Transcriptome profiling of prostatic tumours from ENACT trial patients with or without enzalutamide

Abstract

Objectives

To evaluate the longitudinal transcriptomic changes that occurred over time in patients from the ENACT trial (ClinicalTrials.gov identifier: NCT02799745). ENACT evaluated patients with low‐ or intermediate‐risk prostate cancer, comparing the efficacy and safety of enzalutamide plus active surveillance (AS) to AS alone.

Patients and Methods

Gene‐expression profiling was conducted using the Decipher Genomic Resource for Intelligent Discovery (GRID) platform (Veracyte, Inc.) on 131 patient samples (enzalutamide plus AS: n = 57; AS alone: n = 74) collected at screening, Year 1, and Year 2. Pre‐defined GRID signatures were analysed for associations with biological and clinical features, including immune activity, treatment sensitivity, metastatic risk, and molecular subtypes. Statistical analyses utilised Cox proportional hazards models and logistic/linear regression, while longitudinal changes were evaluated using linear mixed‐effects models.

Results

At Year 1, patients treated with enzalutamide plus AS showed significant transcriptomic changes, such as downregulation of androgen‐receptor signalling and immune‐suppressor signatures, alongside upregulation of immune‐activated and basal‐like markers. Many of these transcriptomic alterations, including immune‐related changes, were transient and reverted to baseline levels at Year 2 after enzalutamide treatment cessation.

Conclusion

This exploratory biomarker analysis highlights transcriptomic changes in patients undergoing AS and those treated with enzalutamide, providing insights into molecular disease progression on surveillance and the effects of transient exposure to enzalutamide. The observed immune modulation during enzalutamide treatment suggests opportunities to explore immunotherapy strategies.

Abbreviations

ADT

androgen‐deprivation therapy

AE

adverse event

AR

androgen receptor

AR‐A

androgen receptor‐activity

ARPI

AR pathway inhibitor

AS

active surveillance

CRPC

castration‐resistant prostate cancer

CSPC

castration‐sensitive prostate cancer

GRID

Genomic Resource for Intelligent Discovery

HR

homologous recombination

mpMRI

multiparametric MRI

PSC

prostate cancer subtyping classifier

PARP

poly‐ADP ribose polymerase

PTEN

phosphatase and tensin homologue

Introduction

Downregulation of androgen‐receptor (AR) signalling pathways is commonly employed in prostate cancer [1, 2]. Initially, lifelong androgen‐deprivation therapy (ADT) was utilised for advanced disease, but currently, strategies consisting of defined treatment courses of ADT or blockade of the AR are used to enhance clinical outcomes while also attempting to limit therapeutic morbidity [3, 4]. The longitudinal, sub‐cellular effects of a course of AR‐signalling inhibition followed by withdrawal have not been well studied.

Active surveillance (AS) is the preferred management strategy for favourable‐risk prostate cancer. AS that involves regular monitoring through clinical examinations, PSA tests, periodic biopsies, and multiparametric MRI (mpMRI) allows for a delay in or potential avoidance of definitive treatment unless cancer progression is indicated [3, 4, 5]. However, 25–50% of patients who chose conservative management experience disease progression that necessitates further treatment [6, 7, 8]. Additionally, many patients and healthcare professionals will abort AS due to PSA or other concerns, without histopathological changes [9]. Enzalutamide is an oral AR pathway inhibitor (ARPI) that prevents AR translocation from the cytoplasm to the nucleus, inhibiting AR binding to chromosomal DNA and preventing transcription of tumour‐related genes [10]. This comprehensive inhibition of the AR‐signalling pathway effectively reduces tumour growth and proliferation, as demonstrated in clinical responses in patients with prostate cancer [11, 12]. In the United States, enzalutamide has been approved for treatment of castration‐resistant prostate cancer (CRPC), metastatic castration‐sensitive prostate cancer (CSPC, also known as hormone‐sensitive prostate cancer), and non‐metastatic CSPC with biochemical recurrence at high risk of metastasis [10].

