Targeting of the androgen receptor (AR) axis has been the foundation of therapy for advanced prostate cancer for over 75 years. The seminal work by Huggins and Hodges in the 1940s led to the widespread use of surgical orchiectomy for symptomatic metastatic prostate cancer [1]. Medical castration was introduced in the 1980s and has largely supplanted surgical orchiectomy as the preferred method of androgen-deprivation therapy (ADT) [2]. Other recent advances have further increased our ability to inhibit the AR axis: abiraterone is an inhibitor of CYP17 (cytochrome P450 17A1, or 17α-monooxygenase, 17α-hydroxylase, 17,20-lyase) that effectively blocks adrenal production of testosterone [3], and enzalutamide represents a new class of potent and specific antiandrogens [4]. Both agents were initially approved in the metastatic, castration-resistant setting, but are now used for hormone-sensitive disease with significant prolongation in the overall survival. The combination of abiraterone or enzalutamide with standard ADT provides a very effective blockade of the AR axis. Arguably, further clinical benefit from more effective AR inhibition seems unlikely.

The use of more-intensive strategies to block androgen signaling has now been extended into the metastatic, hormone-sensitive setting. For example, in the LATITUDE trial, men with metastatic, castration-sensitive prostate cancer were randomized to ADT with or without abiraterone as initial systemic therapy [5]. Patients included in this study had one of three high risk factors: Gleason score of ≥8, presence of three or more lesions on bone scan, or presence of visceral organ involvement. With a median follow-up of 51.8 months, the overall survival was improved with abiraterone treatment (median 53.3 months), compared to the placebo group (36.5 months), with a hazard ratio of 0.66 (95% CI 0.56−0.78; p < 0.0001). In the combination group, a significant long-term benefit was realized, with the median time to subsequent prostate therapy and time to secondary progression-free survival in excess of 4 years.

Despite more aggressive upfront therapy, the outcomes of metastatic prostate cancer remain poor, especially after the development of castrate-resistant disease. For example, in later-stage patients with castration-resistant prostate cancer who have progressed after docetaxel chemotherapy and an androgen-signaling-targeted inhibitor (e.g., abiraterone or enzalutamide), the use of secondary chemotherapy versus an additional androgen-signaling-targeted inhibitor has been studied. In this setting, the use of secondary chemotherapy was superior to an additional androgen-signaling-targeted inhibitor; however, the median overall survival was 13.6 months with cabazitaxel versus 11.0 months with the androgen-signaling-targeted inhibitor, which highlighted the limits of treatment efficacy in later-stage prostate cancer patients.

While both preclinical and human studies have identified potential mechanisms for resistance to ADT, such as expression of AR variants [6] and overexpression of c-myc [7,8], currently there are no effective strategies to target these mechanisms Thus, there is an unmet need for effective therapies for men progressing on androgen-targeting therapy.

The centrality of androgen signaling in prostate cancer was highlighted by the report of Chen et al. in that a moderate increase in AR gene expression was the only consistent change in gene expression in CRPC, and that this increase in receptor mRNA and protein was required for resistance [9]. There are longstanding observations in preclinical models of CRPC that androgens can have a suppressive effect [8,10,11,12,13,14,15,16,17,18,19]. While periodic suspension of androgen suppression showed promise in animal models, in human trials intermittent androgen suppression did not produce the desired benefit in tumor progression. The most important clinical trial in this area, SWOG9346, was designed as a noninferiority trial of continuous versus intermittent ADT in untreated metastatic patients [20]. The results of this large trial were inconclusive with respect to survival between the treatment arms, and decreased survival with intermittent ADT could not be ruled out. There were initial improvements in quality of life in those treated with intermittent ADT, including with erectile function, but these largely dissipated beyond the first 9 months of starting the trial. Importantly, animal studies differed from the human trials in that exogenous testosterone was administered in the animal studies, whereas in humans, testosterone was allowed to recover simply by suspending the chemical suppression of androgens. In this older population of men, reestablishment of physiologic testosterone levels can be relatively slow, more than a year in some patients, and incomplete. This may allow tumor cells sufficient time to adapt to the changing androgen environment in contrast to the animal studies where treatment protocols permit rapid alteration of androgen levels. However, the translation of the preclinical findings to the clinical setting has been delayed in large measure due to concerns about the potential stimulatory effect of androgen in advanced prostate cancer patients. Indeed, in the landmark paper by Huggins and Hodges detailing the benefit of surgical castration, the effect of androgen administration was also reported in a small number of prostate cancer patients, establishing the long-held clinical concern about this approach: “Testosterone propionate caused an increase of serum acid phosphatase (tumor marker of that era)] above the pre-injection level in these patients. Following the cessation of injections, there was a decline to the preliminary level.” [1].

