Enhancing Abiraterone Acetate Efficacy in Androgen Receptor–positive Triple-negative Breast Cancer: Chk1 as a Potential Target

Gene expression array (GEA) studies identified a breast cancer subtype characterized by the expression of the androgen receptor (AR) and the absence of the estrogen receptor α (ER; refs. 1–3). Although AR expression in breast cancers has been known for a long time, GEA studies were instrumental in highlighting the increased androgen signaling and the apocrine-tumor morphologic hallmarks of the “molecular apocrine” subgroup (1). These tumors could either be HER2-positive or -negative. In this report, we focus on the latter that are part of the triple-negative breast cancer (TNBC) group (4, 5).

Although this subtype has been identified by GEA, the following IHC definition is classically used in prospective clinical trials to identify these tumors: AR-positive and ER-, progesterone receptor (PR)- and HER2-negative. Despite no justified objective biological evidence, the cutoff retained for AR-positivity by the three clinical trials published so far is ≥10% (6–8). The frequency of AR-positive TNBC ranges from 22%–35.9% of TNBC (9, 10).

Preclinical data show androgen-dependent growth of the classic AR-positive TNBC subtype model, the MDA-MB-453 cell line (2, 11, 12). Three clinical trials assessed AR-antagonists (bicalutamide or enzalutamide) and an androgen synthesis inhibitor (abiraterone acetate) in patients with metastatic AR-positive TNBC (6–8). In these trials, clinical benefit rates (CBR) at 6 months range from 19%–29% with excellent toxicity profiles (6–8). Seven and two objective responses were reported in the enzalutamide and in the abiraterone acetate trial, respectively (7, 8). The complete response (CR) obtained with abiraterone acetate is still ongoing after 4 years of treatment (13). However, the majority of patients with AR-positive TNBC do not benefit from antiandrogen treatment.

Therefore, we decided to perform a biological study using samples from patients included in the UCBG 2012-1 trial with the aim to identify: (i) predictive markers of response to abiraterone acetate and (ii) new druggable targets. The first aim is based on the published observation that IHC assessment of AR expression at a 10% threshold is the currently accepted predictor of response, although it had a modest (30%) positive predictive value (PPV) in the enzalutamide trial (14). Regarding the second aim, we were interested by the finding that, in a classic AR-driven prostate cancer model, the use of antiandrogens in combination with drugs targeting the DNA damage repair (DDR) genes is especially effective, with possible therapeutic implications (15). Our study suggests that an apocrine-based IHC marker score predicts the abiraterone acetate response. It also identifies Chk1, a serine/threonine kinase protein essential for the maintenance of genomic integrity, as a potential new target in AR-positive TNBC.

This study was a planned analysis within the UCBG 2012-1 trial. Patients eligible for the UCBG 2012-1 trial were women with metastatic or unresectable locally advanced breast cancer centrally confirmed as AR-positive (≥10%), ER, PR-negative (≤10%), and HER2-negative (0 or 1+ in IHC, 2+ IHC with negative FISH), on a formalin-fixed, paraffin-embedded (FFPE) tumor sample. Treatment with abiraterone acetate (1,000 mg once a day) was administered orally on a continuous daily schedule plus prednisone (5 mg twice a day). This treatment continued until disease progression or unacceptable toxicity. The primary endpoint was the CBR, defined as the proportion of patients presenting either a CR, a partial response (PR), or a stable disease (SD) at 6 months.

For the substudy that is the subject of this report, a subgroup of the UCBG 2012-1 trial initial population was included on the basis of the following criteria: (i) patients eligible and evaluable for treatment efficacy defined as having received at least one complete cycle of treatment and at least disease assessment recorded at 8 weeks from the start of the treatment and (ii) patients for whom tumor material (primary tumor or metastasis, depending on the patients) contained more than 20% of tumor cells (Supplementary Fig. S1).

Responders were defined as patients whose tumor achieved a clinical benefit at 6 months (CR, PR, and SD ≥ 6 months). Nonresponders were defined as patients whose tumor progressed before 6 months of abiraterone treatment.

A cohort of patients for whom the central pathology assessment showed an AR-negative TNBC was included in the IHC part of this substudy. These patients were screened for the UCBG 2012-1 trial but not treated with abiraterone because of the AR negativity. These patients have consented for IHC central review and for additional research on their samples.

The trial was registered at ClinicalTrials.gov (NCT01842321) and approved by a national ethics committee (CPP Bordeaux) and by local institutional review board of all participating centers. Before registration, all patients signed an informed consent for the trial and for research on tumor samples collected. The study was performed in accordance with the Declaration of Helsinki.

