The Bioreductive Prodrug PR-104A Is Activated under Aerobic Conditions by Human Aldo-Keto Reductase 1C3

Bioreductive prodrugs are designed to provide targeted release of toxins in tumors. Enzymatic addition of one (1e⁻) or two (2e⁻) electrons initiates the formation of DNA reactive species. 2e⁻ reduction of quinone-based prodrugs (e.g., mitomycin C, apaziquone) could provide tumor selectivity because of the elevated expression of enzymes such as NAD(P)H: quinone oxidoreductase (NQO1) in some tumors (1). Nitroheterocyclic compounds are rarely substrates for 2e⁻ reduction, with the notable exception of tretazicar (CB 1954), an aziridinyl dinitrobenzamide substrate for both NQO1 and NQO2 (2). More typically, nitroreduction in human tissues occurs via 1e⁻ addition in a process that can be inhibited by molecular oxygen. This redox relationship serves to restrict net metabolism to hypoxic cells which, together with the absence of oxygen-insensitive (2e⁻) nitroreduction, underlies the utility of nitro compounds as hypoxia imaging agents (e.g., PIMO, F-MISO, EF5, and FAZA; ref. 3) and as hypoxic cytotoxins (e.g., CI-1010, PR-104, TH-302, NLCQ-1, and KS119W; ref. 4). Clinical development of this class of agents reflects the prevalence of hypoxia in solid tumors, the recognition of its negative effect on treatment outcome and the strong associations with malignant progression (5, 6).

PR-104 is a water-soluble phosphate ester that is hydrolyzed rapidly in vivo to the corresponding alcohol PR-104A, a bioreductive prodrug (7). PR-104A is a dinitrobenzamide mustard that undergoes nitro reduction to hydroxylamine PR-104H and amine PR-104M, which are DNA cross-linking metabolites able to diffuse locally and kill neighboring cells. Cytochrome P450 reductase (POR) is an important 1e⁻ oxidoreductase that accounts for the majority (∼60%) of anaerobic PR-104A activation in anoxic SiHa cells in vitro (8). Back-oxidation of the initial nitro radical is an efficient process, with 50% inhibition by as little as 0.1 μmol/L of oxygen (9). Consistent with this oxygen-sensitive mechanism, PR-104A is 5 to 120 times more cytotoxic under anoxia for all neoplastic cell lines tested and PR-104 has striking antitumor activity in human tumor xenograft models when combined with anticancer agents that spare hypoxic cells (7). These properties have led to the clinical evaluation of PR-104 (10).

Despite clear evidence for hypoxia-selective activation, the oxygen concentration dependence of PR-104A cytotoxicity in SiHa cultures shows an asymptote at high oxygen levels (9), consistent with a residual oxygen-independent mechanism, with metabolism (8) and DNA crosslinking (11), being due to aerobic bioreduction to PR-104H/M, as for the hypoxic mechanism. In addition, xenograft growth inhibition following single-agent PR-104 treatment seems to track with aerobic sensitivity of the corresponding cell lines in vitro (7). Taken together, these observations suggest the existence of an oxygen-independent 2e⁻ metabolic pathway.

In a previous study we tested whether NQO1 might be responsible for this aerobic activation of PR-104A. Cellular sensitivity to PR-104A weakly correlated with NQO1 enzymatic activity in a panel of eight cell lines, but overexpression of the enzyme showed that NQO1 itself is not a PR-104A reductase (8). This led us to suggest that NQO1 is coordinately regulated with an unknown oxidoreductase responsible for 2e⁻ reduction of PR-104A. Given the NQO1 gene is regulated by nuclear factor erythroid 2-related factor 2 (Nrf2; refs. 12, 13), other 2e⁻ oxidoreductases regulated by the Keap1-Nrf2-ARE antioxidant response pathway would be likely candidates.

