Serum Sex Steroids as Prognostic Biomarkers in Patients Receiving Androgen Deprivation Therapy for Recurrent Prostate Cancer: A Post Hoc Analysis of the PR.7 Trial

Biomarkers to categorize the phenotype and predict the prognosis of patients with lethal prostate cancer are a high priority, particularly given the expanding number of therapeutic options. An emerging strategy to delay the morbidity of castration-resistant prostate cancer (CRPC) is to combine effective therapies, such as docetaxel or abiraterone, with the initiation of androgen deprivation therapy (ADT) for metastatic disease. In men initiating ADT, multiple studies demonstrate that serum testosterone is a prognostic biomarker of time to castration resistance and prostate cancer survival.

Our prior study in men receiving continuous ADT postradiotherapy for recurrent prostate cancer demonstrated that serum testosterone levels within the first year of ADT initiation predict the time to development of CRPC, which occurred about 5 to 7 years later (1). Multiple studies indicate the importance of sex steroids such as androgens in the development of castration-resistant prostate cancer (2). Their accessibility for measurement and their association with the biology of resistance thus make serum sex steroids ideal candidates as prognostic or predictive biomarkers. This includes their potential utility for predicting response to combined docetaxel or abiraterone at the time of ADT initiation for men with metastatic castration-sensitive prostate cancer. However, beyond testosterone, sex steroids have received little investigation to date for patients with prostate cancer receiving ADT.

The PR.7 study of intermittent versus continuous ADT established intermittent ADT as an equivalent option for men with biochemically recurrent prostate cancer following radiotherapy (3). Our prior study demonstrated that serum nadir testosterone levels in the first year were prognostic for time to castration resistance and prostate cancer–specific survival (1). Initiated in 1997, the study incorporated a correlative science component, which mandated the collection of longitudinal serum samples for all the Canadian patients in the study, beginning at study entry. Long-term follow-up is now available on these men. Using these cryopreserved samples from the PR.7 study, we determined to assess the prognostic role of serum sex steroids in men treated with ADT. We assessed both upstream androgen precursors, androgen metabolites, and estrogens using previously described methods (4). Thus, we seek to understand whether the prognostic ability of nadir serum testosterone is related to the availability of androgens to stimulate prostate cancer cell growth, or whether this may be reflective of other microenvironmental factors. Furthermore, we explore the changes in the levels of sex steroids that occur over time among men receiving continuous ADT and their correlation with outcome.

The PR.7 study opened for accrual in Canada in 1999 and was supported by the Southwest Oncology Group, the Radiation Therapy Oncology Group, the Cancer Trials Support Unit, and the Institute of Cancer Research Clinical Trials and Statistics Unit (United Kingdom). Details of the protocol and study results have been published (3). Briefly, patients with prostate cancer with a PSA level above 3 ng/mL and no metastases 12 months or more after definitive radiotherapy (primary or salvage) were randomly assigned in a 1:1 ratio to the two treatment groups. At baseline, men had a serum testosterone level above 5 nmol/L. Continuous ADT consisted of a luteinizing hormone–releasing hormone (LHRH) agonist, combined with a nonsteroidal antiandrogen, with the latter continued for a minimum of 4 weeks, or orchiectomy. Any LHRH agonist preparation was acceptable in any of the 1-month or longer depot formulations. Follow-up was until death. After the development of castration-resistant disease, management was determined by the local investigator. The primary endpoint was overall survival. Secondary endpoints included time to CRPC and quality of life.

Following informed consent, Canadian patients in the study had serum prospectively collected and stored at annual intervals starting at randomization for up to 5 years. This study was conducted in accordance with the Declaration of Helsinki and institutional ethics approval for this ancillary study was obtained (2016–2835). All samples were aliquoted and stored in −80°C. Patients were included in this study who were randomized to and received continuous ADT. Patients with insufficient cryopreserved sample for gas chromatography/mass spectrometry (GC-MS) analysis (250 μL serum) were excluded as were patients with less than 2 years of follow-up. Patients were excluded who received exogenous corticosteroids or systemic estrogens within 3 months of the sample collection. Our cohort of initial nadir ADT samples consisted of the first available sample after randomization; the mean time from randomization was 14 (median 12, range 10–33) months.

