Prognostic Value of Low-Pass Whole Genome Sequencing of Circulating Tumor DNA in Metastatic Castration-Resistant Prostate Cancer

Abstract

Background

Multiple treatments are available for metastatic castration-resistant prostate cancer (mCRPC), including androgen receptor signaling inhibitors (ARSI) enzalutamide and abiraterone, but therapy resistance remains a major clinical obstacle. We examined the clinical utility of low-pass whole-genome sequencing (LPWGS) of circulating tumor DNA (ctDNA) for prognostication in mCRPC.

Methods

A total of 200 plasma samples from 143 mCRPC patients collected at the start of first-line ARSI treatment (baseline) and at treatment termination (n = 57, matched) were analyzed by LPWGS (median: 0.50X) to access ctDNA% and copy number alteration (CNA) patterns. The best confirmed prostate specific antigen (PSA) response (≥50% decline [PSA₅₀]), PSA progression-free survival (PFS), and overall survival (OS) were used as endpoints. For external validation, we used plasma LPWGS data from an independent cohort of 70 mCRPC patients receiving first-line ARSI.

Results

Baseline ctDNA% ranged from ≤3.0% to 73% (median: 6.6%) and CNA burden from 0% to 82% (median: 13.1%) in the discovery cohort. High ctDNA% and high CNA burden at baseline was associated with poor PSA₅₀ response (P = 0.0123/0.0081), poor PFS (P < 0.0001), and poor OS (P < 0.0001). ctDNA% and CNA burden was higher at PSA progression than at baseline in 32.7% and 42.3% of the patients. High ctDNA% and high CNA burden at baseline was also associated with poor PFS and OS (P ≤ 0.0272) in the validation cohort.

Conclusions

LPWGS of ctDNA provides clinically relevant information about the tumor genome in mCRPC patients. Using LPWGS data, we show that high ctDNA% and CNA burden at baseline is associated with short PFS and OS in 2 independent cohorts.

Introduction

Standard treatment for patients with metastatic prostate cancer includes androgen deprivation therapy, leading to disease remission in most patients. However, despite castrate levels of testosterone, patients eventually develop resistance and progress to metastatic castration-resistant prostate cancer (mCRPC). Multiple life-prolonging treatments are available for mCRPC, including second-generation androgen signaling inhibitors (ARSI) enzalutamide and abiraterone. Still, primary resistance is common (1), and most patients who initially respond will acquire resistance during treatment. Thus, there is an urgent need for novel biomarkers that can predict treatment response and outcome and thereby guide a more personalized treatment selection through inclusion of information about the patient’s tumor genome.

Analyses of circulating cell-free DNA (cfDNA) in plasma provides an opportunity for minimally invasive tumor profiling, since a fraction of plasma cfDNA in cancer patients is tumor-derived (circulating tumor DNA [ctDNA]). Such a liquid biopsy approach is attractive in mCRPC patients, where metastasis biopsies (often located in bone) are not easily obtained. Recent studies have shown that analysis of ctDNA can identify genomic driver alterations present in matched metastatic prostate cancer (mPC) tissue (2, 3) and that ctDNA has potential as a novel biomarker in mCRPC (4–7). Furthermore, as compared to a single tissue biopsy, ctDNA seems to better reflect the entire tumor and metastasis burden in each patient (2).

Deep whole-genome sequencing, whole-exome sequencing, and deep targeted panel sequencing allow for extensive genome profiling and identification of biomarkers. However, these methods are cost-prohibitive for high-throughput analyses. On the other hand, low-pass WGS (LPWGS) allows estimation of ctDNA% and tumor copy number alteration (CNA) analyses in a cost-effective manner. As prostate cancer (PC) is characterized by genomic instability (8), CNAs are relevant for biomarker discovery. Previously, AR amplification has been associated with ARSI treatment resistance (7, 9–12) and deletion of TP53, PTEN, or RB1 with poor prognosis on ARSI (5–7, 13–16). Moreover, recent studies have reported that a high ctDNA% at mCRPC diagnosis may be linked with poor progression-free survival (PFS) and poor overall survival (OS) (15, 17, 18).

In this study, we therefore aimed to test the clinical utility of using LPWGS for blood-based biomarker discovery in mCRPC. Using LPWGS, we profiled a total of 200 plasma samples from 143 mCRPC patients undergoing first-line ARSI treatment in Denmark. We found that a high ctDNA% and a high CNA burden was associated with poor best confirmed prostate specific antigen response (≥50% decline [PSA₅₀]), poor PFS, and poor OS. The results were validated in an independent cohort of 70 mCRPC patients from Belgium (19). Our study demonstrates that LPWGS of plasma ctDNA can provide clinically relevant information about the mCRPC tumor genome and hence may be used as a cost-effective, high-throughput method to help inform better mCRPC patient management in the future.