More recently, there has been renewed clinical interest in the potential therapeutic and quality-of-life benefit of HDT in CRPC. Much of the recent work has been spearheaded by Sam Denmeade and colleagues at Johns Hopkins University, including a small pilot clinical study in patients with CRPC published in 2015. In this study, investigators administered HDT as bipolar androgen therapy (BAT), in which patients were given a high-dose injection of testosterone cypionate monthly [21]. This produced transient supraphysiologic testosterone levels of more than 1500 ng/dl in some patients. Although etoposide was also used in the early studies, subsequent clinical trials of BAT have not utilized etoposide due to its unclear contribution to the clinical effect noted. In this pilot, prostate-specific antigen (PSA) reductions were seen in 7 of 14 evaluable patients, with radiographic responses also noted in 5 of 10 evaluable patients with radiographically measurable disease. Of the 7 patients with a PSA response, the median time to PSA progression was 221 days. Subsequent studies of BAT have been performed in CRPC patients progressing after treatment with enzalutamide, with 9 of 30 (30%) demonstrating a PSA response [22]. Building on this initial experience, a multicenter randomized trial of BAT has now been completed [23]. In this trial, men with metastatic CRPC previously treated with abiraterone, but without previous potent next-generation antiandrogen (e.g., enzalutamide) exposure, were randomized to enzalutamide versus high-dose intramuscular testosterone with crossover at progression. A PSA decline of 50% was noted in 28% with BAT versus 25% with enzalutamide, demonstrating the activity of BAT in this setting. With the crossover design, it was also noted that BAT followed by enzalutamide did sensitize some patients to subsequent antiandrogen therapy [23].

In these clinical reports, several patterns of response to BAT have been described [22,23]. Some patients had no response, other had a prolonged response, while others yet had an initial response followed by the development of resistance and progression. These descriptions are consistent with a substantial proportion of CRPC patients with de novo resistance to HDT and others who acquire resistance over time, after an initial response. Through understanding the mechanisms that produce de novo or acquired resistance, there is an opportunity to identify patients likely to respond to HDT or to prolong responses in those with an initial response to HDT. Moreover, these studies have identified a new clinical state: high dose testosterone resistant CRPC. To put this into context requires an understanding of how cancer adapts to a changing androgen environment, and whether these adaptations may reveal new vulnerabilities that can be exploited clinically.