The ENACT trial (ClinicalTrials.gov identifier; NCT02799745), a phase II randomised trial conducted from June 2016 to August 2020 in patients with low‐ or intermediate‐risk localised prostate cancer, compared the efficacy and safety of enzalutamide plus AS vs AS alone, and demonstrated that patients receiving enzalutamide had a 46% reduced risk of disease progression at 1 year vs those who continued AS without therapy (hazard ratio 0.54, 95% CI 0.33–0.89) [13]. Exploration of gene expression in diagnostic biopsies from patients enrolled in ENACT found that patients with transcriptomically defined luminal prostate cancer or with higher AR‐activity (AR‐A) appeared more likely to have oncological response to enzalutamide, although the benefit did not persist 1 year after treatment discontinuation [14]. As serial profiling of biopsy specimens was performed, ENACT additionally allowed for an understanding of transcriptomic changes that can occur over time, both during AS, and during and after exposure to enzalutamide. In this biomarker study, we sought to examine longitudinal transcriptome profile changes that occur over time in patients on AS alone or on enzalutamide plus AS treatment in the ENACT study.

Patients and Methods

Study Design and Participants

The ENACT study design and eligibility criteria were described previously by Shore et al. 2022 [13]. The study adhered to the International Council for Harmonisation guidelines, and all applicable laws, regulations, and directives governing clinical study conduct, as well as ethical principles detailed in the Declaration of Helsinki. Approval was obtained from the independent Ethics Committee or Institutional Review Board at each participating study site [13]. Briefly, patients with clinically localised low‐ or intermediate‐risk prostate cancer on AS alone were randomised to either 1 year of treatment with enzalutamide (160 mg) plus AS or AS alone, as stratified by cancer risk (low vs intermediate) and type of biopsy (TRUS‐guided prostate biopsy [mpMRI]‐targeted vs non‐mpMRI‐targeted) [13]. Patients were monitored until the last patient completed the 24‐month post‐treatment visit (1 year of follow‐up and 1 year of continued follow‐up) [13]. Patients who achieved complete response did not undergo subsequent Decipher testing because there was no residual tumour present for testing, resulting in a gradual decrease in the number of patients with repeat Decipher testing over time.

Transcriptional analyses were conducted on formalin‐fixed, paraffin‐embedded tissue from enrolled patients who consented to the exploratory biomarker and transcriptome analysis to examine enzalutamide's effect. Biopsies were taken at predetermined intervals: screening, Year 1, and Year 2. The biopsy specimen sectioning was analysed centrally, and the samples were subsequently stored at the University of Michigan until they were shipped for transcriptome analysis. To ensure blinded evaluations, de‐identified biopsy specimens were sent to Decipher for central pathology review and transcriptomic profiling. Tumour expression profiles were generated from biopsies of ≥0.5 mm of tumour linear length, with the region of interest designated for microdissection having ≥25% cancer cells and <15% benign cell contamination. Tumour and benign cellularity were quantified by calculating the percentage of cells within the tumour relative to the total area occupied by cells across the entire field of view. RNA was extracted from microdissected tumour samples collected from each patient and at each time point, and then amplified for the Decipher transcriptome‐wide oligonucleotide microarray assay in a College of American Pathologists (CAP)/Clinical Laboratory Improvement Amendments (CLIA) laboratory (Veracyte, San Diego, CA, USA). Gene‐expression signatures were generated using locked models stored in the Decipher Genomic Resource for Intelligent Discovery (GRID) database (Veracyte) [14, 15].

