (PSMA)-radioligand therapy (PSMA-RLT) targets PSMA and is labeled with radioisotopes such as the beta-emitter, lutetium-177 (Lu-177), or the alpha-emitter, actinium-225 (Ac-225) [163,164]. The phase 3 VISION trial evaluated the response on PSMA-Lu-177 in patients with mCRPC after treatment with at least one ARSI and at least one taxane-based chemotherapy and found a > 5 months prolonged PFS and a 4 months prolonged OS when PSMA-Lu-177 was added to the standard of care [165]. This led to the FDA approval of PSMA-Lu-177 on 23 March 2022.

PSMA-RLT response is dependent on PSMA expression, which is higher in HRD mCRPC patients [166,167]. This increased PSMA expression may be an adaptive cellular response to HRD, because HRD causes replication stress and thereby increases the demand for several enzymatic products of PSMA. Consequently, HRRm mCRPC patients may exhibit stronger responses to PSMA-RLT, as observed in a case report [168]. The largest cohort of patients who received PSMA-RLT and underwent genetic profiling included 40 patients [169]. No differences in PFS or biochemical responses rates were observed between patients with and without DDR gene defects or BRCA2 defects specifically. A study specifically focusing on patients treated with PSMA-Ac-225 found two deleterious BRCA1 aberrations in 13 patients; these patients had a longer OS (16.1 vs. 7.6 months). Because Ac-225 is an alpha-emitter, it could be more beneficial than Lu-117 in HRRm mCRPC patients, because alpha radiation causes more DSBs compared to beta radiation [170].

HRR alteration detection in liquid biopsies

HRR status can be analyzed using solid and liquid biopsies. The importance of liquid biopsy testing in addition to tissue biopsy testing is shown by the PROpel trial in which 23% (186/796) of patients’ tissue biopsies could not be performed and only liquid biopsies were available [106]. Also, in the TRITON2 and GALAHAD trials, liquid biopsies were assessed when tissue was unavailable [83,84, 87].

Challenges associated with liquid biopsies, especially for detecting HRR alterations, include the occurrence of clonal hematopoiesis (false positive results) and the detection of alterations in low ctDNA fractions, particularly homozygous deletions (false negative results). Jensen et al. reported that 10% of mCRPC patients had clonal hematopoiesis of indeterminate potential (CHIP) variants with >2% allelic fraction in genes associated with DDR, including BRCA2 [171]. Moreover, as the prevalence of CHIP increases with age, false positive mutation calling in the plasma of mCRPC patients, who are often >70 years old, should be taken seriously [172]. In addition to false positive results, false negative results can occur. Tukachinsky et al. reported concordant BRCA alteration detection between tissue and plasma in 67/92 (73%) patients with a BRCA alteration detected in either specimen [47]. Only 5/92 (5%) of alterations were found exclusively in tissue, with 4/5 (80%) patients having a ctDNA fraction <1%. Nevertheless, this study did not report on BRCA copy losses. Chi et al. reported a lower concordance between tissue and liquid biopsies in the PROfound study, which included information on copy losses, with 146/258 (57%) BRCA and ATM alterations detected in both specimen [173]. Only 8/30 (27%) homozygous losses were detected in ctDNA, with low ctDNA fraction being the limiting factor. Consequently, low ctDNA fraction highly impacts false negative results and such test results should be treated with caution.

Interestingly, Tukachinsky et al. [47] and Chi et al. [173] reported 20/92 (22%) and 61/258 (24%) BRCA/ATM alterations, respectively, to be detected exclusively in ctDNA. These variants were more likely to be subclonal and might therefore be related to alterations arising later in the tumor evolution. Liquid biopsies are also crucial to monitor the emergence of the previously discussed reversion mutations during treatment with HRD-targeting agents, because repetitive tissue biopsies are impractical.

Finally, liquid biopsies may also be prognostic of outcome. For PARPi in particular, a decrease in cfDNA concentrations or CTCs after 8 weeks of olaparib in mCRPC was associated with better outcomes independent of prognostic clinical characteristics [128].