At initial clinical presentation, most prostate cancer patients’ tumors are dependent on androgens for growth. Unlike hormone-dependent mammary cancers, there is a paucity of reliable cell models of hormone-dependent prostate cancer. Indeed, key early studies on androgen-dependency employed the androgen-dependent Shionogi mouse mammary tumor model [24]. Many subsequent studies have employed the LNCaP human prostate line cell line [25], the growth of which, both in culture and as xenografts, is androgen-dependent. Many investigators have used LNCaP cells to explore the basis of androgen dependence and the development of androgen-independent growth. While it is beyond the scope of this paper to exhaustively review all the worthy work in this area, we acknowledge the contributions of the late Nicholas Bruchovsky [10,24], the late Shutsung Liao [8,11,12,13,14], and their collaborators in Vancouver and Chicago, respectively, whose use of preclinical models has been critical to the development of concepts that underlie our current understanding of androgens and androgen resistance in prostate cancer. Seminal studies from John Isaacs and his collaborators at Johns Hopkins have promoted the concept of testosterone administration for CRPC and developed a mechanistic understanding of how androgens can inhibit the growth of prostate cancer cells [15,16,17,18,19,26,27]. Recent studies out of the University of Washington have implicated DNA damage mediated by liganded receptors as a mechanism for the tumor-suppressive activity of supraphysiologic testosterone and suggest that patients whose tumors exhibit defects in DNA repair pathways may be particularly responsive to testosterone therapy [28,29]. The once-heretical idea that prostate cancers adapted to a low-androgen environment may be growth-suppressed by high-dose testosterone has now led to successful clinical trials [23].

In the sections that follow, we offer illustrative examples utilizing unpublished preclinical studies from our own laboratories that emphasize the observations we and others have made that have led to insights into the role of androgens in prostate cancer, and how these insights may be exploited clinically.

The LNCaP prostate cancer cell line is derived from a lymph node metastasis and is dependent on androgens for growth in cell cultures and in xenografts [25]. When switched from growth medium (10% fetal bovine serum) to low-androgen medium (5% charcoal stripped fetal bovine serum), LNCaP cell growth ceases. A careful dose response of early passage LNCaP (≈30 from derivation) to the nonmetabolizable androgen, R1881, reveals an important but underacknowledged feature: growth of LNCaP cells exhibits a biphasic dose response to androgen exposure. Growth is maximally stimulated between 10 and 30 pM R1881 to a level approaching that seen in growth medium, but falls off dramatically at higher doses (Figure 1, red). We have observed that a biphasic dose response is also seen with dihydrotestosterone instead of R1881, and in the androgen-responsive VCaP prostate cancer cell line. The implication of data such as these is that growth of LNCaP cells, while requiring androgens, is carefully balanced between insufficient and excessive androgen signaling for optimal growth.

When growth medium is supplemented with a high level of R1881 (10 nM), LNCaP cell growth is partially inhibited, and a subset of cells undergo apoptosis. Nonetheless, LNCaP cells adapt to continuous exposure to high levels of R1881 in growth medium and will resume growth. R1881-adapted cells exhibit a dose response that is right-shifted to become a classic S-shaped curve (Figure 1, blue), unlike the biphasic response of the parental line (Figure 1, red). LNCaP cells have been widely used and often passaged extensively. We have observed a propensity for LNCaP cells to exhibit some right shift in their biphasic growth response to androgen after repeated passaging. Together, such observations imply that LNCaP cells have a well-defined androgen-sensing system that regulates growth in response to the environment.

Like many investigators, we selected LNCaP cell variants as models of CRPC through long-term culturing in low-androgen medium. The CRPC LNCaP variant, P1, was derived from a parental LNCaP line (P0) that had previously been engineered to contain an androgen-responsive luciferase reporter gene. Two other independent lines (M1, M2) were selected similarly from the M0 parental line (see the Materials and Methods section). A fourth CRPC model, 4D1enz1, was derived by selection of M0 in the androgen antagonist, enzalutamide, followed by further selection in low-androgen medium (Table 1). As models, these CRPC variants reflect many of the properties of human CRPC, including overexpression of AR mRNA and protein that enables the cells to adapt to a low-androgen environment (Figure 2 and Figure S1). Notably, we did not observe expression of mRNA, nor protein, for the truncated V7 version of the AR. In addition, Table 1 lists the variant lines derived from the CRPC variants by further manipulation of their androgen environment. The characterization of these high-testosterone-resistant lines and androgen-independent lines will be addressed below.

