Contribution of Adrenal Glands to Intratumor Androgens and Growth of Castration-Resistant Prostate Cancer Free

Androgen deprivation therapy (ADT) remains front-line treatment for patients with locally recurrent or metastatic prostate cancer, but patients uniformly progress to castration-resistant prostate cancer (CRPC). Residual intratumoral androgen are believed to play a critical role in maintaining ligand-dependent mechanisms of androgen receptor (AR) activation (1). The source of residual tissue androgens in castrated patients is believed to reflect the tumoral uptake and intracellular conversion of the adrenal androgens such as DHEA-sulfate (DHEAS) to testosterone (T) and dihydrotestosterone (DHT), and/or the de novo intratumoral synthesis of androgens from cholesterol or progesterone precursors (2, 3). CYP17A, expressed in the human adrenal gland, testis and ovary, is a single enzyme with one active site that catalyzes sequential but independent hydroxylase and lyase reactions both of which are required for converting pregnenolone or progesterone to the adrenal androgens DHEA or androstenedione (AED), respectively. In particular, circulating levels of DHEA-S (the primary circulating form of DHEA) in eugonadal men are extremely high and are not reduced by standard castration (CX) therapy (4–6).

The contribution of adrenal steroids to CRPC progression is directly supported by historical case reports documenting clinical responses to adrenalectomy (ADX) in men with CRPC (7), and indirectly supported in the modern era by the efficacy of the CYP17A inhibitor abiraterone (ABI) in decreasing circulating adrenal androgens and improving overall survival for men with metastatic prostate cancer in the androgen sensitive and CRPC settings (8, 9). A direct impact of adrenal CYP17A inhibition on suppressing prostate tissue androgens can be inferred from the decrease in prostate androgen levels achieved by addition of ABI to standard ADT in the neoadjuvant setting, wherein tumoral CYP17A expression and de novo intratumoral androgen synthesis has not yet been upregulated (4, 10, 11). However, the antitumor activity of ABI in the setting of CRPC may also occur at the level of the tumor via inhibition of intratumoral CYP17A activity or via the direct anti-AR activity of ABI (2, 3, 12). Thus, the antitumor activity of ABI cannot necessarily be ascribed to inhibition of adrenal androgen synthesis alone, and the specific contribution of adrenally-derived steroids to intratumoral androgens and CRPC progression remains unproven.

The role of adrenal steroids in xenograft models of CRPC grown in castrated mice has not been previously investigated due to early literature suggesting a lack of functional CYP17A in rodent adrenal glands (13–17). Accordingly, the source of tumor androgens detected in patient-derived xenograft (PDX) models of CRPC grown in castrated male mice has been attributed solely to de novo intratumoral androgen synthesis (18). However, both historical studies using radioimmunoassays (RIA) and more recent studies using mass spectrometry (MS) have suggested functional adrenal CYP17A activity in rats and mice, including residual prostate androgens in castrated rats that were eliminated by ADX (19–23). Although the physiologic contribution of these steroid levels to normal prostatic biology was considered negligible (22, 23), this does not rule out a potential pathologic role in stimulating prostate tumor cells that are hypersensitive to androgen stimulation. We sought to experimentally demonstrate the contribution of adrenal steroids to intratumoral androgens and tumor progression in PDX models of CRPC grown in castrated male C.B-17 SCID mice in which we demonstrate adrenal CYP17A expression and activity.

All experiments involving animals were performed in accordance with protocols approved by the Fred Hutchinson Center Institutional Animal Care Use Committee (file 1775) and recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All surgery was carried out under isofluorane anesthesia. All studies utilized male C.B-17 SCID mice (Taconic). Serum samples were collected by retro-orbital bleed from non-tumor bearing 18- to 20-week-old intact (eugonadal) male mice, an age-matched cohort of mice at 12 weeks after CX (CX performed at 6–8 weeks), and an age-matched cohort at 12 weeks after both CX and ADX. Serum was obtained from a separate cohort of intact and castrated male mice for independent measurement of serum steroids in the laboratory of Dr. Penning. Adrenal glands, kidney, liver, and quadriceps muscle were resected for tissue steroid measurements. Drinking water was replaced with normal saline supplemented with corticosterone (Sigma; 25 μg/mL in 0.2% ethanol–0.5% NaCl) at time of ADX (24). Adrenal glands for assessment of CYP17A methylation and transcript profiling were resected from intact and castrated male mice ranging in age from 12 to 36 weeks (3–8 months) to mirror the ages spanning the course of a typical xenograft study.