Mismatch repair

Within the DDR machinery, MMR is another key druggable pathway. MMR facilitates the repair of erroneous insertions/deletions or mismatched nucleotides in the DNA that arise from exogenous or endogenous damage, mismatch following other repair mechanisms or, most often, replication errors [174]. A dysfunctional MMR results in an increased mutation rate.

MMR consists of at least two key proteins of the Mut class: a detector (MutSα) and a recruiter (MutLα) [97]. MutSα is a dimer composed of the two MutS homologs (MSH), MSH2 and MSH6, and is responsible for identifying the mismatches and initiating the MMR process. MutLα is a dimer composed of MutL homolog 1 (MLH1) and PMS2, and has endonuclease activity [175]. MutSα activates MutLα, which then incises and removes the erroneous strand through exonuclease and restores the mismatch in an error-free manner.

Deficiency of the canonical MMR proteins, MSH2, MSH6, MLH1, and PMS2, leads to a hypermutator phenotype in which frequent frameshift mutations are observed in coding and non-coding repetitive DNA sequences. These repetitive DNA sequences are called microsatellites and due to the frequent frameshift mutations, they have variable sizes, which are called MSI. The promoter of MSH2 may also become hypermethylated as a result of deleterious changes in the last exons of the EPCAM gene, leading to epigenetic inactivation of MSH2 and thus MSI [176]. MSI is considered a surrogate marker for deficient MMR machinery. The abundance of mutations in MSI/MMRd tumors lead to more tumor-specific neoantigens and can provoke an increased immune response [175]. Consequently, MSI mCRPC is immunologically more “hot” than MSI stable mCRPC.

Aberrations in MMR genes can be either germline or somatic. It is well known that Lynch syndrome, characterized by germline aberrations in MMR genes, increases the risk for various types of cancer, including colorectal, endometrial, ovarian, and upper tract urothelial carcinomas. However, recent studies have suggested that individuals with mutations in the genes, MSH2 and MSH6, may also be at an elevated risk for PCa [177–179].

The prevalence of deleterious aberrations in the canonical MMR genes in advanced PCa is estimated to be around 5% [180]. Most aberrations are thought to be somatic. In the landmark publication by Robinson et al. somatic mutations in MLH1 and MSH2 were found in 0.7 and 2.0% of 150 mCRPC patients, respectively [24]. Antonarakis et al. reported a pathogenic MMR aberration in 4.2% of 236 PCa patients undergoing somatic sequencing and in 0.9% of 348 PCa patients undergoing germline sequencing [62]. In a study with 3338 PCa patients, 0.7% had a germline alteration in one of the canonical MMR genes, resulting in an overall MMR deficiency prevalence of 3.6% when combined with somatic sequencing results [181]. Abida et al. identified 2.4% of 1033 PCa patients with MSI/MMRd from somatic origin and 0.7% from germline origin [64]. Ritch et al. detected MMR deficiency in 3.7% of 433 metastatic PCa patients through ctDNA sequencing, all confirmed by immunohistochemistry [65].

Response to immune checkpoint inhibitors

MMRd cancers have an abundance of neoantigens because of nucleotide mismatches and frameshift mutations. As stated earlier, these neoantigens can be detected by the immune system as non-self, and therefore could enhance immune responses and result in better outcomes to immune therapies. Subsets of mCRPC patients have demonstrated remarkable responses to various checkpoint inhibitors. Follow-up studies have shown that these subsets were enriched in patients with MMRd tumors. These studies have limited numbers of responders because of the paucity of MMRd in mCRPC. A summary of studies involving checkpoint inhibitors for MMRd mCRPC patients is presented in Table 5. In total, 56 patients from 5 studies were combined for PSA50 evaluation and 31 (55%) showed a PSA50. Regarding radiographic response, 17 patients, of which 10 (59%) showed an objective radiographic response. were included.

Table 5. Response of mismatch repair deficient patients on immune checkpoint inhibitors.