The adaptation of CRPC LNCaP variant lines to a low-androgen environment was evidenced by their growth properties in vivo. The sensitivity of CRPC cells to androgens is highlighted in the xenograft data shown in Figure 3. The P1 CRPC model was injected into both flanks of both male and female NOD/SCID and nude mice. Tumor growth of the P1 CRPC model was slower in male mice of both strains compared to female hosts, suggesting that androgens may inhibit their growth, even though the mean testosterone levels of the males in these immunocompromised mouse models were more similar to levels of hypogonadal, not eugonadal, men [30]. Supporting the suggestion that androgens mediate growth inhibition of this CRPC tumor model was the observation that testosterone levels in female nude mice were less than one-tenth that of the males [31]. Interestingly, tumor growth was faster in nude females than in NOD/SCID females, though in both, tumor growth was faster than in the males. Thus, LNCaP variants selected in vitro for growth in a low-androgen environment were growth-inhibited even by the relatively low level of androgens present in intact male immunocompromised mice, predicting the potential utility of testosterone supplementation in men whose disease has progressed in the face of ongoing androgen suppression.

The response of the P1 CRPC variant to androgen supplementation was evaluated further in xenograft studies. In the experiment shown in Figure 4, castrate-resistant P1 cells were injected into one flank of female nude mice. Tumors were permitted to grow to a readily palpable volume, at which time mice with tumors of a similar volume were paired; one received a testosterone pellet implant and the other remained untreated. The tumors in untreated control mice continued to grow, whereas in the T-treated animals, all tumors ceased growth or regressed. In three of six mice, tumors regressed to the point of no longer being palpable. The tumors of the remaining three animals regressed somewhat or showed no progression for almost 2 months. One control mouse received a testosterone implant after a large tumor had formed (arrow, Figure 4). The tumor regressed by 75% over the next 6 weeks. The tumor from the one testosterone-treated mouse that exhibited little or no regression eventually relapsed, and the tumor began to grow despite ongoing testosterone treatment. Thus, cell-line-derived tumors exhibited heterogeneity in their response to testosterone, as has been seen in the human prostate cancer trials employing BAT.

With the exception of the relapsed tumor in the experiment above, there was insufficient tumor to analyze in the testosterone-implant cohort. Where palpable, the tissue appeared to consist predominantly of stromal elements and fibrous extracellular material. The relapsed tumor was still relatively modest in size, having been one of the pair with the smallest tumor volumes at the initiation of treatment. Analysis of the relapsed tumor by hematoxylin and eosin staining and immunohistochemistry (IHC) for AR and Ki67 revealed some notable features (Figure 5, treated versus untreated). Compared to a tumor from its paired control, the relapsed tumor volume was due in part to fibrotic and necrotic components devoid of tumor cells. AR expression in the relapsed tumor was lower than in the untreated tumor with some cells or regions expressing little or no receptor. Unlike the control tumor, the testosterone-treated tumor exhibited low levels of DNA replication in much of the tumor. Thus, the relapsed tumor appeared to still be in the process of adapting to high-dose testosterone and was not yet fully resistant. Similar experiments were performed in castrated male nude mice and yielded similar results, with the exception that no relapses were seen following testosterone treatment. Earlier studies observed a higher frequency of adaptation to treatment, as evidenced by tumor progression in some animals in 5 to 6 weeks [8,12,17]. This may be due to the use of different host strains or different properties of the CRPC cell lines used, and likely reflects that these models, in the same manner as human disease, can show heterogeneous characteristics, while the overall tumor behavior with respect to androgen signaling and modulation of the androgen environment faithfully reproduces human disease. This offers the opportunity to exploit the flexibility of preclinical models to understand the mechanisms underlying the response to testosterone and the development of resistance to testosterone-based therapy.