All studies utilizing human tissue sources were conducted in accordance with recognized ethical guidelines (Belmont Report and U.S. Common Rule). The LuCaP35CR and LuCaP96CR lines are CX-resistant (CR) prostate cancer PDX models expressing wild-type AR established as part of the University of Washington Medical Center (UWMC) Genitourinary Biorepository as previously described (21). Cell line authentication is regularly performed via STR profiling. Castrated male C.B-17 SCID mice aged 8 to 10 weeks were subcutaneously implanted in bilateral flanks with 30 mm³ tumor pieces. Tumor volume was measured three times per week and calculated as length × (width²)/2. When tumors reached 200 to 250 mm³, drinking water was replaced with normal saline supplemented with corticosterone as described above, and mice were randomized to no further treatment (CX alone) or to undergo ADX (CX plus ADX). Tumors from n = 4 mice in each PDX model were harvested at 21 to 30 days following ADX. Mice were monitored until tumors reached approximately 1,000 mm³ (end of study, EOS) or animals became compromised, at which point animals were euthanized according to institutional protocol and tumors harvested for flash freezing and formalin fixation. Because of their prolonged time to tumor regrowth, a subset of mice in the ADX plus CX arms were euthanized and the data censored prior to reaching the size endpoint due to concerns for animal health.

In the LuCaP96CR study, n = 8 mice were bilaterally implanted to yield approximately nine evaluable tumors per arm for long term follow-up (∼60% take rate). Based on 25% variation in re-growth rates, nine tumors per arm is predicted to provide 80% power with two-tailed α of 0.05 to detect a 35% difference in mean time to regrowth (defined as 750 mm³). Because of the unexpectedly prolonged time to regrowth following CX plus ADX in this model (with interval attrition of mice due to health reasons) animal number in the combined treatment arm of the LuCaP35CR study was increased to n = 12.

Metastatic CRPC tissues for creation of the CRPC tissue microarrays (TMA) were obtained as part of the UWMC Prostate Cancer Rapid Autopsy Program, and comprised metastases from 42 patients (including 65 soft tissue metastases and 120 bone metastases, with up to four metastatic sites from each patient) as previously described (25, 26). The UWMC Institutional Review Board approved all procedures involving human subjects, and all subjects signed written informed consent.

Detailed methods for assessment of adrenal CYP17A methylation status by pyrosequencing, adrenal CYP17A protein expression by Western blotting, and quantitative RT-PCR in murine adrenal glands and xenograft tissues are given in the Supplementary Methods and Supplementary Table S1.

Detailed methods for determination of steroids in serum and tissue samples by mass spectrometry are given in the Supplementary Methods. The lower limits of quantitation (LLOQ) for steroids in serum were 0.01 ng/mL for pregnenolone, progesterone, AED, T, and androsterone; 0.02 ng/mL for DHEA and cortisol, 0.04 ng/mL for DHT and corticosterone, and 0.002 ng/mL for deoxycorticosterone. The LLOQ for steroids in tissue was 0.02 pg/mg for progesterone, AED, DHT, and T; 0.04 pg/mg for DHEA; and 0.08 pg/mg for pregnenolone. Methods for determination of steroids in serum by the Penning Laboratory are as previously published (27).

Details for creation of the PDX TMA are given in the Supplementary Methods. The CRPC TMA was created as previously published (26). Detailed methods for immunostaining and specific antibody clones are given in the Supplementary Methods and Supplementary Table S2. Quantitative image analysis of PDX TMA slides stained for AR and ARV7 was performed using metrics we have previously described and further detailed in the Supplementary Methods (28). Scoring for all other stains carried out in a blinded fashion by an experienced pathologist (L.T. or M.T.). A quasi-continuous IHC score was calculated by multiplying each intensity level (0 for no stain, 1 for faint stain, and 2 for intense stain) by the corresponding percentage of cells (0–100%) at the corresponding intensity and totaling the results. For all stains data from the two TMA spots per sample was averaged.

The nonparametric Mann–Whitney rank test was used for all two sample comparisons. P values < 0.05 were considered significant and those <0.10 as trending toward significance. Progression-free survival in CX versus CX plus ADX treated mice (defined as tumor size <750 mm³) was determined via Kaplan–Meier analysis with comparison of curves using the Mantel–Haenszel log-rank test. Statistical analyses were carried out using GraphPad Prism Software.