Here, we identify candidate aerobic PR-104A reductases by correlating gene expression with aerobic PR-104A reduction. By expressing the most highly correlated of these genes in HCT116 cells, we show that the PR-104A reductase is aldo-keto reductase 1C3 (AKR1C3), a ketosteroid reductase (14) not previously described as a nitroreductase. We confirm that AKR1C3 is a bona fide oxygen-insensitive PR-104A reductase using purified enzyme, that it does not activate other bioreductive prodrugs, and that it contributes to the activity of PR-104 in human tumor xenografts. In a tissue microarray (TMA) survey of AKR1C3 expression in surgical samples, we show that AKR1C3 is highly expressed in some human tumors.

PR-104 was supplied by Proacta, Inc. PR-104A, PR-104H, and tetradeuterated derivatives were synthesized, purified, and stored as described previously (15, 16). Sources of other compounds are in Supplementary Table S1.

Cells were maintained in culture as described (7, 17) with <6 mo cumulative passage from sources (see Supplementary Table S2). Antiproliferative and clonogenic survival assays were performed as previously described (7, 17) and measurement of metabolism of PR-104A to hydroxylamine PR-104H and amine PR-104M by liquid chromatography-tandem mass spectrometry (LC/MS/MS) as before (11).

Microarray-based RNA expression profiles covering 38,500 probes (Affymetrix HG-U133 Plus 2.0) were obtained for cultures of 23 human tumor cell lines. RNA was purified using Trizol-chloroform extraction (Invitrogen) and RNeasy columns (Qiagen). Both RNA purification and Affymetrix array analysis were conducted according to the manufacturers' instructions. The raw data are available from ArrayExpress.⁶

Plasmids encoding sequence-confirmed open reading frames for AKR 1B1, 1B10, 1C1, 1C2, 1C3, 1C4, and NQO1 were purchased (Supplementary Table S3) and cloned into the Gateway compatible vector F527-V5, constructed from pcDNA6.2V5DEST (Invitrogen) and a modified version of the pEFIRES-P plasmid (20). F527-V5 provides the transcription of a bicistronic mRNA which encodes the oxidoreductase and pac (puromycin resistance) open reading frames, the former harboring an occult COOH-terminal V5-tag inducible by infection with AdV5 expressing a TAG suppressor tRNA (Tag-on-Demand, Invitrogen). HCT116 and H1299 cells were transfected using Fugene6 and pooled stable populations selected with puromycin as previously described (21). Functional activity of clones was established by incubating cells with the fluorogenic AKR1C probe coumberone (ref. 22; 5 μmol/L, 37°C, 3 h) using a fluorescent platereader (Molecular Devices SpectraMax M₂; Ex/Em 385/510 nm) and standard curves of authentic reduced coumberone.

Cell lysates were prepared in radioimmunoprecipitation assay buffer and 10 to 20 μg of protein was loaded on SDS-PAGE gels (4–12% gradient gels or 12% gels), transferred, blocked and probed with primary antibodies (Supplementary Table S4) and detected using chemiluminescent ECL detection (Supersignal, Thermoscientific).

Stealth short interfering RNA (siRNA; Invitrogen) against Nrf2 (HSS107128) were tested in A549 and H460, or against Keap1 (HSS114801) in HCT116 and MDA231 cells. Cells (10⁶/10 cm dish) were transfected with 10 nmol/L of siRNA and LipofectAMINE RNAiMAX transfection reagent (Invitrogen), harvested 72 h later by trypsinization, and immunoblotted and assayed for PR-104A metabolites as above.

Recombinant AKR1C3 was diluted to 2 μmol/L in potassium phosphate buffer [100 mmol/L (pH 7.4), 37°C] containing NADPH (100 μmol/L), bovine serum albumin (0.5 mg/mL), and PR-104A (0–150 μmol/L). Reactions (400 μL) were monitored by absorbance (340 nm, Agilent 8453E diode array spectrophotometer) and initial rates determined. Products were analyzed with an Agilent LC/MS (7), following incubation for 10 min in the same (aerobic) reaction mix or under anoxic conditions (Coy anaerobic chamber).