Endpoints

Longitudinal Transcriptome Profiles Observations

Differences in transcriptome profiles at baseline, Year 1, and Year 2 in the AS‐alone arm and pre‐, peri‐, and posttreatment in the enzalutamide plus AS arm were assessed. In order to reduce the chances for false discovery, hypothesis testing was limited to a panel of transcriptomic signatures including: the Decipher score (with locked cut‐offs defining low [<0.45], intermediate [0.45–0.60], and high [>0.60] risk), a highly validated prognostic signature [16]; AR‐A signature (with a locked cut‐off determining lower [≤11] and higher [>11] AR‐A), which measures downstream transcriptional activity of the AR [17]; phosphatase and tensin homologue (PTEN) loss signature (with a locked cut‐off determining wildtype [≤0.54] or PTEN loss [>0.54]) [18]; AR‐induced and ‐repressed genes, which measures AR downstream activity [19]; immune190 signature [20], which measures overall immune infiltration into tumours; immunophenoscore [21], which evaluates expression of pan cancer‐associated immune checkpoint gene signatures as part of tumour microenvironments; small‐cell genomic score (with a locked cut‐off determining adenocarcinoma [≤0.25] or neuroendocrine‐like [>0.25]), a signature of small‐cell or neuroendocrine carcinoma prostate cancer [22]; homologous recombination (HR) deficiency transcriptomic signature (with a locked cut‐off defining HR intact [≤−0.163] and deficient [>−0.163]) [23]; Zhang basal signature (with a locked cut‐off defining luminal‐like [≤−0.211] and basal‐like [>−0.211]), which encapsulates prostate basal and stem cell genes [24]; and prostate cancer subtyping classifier (PSC), a 215‐gene signature that quantifies a tumour's similarity to the biology of prostate basal or luminal cells [25]. These papers did not disclose the cut‐offs for PTEN, HR, and Zhang basal signatures. For PTEN and Zhang basal signatures, cut‐offs were derived from GRID based on information in their respective publications, which used assays that differed from microarray. During the derivation process, we selected cut‐off points such that the proportion of the class calls aligns with the original publication. Once imported into GRID, their cut‐off points were locked and unchanged.

Statistical Analysis

Baseline patient characteristics and transcriptomic signatures were summarised as medians (range or interquartile range) for continuous variables and counts (percentage) for categorical variables. All formal inferences on transcriptomic signatures were performed using their continuous scores. The PSC produces a probability associated with each of four subtypes (luminal differentiated, luminal proliferating, basal immune, and basal neuroendocrine), with the named subtype being the one with highest probability. The basal immune probability score is used in the analyses that follow for inference purposes. The statistical significance of pairwise differences in transcriptomic signatures within randomisation arms was assessed via Wilcoxon signed‐rank tests. Overall patterns of longitudinal change in transcriptome profiles between arms were assessed via linear mixed‐effects models with a random intercept to account for within‐patient correlation and fixed effects for randomisation arms, sample time points, and their interactions. Specifically, a likelihood ratio test against a null model without intercept terms was used to determine if longitudinal changes in transcriptomic signatures were different between treatment arms. Statistical significance was defined as P < 0.05.

Results

In the ENACT study, 227 patients with low‐ or intermediate‐risk localised prostate cancer were randomised 1:1 to enzalutamide plus AS (n = 114) or AS alone (n = 113) [13]. Patient and tumour characteristics have been previously published [13, 14]. Any patient with two or more serial Decipher tests was included in the transcriptome analysis, totalling 131 samples (enzalutamide plus AS: n = 57; AS alone: n = 74) collected from 60 unique patients (Fig. S1) [14]. Patients’ characteristics were generally balanced between both study arms (Table 1).

Table 1.

Patients’ characteristics by treatment arm in the analytic cohort.

Characteristic	AS alone, n = 34	Enzalutamide + AS, n = 26	P
Age, years, median (range)	64 (57–83)	65 (54–84)	0.55*
Race, n (%)
Black or African American	2 (6)	0 (0)	0.50 †
White	32 (94)	26 (100)
Baseline PSA, ng/mL, median (range)	5.7 (1.5–16.8)	4.6 (2.1–15.1) ‡	0.22*
Baseline PPC, %, median (range)	26 (0–50)	33 (7–54)	0.31*
NCCN risk group, n (%)
Low	15 (44)	11 (42)	0.99 †
Intermediate	19 (56)	15 (58)
Biopsy type, n (%)
mpMRI‐targeted	9 (26)	8 (31)	0.78 †
Non‐mpMRI‐targeted	25 (74)	18 (69)

NCCN, National Comprehensive Cancer Network; PPC, percentage positive core.

^*Mann‐Whitney U test.

^† Fisher's exact test.

^‡ Missing two values.

The distribution of longitudinal observations available are given in Fig. S1, stratified by treatment arm. Samples analysed in the enzalutamide arm at Year 1 were from tumours present at time of biopsy on therapy, and samples analysed at Year 2 were from tumours present a year after the cessation of therapy. Longitudinal changes in transcriptomic signatures categorised as high are summarised in a heat map displayed in Fig. 1. In general, enzalutamide induced shifts in tumour biology, including downregulation of AR‐signalling and immune‐suppressor signatures, alongside upregulation of immune‐activated and basal‐like markers, with many signatures returning to baseline after treatment cessation (Figs 1 and S2).

Fig. 1.