Materials and Methods

mCRPC Patients and Blood Samples

A total of 200 EDTA blood samples were obtained from 143 mCRPC patients treated with either enzalutamide or abiraterone acetate as a first-line treatment at Aarhus University Hospital or Regional Hospital West Jutland (April 2016–February 2020) (Table 1). Samples were collected at start of treatment (baseline, n = 143) and at treatment termination for 57 of the patients (PSA progression in 52/57 and death/side effects in 5/57). The study was approved by the National Committee on Health Research Ethics (#1901101) and notified to the Danish Data Protection Agency (#1-16-02-366-15). All patients provided written informed consent.

Table 1.

Patient overview.

Variable	Discovery cohort (n = 143)
M-stage at initial PC diagnosis, n (%)
ȃ0	59 (41.3)
ȃ1	79 (55.2)
ȃNA/X	5 (3.5)
Treatment with curative intend at initial PC diagnosis, n (%)
ȃRadical prostatectomya	12 (8.4)
ȃRadiation therapy	21 (14.7)
ȃNone	110 (76.9)
Treatment in the hormone-sensitive setting
ȃADTb alone	98 (68.5)
ȃSurgical castration alone	5 (3.5)
ȃADTb + surgical castration	7 (4.9)
ȃADTb + docetaxel	32 (22.4)
ȃADTb + surgical castration + docetaxel	1 (0.7)
Metastatic burden at mCRPC diagnosis (imaging), n (%)
ȃBone only	74 (51.7)
ȃLymph node only	15 (10.5)
ȃBone and lymph node	46 (32.2)
ȃVisceralc	5 (3.5)
ȃNA	3 (2.1)
Blood chemistry at mCRPC diagnosis (baseline)
ȃPSA (ng/mL), median (range) (n = 143)	32.1 (0.4–759.4)
ȃAlkaline phosphatase (U/L), median (range) (n = 133)d	71 (9.6–1448)
ECOG Performance score at mCRPC diagnosis, n (%)
ȃ0	90 (62.9)
ȃ1	45 (31.5)
ȃ2	8 (5.6)
ȃ3–5	0 (0.0)
First-line mCRPC treatment, n (%)
ȃAbiraterone acetate	19 (13.3)
ȃEnzalutamide	124 (86.7)
PSA progression, first-line mCRPC treatment
ȃYes, n (%)	104 (72.7)
ȃȃ Progression-free survival (months), median (range)	8.1 (1.3–47.7)
ȃNo, n (%)	39 (27.3)
ȃȃAvailable follow-up time during first-line treatment (months), median (range)	25.8 (1.0–64.3)
Dead
ȃYes, n (%)	68 (47.6)
ȃȃOverall survival (months), median (range)	19.0 (2.1–49.7)
ȃNo, n (%)	75 (52.4)
ȃȃAvailable follow-up time (months), median (range)	25.8 (2.6–64.3)

Two patients were treated with salvage radiation therapy after RP.

Androgen deprivation therapy; antiandrogen (e.g., Bicalutamide) and gonadotropin-releasing hormone agonist/antagonist.

Specific location available for 2 patients (liver and lung metastasis, respectively). Unknown for 3 patients.

Alkaline phosphatase data not available for 10 patients.

Blood Sample Processing and cfDNA Extraction

All blood samples (30–50 ml) were collected in BD Vacutainer K₂ EDTA tubes (Beckton Dickinson) and processed within 2 hours (stored at 4°C until processing). See Supplemental Materials and Methods for a detailed description of blood processing, cfDNA extraction, and quantification.

Library Preparation and LPWGS

Libraries were prepared using the Kapa Hyper Library Preparation Kit (KAPA Biosystems) and sequenced on an Illumina^Ⓡ Novaseq instrument (S-prime flowcell). All samples met the criteria of a coverage of ≥0.05× (Supplemental Materials and Methods).

Bioinformatics processing and analyses of LPWGS data are described in Supplemental Materials and Methods.

Clinical Outcome and Statistical Analysis

See Supplemental Materials and Methods for details. In brief, the primary endpoint was PSA PFS defined as the time from treatment initiation to PSA progression (20). Secondary endpoint was OS defined as the time from treatment initiation until death (any cause). Treatment effect was evaluated by PSA response, defined as the best confirmed PSA response during treatment (PSA nadir) in each patient (21).

All statistical analyses were conducted in R studio (v.3.6.3).

Validation Cohort

LPWGS data on plasma samples from a total of 70 mCRPC patients was used for independent validation (Supplemental Table 1) (19). See Supplemental Materials and Methods for details.