To verify the clinical relevance of AR overexpression in CRPC cell models, we assessed AR expression by IHC in 12 tumor samples from men with progressive disease in the face of ADT. In 10 of 12 metastases, the tumors overexpressed AR to different degrees. Figure 6 shows an example of moderate overexpression (patient 1) compared to another with little to no expression of AR (patient 2). Substantial heterogeneity of AR expression from cell to cell was observed in the overexpressing tumor. The heterogeneity is most easily seen in the bottom row, where the indicated portions of the field have been magnified. The importance of heterogeneity of AR expression to the process of tumor adaptation to the androgen environment is addressed further below.

Clinical trials of supraphysiologic testosterone therapy have demonstrated a new clinical state: high-testosterone resistance. Xenograft data predicted that tumors can adapt to a high testosterone environment just as they can adapt to a low-androgen environment. However, in preclinical models, treatment of CRPC xenografts differed from the experience with humans thus far, in that very long regression (relative to life span) with no sign of recurrent tumor can be seen [8] (Figure 4). This makes the study of properties of high-testosterone resistance difficult, requiring a long treatment time course to generate tumors in a fraction of mice. We asked whether selection of the testosterone-resistant lines from CRPC lines in culture rather than in vivo might serve as an appropriate and time-efficient approach for model development. Four independent CRPC variant LNCaP lines were further selected for growth in vitro by supplementing the low-androgen medium with 1 nM R1881. While growth was initially severely suppressed, the cells survived and eventually adapted to the high androgen levels in the medium over time and repeated passages until a stable phenotype was achieved that exhibited robust growth in these conditions. These lines were named P1R1/R, M1R1/R, M2R1/R, and 4D1enz1/R (Table 1). These variants all expressed diminished levels of AR (Figure 2C and Figure S1). Examination of AR expression by IHC revealed that these high-testosterone-resistant lines had not simply reverted to a “wild-type” AR phenotype. The variants, models of high-testosterone resistance, exhibited more heterogeneity of AR expression than seen in the original parental lines or the immediate progenitor, the CRPC variant. This can be seen in Figure 2C for the P series of variants and in the Supplementary Materials for others. In the high-testosterone-resistant P1R1/R variant, there were scattered cells that did not appear to express immunoreactive AR at all, while some continued moderate to robust expression of AR. These data supported the clinical observation that there is a limit to the efficacy of repeated reversal of the androgen environment.

To test this hypothesis, we subjected the four high-testosterone-resistant lines to a second round of withdrawal of androgens in the culture medium. Unlike the case with the initial selection of CRPC lines, the second round of selection in low-androgen medium resulted in a rather transient and partial level of growth suppression as the cells adapted readily to the low-androgen environment. These lines were named P1R1/R *, M1R1/R *, M2R1/R *, and 4D1enz1/R * (Table 1). We again assessed receptor expression by IHC (Figure 2C and Supplementary Materials). This revealed that the P1R1/R * cell lines had acquired additional heterogeneity with respect to AR. Significant numbers of cells did not express detectable AR, and many others expressed only moderate levels of AR, unlike the initial CRPC variants in low-androgen medium. At this stage, these variants grew in either high-androgen or low-androgen medium and represent a model of androgen-independence.

To substantiate the in vitro data suggesting the capacity of high-testosterone-resistant P1R1/R cells to respond to a further round of androgen manipulation was limited, P1R1/R cells were injected into the flanks of castrated male mice with implanted testosterone pellets. Tumors were allowed to develop; at which time the testosterone pellets were removed from half of the mice. Androgen withdrawal had only a modest, transient effect on tumor growth, supporting the in vitro observations that these cells were relatively resistant to further rounds of manipulation of their androgen environment. Histochemical analysis (Figure 7) did show evidence for fibrotic tissue within the testosterone-withdrawn tumors, but DNA replication was robust in both the testosterone-replete controls and the testosterone-withdrawn tumors. In this setting, testosterone withdrawal no longer elicited high AR expression. Receptor expression was heterogeneous with some cells and some areas of the tumor expressing little or no receptors. These data indicated that the tumor growth had become largely independent from the androgen environment, and that there was limited capacity to manipulate the androgen environment before growth truly became independent of androgen status.