Longitudinal changes in transcriptomic signatures by randomisation arms, with change from baseline standardised to the SD of each signature at baseline. Dashed lines indicate a statistically significant difference from screening values, as determined by Wilcoxon signed‐rank tests.

We first explored the influence of enzalutamide on androgen‐related signalling pathways. At baseline, most patients in both ENACT arms had higher AR‐A, as expected for the majority of favourable‐risk prostate cancers. Compared to patients on AS alone, patients on enzalutamide demonstrated a significant decrease in AR‐A from baseline to Year 1 in both score (median: 9.8 vs 13.8, respectively; P < 0.001) and class (70% vs 7% lower AR‐A, respectively). Accordingly, enzalutamide caused increased expression of AR‐repressed genes and decreased expression of AR‐induced genes, which reverted to near baseline after treatment withdrawal (Table 2; Figs 2A and S3A). The difference in pattern of longitudinal change in AR‐A score between treatment arms was statistically significant (P < 0.001).

Table 2.

Transcriptomic signature scores over time by randomisation arm.

Score, median (IQR)	Screening	Year 1	Year 2	P screening vs Year 1	P screening vs Year 2	P Year 1 vs Year 2	P arm × time interaction
AS alone
Decipher	0.17 (0.10, 0.17)	0.18 (0.12, 0.18)	0.30 (0.22, 0.30)	0.66	0.002*	0.005*	0.03*
AR‐A	14.0 (13.5, 14.0)	13.8 (12.7, 13.8)	13.5 (12.3, 13.5)	0.65	0.03*	0.10	<0.001*
AR‐induced signature	1.10 (1.05, 1.10)	1.16 (1.09, 1.16)	1.13 (1.06, 1.13)	0.05	0.85	0.19	<0.001*
AR‐repressed signature	0.13 (0.05, 0.13)	0.18 (0.14, 0.18)	0.21 (0.15, 0.21)	0.02*	0.05*	0.74	0.002*
PTEN Liu	0.084 (0.023, 0.084)	0.010 (0.003, 0.010)	0.006 (0.002, 0.006)	0.12	0.004*	0.67	0.38
PSC basal immune probability	0.011 (0.001, 0.011)	0.014 (0.003, 0.014)	0.114 (0.036, 0.114)	0.84	0.03*	0.21	<0.001*
Zhang basal signature	−0.33 (−0.38, −0.33)	−0.34 (−0.43, −0.34)	−0.29 (−0.41, −0.29)	0.96	0.13	0.14	<0.001*
Neuroendocrine signature	0.06 (0.03, 0.06)	0.05 (0.04, 0.05)	0.10 (0.08, 0.10)	0.81	0.01*	0.005*	0.006*
Immune 190	0.12 (0.09, 0.12)	0.14 (0.11, 0.14)	0.20 (0.15, 0.20)	0.13	0.01*	0.13	<0.001*
HR deficient signature	−0.20 (−0.25, −0.20)	−0.24 (−0.27, −0.24)	−0.19 (−0.24, −0.19)	0.11	0.38	0.009*	0.27
Enzalutamide + AS
Decipher	0.19 (0.12, 0.19)	0.44 (0.19, 0.44)	0.36 (0.27, 0.36)	0.002*	0.12	0.74
AR‐A	13.4 (13.0, 13.4)	9.8 (7.0, 9.8)	13.0 (12.2, 13.0)	<0.001*	0.10	0.11
AR‐induced signature	1.11 (0.95, 1.11)	0.71 (0.65, 0.71)	1.08 (0.97, 1.08)	0.007*	0.58	0.25
AR‐repressed signature	0.12 (0.05, 0.12)	0.30 (0.24, 0.30)	0.18 (0.12, 0.18)	<0.001*	0.007*	0.02*
PTEN Liu	0.042 (0.025, 0.042)	0.008 (0.002, 0.008)	0.024 (0.003, 0.024)	0.002*	0.12	0.31
PSC basal immune probability	0.003 (0.001, 0.003)	0.914 (0.426, 0.914)	0.181 (0.036, 0.181)	<0.001*	0.10	0.08
Zhang basal signature	−0.33 (−0.38, −0.33)	−0.04 (−0.21, −0.04)	−0.26 (−0.34, −0.26)	0.01*	0.03*	0.46
Neuroendocrine signature	0.05 (0.02, 0.05)	0.11 (0.07, 0.11)	0.11 (0.10, 0.11)	0.01*	0.003*	0.84
Immune 190	0.11 (0.07, 0.11)	0.21 (0.17, 0.21)	0.20 (0.16, 0.20)	<0.001*	0.01*	0.25
HR deficient signature	−0.23 (−0.28, −0.23)	−0.24 (−0.27, −0.24)	−0.19 (−0.24, −0.19)	0.40	0.03*	0.05