Results

mCRPC Cohort

Blood samples from a consecutive cohort of 143 men with newly diagnosed mCRPC starting first-line ARSI (abiraterone/enzalutamide) treatment were prospectively collected at 2 hospitals in Denmark (2016–2020) (Fig. 1, A and Table 1).

Fig. 1.

Study and cohort overview. (A), Flowchart, created in Biorender; (B), Univariate Cox regression (PFS); (C), Kaplan–Meier PFS analysis of ctDNA%; (D), Univariate Cox regression (OS); (E), Kaplan–Meier OS survival analysis of ctDNA%; (F), Association between ctDNA% and PSA response; (G), Waterfall plot of best confirmed PSA change during ARSI treatment.

At mCRPC diagnosis, bone metastases were detected in 83.9% (120/143) of the patients, with 51.7% (74/143) having bone-only and 32.2% (46/143) having bone as well as lymph node (LN) metastases (Table 1). Another 10.5% (15/143) of the patients had LN-only metastases and 3.5% (5/143) had visceral metastases co-occurring with bone and/or LN metastases.

As first-line therapy, the majority of patients received enzalutamide (86.7%, 124/143), while the remaining patients received abiraterone (13.3%, 19/143). At the time of last follow-up, 72.7% (104/143) of the patients had experienced PSA progression on first-line ARSI and 47.6% (68/143) had died (Table 1 and Supplemental Fig. 1, A). PSA PFS and OS was similar for patients treated with enzalutamide and abiraterone, respectively (Supplemental Fig. 1, B).

Plasma cfDNA Level, ctDNA%, and Clinicopathological Parameters

We analyzed plasma samples drawn prior to first-line ARSI treatment (baseline) for the entire cohort (n = 143) and at termination of first-line treatment (progression) for 57 of these patients (Fig. 1, A). Thus, cfDNA was extracted from a total of 200 plasma samples, quantified by ddPCR, and analyzed with LPWGS at a median coverage of 0.50X (range: 0.06–0.102X) (Fig. 1, A). For each sample, ctDNA% was estimated using IchorCNA (22).

At baseline, cfDNA levels ranged from 1.7 to 183.6 ng/mL (median: 7.3 ng/mL) and ctDNA% from ≤3.0 to 72.8% (median: 6.6%). ctDNA% was moderately positively correlated with total cfDNA levels (Spearman rho = 0.605, P < 0.00001. Supplemental Fig. 2) as well as with PSA (rho = 0.390, Benjamini-Hochberg adjusted [BH adj.] P ≤ 0.001) and alkaline phosphatase (ALP) levels (rho = 0.473, BH adj. P ≤ 0.001) (Supplemental Table 2). Furthermore, a significantly higher ctDNA% was seen in patients who had visceral metastases vs only bone and/or LN metastases (BH adj. P ≤ 0.024. Mann–Whitney test; Supplemental Fig. 4, A) and in patients with a worse performance score (1 vs 0: BH adj. P = 0.028; Supplemental Fig. 4, B).

There was no significant change in the median ctDNA% from baseline to progression (treatment termination) in the subset of 57 patients with available matched samples (7.9% vs 7.3%, P = 0.50, Wilcoxon signed rank test). PSA progression was the reason for treatment termination in 52 of the 57 patients (Supplemental Fig. 3). We observed an increase in ctDNA% from baseline to progression in 17/52 (32.7%) patients with PSA progression, whereas ctDNA% was similar at baseline and progression in 13/52 (25%) patients and lower at progression in 22/52 (42%) patients (Supplemental Fig. 3).

High ctDNA% Is Associated With Shorter PFS

Univariate Cox regression analyses showed that presence of visceral metastases [hazard ratio (HR) = 7.49, BH adj. P = 0.0002), high PSA (HR = 1.002, BH adj. P = 0.030), and high ALP levels (HR = 1.002, BH adj. P = 0.0002) were significantly associated with worse PFS (Fig. 1, B). Furthermore, a high ctDNA% at baseline was associated with poor PFS (HR = 4.08, BH adj. P < 0.00001. Fig. 1, B), which was confirmed in Kaplan–Meier analysis (BH adj. P < 0.0001, log-rank test; Fig. 1, C). Consistent with this, patients with short vs long PFS (<6 vs >18 months) had higher ctDNA% at baseline (median: 24.4% vs 3.0%, BH adj. P < 0.0001, Kruskal–Wallis test. Supplemental Fig. 4, C). In multivariate step-backward Cox regression analysis, including all variables significant in univariate analysis, a high ctDNA% (HR = 3.63, BH adj. P < 0.00001) and presence of visceral metastases (HR = 3.63, BH adj. P = 0.022) were the only variables that remained significantly associated with poor PFS (Supplemental Table 3).