Measurement of steroid levels was performed according to a previously described validated gas chromatography selected reaction monitoring–tandem mass spectrometry assay (4). The lower limit of quantification (LLOQ) was 100 pg/mL for dehydroepiandrosterone (DHEA), 50 pg/mL for progesterone, androstenediol (A5diol), androstenedione (AD), and androsterone (AST), 30 pg/mL for testosterone, 10 pg/mL for androstane-3β,17βdiol (3βdiol), and DHT, 5 pg/mL for estrone (E₁), and 1 pg/mL for estradiol (E₂) using a sample volume of 250 μL. Internal standards (deuterated steroids) were added to samples and quality controls were included in each run (4). All metabolite coefficients of variation (CV) were <10%.

For our statistical analysis, our primary outcome was time to castration resistance as previously defined (1) and secondary outcomes included prostate cancer–specific survival and overall survival. Individual steroids values were categorized in tertiles; in cases where the range was small, a binary categorization was used. χ² tests were used to check the association between each steroid marker's levels and baseline factors for categorical variables, using the ANOVA test for continuous variables. Kaplan–Meier curves were used to estimate distributions of time to event outcomes, and Cox regression models were used to test the steroid marker's influence on outcome. All Cox regression analyses were adjusted for potential confounding factors previously found significant in this population (1, 3), namely age (continuous), initial Gleason score (≤7 vs. ≥8), prior prostatectomy (yes vs. no), baseline Eastern Cooperative Oncology Group (ECOG) status (0 vs. 1), initial PSA (≤15 vs. >15 ng/mL), and time from radiotherapy to randomization (>3 years vs. 1–3 years). These were included in the model using backwise selection with P = 0.3 as a cutoff. For evaluation of longitudinal changes in serum sex steroids, we sought to evaluate our hypothesis that increasing serum sex steroids were associated with worse outcomes. Patients with an increase of >0.5 SDs (of each respective steroid value, excluding outliers) on the 2nd measurement, which was sustained or on the 3rd measurement relative to the first, were categorized as increasing and compared with the remainder that remained stable or decreased. Comparison of testosterone measurements reported by participating sites in the original trial, performed in the hospital clinical laboratory, were compared with those measured in our study using a paired t test when the date of the respective sample collections was within 4 weeks.

A total of 219 patients met our inclusion criteria and had mass spectrometry analysis of cryopreserved serum performed. A subset of 101 patients had two subsequent annual samples, which were concurrently analyzed. For patients included in this analysis, the consort plot is shown in Supplementary Fig. S1, with clinical characteristics summarized in Table 1. The relationship of the measured sex steroids in the steroidogenic pathway is presented in Supplementary Fig. S2, with a summary of measured values in Supplementary Table S1.

Analysis of each of the initial values for individual steroids showed several associations with baseline clinical variables. Older age was commonly associated with lower levels of sex steroids (Table 2). ECOG performance status (1 vs. 0) was also significantly associated with lower DHT and DHEA values (χ² test, P = 0.03 and P = 0.009, respectively). Higher tertiles of estradiol E₂ measurements were significantly associated with a longer time since radiotherapy (>3 years vs. 1–3 years, P = 0.007). There was no significant association of E₁, progesterone, A5diol, 3βdiol, or AD with any baseline factors.

High intrapatient correlation between all measured steroids values was observed in our initial samples (Table 3), with the exception of progesterone and 3βdiol for which values were frequently below the LLOQ (Supplementary Table S1). Pearson correlation coefficients were significant among steroids more closely related, with the highest correlation between DHEA and A5diol (r = 0.81; Supplementary Table S2). In our longitudinal analysis of annual sex steroid measurements in 101 patients, the correlation of DHT with DHEA, A5diol, and AD decreased over time, but its correlation with E₂ increased (Supplementary Table S3). Furthermore, we found a high correlation between the changes of DHT and testosterone over time (Supplementary Table S4).

Adjusted regression analysis demonstrated the independent prognostic value of individual sex steroids on outcomes of time to castration resistance, prostate cancer survival, and overall survival (Table 4). Of the 10 steroids measured, estrone (E₁) and estradiol (E₂) levels demonstrated the strongest association with time to castration resistance, with significantly earlier time to castration resistance in the highest tertile. Similarly, the highest tertile of AD levels almost reached statistical significance in predicting earlier time to castration resistance. For the highest tertile of estrone, similar nonsignificant associations were also seen with prostate cancer–specific and overall survival. The relationship between higher castrate serum testosterone and sooner time to castration resistance or cancer-specific survival did not achieve statistical significance, although the trends were similar to our previous report. Kaplan–Meier curves demonstrate similar trends for testosterone, DHT, AD, E₁, and E₂ (Fig. 1; Supplementary Fig. S3) on all three outcomes assessed.