	Type of ICI	Pts analyzed for MMRd	MMRd pts		With ICI	PSA50		PSA99		ORR
			Pts	Fraction	Pts	Pts	Fraction	Pts	Fraction	Pts	Fraction
Abida et al. [64]	Anti-PD-L1 (n = 4), anti-PD-1 (n = 7)	1346	32	2.4%	11	6/11	54.5%	4/11	36.4%	4/8	50.0%
Antonarakis et al. [62]	Anti-PD-1	584	13	2.3%	4	2/4	50.0%	NR	NR	3/4	75.0%
Graham et al. [63]	Anti-PD-1	NR	27	NR	17	8 /15	53.3%	5/15*	33.3%	NR	NR
Barata et al. [182]	Anti-PD-1	460	15	3.3%	9	4/9	44.4%	3/9	33.3%	3/5	60.0%
Sena et al. [183]	Anti-PD-1	NR	65	NR	19	11/17	64.7%	4/17*	23.5%	NR	NR
Total						31/56	55.4%	16/52*	30.8%	10/17	58.8%

* Estimated.

ICI, immune checkpoint inhibitors; MMRd, mismatch repair deficient; NR, not reported; ORR, objective radiologic response; PD-1, programmed cell death protein 1; PD-L1, Programmed cell death protein ligand 1; PSA50, ≥50% prostate specific antigen decline; PSA90, ≥90% prostate specific antigen decline; Pts, number of patients.

In a study involving 19 patients with MMRd mCRPC, the response to anti-PD-1 therapy was most strongly correlated with the proportion of frameshift mutations [183]. This correlation was even stronger than the correlation with tumor mutational burden. MMRd results in a higher proportion of frameshift mutations than HRD, which explains why responses to checkpoint inhibitors are better for MMRd than HRRm mCRPC patients. As of now, pembrolizumab (a PD-L1 inhibitor) is the only approved checkpoint inhibitor based on MSI/MMR status.

In addition to MMR genes and those linked to HRR, other genetic factors may play a role in determining a patient’s response to checkpoint inhibitor therapy. Further research is needed to fully understand the impact of these other genetic aberrations, which may include the aforementioned aberrations in CDK12, mutations in the exonuclease domain of the DNA polymerase genes, POLE and POLD1, and deletion of the 3′ untranslated region of CD274 (encoding for PD-L1) [180].

The question remains whether mCRPC patients with co-occurrence of MSI and pathogenic BRCA aberrations should receive treatment with PARPi or checkpoint inhibitors. A recent study evaluating this co-occurrence in 213,199 genetically profiled cancer patients [184] found that BRCA alterations were more often present in MSI patients than microsatellite stable patients (20 vs. 5%, p < 0.001). Limited cases of BRCA mutations in MSI patients were identified for breast and ovarian cancer, but in PCa, 13% of BRCA1 and 3% of BRCA2 aberrations co-occurred with MSI. Analysis of all 14 genes from the PROfound trial revealed an even higher overlap, with 46% of MSI PCa samples harboring at least one alteration in a gene associated with HRR. However, when examining HRD at a functional level and considering genomic loss of heterozygosity ≥ 16% to be HRD, MSI and HRD were found to be mutually exclusive across all samples, with only 0.07% showing co-occurrence. In analyzing only PCa samples, the same mutual exclusiveness was found. The BRCA alterations co-occurring with MSI were thus most likely (mono-allelic) bystander mutations. The study also reported two mCRPC patients with concurrent MSI and HRRm; neither of these patients showed a PSA response after PARP inhibition whereas both showed a PSA response after PD-1 inhibition. Hence, it is important to avoid a narrow focus on a single pathway and instead to consider the broader framework of multiple aberrant pathways to elucidate the driver mechanism of the cancer and to determine the best therapy.