High ctDNA% Is Associated With Shorter OS

In univariate cox regression analysis, upfront treatment with docetaxel (HR = 2.18, BH adj. P = 0.011), high PSA (HR = 1.003, BH adj. P = 0.036), and high ALP levels (HR = 1.002, BH adj. P < 0.0001) were significantly associated with poor OS, whereas metastatic location was not (BH adj. P = 0.171) (Fig. 1, D). Moreover, a high baseline ctDNA% was associated with poor OS (HR = 5.08, BH adj. P < 0.00001; Fig. 1, D), which was confirmed in Kaplan–Meier analysis (BH adj. P < 0.0001; Fig. 1, E). In line with this, patients with shorter OS (<9 vs 9–24 and >24 months) had significantly higher ctDNA levels (median: 32.3% vs 9.2% and 3.0%, P = 0.00002, Kruskal–Wallis test; Supplemental Fig. 4, D). In multivariate step-backward Cox regression analysis, including all variables significant in univariate analysis, high ctDNA% (HR = 3.91, BH adj. P = 0.0007) and upfront docetaxel (HR = 2.09, BH adj. P = 0.018) were the only variables that remained significantly associated with poor OS (Supplemental Table 3).

High ctDNA% Is Associated With Poor Response to First-Line ARSI Treatment

We also evaluated PSA response to first-line ARSI, defined by the absolute change from baseline to PSA nadir (see Materials and Methods section). Overall, a lower ctDNA% at baseline was associated with better PSA response (Fig. 1, F). Furthermore, a significantly larger proportion of patients with low vs high ctDNA% reached a best confirmed PSA decline of at least 50%, that is, PSA₅₀ (95.2% vs 75.0%. BH adj. P = 0.0075, Fisher’s exact test; Fig. 1, G). Similar results were obtained when restricting the PSA₅₀ analysis to patients treated with enzalutamide (Supplemental Fig. 5, A).

A High CNA Burden Is Associated With Poor Treatment Response and Poor Outcome

Next, tumor copy number alteration (CNA) patterns were determined genome-wide for all 143 mCRPC patients using IchorCNA calls based on LPWGS data (Fig. 2, A). Several recurrent CNAs were identified at baseline, including chr8p deletion (64/143 patients, 47.8%), chr8q amplification (71/143, 49.7%), AR amplification (51/143, 35.7%), and chr13q deletion (57/143, 39.9%). These findings are consistent with previous reports on recurrent somatic CNAs in PC/mCRPC (4, 23–27), indicating that this is a representative mCRPC cohort.

Fig. 2.

Copy number alterations. (A), Frequency of gains/losses at chr1-X; (B), Waterfall plot of best confirmed PSA change during ARSI treatment; (C), Kaplan–Meier PFS analysis of CNA burden; (D), Kaplan–Meier OS analysis of CNA burden; (E), Kaplan–Meier PFS analysis of CNA burden in validation cohort (n = 70); (F), Kaplan–Meier OS analysis of CNA burden in validation cohort (n = 70).

The median CNA burden (percent genome altered) at baseline was 13.1% (range: 0.0%–81.9%) in the whole cohort and did not change significantly from baseline to progression (treatment termination) in the subset of 57 patients with matched samples (median 16.6% vs 15.9%, P = 0.70, Wilcoxon signed rank test). Out of the 52 patients who had PSA progression at treatment termination, the CNA burden increased from baseline to progression in 22 (42.3%) patients, was similar at baseline and progression in 10 (19.2%) patients, and was lower at progression in 20 patients (Supplemental Fig. 3).

The proportion of patients obtaining a PSA₅₀ response on ARSI was significantly lower for those with a high vs low CNA burden at baseline (80.6% vs 95.8%; BH adj. P = 0.0081, Fisher exact test; Fig. 2, B). Similar results were seen if restricting the analysis to patients treated with enzalutamide (Supplemental Fig. 5, B).

In addition to being associated with poor PSA₅₀ response, a high CNA burden at baseline was significantly (BH adj. P < 0.0001) associated with poor PFS (HR = 2.38) and poor OS (HR = 3.21) in univariate Cox regression (Table 2) as well as in Kaplan–Meier analyses (Fig. 2, C and D). CNA burden remained borderline significantly associated with poor PFS (HR = 1.56, BH adj. P = 0.0763) and poor OS (HR = 1.86, BH adj. P = 0.0534) after adjustment for ctDNA% in multivariate analyses (Table 2). Sensitivities and specificities of ctDNA% and CNA burden in our discovery cohort can be found in Supplemental Table 4.

Table 2.

Uni- and multivariate Cox regression analyses in the discovery cohort (n = 143).