FFPE tumor samples were collected for central pathology assessment before inclusion in the clinical trial. Tumor cellularity was assessed by a senior pathologist (G. MacGrogan, Bergonie Cancer Institute, Bordeaux, France) on a hematoxylin and eosin slide. For each case, three different areas were isolated, one for tissue microarray (TMA) drilling, one for DNA extraction, and one for RNA extraction.

Three representative triplicate spots of 1-mm cores of tumor FFPE tissue were taken for TMA construct. TMAs were produced using a Tissue Arrayer MiniCore 3 (Excilone). When surgical specimens were not available, samples were analyzed on whole sections of Tru-Cut biopsies. IHC was performed on 4-μm–thick paraffin-embedded tissue sections. Slides were treated on BenchmarkULTRA (Ventana) with UltraView Universal DAB (760-500) detection kit.

A total of 15 apocrine features or cancer-associated markers were assessed: AR, FOXA1, GCDFP15, GGT1, L1CAM, KI-67, GATA3, EGFR, CK14, CK17, CK5/6, PTEN, MAPK (ERK1/2), p-S6, and CCND1. Primary antibodies used are listed in Supplementary Material S1. A senior pathologist (G. MacGrogan, Bergonie Cancer Institute, Bordeaux, France) performed the analyses. Twelve markers were semiquantitatively assessed using the H score which combines staining intensity and the percentage of positive cells, determined by the following formula: 3 × percentage of highly staining cells + 2 × percentage of moderately staining cells + percentage of weakly staining cells, as described previously (16). The percentage of positive nuclei was quantified for Ki67 immunostaining. AR was quoted as the percentage of positive cells (whatever the intensity of nuclei staining). EGFR was quoted by multiplying the strongest intensity of staining and percentage of that staining. All IHC results are presented as median and extreme values.

Selected zones of FFPE blocks were macrodissected. After deparaffinization, tissues were incubated overnight at 56°C with proteinase K; genomic DNA was extracted using a GeneRead DNA FFPE Kit (Qiagen); and quantified with the Qubit dsDNA BR Assay Kit (Thermo Fisher Scientific), according to manufacturer's instructions. Targeted sequencing of 91 cancer-associated genes was performed using Illumina Hiseq2500 technology on a custom made panel (Supplementary Material S2). Data were aligned to the human reference genome (hg19) using Bowtie2 algorithm (17). Two samples (2/26) were removed from the analyses because of low coverage (<30% of targeted genes covered at a depth of 100×). Coding variants occurring in 5% or more and at a depth of coverage of at least 100× were considered. Single-nucleotide variants (SNV) and indels were identified using GATK UnifiedGenotyper. We retained COSMIC-confirmed nonsynonymous, exonic/splice variants observed at a frequency lower than 0.1% in population. The potential pathogenicity of the identified variants was estimated using different public databases (cancer hotspot, ref. 18; oncoKB, ref. 19; cBioPortal, ref 20; Tumorportal, ref. 21; IARC TP53, ref. 22; and UMD BRCA, ref. 23). We only retained confirmed pathogenic variant and variant of unknown significance if they were reported at least once in cBioPortal and/or Tumorportal for this analysis.

Affymetrix OncoScan FFPE assay was carried out, as described previously (24), to obtain genomic information including copy number variations and SNPs from the DNA of 4 of 6 responder patients. Analyses were performed with Chromosome Analysis Suite 3.3 (ChAS 3.3, Thermo Fisher Scientific).

Samples macrodissected from FFPE blocks were deparaffinized in xylene and precipitated in ethanol. Tissues were digested for 3 days at 55°C in lysis buffer containing proteinase K and Triton X100, and total RNA was then extracted following the manufacturer's protocol (High Pure FFPE RNA Micro Kit, Roche Applied Science). Quantifications were performed using a NanoDrop Spectrophotometer (Thermo Fisher Scientific). NanoString nCounter Platform (version 3.0, NanoString Technologies) was used to measure the expression levels of transcripts (25). The nCounter PanCancer Pathways panel includes 730 genes representing 12 cancer-associated canonical pathways including: MAPK, PI3K, cell cycle, apoptosis, or DNA damage control. Thirty supplementary genes were added because of their relevance in transcriptional studies for identifying AR-positive ER-negative breast cancer (refs. 1, 2; Supplementary Material S3). RNA labeling and hybridization reactions were performed according to the manufacturer's instructions. A gene was kept in the analysis if its expression was at least 50 counts, in at least 4 samples. Data normalization was performed internally using positive and negative controls and housekeeping genes (26). The R package NanoStringDiff V1.4 (27) that estimates a generalized linear model with a negative binomial family (26), was used to identify differentially expressed genes and to estimate fold change and compute significance. P values were corrected for multiple testing by the Benjamini Hochberg procedure (28). For the responder versus nonresponder supervised analysis, we retained genes with significant differential expression, at FDR 5%. All heatmaps were generated on filtered data. For heatmaps with selected genes, only the intersections with expressed genes were used. Clustering was generated with Pearson correlation distance and Ward criterion. Heatmaps displayed scaled expression values by gene. Pathway and network analyses were performed with Ingenuity Pathway Analysis software (IPA, Qiagen; ref. 29).

