Abstract: | BackgroundFactors influencing differential responses of prostate tumors to androgen receptor (AR) axis-directed therapeutics are poorly understood, and predictors of treatment efficacy are needed. We hypothesized that the efficacy of inhibiting DHT ligand synthesis would associate with intra-tumoral androgen ratios indicative of relative dependence on DHT-mediated growth.MethodsWe characterized two androgen-sensitive prostate cancer xenograft models after androgen suppression by castration in combination with the SRD5A inhibitor, dutasteride, as well as a panel of castration resistant metastases obtained via rapid autopsy.ResultsIn LuCaP35 tumors (intra-tumoral T:DHT ratio 2∶1) dutasteride suppressed DHT to 0.02 ng/gm and prolonged survival vs. castration alone (337 vs.152 days, HR 2.8, p = 0.0015). In LuCaP96 tumors (T:DHT 10∶1), survival was not improved despite similar DHT reduction (0.02 ng/gm). LuCaP35 demonstrated higher expression of steroid biosynthetic enzymes maintaining DHT levels (5-fold higher SRD5A1, 41 fold higher, 99-fold higher RL-HSD, p<0.0001 for both), reconstitution of intra-tumoral DHT (to ∼30% of untreated tumors), and ∼2 fold increased expression of full length AR. In contrast, LuCaP96 demonstrated higher levels of steroid catabolizing enzymes (6.9-fold higher AKR1C2, 3000-fold higher UGT2B15, p = 0.002 and p<0.0001 respectively), persistent suppression of intra-tumoral DHT, and 6–8 fold induction of full length AR and the ligand independent V7 AR splice variant. Human metastases demonstrated bio-active androgen levels and AR full length and AR splice-variant expression consistent with the range observed in xenografts.ConclusionsIntrinsic differences in basal steroidogenesis, as well as variable expression of full length and splice-variant AR, associate with response and resistance to pre-receptor AR ligand suppression. Expression of steroidogenic enzymes and AR isoforms may serve as potential biomarkers of sensitivity to potent AR-axis inhibition and should be validated in additional models. |