Scores are reported as their median (interquartile range [IQR]; first quartile, third quartile), with the statistical significance of paired longitudinal differences determined by Wilcoxon signed‐rank tests and an overall test of differential longitudinal patterns by arms determined by linear mixed‐effects models.

^* Statistically significant at P < 0.05.

Fig. 2.

Stacked bar plots of genomic signature class calls over time by signature arm for (A) AR‐A, (B) Zhang basal signature (C) PSC, and (D) Decipher.

Next, we analysed the effect of enzalutamide on cell‐of‐origin classifiers. These classifiers relate to AR‐regulated genes in luminal subtypes associated with higher AR signalling. By the Zhang basal signature, at baseline, most patients in both arms had luminal‐like tumours (88% AS alone and 83% enzalutamide plus AS; Figs 2B and S3B). For patients on AS alone, there was a slight increase in basal‐like proportion from 12% at baseline to 23% at Year 1. With enzalutamide treatment at Year 1, basal‐like signature was substantially enriched, with an increase in prevalence to 75% from 17% at baseline. The increase in basal‐like gene expression among patients treated with enzalutamide was significantly greater than that seen in the AS‐alone population (mixed‐model arm/time point interaction P value at Year 1: P < 0.001; Table 2). At Year 2, the prevalence of basal‐like subtype decreased from Year 1 to 43% in the enzalutamide plus AS arm and remained elevated at 20% in the AS‐alone arm (Figs 2B and S3B). The difference in pattern of longitudinal change in basal‐like scores between treatment arms was statistically significant (P < 0.001).

Additionally, we explored effects on PSC, another commonly assessed cell‐of‐origin classifier. In the PSC basal‐luminal subtyping classifier, a continuous increase in the prevalence of the basal immune subtype was observed in the AS‐alone arm from baseline (4%) to Year 1 (17%) and Year 2 (25%), while a decrease in prevalence of the luminal differentiated subtype was observed from baseline (83%) to Year 2 (50%). Prevalence of luminal proliferating subtype also increased from baseline (8%) to Year 1 (23%) in the AS‐alone arm. Like in the AS‐alone arm, the enzalutamide plus AS arm showed a significant increase in prevalence of the basal immune subtype from baseline (4%) to Year 1 (75%). However, this prevalence partially decreased at Year 2 (43%) after treatment was discontinued. Again, as in the AS‐alone arm, the luminal proliferating subtype increased from baseline (13%) to Year 2 (36%) in the enzalutamide plus AS arm. No major change in basal neuroendocrine‐like subtype was observed in either arm (Fig. 2C). The difference in pattern of longitudinal change in basal immune probability scores between treatment arms was statistically significant (P < 0.001; Table 2).

Treatment with enzalutamide did not enrich for neuroendocrine signature. By neuroendocrine signature score, there was a slight increase in the proportion of patients with neuroendocrine‐like expression at Year 2 (5%) compared to baseline (0%) in the AS‐alone arm. There were no significant changes in the neuroendocrine signature score in the enzalutamide arm at Year 1 or Year 2 (Table 2).

The tumour immune microenvironment was also assessed using the Immune190 and regulatory T‐cell expression signatures. Immune190 score was increased from baseline at Year 1 (P < 0.001) and remained elevated compared to baseline at Year 2 (P = 0.009) in the enzalutamide plus AS arm. At Year 1, the increase was significantly greater than the AS‐alone arm (mixed‐model arm/time point interaction at Year 1: P < 0.001; Fig. S3C). In the enzalutamide plus AS arm, levels of regulatory T‐cell signature decreased at Year 1 (P < 0.001) and then returned to near‐baseline levels at Year 2 after treatment discontinuation. The downregulation at Year 1 in patients treated with enzalutamide was statistically significant vs patients on AS alone (mixed‐model arm/time point interaction at Year 1: P < 0.001; Table 2).