Increases in measured serum steroids occurred over time in a minority of patients (range, 16–43) among the 101 patients with longitudinal measurements (Supplementary Fig. S4). Kaplan–Meier curves for time to castration resistance for all steroids are shown in Fig. 2; results for A5diol are not included due to technical problems precluding measurement among most of the second longitudinal samples. Increases in sex steroids over time corresponded with sooner time to castration resistance or worse prostate cancer mortality on regression analyses adjusted for baseline steroid tertile and clinical variables for DHEA, AST, DHT, and AD. Increasing DHEA was associated with an HR of 3.94 [95% confidence interval (CI), 1.64–9.46; P = 0.002] for time to castration resistance, with a similar trend seen for time to prostate cancer mortality (HR = 3.62; 95% CI, 0.78–16.73; P = 0.099). Increasing AST was also associated with sooner time to castration resistance (HR = 2.26; 95% CI, 1.04–4.90; P = 0.039), but not prostate cancer mortality. Increasing AD corresponded to significantly worse prostate cancer mortality (HR = 2.45; 95% CI, 0.97–6.46; P = 0.011), but no significant differences in time to castration resistance were found. Finally, increasing DHT values corresponded to sooner time to castration resistance (HR = 1.88; 95% CI, 0.87–4.06; P = 0.11) and prostate cancer mortality (HR = 3.79; 95% CI, 0.89–16.10; P = 0.07), but these did not reach statistical significance.

Finally, we compared our testosterone values measured by mass spectrometry with those reported in the trial (assay according to each hospital site not specified). In 74 samples that had concomitant values reported in the trial (<4-week difference), we found our mass spectrometry testosterone values were significantly lower by a mean of 0.31 nmol/L (paired t test, P = 0.05). Testosterone values in our samples were also lower than those reported in the trial at the two subsequent annual longitudinal measurements by a mean of 0.47 nmol/L (n = 56, P < 0.001) and 0.46 nmol/L (n = 52, P < 0.001), respectively.

This study demonstrates that levels of sex steroids beyond testosterone hold significant prognostic ability to predict the development of castration resistance in men with a biochemical recurrence postradiotherapy. In particular, it appears that serum estrone and estrogen levels act as important prognostic biomarkers following ADT initiation and the increasing levels of some androgens during ADT identify men at higher risk of resistance.

Development of reliable and accessible biomarkers in patients with castration-sensitive prostate cancer presents an opportunity to maximize the initial response to ADT by employing appropriate combination strategies. The initial PSA response to ADT is an important biomarker, but in men with recurrent, nonmetastatic cancer, it usually falls to <0.2 ng/dL. The biology associated with sex steroids and prostate cancer progression naturally lends to their use to distinguish not only prognosis, but also disease phenotype. The moderate to high correlation we observed among all sex steroids measured suggests that it is not the direct effects of testosterone (or suboptimally suppressed testosterone) on prostate cancer cells that mediate its prognostic value. Rather, our data suggest that elevated levels of various circulating sex steroids may directly or indirectly influence the tumor to favor the development of castration resistance. This may be mediated though intratumoral conversion of androgen precursors, but also through effects on stromal, immune, and other cell types in the tumor microenvironment. The recent demonstration of a survival benefit with the addition of abiraterone acetate to ADT highlights a potential application as relatively higher circulating sex steroids may describe a phenotype more susceptible to abiraterone acetate in men initiating ADT (5, 6).

Estrone and estradiol values have not been previously studied as prognostic biomarkers during ADT, but germline variation in estrogen-related genes have been implicated in prostate cancer prognosis (7). In men, serum estrogens are largely derived from peripheral conversion of AD and testosterone. If most serum estrogens in men on ADT are derived from conversion of testosterone, subdividing their relative influence on cancer progression and resistance compared with testosterone is challenging in this type of cohort study. However, we found the correlation between AD and estrogens higher than between testosterone and estrogens, suggesting more direct effects of estrogens may explain their prognostic value (2, 8). Interestingly, we found E₁ and E₂ both reached statistical significance, whereas serum testosterone did not. Further validation studies are needed to confirm the prognostic role of serum estrogens in men treated with ADT and to investigate potential mechanisms of action.