MMRd detection in liquid biopsies

Although the prevalence of MMRd is low for PCa, the clinical implications are important. MSI status, tumor mutational load and mutational signature analysis can all be identified using ctDNA [65, 185–187]. Willis et al. showed that ctDNA-based MSI status evaluation has a high sensitivity (86.6%) and specificity (99.5%) in plasma samples with at least 0.1% ctDNA fraction [187]. That study used a ctDNA assay that integrated putative microsatellite loci selected to encompass sites prone to instability. Although the threshold for a relevant tumor mutational load may be different in plasma compared to tissue, Ritch et al. reported an average plasma tumor mutational load for metastatic PCa patients that was similar to that suggested by previous tissue-based data [24, 65, 188]. Out of 433 patients, 24 had >11 mutations per Mb, of which 16 were confirmed MMRd, 4 were a deleterious BRCA2 alteration and 2 were localized hyper­mutations.

The clonal diversity in patients with MMRd is high, with low concordance between same patient tissue biopsy samples [65]. With a high rate of newly acquired mutations over time, a single primary tissue sample would not be representative of posttreatment metastatic PCa genomic features. CtDNA could provide more insight into the dynamic landscape of MMR-deficient tumors.

Androgen receptor axis

PCa is primarily androgen-sensitive, and endocrine therapies that target the AR have been developed to treat advanced cases. The AR is a transcription factor that regulates gene expression in response to androgens such as testosterone and dihydrotestosterone. Its upregulation drives the uncontrolled proliferation of PCa cells [189]. The AR consists of three distinct domains: an N-terminal domain, a highly conserved DNA-binding domain, and a ligand-binding domain. The previously-held belief that mCRPC is independent of AR signaling has been disproven. Residual androgens and the AR itself continue to play a significant role in both the progression to castration-resistant PCa and its growth [190]. The occurrence of AR amplification in late-stage mCRPC tumors, but not in pre-ADT patient-matched samples, serves as early evidence of AR's contribution to the progression to castration-resistance [191]. The ongoing dependence on the AR-axis has led to the development of novel ARSIs such as abiraterone, enzalutamide, darolutamide, and apalutamide [192]. Treatment with an ARSI is considered standard of care for unselected mCRPC patients and will therefore not be discussed in this review. However, resistance to ARSI is most often caused by aberrations in AR. Precision oncology in this context entails knowing when response to an ARSI may be limited. To counteract the resistance mechanism, new third-generation ARSIs that show promise in patients with specific AR alterations are being developed, although they are still in their infancy. AR aberrations are becoming a crucial aspect in the field of precision medicine for PCa.

Resistance mechanisms

There are three main mechanisms that cause resistance to ARSI: restoring AR signaling, bypassing AR, and achieving AR independence.

Restoring AR signaling

AR mutations, found in 20-25% of mCRPC cases, have been shown to contribute to treatment resistance and disease progression [193,194]. Recurrent hotspot mutations in AR, such as T878A, H875Y, F877L, and W742C, have been reported to have clinical significance in the development of resistance to ARSI and can be detected in ctDNA [194–196]. AR overexpression due to gene amplification is present in up to 50% of mCRPC patients [24]. It enables PCa cells to maintain cell growth in low-androgen conditions [197]. Indeed, the detection of AR amplification in ctDNA has been linked to resistance to ARSI [196, 198,199]. A meta-analysis showed that PFS and OS in patients on ARSI were significantly shorter in the presence of AR gene amplifications detected in ctDNA (HR 2.33 and HR 3.82, respectively) [200].

Alternative splicing of AR results in numerous isoforms. The splice-variants typically lack the ligand binding domain but retain the N-terminal and DNA binding domains, making them constitutively active in the absence of androgens and capable of promoting PCa cell survival [201,202]. AR-V7, detectable in liquid biopsies, is the most widely studied AR splice variant and has been linked to resistance to ARSI [203].

The residual androgens after ADT and ARSI treatment may still contribute to PCa growth. This is particularly true in cases with the 1245A>C polymorphism or other gain-of-function variants in the gene encoding for the enzyme, HSD3B1, which leads to an increase in the metabolic conversion of dehydroepiandrosterone to intraprostatic dihydrotestosterone. In the presence of a homozygous 1245A>C variant, the clinical benefit of ARSI is limited [204,205], although one study suggested that response to abiraterone was better for heterozygous HSD3B1 1245A>C carriers when compared to wildtype HSD3B1 patients [206]. This highlights the importance of inhibiting androgen biosynthesis above the level of CYP17A1. However, these variants did not surpass standard clinical variables as a predictive marker for response on ARSI.