MDA-MB-453 were purchased from ATCC (ATCC HTB-131) and cultured in Advanced DMEM (Thermo Fisher Scientific) with 1% FCS, 1% Glutamax, and 1% penicillin/streptomycin. SUM185PE were kindly provided by Lehmann-Che and De Cremoux (CNRS UMR7212/INSERMU944; Hôpital Saint-Louis; Paris, France) and cultured in DMEM (Thermo Fisher Scientific) with 10% FCS and penicillin/streptomycin 1%.

The different antiandrogens used were: a CYP17A1 inhibitor (abiraterone acetate) and two AR antagonists, enzalutamide and darolutamide (ODM-201). All three were purchased from Selleckchem (ref. S2246, S1250, and S7569). Two Chk1 inhibitors were used: AZD-7762 purchased from Selleckchem (ref. S1532) and GDC-0575 that was obtained from Genentech (MTA ID: OR-216276). Abiraterone acetate was diluted in ethanol while enzalutamide, ODM-201, AZD-7762, and GDC-0575 were diluted in DMSO.

Cells (∼1,800) were seeded in 96-well plates for 24 hours and then treated for 5 days with a range of increasing concentrations of drugs, used either alone or in combination. Methyl thiazolyl tetrazolium (MTT, Sigma Aldrich) was applied for 3 hours (final concentration: 0.5 mg/mL). Then, supernatants were discarded and 100 μL of dimethylsufoxyde was added to the wells. Absorbance was measured on a Microplate Photometer (Bio-Tek Instruments) using a test wavelength of 570 nm and a reference wavelength of 630 nm. Concentration–response curves were plotted and the concentrations causing 50% inhibition of viability compared with vehicle control (IC₅₀) were interpolated from nonlinear regressions of the data (Prism 5; GraphPad Software).

Drug combination assays were carried out following the Chou–Talalay method, based on the median-effect equation (30). Cell viabilities were assessed using MTT, following the same protocol used for single drugs assays. IC₅₀ were determined, and the effects of drug combinations were analyzed by calculating the combination index (CI) with the following equation, |CI = \displaystyle{{\frac{a}{A}} + {\frac{b}{B}$| (where A and B are the single-agent IC₅₀s and a and b are the respective combination IC₅₀s of drug A and B). Synergistic, additive, and antagonistic effects are characterized by CI < 1, CI = 1, and CI > 1, respectively.

Cells were seeded in 12-well Chamber Microscopy Glass Slide (Ibidi, Planegg/Martinsried) and incubated with drugs for 72 hours. Slides were then washed twice with PBS, fixed in 4% formaldehyde, permeabilized with Triton X-100, and incubated with anti–Phospho-Histone H2A.X (Ser139) mAB (Rabbit mAb #2577, Cell Signaling Technology, Inc.) overnight, and then with a goat anti-rabbit Alexa Fluor 488 antibody (Invitrogen). Slides were then counterstained by 4,6-diamidino-2-phenylindole. Quantification was performed with ImageJ software version 1.51n.

Apoptosis and cell cycle were evaluated using FACS analysis. Cells (1 × 10⁶) were seeded in 6-well plates and incubated with drugs for 72 hours. For apopotosis/necrosis assay, cells were exposed to FITC-Annexin and propidium iodide (PI) according the manufacturer's protocol (BD Biosciences). For cell-cycle analysis, cells were fixed and permeabilized in absolute ethanol with PBS overnight at −20°C, then rinsed and incubated with RNase and PI (50 μg/mL; Sigma Aldrich).