Animal studies were approved by the University of Auckland Animal Ethics Committee and international guidelines were followed. Tumors were grown subcutaneously in the flank of female NIH-III or CD1 nude mice (Charles River Laboratories) by inoculating 5 × 10⁶ cells in 100 μL αMEM. To evaluate clonogenic survival, tumors (800–1500 mm³) were dissected 24 h after single dose PR-104 and plated as described (7). Tumor growth delay experiments were undertaken as described previously (7). Median time for tumors to increase in volume 4-fold (RTV⁴) was determined, and tumor growth inhibition (TGI) was calculated as the percentage of increase in RTV⁴ for treated over control (statistical difference tested by Mann-Whitney U test using SigmaStat v3.5).

Formalin-fixed tumors were sectioned (5 μm) and immunostained using anti-AKR1C3 monoclonal antibody (ref. 23; Sigma) and visualized with the EnVision Dual Link-HRP/DAB kit (Dako). For detection of hypoxia, mice were dosed i.p. with 60 mg/kg of pimonidazole (Hypoxyprobe-1 kit; Chemicon Int) 90 min before removing tumors. To determine the pimonidazole adducts, half of each tumor was formalin-fixed for immunofluorescence microscopy and the other half dissociated enzymatically for flow cytometry according to the manufacturer's instructions.

TMAs were sourced as shown in Supplementary Table S5. A total of 3,932 individual cores representing 19 cancer types (2,490 cases) were analyzed across 38 TMAs with an average of 207 cores per disease (median, 174; range, 31–452). Methodology optimization was carried out and cross-validated against paired frozen samples by Western blot. Slides were immunostained for AKR1C3 as for tumor xenografts and cores were scored for staining intensity and proportion of AKR1C3-positive neoplastic cells by a certified pathologist (S.B. Fox) using a semiquantitative measure on a 7-point scale ranging from negative (score 0) to diffuse strong staining (score 6). This measure was applied to the neoplastic cell element of the tumors within the TMAs and the epithelial elements within the normal TMA. An unrelated series of normal tissue, NSCLC and breast cancer sections were evaluated independently by Mosaic Laboratories, LLC.

As reported previously for SiHa cells (8), the flavoenzyme inhibitor diphenyliodonium prevented the anoxic cytotoxicity of PR-104A in A549 cells (Fig. 1A). However, it had no effect on the aerobic cytotoxicity of PR-104A. In contrast, ketoconazole (identified from a chemical inhibitor screen; Supplementary Fig. S1), blocked aerobic but not anoxic cytotoxicity (Fig. 1A), suggesting that different enzymes mediate the activation of PR-104A under these two conditions. To distinguish metabolism-dependent from intrinsic cell sensitivity, inhibition of aerobic cell proliferation by PR-104A and its cytotoxic metabolite PR-104H were compared across 12 cell lines (Fig. 1B). A consistent ∼20-fold PR-104 A/H deactivation ratio was found for seven cell lines (r² = 0.96), whereas in five cell lines (A549, H460, SiHa, 22Rv1, and HT29) PR-104A showed anomalously high toxicity. Notably, these five cell lines were recently shown to have high rates of aerobic reduction of PR-104A (11). Taken together, these results suggest variable expression of an aerobic PR-104A reductase in human tumor cell lines.