To determine the contribution of adrenal androgens to circulating serum levels in castrated mice, we evaluated steroid levels in serum from intact (INT) male C.B-17SCID mice, and from age-matched mice 12 weeks after CX or after CX plus ADX (Fig. 1). As expected, median levels of pregnenolone and progesterone in serum were similar in castrated versus intact mice [0.52 and 0.48 ng/mL pregnenolone, and 1.9 ng/mL vs. 3.2 ng/mL progesterone, p = not significant (ns) for both] and were significantly decreased following ADX [to 0.12 ng/mL, P = 0.002 pregnenolone, and to below detectable limits (bdl) for progesterone, P < 0.001; Fig. 1A and B], consistent with the adrenal origin of these steroids. Notably, median levels of DHEA in castrate mice were also similar to those in intact mice (0.03 ng/mL vs. 0.04 ng/mL, p = ns) and decreased to the limit of detection following ADX (to 0.02 ng/mL P = 0.001; Fig. 1C). Following CX median levels of AED and T observed in intact mice were decreased but detectable (from 0.03 to 0.01 ng/mL, P = 0.08 AED; from 1.9 to 0.02 ng/mL, P < 0.001 T), only becoming undetectable following ADX (P = 0.05 and 0.004, respectively; Fig. 1D and E), whereas levels of DHT in intact mice (0.15 ng/mL) became undetectable following CX alone (P = 0.05; Fig. 1F). Serum levels of DHEA, AED, T, and DHT from a separate cohort of intact and 6-week castrated male C.B-17 mice, were independently measured in the laboratory of Dr. Penning, and were consistent with our measurements, with the higher castrate T levels in the latter likely reflecting the shorter duration of CX (Supplementary Table S3).

Consistent with literature documenting corticosterone as the primary circulating mineralocorticoid in mice (13, 14), circulating levels of cortisol were undetectable (not shown), whereas levels of 11-deoxycorticosterone were similar in intact and castrated mice (2.6 ng/dL vs. 4.8 ng/mL, p = ns) and became significantly decreased following ADX (0.005 ng/dL, P < 0.001; Supplementary Fig. S1A). Because of the requirement for mineralocorticoid replacement (administered as corticosterone in drinking water), levels of corticosterone in the adrenalectomized mice were similar to those in the intact and castrated mice; Supplementary Fig. S1B).

Previously published data in rats and in the mouse adrenocortical Y1 cell line suggested epigenetic regulation of CYP17A expression in adrenal cells via methylation, possibly via activity of cAMP responsive element modulator (CREM) isoforms (29–31). However, we show that CpG islands in the CYP17A promoter DNA were unmethylated (Fig. 1G) and protein expression was detectable by Western blot analysis (Fig. 1H) in adrenal glands of intact and castrated mice ranging in age from 12 to 36 months (3–8 months; to mirror the ages spanning the course of a typical xenograft study). Transcripts of steroidogenic genes required for de novo and adrenal androgen synthesis (Star, Cyp11a, Cyp17a, Hsd3b1) and of Srd5a (required for production of DHT) were detectable in adrenal glands of intact mice and were significantly increased after CX (Supplementary Fig. S1C), as observed previously in some rodent studies (19, 32, 33). Notably, adrenal expression of melanocortin 2 receptor (Mc2r; also known as adrenocorticotropic hormone receptor; Acth) was also increased following CX, and is consistent with response to Acth as a driver of adrenal steroidogenesis in the castrate setting (sufficient serum for ACTH measurements was not available).

To confirm the functional significance of adrenal steroidogenic enzyme expression, we evaluated steroid levels in adrenal glands resected from eugonadal mice and age-matched castrated mice (12 weeks after CX; Fig. 1I). As expected, median levels of pregnenolone (4,238 pg/mg) and progesterone (1,876 mg/pg) were very high in adrenal glands of intact mice and remained unchanged by CX. Notably, median levels of AED (24.6 pg/mg) and T (10 pg/mg) in adrenal glands from intact mice were substantial, while levels of DHEA (0.7 pg/mg), DHT (1.7 pg/mg), and androsterone (1.4 pg/mL) were easily measurable. Similar to pregnenolone and progesterone, adrenal levels of AED, T, and DHT were also unchanged by CX, suggesting that testicular steroids do not serve as a source of these androgens detected within the adrenal gland. In contrast, median levels of androsterone in the adrenal gland were substantially decreased after CX, suggesting a testicular contribution to this steroid level (P = 0.095).

As serum androgen levels in castrated mice may be influenced by steroid production in other organs with steroidogenic potential, we assessed steroid levels in liver, kidney and muscle (Supplementary Fig. S2; refs. 34, 35). Notably, levels of pregnenolone, progesterone, AED, T, and DHT in the adrenal gland of castrated mice were 2–3 orders of magnitude higher than levels in liver, kidney and muscle (Supplementary Fig. S2A–S2F), suggesting active production specific to the adrenal gland. Moreover, while CX did not affect steroid levels in the adrenal gland (other than androsterone), it significantly decreased levels of AED in liver (P = 0.10; Supplementary Fig. S2D), of T in liver, kidney and muscle (P = 0.013, 0.05, 0.028; Supplementary Fig. S2E), of DHT in kidney (P = 0.028; Supplementary Fig. S2F), and of androsterone in liver (P = 0.039; Supplementary Fig. S2G). These observations demonstrate the testicular contribution to steroid levels detected in liver, kidney, and muscle, and support the hypothesis that circulating steroid levels observed in castrated mice are primarily derived from the adrenal gland rather than other organs.