Approval for the animal experiments was granted by the Comité d'éthique pour l'expérimentation Animale (CEEA50) ethics committee; Bordeaux (project number 2017051116491638 –V2 APAFIS #9883). The study was performed in accordance with European Community Standards of Care and Use of Laboratory Animals under level 2 containment (animal house authorization number A33063916).

MDA-MB-453 cells were transduced at a multiplicity of infection of 5 with a lentiviral vector allowing the expression of firefly Luciferase (MND-Luc, kindly provided by the Vectorology platform, SFR Transbiomed, Bordeaux University, Bordeaux, Nouvelle-Aquitaine, France). A total of 2 × 10⁵ cells were injected through the nipple into the mammary ducts of the fourth (inguinal) mammary glands of female NSG mice (Jackson Laboratory strain number 005557), as described previously (31). A total of 40 mice were injected. Tumor-bearing mice were randomized into four groups (10 mice in each group) after 10 days and treated by oral gavage as follows: (i) vehicle control (in a solution supplied by Roche/Genentech), (ii) abiraterone acetate (400 mg/kg) 5 of 7 days, (iii) GDC-0575 (25 mg/kg) 2 of 7 days, and (iv) combination of abiraterone acetate plus GDC-0575, using the same doses. Tumor growth was measured by bioluminescence imaging twice a week using a cryogenically cooled imaging system (PhotonIMAGER, Photo-acquisition and 3D Vision Software, Biospacelab). Tumor progression and doubling time were analyzed with GraphPad Prism software. After 3 weeks of treatment, mice were sacrificed and mammary glands were retrieved, and then fixed in 4% buffered formaldehyde for IHC. Human-specific CK7 staining (SP52, Roche Ventana) was applied on whole gland section. High quality images were made with Slides Scanner Nanozoomer (Hamamatsu NANOZOOMER 2.0 HT). Areas of invasive carcinoma were delimited on each whole-gland section by image analysis (NDP.view 2 version 2.6.13); the ratios of the infiltrating carcinomas to the whole-gland areas were calculated.

The Exact Wilcoxon two-sample test was performed to assess comparisons of values between groups of patients for the IHC markers analysis (SAS software version 9.4). The effects of drugs or drug combinations on cell death, phospho-H2AX–labeling, and tumor growth in vivo were all analyzed using GraphPad Prism 5.00 (GraphPad Software). Histograms and error bars represent mean ± SEM, respectively, except otherwise stated in the figures legends. The tests used to assess the statistical significance of the differences are indicated in the figures legends (*, P < 0.05; **, P < 0.01; and ***, P < 0.001).

Tumor material was available from 28 eligible and evaluable patients included in UCBG 2012-1 trial (flow chart, Supplementary Fig. S1). Clinical and pathologic characteristics are reported in Table 1.

Responders were older than nonresponders (median age: 76 years vs. 62 years, respectively). Other patient and tumor characteristics were similar in the two groups although the numbers are small.

Tumor material was available for TMA construct in 24 patients, for DNA analysis in 26 patients, and for RNA analysis in 22 patients (flowchart, Supplementary Fig. S1).

Fourteen markers were assessed by IHC in patients with AR-positive TNBC included in the trial for whom enough tumor material was available (n = 24; 4 responders and 20 nonresponders) and also in screened but nontreated patients with AR-negative TNBC (n = 17). Markers were selected because they are commonly studied in breast cancer and/or reported in AR-positive TNBC.

First, we compared the selected markers in AR-positive (n = 24) and AR-negative (n = 17) TNBC groups (Table 2). The median percentages of positivity of GCDFP15, FOXA1, GGT1, CCND1, and GATA3 were statistically higher in the AR-positive TNBC group (Table 2). These results are in line with previously published studies (32–34), because most of these proteins regulate, are regulated by, or are coexpressed with AR.

Second, we compared in the AR-positive group the median positivity of AR and the 14 markers between responders (n = 4) and nonresponders (n = 20). Only AR (P = 0.0067), FOXA1 (P = 0.0048), and GGT1 (P = 0.0195) showed statistically significant differential overexpression in the responders group (Table 2; Fig. 1A). GCDFP15 median expression was numerically higher in the responders group but did not reach statistical significance (P = 0.062) (Table 2). We then created an IHC score by adding the percentage of AR-positive cells and H scores of FOXA1, GGT1, and GCDFP15. We named this IHC4 apocrine score. With median values of 79.6 (min–max: 75–82.5) versus 46.0 (Min–Max: 2.5–75), this IHC4 apocrine score significantly distinguished responders (n = 4) from nonresponders (n = 20; P = 0.0025, Mann–Whitney test; Fig. 1B). Morphologic apocrine characteristics (abundant eosinophilic and granular cytoplasm, large nuclei with prominent nucleoli) were present in all responders (6/6) and in majority of nonresponders (66%, 15/22; data not shown).