Bypassing AR

The glucocorticoid receptor is upregulated in response to treatment with enzalutamide, and glucocorticoids have been implicated in bypassing the AR-axis through glucocorticoid receptor upregulation, thereby maintaining expression of AR target genes despite AR blockade [207,208]. Other steroid hormone nuclear receptors, such as the progesterone and mineralocorticoid receptors, have also been linked to resistance to ARSI, but the evidence supporting the role of these receptors is less compelling [195]. Targeting these steroid hormone nuclear receptors in PCa as a therapeutic strategy is difficult because most are essential for life. Bypassing AR signaling also occurs through reciprocal activation of the PI3K pathway (discussed below).

AR independence

Aggressive variant PCa (AVPC) constitutes a complex spectrum of phenotypes. These atypical and aggressive PCas have been associated with small cell/neuroendocrine PCa and anaplastic PCa [209]. AVPC is frequently AR-negative and thus independent of AR. Molecularly, this subtype is characterized by combined aberrations in RB1, TP53, and/or PTEN [210]. In a post hoc analysis of the earlier described phase 1–2 study by Corn et al. that compared cabazitaxel alone with cabazitaxel plus carboplatin, the authors found that patients with a molecular AVPC signature, identified by ctDNA sequencing, benefited more from the addition of carboplatin to cabazitaxel than unselected patients [121]. Patients with AVPC had an increased median PFS from cabazitaxel versus the combination (2.2 vs. 5.1 months, p = 0.030) while non-AVPC mCRPC patients did not show a significant increase in PFS (5.7 vs. 6.0 months, p = 0.65).

Novel AR-axis targeting therapies

Although the four second-generation ARSIs have improved patient outcomes, their efficacy is limited when administered sequentially because of the resistance mechanisms discussed above [211]. Therefore, efforts are ongoing to develop third-generation ARSIs that can reverse or overcome resistance to the second-generation agents.

ODM-208, a promising new drug, has the potential to block steroidogenesis higher up than abiraterone by inhibiting CYP11A1, an enzyme that converts cholesterol to pregnenolone. ODM-208 is being tested in the phase 2 CYPIDES trial, with preliminary results showing 68% of mCRPC patients with AR ligand binding domain mutations experiencing a PSA50, while only 8% of those without such mutations had a similar response [212]. ARV-110 is an oral, first-in-class PROteolysis TArgeting Chimera (PROTAC) protein degrader that selectively targets the AR and has shown improved efficacy in patients with specific AR mutations. This led to the initiation of the phase 2 ARDENT trial in which 46% of the 26 patients with a T878A/S and/or H875Y mutation had a PSA50 compared to 10% of 114 patients without these mutations [213]. TAS3681 downregulates AR and AR splice variants and has demonstrated antitumor activity in preclinical castration-resistant PCa models with AR splice variants or enzalutamide-resistance. In an ongoing phase 1 trial, some responses have been observed, but the data are still limited [214].

Another strategy to overcome resistance due to splice variants is to develop drugs that bind to the N-terminal domain instead of the ligand binding domain. Such drugs can overcome constitutive activity of the AR as they interfere with AR’s regulatory region instead of the ligand binding region. EPI-506 and EPI-7386 are two inhibitors targeting the AR N-terminus. A phase I trial with the first-generation EPI-506 showed only minor PSA declines in mCRPC patients, with none of the 32 patients having a PSA50 [215]. EPI-7386, a more potent and stable N-terminus inhibitor, is currently being investigated in a phase I trial (NCT04421222). Niclosamide, an anti-helminthic drug, has shown promise in overcoming resistance due to the AR-V7 splice variant and synergy with abiraterone. A phase 1 study combining niclosamide and abiraterone resulted in a PSA50 and radiographic response in 5 of the 8 evaluable patients [216]; however, this effect could be solely from abiraterone because the patients were not selected for AR-V7 splice variants. A phase 1 study of niclosamide in combination with enzalutamide was prematurely closed for futility [217]. A phase 2 study investigating its combination with abiraterone in CRPC is underway (NCT02807805).