Transcriptomically, PTEN appeared intact in most patients, and this was not significantly modified by treatment with enzalutamide (mixed‐model arm/time point interaction at Year 1: P = 0.38; Table 2; Fig. S3D). The proportion of patients with HR deficiency progressively increased in the enzalutamide plus AS arm from baseline (9%) to Year 1 (15%), and again to Year 2 (29%). In contrast, in the AS‐alone arm, the prevalence of HR deficiency initially decreased at Year 1 (15%) and then slightly increased at Year 2 (20%) but remained below the baseline prevalence (33%). The difference in pattern of longitudinal change in HR deficiency score between treatment arms was not statistically significant (mixed‐model arm/time point interaction at Year 1: P = 0.27; Table 2).

Examining prognostic Decipher scores, we noted an increase in patients with high‐risk Decipher scores in both arms from baseline to Year 2 (Table 2, Fig. 2D). In the AS‐alone arm, Decipher scores and risk groups increased progressively from baseline (8% intermediate and high risk) to Year 2 (25% intermediate and high risk), with similar risk between baseline and Year 1. In the enzalutamide plus AS arm, an immediate increase in Decipher score and risk groups was observed at Year 1 (45% intermediate and high risk) compared to baseline (4% intermediate risk, 0% high risk), which remained elevated among evaluable patients at Year 2 (35% intermediate and high risk). The difference in pattern of longitudinal change in Decipher scores between treatment arms was statistically significant (P = 0.03; Table 2; Figs 2D and S3E).

Finally, the incidence of adverse events (AEs) was previously reported [13]. Briefly, throughout the 1‐year treatment period, patients who received enzalutamide experienced a higher incidence of AEs (92.0%) compared to those who received AS (54.9%). During the subsequent 1‐year follow‐up period, the corresponding numbers were 39.3% and 23.0%, respectively, with a further reduction to 14.3% and 10.6%, respectively, during the 1‐year continued follow‐up period. Of the patients who received enzalutamide, 5.4% experienced a drug‐related AE of Grade ≥3. The incidence of serious AEs remained low across all treatment arms and time periods [13].

Discussion

This study analysed longitudinal transcriptome changes, transcriptomic profiles, and patterns of effects in tumours of patients with clinically localised low‐ or intermediate‐risk prostate cancer on AS, with or without enzalutamide treatment, to examine castrate‐sensitive tumours in the setting of a phase II clinical trial. The results demonstrated that after a year of enzalutamide treatment, AR‐signalling signatures remained downregulated, activated immune markers were upregulated, and suppressed immune markers were downregulated in prostate cancer tumours. These transcriptomic profiles in large part appear to revert to baseline after cessation of treatment with enzalutamide. A key clinical implication of these findings, guiding future research, is that the identified relationship between transcriptome changes and drug sequencing suggests the potential for developing targeted therapies for patients with CSPC. Thus, our findings showcase the utility of expression profiling before therapy to establish a basis for biomarker‐guided selection of therapeutic approach for patients with prostate cancer. We also add to the evidence that therapy interruption after a defined course of AR blockade does not select for resistant clones.

The AS‐alone group in the study did not receive any treatment intervention, allowing us to test if changes occur in tumour transcriptomic expression patterns over time. Future research may investigate prognostic signatures to better understand if transcriptome testing alone can be used to inform clinical decision making about continuing surveillance or progressing to treatment.

In tumours responding to enzalutamide treatment, we observed a significant shift in AR signalling, characterised by a decrease in AR‐induced gene expression and an increase in AR‐repressed gene expression. These changes are consistent with the mechanism of action of enzalutamide as an ARPI [10]. The expression of AR signatures returned to near‐baseline levels upon treatment discontinuation, which indicates that while enzalutamide suppresses AR signalling, after treatment cessation its effects on molecular activity of this pathway may diminish over time. Although enzalutamide was effective in initially treating the cancer, it was plausible that the 1‐year course of enzalutamide could select for androgen‐insensitive subclones. However, once the selective pressure of enzalutamide was removed, tumoral transcriptomic expression of androgen sensitivity reverted to near‐baseline levels, suggesting that finite therapy does not select for long‐term decreased androgen sensitivity. AR signalling by prostate cancer tumours in this setting may act similarly to the sensitisation to subsequent androgen ablative therapy observed in the bipolar androgen therapy study, but further investigation is warranted [26]. Understanding the sustainability of the effects of enzalutamide may assist in exploring combined therapies that might enhance or prolong its efficacy in managing prostate cancer and may inform protocols for pulse‐dose treatments with enzalutamide if they are pursued for management going forward.