Endpoint/variable		Univariate		Multivariate
		HR (CI)	BH adj. P value	HR (CI)	BH adj. P value
PFS
ȃCNA burden	Low vs high	2.38 (1.60–3.54)	<0.0001	1.56 (0.97–2.52)	0.0763
ȃctDNA%	Continuous	1.03 (1.02–1.04)	<0.00001	1.02 (1.01–1.04)	0.0002
OS
ȃCNA burden	Low vs high	3.21 (1.93–5.32)	<0.0001	1.86 (1.02–3.40)	0.0534
ȃctDNA%	Continuous	1.04 (1.03–1.05)	<0.00001	1.03 (1.02–1.05)	<0.0001
PFS
ȃChr13q deletion	Wta vs deletion	2.68 (1.80–3.98)	0.00001	1.82 (1.14–2.90)	0.0172
ȃctDNA%	Continuous	1.03 (1.02–1.04)	<0.00001	1.02 (1.01–1.03)	0.0002
OS
ȃChr7 amplification	Wta vs amplification	2.82 (1.73–4.60)	0.0001	1.91 (1.13–3.211)	0.0181
ȃctDNA%	Continuous	1.03 (1.02–1.04)	<0.00001	1.04 (1.02–1.05)	<0.00001
OS
ȃChr8p deletion	Wta vs deletion	3.10 (1.89–5.07)	0.00005	2.05 (1.21–3.46)	0.0117
ȃctDNA%	Continuous	1.03 (1.02–1.04)	<0.00001	1.04 (1.02–1.05)	<0.00001
OS
ȃChr9q amplification	Wta vs amplification	2.53 (1.51–4.23)	0.0011	2.28 (1.35–3.85)	0.0035
ȃctDNA%	Continuous	1.03 (1.02–1.04)	<0.00001	1.04 (1.03–1.05)	<0.00001

Significant P values are in bold.

Wild-type.

As CNA burden was positively correlated with ctDNA% (Spearman rho = 0.838, P < 0.00001; Supplemental Fig. 6 and Supplemental Table 5), we performed a subgroup analysis for patients with ctDNA% >10 (n = 55). Here, a high CNA burden (≥65.0% [median]) was associated with a significantly worse PFS independent of ctDNA% (HR = 2.33, BH adj. P = 0.0149; Supplemental Table 6). Although statistical significance was not reached, we also observed a trend toward a high CNA burden being associated with poor OS in this subgroup (HR = 1.74, BH adj. P = 0.1690, univariate Cox regression; Supplemental Table 6).

External Clinical Validation

To the best of our knowledge, tumor CNA burden determined by plasma LPWGS has not previously been shown to harbor prognostic biomarker potential in mCRPC. Hence, for external validation we used IchorCNA to analyze plasma LPWGS data from an independent cohort of 70 mCRPC patients undergoing first-line ARSI treatment in Belgium (19). In the validation cohort, a high CNA burden (>14.5% [median]) was significantly associated with shorter PFS (HR = 3.83, BH adj. P = 0.0002; Table 3) and shorter OS (HR = 13.01, BH adj. P = 0.0016; Table 3) in univariate Cox regression analyses. Similar results were seen in Kaplan–Meier analyses (PFS/OS: BH adj. P < 0.00001/P = 0.00001; log-rank test; Fig. 2, E and F). This confirmed our findings from the discovery cohort (Fig. 2, C and D and Table 3). Sensitivities and specificities of ctDNA% and CNA burden in our validation cohort can be found in Supplemental Table 4.

Table 3.

Uni- and multivariate cox regression analyses in the validation cohort (n = 70).

Endpoint/variable		Univariate		Multivariate
		HR (CI)	BH adj. P value	HR (CI)	BH adj. P value
PFS
ȃCNA burden	Low vs high	3.83 (2.03–7.24)	0.0002	2.77 (1.37–5.58)	0.0068
ȃctDNA%	Continuous	1.05 (1.03–1.07)	<0.00001	1.03 (1.01–1.05)	0.0019
OS
ȃCNA burden	Low vs high	13.01 (2.93–57.85)	0.0016	6.43 (1.33–31.19)	0.0272
ȃctDNA%	Continuous	1.07 (1.04–1.09)	<0.00001	1.05 (1.02–1.08)	0.0007
PFS
ȃChr13q deletion	Wta vs deletion	2.86 (1.52–5.36)	0.0019	1.53 (0.71–3.32)	0.3179
ȃctDNA%	Continuous	1.05 (1.03–1.07)	<0.00001	1.04 (1.02–1.06)	0.0009
OS
ȃChr7 amplification	Wta vs amplification	4.53 (1.68–12.2)	0.0046	1.50 (0.44–5.03)	0.5579
ȃctDNA%	Continuous	1.07 (1.04–1.09)	<0.00001	1.06 (1.03–1.09)	0.0002
OS
ȃChr8p deletion	Wta vs deletion	3.86 (1.39–10.7)	0.0131	0.65 (0.14–3.13)	0.6198
ȃctDNA%	Continuous	1.07 (1.04–1.09)	<0.00001	1.07 (1.04–1.11)	0.0002
OS
ȃChr9q amplification	Wtc vs amplification	2.64 (1.02–6.81)	0.0526	1.12 (0.41–3.04)	0.8320
ȃctDNA%	Continuous	1.07 (1.04–1.09)	<0.00001	1.07 (1.04–1.09)	<0.00001