To identify candidate PR-104A oxidoreductases, we examined gene expression by microarray profiling and compared these data with aerobic reduction of PR-104A to PR-104H/M, adding a further 11 cell lines to the above 12 cell line panels to increase statistical power. Unsupervised hierarchical clustering of oxidoreductase gene expression identified two distinct groups, defined largely by members of the AKR superfamily and other Keap1/Nrf2/ARE regulated genes such as NQO1 (“Nrf2 cluster”; Fig. 1C; Supplementary Figs. S2 and S3). Using an LC/MS/MS assay to quantify PR-104A reduction (Fig. 1D) we showed large differences between the same cell lines (160-fold range) with the fast metabolizers associated with the “Nrf2 cluster.” A principal component analysis, accounting for false discovery (adjusted P value < 0.1), identified 20 probes positively correlated with PR-104H/M formation (Supplementary Table S6), 8 of which corresponded to five members of the AKR family (1C1, 1C2, 1C3, 1C4, and 1B10).

These five AKR1 candidates, along with AKR1B1 and known aerobic nitroreductase, NQO1, were expressed by plasmid transfer into HCT116 cells and stable populations tested for aerobic PR-104A metabolism. Only AKR1C3 transfectants were able to generate PR-104H/M (Fig. 2A), despite all the candidate reductases being immunodetected following induced translation of a cryptic COOH-terminal V5-tag (Fig. 2B). Expression of AKR1C3, AKR1B10, and NQO1 was confirmed independently using monoclonal antibodies and an AKR1C1 polyclonal with cross-reactivity for other AKR1C family members (Fig. 2B; antibody details Supplementary Table S4). All AKR1C family transfectants reduced the fluorogenic substrate coumberone (22), confirming the functional expression of these enzymes (Fig. 2C). We next confirmed that AKR1C3 itself is a 2e⁻ PR-104A reductase by demonstrating that purified recombinant AKR1C3 catalyses NADPH-dependent formation of PR-104H in the presence or absence of oxygen (Fig. 2D). The reaction followed Michaelis-Menten kinetics with an apparent K_m of 20.6 ± 2.6 μmol/L and K_cat of 0.800 ± 0.025 min⁻¹ (Supplementary Fig. S4).

Evaluation of AKR1C3 expression in the 23 cell line panel by Western blotting (Fig. 3A) showed highly variable protein levels; linear regression of PR-104H/M formation (Fig. 1D) versus AKR1C3/actin (from Fig. 3A) showed a highly significant correlation (r² = 0.83; P < 0.001), which was stronger than the correlations with AKR1B10/actin (r² = 0.27; P = 0.011) or NQO1/actin (r² = 0.17; P = 0.053).

Given the discordant relationship between NQO1 and AKR1C3 levels across the 23 cell line panel in vitro (r² = 0.19), we evaluated the role of the Nrf2 in the expression of AKR1C3. RNA interference of Keap1, a specific Nrf2 repressor, by siRNA transfection in MDA231 cells induced AKR1C3 protein levels 5.7-fold (Fig. 3B) and aerobic PR-104A metabolism 7.0-fold (Fig. 3B), with attendant 4.5-fold induction of an ARE-luciferase reporter (Supplementary Fig. S5A). In analogous experiments, Nrf2 siRNA transfection of A549, in which Nrf2 is constitutively activated by a Keap1 point mutation (24, 25), suppressed ARE reporter activity (Supplementary Fig. S5B), AKR1C3 and NQO1 expression, and aerobic PR-104A metabolism (Fig. 3B). Similar results were obtained with H460 cells (Supplementary Fig. S5C). However, Keap1 siRNA increased NQO1 but not AKR1C3 (or PR-104A reduction) in HCT116 cells (Supplementary Fig. S5D). These results suggest a role for Nrf2 in induction of AKR1C3 expression, but that additional cell line–specific factors result in differential regulation relative to NQO1.

To test whether AKR1C3 expression enhances the cytotoxicity of PR-104A, we isolated two clones from the pool of HCT116 cells transfected with AKR1C3. Clonogenic survival curves showed clone no. 1 to be 10-fold more sensitive to PR-104A than the parental cells under aerobic conditions, and was further sensitized (44-fold) under anoxia (Fig. 3C). Using aerobic IC₅₀ assays, we compared the sensitivity of clone no. 1 and no. 2, relative to the parental line, to PR-104A and 10 other bioreductive agents including 6 other nitro compounds, 3 quinones and a tertiary amine N-oxide (Fig. 3D). This confirmed the ∼10-fold sensitization to PR-104A, and showed that this activation by AKR1C3 is unique among the bioreductive drugs tested.