While evidence suggests that certain transcriptomic changes are reversible, it is unclear whether they result in a reversible responsiveness by tumours that remained or grew following treatment suspension. To address this uncertainty, further research is needed to investigate the correlation between transcriptomic reversion and tumour dynamics.

The ENACT trial's finding that luminal tumours may predict responsiveness to enzalutamide is consistent with prior studies suggesting that luminal tumours are most likely to be responsive to ADT [14, 23]. Therefore, to optimise treatment outcomes with enzalutamide or other ARPIs, selection of patients with luminal as opposed to basal subtypes by PSC may be preferred. One concern for treating lower‐risk prostate cancers with finite or pulse‐dosed ARPI therapy is that it could select for more aggressive prostate cancers or accelerate progression to CRPC. Importantly, in this study, the incidence of neuroendocrine prostate cancer did not increase following enzalutamide treatment, suggesting a favourable resistance profile in tumours present after enzalutamide. While further research is necessary to understand the long‐term tumour dynamics, luminal subtypes responded to enzalutamide treatment, and there was no evidence to suggest that finite enzalutamide treatment selects for more aggressive subtypes.

Additionally, there were changes in the immune microenvironment with enzalutamide that persisted even after treatment cessation in tumours present after treatment. The Immune190 signature, a bulk measure of tumour immune infiltration, demonstrated that enzalutamide plus AS treatment led to a statistically significant increase in tumour immune infiltration at Year 1 compared to patients on AS alone, and regulatory T‐cell signature expression decreased, suggesting the ability for enzalutamide to modulate the immune system. At Year 2, Immune190 remained upregulated in patients who had received enzalutamide, suggesting that the immune modulation is sustained. These findings align with a previous study evaluating the clinical and immunological impact of short‐course enzalutamide in non‐metastatic CSPC, where enzalutamide was found to be independently associated with immunological changes, such as increasing natural killer cells and naïve T‐cells and decreasing myeloid‐derived suppressor cells [27]. This suggests how enzalutamide's effects on immunity may benefit patients beyond the direct effect on tumour biology. Enzalutamide may therefore also be used to prime tumours for subsequent or concurrent therapy with immune modulators.

Enrichment of HR deficiency with enzalutamide treatment supports consideration for its potential in combined therapies, specifically, utilising enzalutamide to sensitize tumours to poly‐ADP ribose polymerase (PARP) inhibitors. Together, PARP inhibitors, which block single‐strand DNA repair, in combination with HR deficiency, which causes abnormalities in a separate mechanism for DNA repair, cause cell death [28]. Several studies have shown that PARP inhibitors are most effective in the setting of HR deficiency, and although they were initially approved in the metastatic CRPC space, there has been an ongoing effort to investigate the use of PARP inhibitors in earlier stages of the disease [29]. Given that, after treatment with enzalutamide, tumours were enriched for HR deficiency signatures, they may also exhibit increased sensitivity to PARP inhibition.

Focusing on patients who received AS alone, there were some interesting trends. For one, there was a trend toward higher Decipher scores over time while on AS. There was also a trend toward increased basal immune subtype by PSC. These shifts in the absence of treatment indicate that low‐ and intermediate‐risk tumours are dynamic and can evolve to have more aggressive molecular features over time. It suggests a potential role for repeat biopsy Decipher testing for patients on AS to aid in management decisions, such as determining which patients should transition from surveillance to definitive treatment.

This study does have limitations to consider when interpreting the results. Patients who completely responded to the treatment were excluded from further analysis as there was no tumour present, and thus tumour transcriptome analysis could not be examined. Furthermore, 23% of patients in Year 1 and 55% of patients in Year 2 were excluded from the analysis due to processing failure or patients who did not consent to transcriptome analysis. Therefore, both arms exhibited selection biases in the subsequent profiling at Years 1 and 2. While this study was limited by a small sample size, Ross et al. [14] previously showed that higher Decipher scores, luminal subtype by PAM50, and higher AR‐A scores were associated with a better response to enzalutamide. Thus, future studies of transcriptomic data and predictive signatures are warranted to offer insights into potential treatment responses, including complete response.