Significant P values are in bold.

Wild-type.

After adjusting for ctDNA% in multivariate Cox regression, a high CNA burden remained significantly associated with both poor PFS (HR = 2.77, BH adj. P = 0.0068; Table 3) and poor OS (HR = 6.43, BH adj. P = 0.0272; Table 3) in the validation cohort, while this was borderline significant in the discovery cohort (Table 3). Together, these results indicate that ctDNA% and CNA burden, determined by low-cost LPWGS, may be useful for prognostication of mCRPC patient outcome and, furthermore, that CNA burden may provide additional prognostic information beyond ctDNA%.

Deletion on Chr13q and Chr8p and Amplification on Chr7 and Chr9q Associate with Poor Outcome and Response

Next, we examined the biomarker potential of CNA in specific genomic regions. To ensure statistical power, we focused on recurrent amplifications/deletions detected at baseline in at least 25% of the patients in the discovery cohort based on CNA calls from IchorCNA (Fig. 2). Based on these criteria, we identified 12 genomic regions with recurrent amplifications or deletions (Supplemental Table 7).

Four of the 12 regions (chr3q amplification, chr13q deletion, chr18q deletion, and chrXq amplification) were significantly associated with poor PFS in univariate Cox regression (BH adj. P ≤ 0.0044; Supplemental Table 8), which was confirmed in Kaplan–Meier analysis (BH adj. P ≤ 0.0038; Supplemental Table 8). After adjustment for ctDNA%, only chr13q deletion remained a significant independent marker of PFS (HR = 1.82, BH adj. P = 0.0172; Fig. 3, A and Table 2). In line with this, a smaller proportion of patients with chr13q deletion (vs wild-type) reached a PSA₅₀ response (78.9% vs 94.2%, P = 0.0127; χ² test; Fig. 3, E). The significant association between chr13q deletion and poor PFS was validated in the external mCRPC cohort in univariate analysis (log-rank BH adj. P = 0.0016; Fig. 3, I; univariate Cox regression, HR = 2.86, BH adj. P = 0.0019 [Table 3]) but failed in multivariate analysis after adjustment for ctDNA% in this smaller cohort (HR = 1.53, BH adj. P = 0.3179; Table 3).

Fig. 3.

Biomarker potential of CNA in specific genomic regions. (A–D), Kaplan–Meier analysis in discovery cohort (n = 143); (E–H), Waterfall plot of best confirmed PSA change during ARSI treatment. Patients were grouped by CNA (wt vs deletion/amplification, respectively); (I–L), Kaplan–Meier analysis in validation cohort (n = 70).

In the discovery cohort, all 12 genomic regions were associated with significantly worse OS in univariate cox regression (BH adj. P ≤ 0.0246; Supplemental Table 8) as well as in Kaplan–Meier analysis (BH adj. P ≤ 0.0227; Supplemental Table 8). After adjustment for ctDNA%, chr7 amplification (HR = 1.91, BH adj. P = 0.0181), chr8p deletion (HR = 2.05, BH adj. P = 0.0117), and chr9q amplification (HR = 2.28, BH adj. P = 0.0035) remained significant independent markers of poor OS (Fig. 3, B and D and Table 2). For all 3 regions, a significantly smaller proportion of patients with CNA (vs wild-type) reached a PSA₅₀ response (chr7 amplification: 76.5% vs 94.6%, P = 0.0034, χ² test; chr8p deletion: 79.7% vs 94.9%, P = 0.0080, Fisher exact test; chr9q amplification: 78.8% vs 94.1%, P = 0.0018, χ² test) (Fig. 3, F–H). Similar results were found for the enzalutamide-treated subgroup of the patients (Supplemental Fig. 5, C–F). The significant association between chr7 amplification, chr8 deletion, and chr9 amplification and poor OS was validated in the external mCRPC cohort in univariate analyses (Fig. 3, J–L, Kaplan–Meier analysis, BH adj. P ≤ 0.0462, Fig. 3, J–L; univariate Cox regression, BH adj. P ≤ 0.0526. Table 3) but failed in multivariate analyses after adjustment for ctDNA% (Table 3).