Aside from these methods, other avenues to target the AR-axis, such as targeting AR chaperones, AR cofactors, and key components of the spliceosome, are being pursued. Other novel agents that could reverse endocrine resistance by targeting the glucocorticoid or progesterone receptor are being developed [195, 218,219].

AR-axis alteration detection in liquid biopsies

For the evolutionary assessment of AR resistance mechanisms, repeated sampling is necessary during treatment or at progression. Utilizing serial ctDNA samples before and after progression on ARSI, Annala et al. showed that the AR genotype continuously evolved during sequential lines of AR inhibition, with an increase in AR mutations per sequential line of treatment. These results were corroborated by the observation that a greater proportion of patients had AR amplifications in ctDNA in later lines of treatment compared to first-line treatment [200]. While the AR genotype is continuously evolving, the effect of AR alterations on ARSI outcomes is similar across treatment lines [42, 200]. Although the association of AR alterations on other therapy types such as chemotherapy is less well described, patients with AR alterations might benefit more from chemotherapy, highlighting the potential predictive value of liquid biopsies [45, 200]. Interestingly, Nakazawa et al. observed changes from AR-V7-positive CTC to AR-V7-negative CTC status in patients treated with chemotherapy, but not with ARSI [220]. In a study analyzing ctDNA of 168 mCRPC patients prior to ARSI initiation, the authors found that TP53 status outperformed AR aberrations and splice variant expression as a prognostic factor for a short PFS on ARSI [221]. Liquid biopsies can also be used to identify AVPC patients based on unique morphologic characteristics of CTCs [222].

In addition to the identification of resistance mechanisms, liquid biopsies may also serve as a prognosticating tool and for response measurements. Several groups have shown the prognostic value of ctDNA and CTC quantity at the start of ARSI and other treatment types [43, 45, 223–225]. Additionally, changes in ctDNA and CTCs have been shown to predict durability of response to treatment [128, 226–228].

Alternative targets

PI3K-AKT-mTOR

The PI3K-AKT-mTOR signaling pathway plays a crucial role in regulating cell growth and survival [229]. It starts with the activation of the phosphatidylinositol 3-kinase (PI3K) enzyme through a broad range of upstream cell surface receptors coupled to specific PI3K isoforms. The gene, PIK3CA, encodes for the p110α subunit of PI3K, and mutations in this gene can increase the activity of PI3K. The activation of PI3K leads to the recruitment and activation of other signaling proteins such as members of the AKT sub-family (AKT1, AKT2, and AKT3). Once activated, AKT can then phosphorylate several downstream effectors such as mTOR. In normal cells, the PI3K pathway is regulated by several negative feedback mechanisms. PTEN is a key protein here as it dephosphorylates one of the messenger molecules and thereby inhibits the pathway. Hyperactivation of this pathway, observed in 49% of the mCRPC patients, can lead to uncontrolled cell division and survival, promoting the development and progression of the disease [24]. The most common alterations are loss-of-function aberrations in PTEN (present in 41% of mCRPC patients) as well as activating alterations in PIK3CA and AKT1. Moreover, the PI3K pathway is activated by reciprocal crosstalk with the AR pathway and thus can function as a resistance mechanism to ARSIs [230]. Co-targeting the AR-axis and PI3K pathway may potentially lead to enhanced responses compared to AR-axis targeting alone.