Little data exist on how serum levels of sex steroids evolve over time during ADT (9, 10). Our data indicate that declines in DHEA and other steroids, which occur over time in the general population (11, 12), also continue to occur under ADT. Contrary to the trend that occurs with age, increases in DHEA and AST levels during ADT were found on multivariate analysis to predict sooner time to castration resistance and may act as an indicator of more aggressive disease. This is in line with reports that indicate adrenal androgens are the principle source of intratumoral androgens in CRPC (13, 14). Furthermore, the more rapid production of androgens during ADT due to genomic variants has also been shown to correspond to poorer prognosis (15–17). The reasons for these changes over time cannot be determined from our data, but similar changes in patient sera have been reported following enzalutamide treatment (18).

The levels of testosterone assessed by mass spectrometry in these patients were significantly lower than the levels reported by the clinical laboratories using an immunoassay done at the time of the study. This is consistent with literature reporting that immunoassay-based measurement of low levels of testosterone tends to overestimate the levels (19–21). Although the levels differed from the clinical laboratory, both sets of testosterone levels correlated with outcome. This reinforces the importance of accurate measurement of serum testosterone in these patients.

The lack of significance of testosterone values, which were previously found to be prognostic in this population, may reflect the limited sample size. However, other possibilities exist to explain this discrepancy. The overestimation of testosterone values on immunoassay-based measurement could be related to other steroids that we found significant on our analyses, such as estrogens, causing nonspecific binding to the testosterone detection antibody. Thus, the testosterone immunoassay could at a certain point (the limitation of antibody specificity) be providing an aggregate measurement of similar steroids. The measurement of other key adrenal androgen sources, such as DHEA-sulfate by mass spectroscopy methods, could also help further clarify the importance of circulating androgen precursors relative to other sex steroids on the prognosis of men starting ADT.

This is a post hoc analysis, and prospective studies are required to validate the use of these steroids as biomarkers. There have been significant advances in the way CRPC is managed since the completion of the PR.7 study, which may influence the generalizability of our results regarding prostate cancer survival and overall survival. Another limitation of our study is the lack of information about body mass index (BMI) in our cohort. It is possible that relative differences in serum estrogen values are related in part to the peripheral aromatization of circulating testosterone, but with no information on BMI, this hypothesis cannot be evaluated. Finally, with the panel of steroids highly correlated and each biologically linked to our outcome tested, we elected a priori not to adjust for multiple testing because these were not random hypotheses to be tested. Notwithstanding these limitations, the detailed mass spectrometry measurements, longitudinal values, and long follow-up present the most detailed hormonal evaluation to date of men treated with ADT.

Serum estrone and estradiol, as well as increasing DHEA and AST levels, hold significant prognostic value for men treated with ADT for recurrent prostate cancer. The high intrapatient correlation suggests it is host factors that explain the long-term prognostic value of serum sex steroids. Further research is needed to understand how these sex steroids impact the intratumoral biology of prostate cancer resistance and how these prognostic biomarkers can help personalize therapy.

P. Toren reports receiving speakers bureau honoraria from Ferring, is a consultant/advisory board member for Abbvie, Astellas, Ferring, Pfizer and Sanofi, and reports receiving commercial research grants from Innocrin Pharma and Janssen. F. Pouliot reports receiving speakers bureau honoraria from Astellas, Bayer, and Sanofi, is a consultant/advisory board member for Abbvie, Amgen, Astellas, Janssen, Pfizer, Progenics, and Roche, and reports receiving commercial research grants from Astellas and Sanofi. L. Klotz is a consultant/advisory board member for Sanofi Genzyme and TerSera. No potential conflicts of interest were disclosed by the other authors.

Conception and design: P. Toren, K. Ding, F. Pouliot, L. Klotz

Development of methodology: P. Toren, K. Ding

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Hoffman, K. Ding, F.-H. Joncas, V. Turcotte, P. Caron, C. Guillemette, L. Klotz

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): P. Toren, K. Ding

Writing, review, and/or revision of the manuscript: P. Toren, A. Hoffman, K. Ding, F.-H. Joncas, F. Pouliot, Y. Fradet, É. Lévesque, C. Guillemette, L. Klotz

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Hoffman, K. Ding, F.-H. Joncas, L. Klotz

Study supervision: P. Toren, L. Klotz

We thank Mylène Vaillancourt for her help with organizing patient samples and the Canadian Cancer Trials Group for their support of the PR.7 clinical data repository and biobank. This project was funded by a Fonds de Recherche du Québec – Santé Clinician-Scientist Award (#32774), a 2016–2017 Urology Care Foundation Scholars Award, and a Prostate Cancer Canada Movember Discovery Grant (D2016-1393).

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