To evaluate the functional relevance of adrenally-derived steroids in CRPC tumor models in vivo, we determined the impact of ADX on suppressing tumor growth and intratumoral androgens in the LuCaP96CR and LuCaP35CR PDX models in castrated C.B-17 SCID mice. In the LuCaP96CR PDX model (Fig. 2A), tumor regrowth was markedly delayed by CX plus ADX (CX + ADX, blue curves) compared with CX alone (CX, black curves), with an increase in median survival (to tumor endpoint of 750 mm³) from 25 days (in the CX group) to 300 days (in the CX + ADX group; P = 0.0002; Fig. 2B). In the LuCaP35CR model (Fig. 2C), tumor growth was markedly delayed by CX + ADX in a subset of tumors (slow, blue curves), whereas a subset (rapid, red curves) regrew with kinetics similar to the CX only group (black curves). Median survival in the CX only versus the entire CX + ADX cohort was delayed from 33 to 179 days (P = 0.005; Fig. 2D), with the growth prolongation primarily contributed by the slow subset of tumors (median survival not reached vs. 66 days in the Rapid group, P = 0.008; Fig. 2E).

In LuCaP96CR median intratumor levels of pregnenolone, progesterone, T, and DHT were decreased by an order of magnitude at 30 days after ADX + CX (ADX d30) compared with CX alone (pregnenolone 24.3 pg/mg vs, 2.7 pg/mg, P = 0.008 (not shown); progesterone 1.4 pg/mg vs. 0.03 pg/mg, P = 0.001; T 0.9 pg/mg vs. 0.03 pg/mg, P = 0.001; DHT 1.2 pg/mg vs. 0.03 pg/mg, P = 0.001; Fig. 3A). Tumor levels of DHEA and AED were lower at ADX d30 versus CX alone, although the decreases did not reach statistical significance (p = ns for both). Although levels of DHEA, AED, T, and DHT remained suppressed near the limit of detection in the end of study (EOS) tumors that regrew after CX + ADX (ADX EOS), progesterone levels in these EOS tumors were increased compared with tumors resected at day 30 after ADX (0.03 pg/mg vs. 0.125 pg/mg, P = 0.0317), although still lower than tumors in the CX only group.

In LuCaP35CR median intratumor levels of progesterone, T and DHT were also decreased by an order of magnitude by 21 days after ADX (ADX d21) compared with CX alone (progesterone 0.21 pg/mg vs. 0.02 pg/mg, P = 0.002; T 0.6 pg/mg vs. 0.02 pg/mg, P = 0.0004; DHT 3.2 pg/mg vs. 0.25 pg/mg, P = 0.0004; Fig. 3B), whereas decreases in AED trended toward significance (0.03 pg/mg vs. 0.02 pg/mg, P = 0.07, respectively). Steroid levels in tumors that regrew after CX + ADX (ADX EOS) were significantly below those in the CX alone group for progesterone, DHEA, T, and DHT (P = 0.056, 0.02, 0.001, and 0.001, respectively); however, median levels in tumors that regrew rapidly (red dots in Fig. 3B) trended toward being higher than levels in tumors that regrew slowly (blue dots) for progesterone (P = 0.07), T (P = 0.03), and DHT (P = 0.17), suggesting the more rapid regrowth in these tumors was in part stimulated by the elevation in tumor androgens. In both models, serum androgens levels following ADX were largely below detectable limits (not shown), precluding assessment of correlation with tumor androgens.

Notably, surgical ADX showed a significantly more profound impact on suppressing tumor androgens and CRPC tumor growth than did previously published studies from our group using ABI, in which LuCaP96CR was classified as an intermediate responder, and LuCaP35CR showed no response in one study and moderate response in another (18, 36). Compared with these studies, ADX more consistently suppressed tumor levels of pregnenolone, progesterone, DHEA, and AED (blue bars, Supplementary Fig. S3A–S3D) than did ABI (red bars), whereas in at least in one model (LuCaP35CR) ABI was associated with increases in pregnenolone and progesterone (immediately upstream of the first enzymatic function of CYP17A inhibited by ABI), and DHEA (immediately upstream of HSD3B1, which is also inhibited by ABI; ref. 37). Similarly, ABI clearly suppressed EOS levels of T and DHT (red bars, Fig. 3C and D); however, ADX (blue bars) appeared to do so more consistently.

We determined the impact of ADX on key mechanisms of resistance identified in treatment refractory CRPC including expression of AR, the prevalent AR splice variant—ARV7, glucocorticoid receptor (GR, NR3C1) and steroidogenic enzymes, as potential mechanisms driving tumor re-growth under the stringent suppression of ligand levels observed in the setting of CX + ADX (3, 18, 36).