This concept was tested in the phase 3 IPATential150 trial, where 1101 mCRPC patients were randomized to receive first-line treatment with abiraterone and prednisolone or abiraterone and prednisolone plus ipatasertib, a selective inhibitor of all three AKT isoforms [231]. The median PFS was 19.2 months in the ipatasertib-abiraterone group and 16.6 months in the placebo-abiraterone group, though this difference was not statistically significant at the prespecified α of 0.01 (HR 0.84, p = 0.043). The objective response rate was higher in the ipatasertib-abiraterone group (61%) compared to the placebo-abiraterone group (44%). By immunochemistry, 521 patients had a PTEN loss. The median PFS in these patients was 18.5 months in the ipatasertib-abiraterone group and 16.5 months in the placebo-abiraterone group (HR 0.77, p = 0.034), and the objective response rate was 61 versus 39%, respectively. In assessing the PI3k pathway for pathogenic PIK3CA and AKT aberrations, the HR for progression or death was 0.63. OS data were immature at the time of data cutoff.

Other inhibitors of the PI3K pathway as monotherapy or in combination were also investigated in early phase trials, as summarized by Pungsrinont et al. [232].

Other genetically matched kinase inhibitors

MAPK pathway hyperactivity was observed in mCRPC, with 32% of patients showing amplifications in MAPK pathway genes [233]. Trametinib, a MEK (MAPK/ERK kinase) inhibitor, demonstrated promising results in case reports with mCRPC patients, including one patient who had a PSA decrease of 93% after five months of treatment with trametinib and remained stable until a fatal stroke 18 months later [233]. Another case report showed a PSA decline from over 900 ng/mL to under 0.6 ng/mL within two weeks on trametinib in a patient with a SND1-BRAF gene fusion [234]. Ongoing trials are investigating trametinib in mCRPC, especially in patients with hyperactive MAPK pathways (NCT02881242, NCT01990196).

Cediranib, a small molecule inhibitor targeting VEGFR1-3 and c-Kit, was evaluated in a phase 2 trial involving 59 patients with mCRPC [235]. The median PFS was 3.7 months with 44% being progression-free after 6 months. Seven of the 39 evaluable patients had a radiographic response. Cediranib also seems to suppress HRR by indirectly downregulating BRCA1, BRCA2, and RAD51 [236]. It can therefore sensitize tumors to PARP inhibition and, indeed, a recent phase 2 trial found that in combination with cediranib, olaparib showed an improved radiographic PFS of 8.5 months compared to 4.0 months for olaparib alone (HR 0.62, p = 0.036) [237]. For HRRm mCRPC patients, the median PFS was 10.6 with the combination of cediranib and olaparib compared to 3.8 months with olaparib alone. For BRCA2-deficient patients, the median PFS was 13.8 months with the combination, compared to 11.3 months with olaparib alone. The improved PFS came with the cost of increased rates of grade 3 or 4 toxicities.

Other inhibitors (as monotherapy or in combination), including dasatinib, masitinib, sunitinib, cabozantinib, erlotinib, and pazopanib, and monoclonal antibodies like bevacizumab, have not been extensively tested in biomarker-selected populations and commonly have limited to very limited effect [238–240].

Alternative target detection in liquid biopsies

Alterations in AKT1, PIK3CA and PTEN are consistently detected between pretreatment primary tumor tissue and post-ADT progression plasma samples [42, 241]. ctDNA analysis revealed a distinct molecular subtype of mCRPC in AKT1/PIK3CA mutant patients with low AR copy number levels [241]. Approximately 6% of advanced PCa cases had activating AKT1/PIK3CA mutations detected in ctDNA, making liquid biopsies a potential tool to select patients in prospective clinical trials testing PI3K inhibitors. The IND223/IND224 trial (NCT02905318, NCT03385655) uses ctDNA for treatment selection in mCRPC patients [242]. In total, 9 different treatments are matched with 9 different molecular subtypes; these include ipataserib for patients with an AKT1/PIK3CA activating hotspot mutation, the PLK4 inhibitor, CFI-400945, for PTEN inactivation, and the CDK4/6 inhibitor, Palbociclib, for patients with a CDK4/6 amplification. Recently presented data on this trial support the feasibility of ctDNA screening for clinical trials [242].