Discussion

In this study, we sequenced 200 plasma samples from a discovery cohort of 143 mCRPC patients undergoing first-line ARSI treatment to investigate the potential clinical utility of LPWGS for ctDNA analysis in the management of mCRPC. We found that a high ctDNA% and a high CNA burden at baseline was associated with poor PSA₅₀-response, poor PFS, and poor OS. The significant association between a high ctDNA% and/or a high CNA burden and poor survival (PFS/OS) was validated in an independent cohort of 70 mCRPC patients. Overall, our results demonstrate that LPWGS of plasma ctDNA can provide clinically relevant information about the tumor genome in mCRPC patients and hence has potential to be used to inform better treatment decisions in the future. To the best of our knowledge, this is the first study to show that CNA burden, estimated by ctDNA analysis, is associated with PSA response and outcome in mCRPC patients undergoing first-line ARSI treatment.

Our finding that ctDNA% was associated with significantly shorter PFS and OS in 2 mCRPC patient cohorts confirms and expands on previous studies (7, 17, 18, 28, 29). Our results indicate that LPWGS of plasma cfDNA holds clinical potential for outcome prognostication in mCRPC patients through estimation of ctDNA% and tumor CNA patterns. Major advantages of using plasma LPWGS include that it is cost-effective, allows for high-throughput and rapid analysis, and does not require prior sequencing of each patient’s tumor genome. LPWGS can also be used for fast screening of the ctDNA% in a given plasma sample and thus for determination of applicable downstream analysis (e.g., whole-exome sequencing). The lower limit of ctDNA% estimation by IchorCNA is 3.0%, which in some clinical settings may be insufficient. However, for mCRPC patients, our results indicate that a lower limit of detection of 3.0% is satisfactory, as this identified patients with good outcome. In other clinical settings (e.g., detection of early-stage cancer or minimal residual disease where the ctDNA% is typically lower), it may however be necessary to use more sensitive methods. This can be achieved through other methods such as detection of tumor-derived single nucleotide variants by ddPCR or deep-targeted sequencing. As mCRPC is characterized by few recurrent mutations (6, 7, 15, 16, 26), this approach would require patient-specific assays, which are difficult to design because the tumor genome is not routinely analyzed at this disease stage.

Plasma LPWGS data can also be used for analysis of tumor CNAs. We found that CNA burden was associated with both outcome (PFS/OS) and treatment response in our mCRPC cohort. Although detection of CNA is positively correlated with ctDNA%, our results from 2 distinct mCRPC cohorts suggested that CNA burden may contribute independent prognostic information beyond that given by ctDNA%, although this was statistically significant only in 1 of the 2 cohorts and borderline significant in the other cohort. For clinical translation, further studies in larger cohorts are needed to confirm this. However, our findings are consistent with previous tissue-based reports showing that a high CNA burden in prostate tumor tissue is associated with significantly shorter biochemical recurrence-free and cancer-specific survival in localized and metastatic PC, respectively (30, 31). Moreover, CNA has been suggested to drive lethal PC (32–34), further supporting our results.

Furthermore, we identified 4 genomic regions that were affected by recurrent genomic deletions in mCRPC and were significantly associated with poor PSA₅₀-response and poor PFS/OS in our discovery cohort. The prognostic biomarker potential for these 4 genomic regions was confirmed in our independent validation cohort. More specifically, chr13q deletion was associated with poor PFS, while chr7 and chr9q amplification and chr8p deletion were associated with poor OS. Except for chr9q amplification, all of these genomic regions have previously been investigated for their biomarker potential in tissue-based analyses of radical prostatectomy (RP) cohorts. Consistent with our current findings, one study of 6695 RP patients reported that deletion of chr13q in PC tissue was associated with poor biochemical recurrence (BCR)-free survival (35). Similarly, amplification of chr7 (and/or chr8q) in PC tissue has been associated with BCR in a cohort of 23 patients with localized PC treated by RP (27). Furthermore, previous studies of 2 large RP cohorts including 6375 and 1954 patients, respectively, showed deletion of chr8p in PC tissue was associated with poor BCR-free survival (36, 37). Thus, our results from liquid tumor biopsy analyses in patients with mCRPC confirm and expand on previous tissue-based reports that showed an association to unfavorable outcome also in early-stage PC patients with deletion of chr8p or chr13q or amplification of chr7 or chr9q.