In LuCaP96CR, we observed no change in transcript levels of AR, but a significant induction of ARV7 (4.9-fold, P = 0.005), decrease in PSA (−2.6-fold, P = 0.009), and increase in expression of GR (3.6-fold, P = 0.028) in the EOS tumors (Table 1). IHC staining of a TMA created from this PDX model (Fig. 4A) similarly showed no change in AR following ADX, an increase in nuclear ARV7 staining (P = 0.08 at EOS, P = 0.05 for EOS vs. ADX d30), and a decrease in cytoplasmic PSA (P = 0.016 at ADX d30, P = 0.035 at EOS). However, an increase in nuclear GR staining was not observed, which instead showed an apparent decrease (P = 0.025 at ADX d30). These data suggest induction of ARV7 as a significant mechanism of resistance in this PDX model, potentially accounting for the observation that transcript levels of some androgen regulated genes (e.g., TMPRSS2 and FKBP5) were not necessarily decreased in the EOS tumors.

In LuCaP35CR, we observed a significant induction of AR (3.5-fold, P ≤ 0.001), ARV7 (3.6-fold, P = 0.004), and GR message (1.6-fold, P = 0.017) in the EOS tumors (Table 1). There was no change in the androgen-regulated genes PSA, TMPRSS2, or FKBP5 at the transcript level in this PDX model, although IHC staining (Fig. 4B) did show a decrease in cytoplasmic PSA (at ADX d21 and at EOS, P = 0.0016 for both). Consistent with the transcript results, we observed an increase in nuclear AR, ARV7, and GR staining in this model (Fig. 4B) that was moderately more significant at ADX d21 (P = 0.004, 0.029, and 0.025, respectively) than in the EOS tumors (P = 0.106, 0.095, and 0.15, respectively). We did not observe significant differences in AR, PSA, or GR staining in the EOS tumors that re-grew rapidly (red dots, Fig. 4B) versus those that regrew slowly (blue dots). However, the increase in ARV7 seen at ADX d21 is appeared to be sustained in the EOS tumors that recurred with low androgen levels (blue dots, here and in EOS T and DHT levels, Fig 3B), but not in the EOS tumors that recurred with an increase in tumor androgens (red dots, here and in EOS T and DHT levels, Fig. 3B). These data suggest induction of AR, ARV7, and GR as potentially significant mechanism of resistance in this model, with ARV7 playing a more important role in tumors that regrew without an increase in intratumoral androgen levels, consistent with the known inverse relationship between androgen levels and ARV7 expression (38). Of note, corticosterone supplementation itself had no impact on tumor growth (Supplementary Fig. S4A) or IHC expression of GR (Supplementary Fig. S4B) or of AR and ARV7 (not shown). This suggests the induction of GR expression in the ADX-treated tumors in this model reflects the impact of ADX and not the replacement corticosterone dosing.

Consistent with the increased intratumoral levels of T and DHT observed in a subset of the ADX recurrent LuCaP35CR tumors (red dots, Fig. 3B), we observed increased expression of several genes involved in mediating steroidogenic enzyme activity (Table 1; see Supplementary Fig. S5 for schema of androgen biosynthesis enzymes), including CYB5A (enhances activity of CYP17A and HSD3B, 3.8-fold P < 0.001), NR5A1 (transcriptional regulator of multiple steroidogenic genes, 3.1-fold P = 0.001), AKR1C3 (8.3-fold P < 0.001), HSD17B3 (1.9-fold P = 0.06), and SRD5A1 (1.8-fold P < 0.001; refs. 39, 40). In contrast, although ADX EOS tumors in the LuCaP96 model also showed increases in expression of CYB5A1 and NR5A1 (two-fold for both, P = 0.018 and 0.026, respectively) and SRD5A1 (3.1-fold P = 0.018), they did not show induction of the steroid producing genes AKR1C3 or HSD17B3 but rather induction of UGT2B15 (4.9-fold P = 0.014), which conjugates T and DHT for excretion, potentially explaining the differences in androgen levels observed in the two models.

To further assess the translational relevance of the steroidogenic enzyme transcripts observed in the ADX-treated tumors, we evaluated AR axis proteins (AR, PSA) and key enzymes required for de novo steroid synthesis and/or conversion of adrenal androgens (HSD3B1, CYP17A1, AKR1C3, and HSD17B3; Supplementary Fig. S5) in a TMA of bone and soft tissue CRPC metastases from 43 patients (two to six metastases per patient). Despite some degree of intrapatient heterogeneity, grouping of tumors by patient of origin revealed several distinct staining profiles, including subsets of patients with (i) AR⁻ and PSA⁻ tumors with low expression of steroidogenic enzymes (Fig. 5A), (ii) AR⁺ tumors with negative (Fig. 5B top) or positive (Fig. 5B bottom) PSA expression and low steroidogenic potential (based on generally low expression of HSD3B1, which is required for both de novo synthesis and conversion of adrenal androgen), and (iii) AR⁺ and PSA⁺ tumors with high steroidogenic potential (Fig. 5C, based on moderate to high expression of HSD3B1). Tumors in the latter group can be further subdivided based on potential for adrenal androgen conversion [e.g., coordinate expression of HSD3B1, HSD17B3/AKR1C3 (which catalyze the same reaction) but low CYP17A1; Fig. 5C, top], or potential for de novo steroidogenesis (e.g., coordinate expression HSD3B1, HSD17B3/AKR1C3, and high CYP17A; Fig. 5C, bottom). Representative images of tumors showing coordinate expression of HSD3B1, HSD17B3/AKR1C3, and CYP17A are shown in Supplementary Fig. S6. As we have previously published, a subset of these tumors show loss of C terminal AR staining, consistent with the presence of truncated AR variant species (41).