Conclusion and future perspectives

The treatment landscape for mCRPC has entered the era of precision oncology with the development of several novel GMTs for key molecular pathways including the PI3K pathway, AR-axis, and MAPK pathway. While the development of these agents is ongoing, the most significant progress has been made in patients with DDR deficiency. MMRd is rare in mCRPC, but responses of MMRd patients to checkpoint inhibitors can outperform all registered treatment options. HRRm is relatively common, and the approval of PARPi for the BRCA-altered subgroup has revolutionized PCa care. However, the potential in HRRm is not limited to PARPi treatment, as platinum-based chemotherapy, checkpoint inhibitors, and radioligand therapies have shown beneficial results in this more aggressive genotype. The future may hold even more precise dissection of HRR targets as aberrations in genes such as ATR and ATM are not exploited to their full potential with the current armamentarium.

The implementation of ctDNA sequencing has advanced our understanding of resistance mechanisms, including AR aberrations and reversion mutations. The use of liquid biopsies offers a more accessible approach in the identification of predictive biomarkers to guide patient selection for GMTs. Furthermore, the use of ctDNA sequencing for real-time monitoring of treatment response and the emergence of treatment-resistant clones represents an exciting area of research for improving patient outcomes in mCRPC.

In conclusion, the landscape of high-potential targets within mCRPC is rapidly expanding, and the identification of these targets through ctDNA sequencing provides a promising avenue for improving patient selection and treatment outcomes. However, challenges remain in identifying and targeting all relevant genetic alterations, as the effects of aberrations in every gene are still unclear. To address this challenge, future studies should include more detailed gene-by-gene analyses or explore alternative methods such as low-pass sequencing or a phenotype-based approach to identify biomarker subgroups. Overall, precision oncology has shown great results in mCRPC and the implementation of ctDNA in ongoing research will push the field of precision oncology even further.

Abbreviations
Ac-225	=	Actinium-225
ADT	=	Androgen deprivation therapy
AR	=	Androgen receptor
ARSI	=	Androgen receptor signaling inhibitor
AVPC	=	Aggressive variant prostate cancer
BRCAm	=	BRCA1 or BRCA2 mutated, including copy number losses
cfDNA	=	Cell-free DNA
CTC	=	Circulating tumor cell
ctDNA	=	Circulating tumor DNA
PSA50	=	≥50% PSA decline
CHIP	=	Clonal hematopoiesis of indeterminate potential
CI	=	confidence interval
CTLA-4	=	Cytotoxic T-lymphocyte-associated protein 4
DDR	=	DNA damage repair
DSB	=	Double-stranded break
EMA	=	EUROPEAN Medicines Agency
FDA	=	FOOD and Drug Administration
GMT	=	Genetically matched therapy
HR	=	Hazard ratio
HRD	=	Homologous recombination-deficient phenotype
HRR	=	Homologous recombination repair
HRRm	=	HRR mutated, including copy number losses
Lu-177	=	Lutetium-177
mCRPC	=	Metastatic castration-resistant prostate cancer
MSI	=	Microsatellite instability
MMR	=	Mismatch repair
MMRd	=	Mismatch repair-deficient
OS	=	Overall survival
PARP	=	Poly-(ADP)-ribose polymerase
PARPi	=	Poly-(ADP)-ribose polymerase inhibitor
PD-1	=	Programmed cell death protein 1
PD-L1	=	Programmed cell death protein 1 ligand
PFS	=	Progression-free survival
PI3K	=	Phosphatidylinositol 3-kinase
PCa	=	Prostate cancer
PSMA	=	Prostate-specific membrane antigen
PSMARLT	=	Prostate-specific membrane antigen-radioligand therapy

Disclosure statement

Peter Slootbeek: No potential conflict of interest to declare; Sofie Tolmeijer: No potential conflict of interest to declare; Niven Mehra: Advisory role (compensated and institutional): ‘Roche, MSD, BMS, Bayer, Astellas and Janssen’. Research support (institutional): ‘Astellas, Janssen, Pfizer, Roche and Sanofi’ Genzyme’. Travel support: ‘Astellas, MSD’; Jack Schalken: Speaker honorarium: ‘Astellas, Bayer’.