A panel of nine cell lines were grown as subcutaneous solid tumors in nude mice. Expression of AKR1C3, measured by Western blot and immunohistochemistry (Fig. 4A), was broadly similar to that in culture. Tumor cell survival determined by ex vivo clonogenic assay 24 hours after single-dose PR-104 showed a trend to greater activity in tumors with high AKR1C3 levels (Mann-Whitney, P = 0.063; Fig. 4B). However, there was a wide range of sensitivities among the AKR1C3-positive tumors, leading us to ask whether the extent of hypoxia also influenced PR-104 sensitivity. Pimonidazole binding, as a hypoxia marker, showed similar trends by flow cytometry and immunostaining (Fig. 4C; see Supplementary Fig. S6 for a quantitative comparison of the two assays), but did not correlate with the activity of PR-104. Indeed, some of the least hypoxic tumors (H460 and SiHa) were the most sensitive to PR-104, whereas the most hypoxic (H1299 and A2780) were relatively insensitive. However, these relationships may be confounded by cell line differences in intrinsic sensitivity to the PR-104H/M active metabolites. To more clearly identify the role of AKR1C3 in PR-104 monotherapy, we compared xenografts grown from AKR1C3-overexpressing H1299 transfectants and parental cells by tumor growth delay (Fig. 4D). Parental tumors showed modest growth inhibition (TGI, 77%; P = 0.03), possibly due to the hypoxic activation of PR-104 with attendant metabolite redistribution (bystander effect). In contrast, AKR1C3-expressing H1299 xenografts regressed following PR-104 treatment with a TGI of 314% (P < 0.01; Fig. 4D). Hypoxic fraction was similar between these two models (Fig. 4D; data not shown); this establishes a major role for AKR1C3 expression in the response of this NSCLC tumor model to PR-104. Similar experiments with an isogenic pair of HCT116 xenografts differing in AKR1C3 expression corroborated this observation (Supplementary Fig. S7).

AKR1C3 was evaluated by immunohistochemistry using TMAs. A scoring system, illustrated in Fig. 5A, gave higher ranking to uniform over focal staining with scores of 4, 5, and 6 considered “positive.” Expression of AKR1C3 was present in most tumor types (Fig. 5B). HCC showed the highest frequency of positive cores with most (58%) staining strongly in all cells (score 6). Other disease types with >50% positive cores included bladder, renal, and gastric carcinomas. A summary of all scores is shown in Supplementary Table S7. Examination of subtypes of lung carcinoma showed marked heterogeneity, with expression restricted to NSCLC whereas small cell lung carcinoma was negative; expression was also present in a high proportion (54%) of lung tumor metastases (Fig. 5C). To confirm these TMA observations, a series of lung and breast cancer tissue sections were analyzed by an independent laboratory using a separate optimized staining protocol and scoring criteria (see Supplementary Methods); 48% (10 of 21) of NSCLC and 14% (3 of 21) breast cancers were classified as highly positive for AKR1C3 (Supplementary Tables S8 and S9, respectively).

A survey of 33 normal tissue TMA cores identified small intestine and kidney as containing moderate numbers of cells with strong AKR1C3 immunoreactivity (score 4), with mixed staining seen in the liver core (score nos. 5 and 4; Fig. 5D). Occasional weak positivity (score 1) was seen in bone marrow cell TMAs (morphology undetermined). Notably, the intensity of normal tissue staining was substantially less than that seen with positive neoplasia samples, as illustrated by a side-by-side macro comparison with a set of HCC cores (Fig. 5D). Independently, 23 normal tissue sections were analyzed and showed strong positive AKR1C3 staining in seven tissues; stomach, small intestine, colon, pancreas, kidney, uterus and ovary, with weak/diffuse staining in the majority of liver cells (Table 1). Most specimens showed both nuclear and cytoplasmic staining but adrenal and liver showed cytoplasmic staining only. Thus, the full section histopathology analysis was broadly consistent with the TMA scoring.