Clinical studies of ABI in men with metastatic prostate cancer are consistent with an important role for CYP17A-mediated adrenal androgen production in CRPC progression, particularly as higher pretreatment levels of circulating adrenal androgens associate with better response to CYP17A inhibition (42–44). However, inhibition of intratumoral CYP17A activity is not excluded, and ABI and its metabolites can directly target the AR (12). Thus, these studies do not actually isolate the specific contribution of adrenal steroids in CRPC progression. It has been widely held that rodent adrenal glands do not produce adrenal androgens and that rodent models cannot, therefore, be used to test this facet of prostate cancer biology. We now show that adrenal CYP17A is present in C.B-17 SCID mice, that steroids downstream of CYP17A are detectable in the adrenal glands and serum of castrated mice, and that surgical ADX markedly suppresses serum and tumor steroid levels and delays tumor re-growth beyond that observed with CX alone in two PDX models of CRPC. This is the first study to demonstrate the specific impact of adrenal steroids on intratumoral androgens and CRPC tumor growth in vivo.

Functional CYP17A is transiently present in rodent adrenal glands during the fetal/early postnatal period (45, 46) and then decreases, likely via epigenetic methylation and repression (29). This phenomenon is not universal however, as we show that adrenal CYP17A DNA is unmethylated and detectable in C.B-17 SCID mice, consistent with the spiny mouse (Acomys cahirinus) that is known to express functional CYP17A (46). Gonadectomy induces adrenal gland expression of steroidogenic genes and increases circulating levels of AED and T in certain inbred mouse strains (33). We did not observe differences in CYP17A methylation status or in transcript and protein levels between intact and castrated mice, although we did observe statistically significant increases in transcript expression of STAR, CYP11A, HSD3B1, and SRD5A2 at 12 weeks following CX.

Importantly, we show levels of adrenal androgens (DHEA, AED) and downstream metabolites (T, DHT, and androsterone) are easily detectable in adrenal glands of intact mice and do not decrease following CX. The adrenal levels of DHEA (0.7 pg/mg, range 0.44–1.2 pg/mg) and AED (44 pg/mg, range 4–97 pg/mg) in our study are consistent with the findings of Hu and colleagues who reported levels of DHEA (∼4–7 pg/mg) and AED (∼2–4 pg/mg) in the adrenal glands of rats (Wistar, via GC/MS) 8 weeks after CX, equivalent to levels in intact rats (23). Moreover, we show adrenal levels of AED, T, and DHT in castrated mice to be nearly 2 orders of magnitude higher than in kidney, liver, or muscle (other organs shown to have steroidogenic capacity in mice; refs. 34, 35), suggesting active production specific to the adrenal gland.

An existing body of data, published primarily in the endocrine literature and not well-recognized in the prostate cancer field, supports functional adrenal CYP17A activity in adult rodents and may reflect strain-specific differences. Although some rodent adrenal suspension studies failed to detect production of 17-hydroxyprogesterone or AED (products of CYP17A hydroxylase and lyase activity, respectively) via RIA (13), others did measure AED or DHEA (via RIA; refs. 46, 47) or showed metabolism of [3H]DHEA or [3H]AED to T and DHT in vitro (48, 49). Similarly, serum levels of T, DHT, and AED were undetectable in some strains of castrated rats (Holtzmann, via RIA; ref. 50); Wistar, via LC/MS; ref. 51), but detectable in others (Sprague–Dawley, via LC/MS (19, 52). Kyprianou and Isaacs detected circulating (T 0.10 ng/mL and DHT 0.16 ng/mL) and prostatic levels of T and DHT (via RIA) in castrated Copenhagen rats that were eliminated by ADX (22), although a study in Wistar rats showed a decrease in prostate levels of T and DHT after CX but no sustained decrease in AED or DHEA (via GC/MS), consistent with an adrenal origin (23). In context of this study, the relevance of this literature becomes more apparent and provides significant support for our findings.

Strain-specific differences in adrenal androgen production are suggested by the observation that certain inbred strains of mice (DBA/2J, CE/J, C3H, NU/J, BALB/c, and B6D2F1) are highly sensitive to gonadectomy induced sex-steroid producing adrenocortical neoplasia, whereas others (C57BL/6 and FVB/N) are not (53). The mouse strains associated with adrenal androgen production in the literature are consistent with those susceptible to adrenal neoplasia, with androgens observed in BALB/c mice (via RIA; ref. 13) but not in C57BL/6 mice (via LC/MS; ref. 54). In this respect it is notable that the C.B-17 SCID mice used in our study are derived from the BALB/C strain.