Limitations of our study includes that CNA calling depends on ctDNA% and sequencing coverage, resulting in lower sensitivity in samples with lower coverage and/or lower ctDNA%. To overcome this, we have worked systematically, i.e., only used samples with coverage ≥0.05X, adjusted for ctDNA% in multivariate Cox regression analyses, and performed subset analyses of patients with ≥10% ctDNA in which estimation of ctDNA% and CNA burden are more robust.

In summary, our results indicate that ctDNA% and CNA burden, as determined by low-cost plasma LPWGS, may be useful for prognostication of mCRPC patient outcome and, furthermore, that CNA burden may provide additional biomarker information beyond ctDNA%. Thus, further validation studies are warranted to assess the potential clinical value of using cost-effective plasma LPWGS to guide better patient management in mCRPC. Such future studies should also investigate if slightly higher LPWGS coverage than used in the current study (approximately 0.5X) would better capture the added biomarker value of CNA burden and/or specific genomic regions affected by CNA. Future studies should also investigate the clinical value of using ctDNA% and CNA data, as determined by plasma LPWGS, for minimally invasive monitoring of treatment response in longitudinally collected samples from mCRPC patients to systematically compare PSA vs ctDNA% and/or CNA burden.

Supplementary Material

Supplementary material is available at Clinical Chemistry online.

Nonstandard Abbreviations

mCRPC, metastatic castration-resistant prostate cancer; ARSI, androgen receptor signaling inhibitors; cfDNA, circulating cell-free DNA; ctDNA, circulating tumor DNA; LPWGS, low-pass whole genome sequencing; PC, prostate cancer; CNA, copy number alteration; PFS, PSA progression-free survival; OS, overall survival; LN, lymph node; ALP, alkaline phosphatase; HR, hazard ratio; BH, Benjamini-Hochberg; RP, radical prostatectomy; BCR, biochemical recurrence.

Human Genes

AR, androgen receptor; TP53, tumor protein p53; PTEN, phosphatase and tensin homolog; RB1, RB transcriptional corepressor 1.

Author Contributions

The corresponding author takes full responsibility that all authors on this publication have met the following required criteria of eligibility for authorship: (a) significant contributions to the conception and design, acquisition of data, or analysis and interpretation of data; (b) drafting or revising the article for intellectual content; (c) final approval of the published article; and (d) agreement to be accountable for all aspects of the article thus ensuring that questions related to the accuracy or integrity of any part of the article are appropriately investigated and resolved. Nobody who qualifies for authorship has been omitted from the list.

Authors’ Disclosures or Potential Conflicts of Interest

Upon manuscript submission, all authors completed the author disclosure form. Disclosures and/or potential conflicts of interest:

Employment or Leadership

M. Borre is chairman for the Danish Prostate Cancer Group and a steering committee member for the Danish Comprehensive Cancer Center.

Consultant or Advisory Role

J.B. Jensen has participated on data safety monitoring and/or advisory boards for the following: Roche, Photocure ASA, Olympus, AMBU, Cepheid, and Urotech. H. Grönberg has received consulting fees from Janssen and Astra Zeneca.

Stock Ownership

None declared.

Honoraria

J.B. Jensen has received lecturing fees from Olympus and Medac. H. Grönberg has received lecturing fees from Bayer and Astella. M. Borre has received lecturing fees from Astellas, Jansen, Bayer, and MSD. K.D. Sørensen has received lecturing fees from Sanofi and “Dagens Medicin.”

Research Funding

This project was supported by grants from the Danish Cancer Society, the Central Denmark Region Health Fund, Aarhus University (Graduate School of Health), the Novo Nordisk Foundation, the Danish Cancer Foundation, Direktør Emil C. Hertz og Hustru Inger Hertz’ Fond, KV Fonden, Raimond og Dagmar Ringgård-Bohns Fond, Beckett Fonden, and Snedkermester Sophus Jacobsen og Hustru Astrid Jacobsens Fond. J.B. Jensen has received grants/contracts from Photocure ASA, Medac, Olympus, Ferring, and Cepheid. H. Grönberg has received a grant/contract from Janssen.

Expert Testimony

H. Grönberg, Astra Zeneca; K.D. Sørensen, Astra Zeneca and MSD.

Patents

None declared.

Other Remuneration

J.B. Jensen, support for attending meetings and/or travel from Medac.

Role of Sponsor

The funding organizations played no role in the design of study, choice of enrolled patients, review and interpretation of data, preparation of manuscript, or final approval of manuscript.

Acknowledgments

The authors wish to thank the staff involved in the CRPC project at the Departments of Urology at Aarhus University Hospital and Regional Hospital West Jutland, as well as the laboratory technicians and clinical academics at the Department of Molecular Medicine. Last, we would like to acknowledge the Danish Cancer Biobank for biological material and the Aarhus Genome Data Center for access to the GenomeDK high-performance computing facility.