The observation that certain human neoplastic cell lines are able to convert the bioreductive prodrug PR-104A to its cytotoxic metabolites under oxygenated conditions led us to seek the identity of the putative aerobic oxidoreductase(s). Microarray analysis of 23 cell lines highlighted a cluster of five AKR family 1 members (1B10, 1C1, 1C2, 1C3, and 1C4) coordinately upregulated in a manner that correlated with oxic PR-104A metabolism (compare Figs. 1D and 3A). Plasmid-based cellular expression confirmed AKR family 1 member C3 (AKR1C3) as a functional PR-104A nitroreductase. Recombinant AKR1C3, a ketosteroid reductase better known for its role in the pre-receptor regulation of steroid hormones and prostaglandins (14, 26), was able to catalyze NADPH-dependent reduction of PR-104A to the hydroxylamine metabolite PR-104H in the presence of oxygen (Fig. 2). Western blotting of the cell line panel shown that AKR1C3 was expressed in 30% (7 of 23) of cell lines and is correlated with elevated aerobic PR-104A reduction. Expression of NQO1, a prototypic reporter gene for the Keap1-Nrf2-ARE pathway (12), was only weakly correlated with AKR1C3 expression, although AKR1C3 has recently been reported to be regulated by this pathway (27). Keap1 knockdown induced AKR1C3 protein with attendant elevation of PR-104A metabolism in MDA-231 cells (Fig. 3B), demonstrating that regulation of AKR1C3 by the Nrf2/Keap1 pathway is possible. In support, constitutive AKR1C3 expression in A549 and H460 cells was sensitive to knockdown of Nrf2 activity (as confirmed by ARE-luciferase activity), leading to reduced PR-104A metabolism, and implicating the AKR1C3-positive phenotype as being associated with the known Keap1 mutations in these cell lines (24, 28). Notably, biallelic loss at the Keap1 locus is common in lung cancer cell lines with attendant nuclear accumulation of Nrf2 (24, 29), and expression of this AKR1 gene cluster (including AKR1C3) has been proposed as a biomarker of Nrf2 activation (27). However, Keap1 siRNA induced NQO1 but not AKR1C3 expression in HCT116 cells, indicating that the two aerobic reductases are differentially regulated in certain contexts.

In vitro, PR-104A displays hypoxia-selective toxicity across all cell lines tested (hypoxic/aerobic cytotoxicity ratio of 5–120; ref. 7). However, the role of hypoxia is less apparent for PR-104 monotherapy of tumors in which hypoxic cells are a geometrically constrained minority population (Fig. 4C) and metabolite diffusion is mandatory for single-agent activity. Consequently, tumor xenograft sensitivity to PR-104 monotherapy (measured by clonogenic survival; Fig. 4B) seems to be strongly, but not exclusively, influenced by the more uniform presence of AKR1C3 (Fig. 4A) rather than the heterogeneous patterning of hypoxia (Fig. 4C), suggesting that the bystander cell killing effect is not in itself sufficient to exploit tumor hypoxia in a monotherapy context. This postulate is supported by several lines of evidence, including the modest growth delay achieved with PR-104 treatment of parental H1299 NSCLC xenografts (TGI, 77%), despite the large hypoxic fraction (62% pimonidazole-positive), and an ability of H1299 cells to generate PR-104A metabolites 35-fold more efficiently under hypoxia (11). Even in this hypoxia and hypoxic metabolism–rich setting, expression of AKR1C3 is necessary for substantial tumor control (TGI, 314%; Fig. 4D). This supports our original observation that aerobic cell sensitivity predicts in vivo activity as measured by tumor growth delay (7), leading to the conclusion that AKR1C3 expression in individual tumors may be an important determinant in defining the most responsive patient population. However, given that Nrf2 regulates a plethora of gene products associated with drug resistance (12, 30–32), one possibility is that cellular resistance mechanisms may concurrently oppose AKR1C3-dependent sensitivity to PR-104. Nevertheless, it is evident that AKR1C3-rich cell lines such as A549 and H460 are sensitive in vitro and in vivo (7), despite constitutive upregulation of multiple Nrf2-regulated cytoprotective genes (24, 28, 30) indicating that, on balance, a PR-104 sensitive phenotype prevails.