Notably, surgical ADX showed a significantly more profound impact on suppressing tumor androgens and CRPC tumor growth than did previously published studies from our group using ABI (18, 36). Although ABI clearly suppressed EOS levels of T and DHT in the prior studies (red bars, Fig. 3C and D), ADX (blue bars) appeared to do so more consistently. Differences in growth suppression might also reflect the AR agonist activity attributed to a 5α reduced metabolite of ABI (55), an effect that would be absent in the ADX-treated tumors.

Given the clear impact of both ADX and ABI on tumor androgen levels, we sought to determine whether mechanisms of resistance to ADX mirrored those observed with ABI. Consistent with our previous findings in LuCaP96CR tumors treated with ABI (36), we observed increased transcript levels of ARV7 and GR, and no increase in transcript or nuclear staining for full-length AR in the ADX-treated tumors (although we did not observe increased nuclear expression of GR as noted in that study). In LuCaP35CR tumors, we observed increased transcript expression and nuclear staining of AR and ARV7 and an increase in nuclear GR staining after ADX, consistent with prior data (although increased AR expression was only observed in one of the prior studies likely due to PDX heterogeneity; refs. 18, 36). These data suggest induction of ARV7 as a potentially significant mechanism of resistance following ADX in LuCaP96CR, and induction of AR, ARV7, and GR as potentially significant mechanisms of resistance in LuCaP35CR. Although AR and ARV7 share an overlapping transcriptome, transcripts of AR target genes are not consistently increased in the ADX-treated tumors, consistent with prior observations that ARV7 is not necessarily as potent as full-length AR in inducing the expression of canonical AR genes (56, 57).

Our data suggest that the adrenal gland contributes substantially to levels of T and DHT in CRPC tumors grown in castrated mice, and is consistent with our data demonstrating expression of enzymes required for adrenal androgen utilization in a majority of AR+ CRPC metastases examined (Fig. 5C). However, a subset of ADX-resistant LuCaP35CR tumors demonstrate induction of steroidogenic enzymes and moderate reconstitution of intratumoral progesterone, T and DHT (similar to our prior data in ABI-recurrent LuCaP35CR tumors (18). These suggest that at least some tumor models are also capable of de novo steroid production and is consistent with the concurrent expression of enzymes required for de novo steroidogenesis in a subset of the CRPC tumors we evaluated (Fig. 5C, bottom). As increased progesterone levels were present in a subset of the ADX EOS tumors versus earlier time points in both PDX models, we determined whether these EOS tumors showed induction of the progesterone receptor (PR) as possible mechanism of progression (58). However, staining for PR was low to undetectable in the CX only tumors, without any difference in the ADX-treated tumors (data not shown). A direct effect of progesterone on stimulation of growth via wild-type (WT) AR cannot necessarily be excluded, as Kumagai et al. did not detect downstream conversion of progesterone to T or DHT in VCaP cells (which has WT AR), yet still observed modest activation of growth (59). Activation of WT AR by a metabolite directly downstream of progesterone (20β-hydroxy-5α-dihydroprogesterone) has also been recently demonstrated (60). We did not measure levels of this particular steroid in our assay, but as its production from progesterone is not mediated by CYP17A it suggests another reason why surgical ADX appears to have better anti-tumor efficacy than CYP17A inhibition with ABI.

We and others have previously demonstrated transcript expression of steroidogenic enzymes in subsets of CRPC metastases, but data on concordant protein expression of key enzymes required for androgen synthesis from adrenal androgens or earlier precursors are limited (2, 3, 61–66). Here, we demonstrate clear subsets of tumors with coordinate expression of the critical enzymes required for de novo steroid synthesis and/or conversion of adrenal androgens. Steroidogenic enzyme expression is generally low in AR negative tumors (Fig. 5A), and in AR positive tumors with low PSA expression (Fig. 5B, top), indicative of poor response to agents targeting the AR or ligand synthesis. Low expression of steroidogenic enzymes is also observed in a subset of AR and PSA positive tumors (Fig. 5B, bottom), suggesting AR axis signaling in response to exogenous ligand is maintained, but without intratumoral androgen production. In contrast, tumors from nearly half the patients (Fig. 5C) appear potentially capable of androgen production from adrenal androgens (based on concordant expression of HSD3B1 with AKR1C3 or HSD17B3,), and of these, half also show concordant expression of CYP17A (Fig. 5C, bottom), suggesting capacity for de novo steroidogenesis.