The study was also limited by a relatively short follow‐up period of 2 years, which may limit the ability to fully assess long‐term responses and resistance patterns of enzalutamide treatment on prostate cancer progression. Additionally, while the focus of this study was on transcriptomics, we did not compare transcriptomic and immune signatures at baseline; we also note the absence of an assessment of changes in AR genomics (amplification, mutations, splice variants) associated with acquired resistance [1, 2].

Despite these limitations, the study uncovers intriguing tumour dynamics over time, raising important questions about tumour behaviour while patients are on AS or being treated with a finite course of enzalutamide [30]. Although many transcriptomic changes reverted to baseline after enzalutamide was discontinued, impacts on immune‐related signatures appeared to be more sustained, highlighting the concept that immune cells and cytokines offer a potentially promising target for prostate cancer therapy [30]. Finite treatment with enzalutamide does not appear to select for more aggressive prostate cancers based on molecular signature expression. While this study looked at enzalutamide in a population with low‐ and intermediate‐risk prostate cancer, it is reasonable to believe that higher‐risk tumours may exhibit similar responses to enzalutamide, and, therefore, further investigation of combination or sequential therapy with immune modulators and PARP inhibitors in the high‐risk localised setting should also be pursued.

We conclude that after a year of treatment with enzalutamide, tumours that did not completely respond or that progressed following cessation of treatment showed downregulation of AR‐signalling and immune‐suppressor transcriptomic signatures, and upregulation of immune‐activated and basal‐like biology markers. These changes may serve as therapeutic targets for combined or subsequent therapies. After discontinuation of enzalutamide, most, but not all, signatures returned to baseline levels without evidence of selecting out more aggressive prostate cancers. Changes in transcriptomic signature activity were also observed in the AS‐alone arm, providing insights into the progression of signature changes in localised prostate cancer on AS and suggesting further investigation for the role of serial Decipher testing for these patients to aid in treatment decisions.

Disclosure of Interests

Nicole Handa declares no conflict of interest. Elai Davicioni, Xin Zhao, Yang Liu, and James A. Proudfoot are employees of Veracyte Inc., which received funding from Astellas Pharma for the conduct of this study. Dina Elsouda, Gaston Kuperman, and Kenneth K. Iwata are employees of Astellas Pharma Inc., a co‐sponsor of this study. David Russell is an employee and shareholder of Pfizer Inc., a co‐sponsor of this study. Elai Davicioni has applied for a patent for Cancer Diagnostics using Biomarkers and has a potential financial interest in the outcome of this study. Matthew R. Cooperberg is a speaker for Veracyte, Bayer, Janssen, Merck, Pfizer, ExosomeDx, LynDx, Verana Health, Tempus, and Biomarker; and has participated on a Data Safety Monitoring Board or Ad Board for AstraZeneca. Neal D. Shore is a consultant and participated on a Data Safety Monitoring Board or Ad Board for Amgen, Astellas, Astellas/Medivation, AstraZeneca, Bayer, Dendreon, Ferring, Genetech/Roche, Janssen Scientific Affairs, Merck, Myovant Sciences, Pfizer, and Tolmar; and is a speaker for Bayer, Dendreon, and Janssen. Edward M. Schaeffer is a speaker for Pfizer, Astellas, and Lantheus; and a board member for NCCN prostate cancer. Ashley Ross is a consultant and speaker for Veracyte and Pfizer.

Supporting information

Fig. S1. Distribution of longitudinal observations analysis set, stratified by treatment arm.

Fig. S2. Heatmap of genomic signatures of interest.

Fig. S3. Box plots of longitudinal transcriptomic signatures by treatment arm with statistical significance of paired longitudinal differences superimposed above each comparison (ns, not significant; *P<0.05; **P < 0.01; ***P ≤ 0.001): (A) AR‐A, (B) Zhang basal signature, (C) Immune 190, (D) PTEN, and (E) Decipher.

Data Availability Statement

All data generated or analysed during this study, which supports the findings of this study, are included within this article and its supplementary information files. Researchers interested in further analyses that are not present in the manuscript may contact the corresponding author.