Whether AKR1C3 can be clinically exploited by PR-104 will depend in part on its expression in tumors relative to normal tissues. Immunohistopathology of normal tissues showed the expression of AKR1C3 in a subset of epithelial cells in the stomach, gastrointestinal tract, pancreas, liver, and kidney (Table 1), broadly consistent with literature findings (23, 33–35). Some endothelial cell expression was also evident. Of potential concern is the immunodetection of AKR1C3 in a minority of bone marrow cells, particularly in light of the recent report that CD34⁺ human myeloid progenitor cells are positive for AKR1C3 mRNA (36). This observation may be associated with the myelotoxic effects of PR-104 in humans (10). A population analysis of AKR1C3 expression in tumor cores (2490 cases) showed that it is strongly and frequently upregulated in some carcinomas (Fig. 5B). Across 19 tumor types, the highest frequency of strongly positive biopsies was in HCC, with many other positive tumor types including bladder, renal, gastric, cervix, colon, and NSCLC (Fig. 5B; Supplementary Table S8). Overexpression of AKR1C3 has been documented in carcinomas of the breast (23, 37, 38), prostate (39–41), endometrium (42, 43), and kidney (44). Surprisingly, the prostate and breast carcinoma cores did not score highly in our screen (see Supplementary Fig. S8 for validation and Supplementary Table S9 for independent laboratory analysis of additional breast cancer tissue samples), suggesting these TMAs (Supplementary Table S5) may be sourced from early stage disease. Overall, the immunohistochemical analysis shows that the intensity of AKR1C3 expression is strikingly elevated in certain tumors relative to normal tissues (Fig. 5D). The distribution of AKR1C3 overexpression has prompted the evaluation of PR-104 in HCC and NSCLC (ClinicalTrials.gov identifier NCT00862082 and NCT00862134, respectively). Collectively, these data indicate that AKR1C3 is a novel target for PR-104, and AKR1C3 detection may identify individual patients with PR-104–sensitive tumors.

W.R. Wilson is a stockholder and consultant to Proacta, Inc. A.V. Patterson is a consultant to Proacta, Inc. The other authors disclosed no potential conflicts of interest.

We thank Proacta, Inc. for PR-104, Graham Atwell and Prof. William Denny for PR-104A and tetradeuterated standards, Kashyap Patel for PR-104H, Dianne Ferry and Yongchuan Gu for assistance with bioanalytical methods, Sophie Syddall for construction of the F527-V5 vector, Dan Li for preparation of high titer Tag-on-Demand adenovirus, Susan Pullen for cell growth inhibition experiments, Wouter van Leeuwen for RNA isolation from cultured cell lines, Dr. Jeffrey Smaill for synthesis of coumberone, Dr. Christopher Squire for recombinant AKR1C3, Prof. David Ross for the gift of anti-human NQO1 monoclonal antibody, and Helen Morrin for curating tissues for the Cancer Society NZ tissue microarray.

Grant Support: Health Research Council of New Zealand, Program grant 08/103 (C.P. Guise, G.U. Dachs, W.R. Wilson, and A.V. Patterson) and Proacta, Inc. (M.R. Abbattista, R. Pollock, J. Harvey, P. Guilford, S. Fox, F. Doñate).

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.