Although we did not observe consistent differences in steroidogenic capacity between bone and soft tissue tumors overall, there was more pronounced staining in bone versus soft tissue metastases in several patients (red arrows in Fig. 5C), potentially consistent with the observation that osteoblasts may induce steroidogenic enzymes in CRPC cells (67). Unexpectedly, some tumors with loss of C terminal AR staining (consistent with the presence of ligand-independent AR variants) also demonstrated steroidogenic enzyme expression, suggesting that in certain cases the response to CX may include both induction of AR variants and upregulated steroid production. Alternatively, ARV7 is a strong inducer of the glucuronidating enzyme UGT2B17, which conjugates active androgens for excretion, and could therefore continue to promote AR variant production in the presence of concomitant steroidogenesis (41).

In summary, we show that adrenally-derived steroids are produced in C.B-17 SCID mice and contribute to tumor androgen levels and growth in two PDX models of CRPC, demonstrating that C.B-17 SCID mice are an appropriate model for evaluating the impact of steroidogenesis inhibitors in CRPC xenograft studies. Although the potential impact/availability of low circulating T levels may be magnified in mice compared with humans due to the lack of circulating sex hormone binding globulin (SHBG) in mice (68), castrate levels of DHEA, AED, T, and DHT in this study were similar to those in ABI-treated CRPC patients (as previously reported for T; ref. 68). These suggest a reduction in adrenally-derived androgens beyond those achieved by ABI may have clinical benefit in men with CRPC, and is consistent with our observation that tissue androgens and tumor growth were suppressed more strongly by surgical ADX in this study than in previously reported studies using ABI. These suggest that proof-of-concept studies testing agents capable of achieving true “nonsurgical ADX” are warranted (such as the recently reported CYP11A inhibitor ODM-208; ref. 69) and that optimal clinical efficacy may be obtained by a combination of CYP11A and CYP17A inhibition. Although beyond the scope of the current work, PDX studies testing CYP11A inhibition are planned. We find that mechanisms of resistance after surgical ADX in CRPC models are similar to those observed following ABI, including the induction of AR, truncated AR splice variants, GR, and steroidogenesis. This is consistent with the fact that both approaches act via suppression of tissue androgen levels, and suggests optimal clinical activity will be achieved by combination strategies cotargeting AR or GR with agents capable of fully abrogating adrenal and intratumoral ligand synthesis.

E.A. Mostaghel is a consultant/advisory board member for Orion Pharma. T.M. Penning reports receiving other commercial research support from Forendo, receiving speakers bureau honoraria from Congress Steroid Research, holds ownership interest (including patents) in Penzymes, and is a consultant/advisory board member for Markey Cancer Center, Research Institute for Fragrance Materials, CounterAct-Rutgers University, Columbia University SPH, and Berkeley U-Superfund Research Program. No potential conflicts of interest were disclosed by the other authors.

Conception and design: E.A. Mostaghel, S.P. Balk, P.S. Nelson

Development of methodology: E.A. Mostaghel, A. Zhang, D. Tamae, M. Tretiakova, R. Dumpit, T.M. Penning, E. Corey, L.D. True, P.S. Nelson

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): E.A. Mostaghel, A. Zhang, S. Hernandez, B.T Mark, D. Tamae, J. Bartlett, J. Burns, L. Ang, A.M. Matsumoto, T.M. Penning, C. Morrissey

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): E.A. Mostaghel, A. Zhang, X. Zhang, D. Tamae, H.E. Biehl, M. Tretiakova, A.M. Matsumoto, L.D. True, P.S. Nelson

Writing, review, and/or revision of the manuscript: E.A. Mostaghel, A. Zhang, D. Tamae, M. Tretiakova, A.M. Matsumoto, T.M. Penning, S.P. Balk, E. Corey, P.S. Nelson

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): E.A. Mostaghel, S. Hernandez, E. Corey, L.D. True, P.S. Nelson

Study supervision: E.A. Mostaghel, P.S. Nelson

Other (pathology review): M. Tretiakova

The authors wish to acknowledge reagents and technical assistance in pyrosequencing provided by Dr. William Grady, and expert technical assistance with image analysis provided by Jonathan Henriksen and NWBioSpecimen (a core service for procurement and annotation of research biospecimens), which is supported by National Cancer Institute grant P30 CA015704 (G. Gilliland, principal investigator [PI]), Institute of Translational Health Sciences grant UL1 TR000423 (M. Disis, PI), the University of Washington School of Medicine and Department of Pathology, and Fred Hutchinson Cancer Research Center, NIH Pacific Northwest Prostate Cancer SPORE P50 CA97186 (EAM, EC, LDT, PSN), NIH P01 CA163227 (EAM, TMP, SPB, EC, LDT, PSN), DOD W81XWH-12-1-0208 (EAM), NIH P30 CA015704 (JH), NIH UL1 TR000423 (JH), and Department of Veterans Affairs Puget Sound Health Care System (EAM, AMM).

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.