To identify specific targetable alterations, DNA extracted from the tumors of 26 patients with AR-positive TNBC treated with abiraterone (6 responders and 20 nonresponders) was analyzed. A panel of 91 breast cancer- and general cancer–associated genes was sequenced. Two samples were excluded due to low coverage. We identified 225 variants in 70 of 91 sequenced genes. The median number of variants per case was 3 (min–max: 0–61) and only one patient (nonresponder) had no variant.

Potentially targetable variants (defined by OncoKB database; ref. 19) were identified in 17 of 24 patients (71%). These include, alterations of PIK3CA in 12 (4 responders and 8 nonresponders), AKT1 E17K variant in 3 (1 responder and 2 nonresponders), mTOR S2215F variant in 1 (nonresponder), and ERBB2 L755S variant in 1 (nonresponder).

The TP53 mutation rate was 17% (1/6) in the responders group and 61% (11/18) in the nonresponders group, respectively (P = 0.035). The rate of PIK3CA mutations was 67% (4/6) in the responders group and 44% (8/18) in the nonresponders group (Fig. 1C), a difference not reaching statistical significance. Other identified variants did not allow segregating patients based on the status of abiraterone response.

RNA extracted from 22 patients' tumors (4 responders and 18 nonresponders) was analyzed to address potential gene expression differences among abiraterone acetate response status group of patients. NanoString platform was selected as it allows working on RNA extracted from samples with high variability, low yield, and eventually, degradation (24). A total of 576 of 760 transcripts were retained in the analysis after filtering the results and 3 nonresponders patients were removed (one was flagged at the first quality control step and two were outliers in the principal component analysis). Therefore, we further analyzed the gene expression profiles from 19 patients (4 responders and 15 nonresponders).

Tumors did not cluster according to their response status using an unsupervised analysis (data not shown). The analysis was repeated using a restricted list of 41 transcripts present in our custom NanoString panel and in the list of transcripts of the first two studies which identified the molecular apocrine subtype (1) and the ER- class A subtype (2). This did not allow clustering responders from nonresponders either (Supplementary Fig. S2).

A supervised analysis (responders vs. nonresponders) identified 31 genes with differential expression in the two groups, with a FDR < 5% (Fig. 2A). Among them, 9 genes, underexpressed in responders, were particularly interesting: 4 are linked to proliferation (CDC6, CCNE2, CDC25C, and E2F5) and 5 to DDR (FANCA/PALB2, FANCB, BRCA1, CHEK1, and RAD21). This was indirectly confirmed by a pathway and network analysis (Fig. 2B) which showed that the most differentially expressed pathways (P ≤ 0.001) between responders and nonresponders concern proliferation (4 pathways concerning cell-cycle checkpoint control and regulation) and DDR (2 pathways concerning role of CHK proteins and ATM signaling pathways).

CHEK1 was one of the most differentially expressed genes (Fig. 2A) and was underexpressed in responders versus nonresponders (fold change = 0.28, P = 0.0000041). To gain an insight into the mechanism of this underexpression, we decided to focus on case number 02. This patient achieved a CR to abiraterone acetate, lasting for more than 4 years (13). We assessed the copy number profile of this tumor by OncoScan comparative genomic hybridization-like (CGH) microarray assay. The CGH profile of the tumor revealed a LOH of several loci, including a deletion of CHEK1 (11q24.2), as well as ATM (11q22.3) and CDC6 (17q21; Fig. 2C). This LOH could explain the observed underexpression of the CHEK1 transcript in this patient. Interestingly, one of three other responders analyzed also presented a LOH of CHEK1 (not shown), indicating that CHEK1 underexpression could rely on genomic abnormalities. However, we cannot rule out the existence of underlying transcriptional regulatory mechanisms for CHEK1 dysregulation. As CHEK1 is overexpressed in nonresponders, we tested the combination of antiandrogen and Chk1 inhibitors with the aim to improve response rate.

Our gene expression analysis suggests that CHEK1 could play a role in the sensitivity of AR-positive TNBC to abiraterone acetate. Consequently, we decided to test the activity of Chk1 inhibitors used alone or in combination with antiandrogens in AR-positive TNBC preclinical models.

We first studied the effects of three antiandrogens [abiraterone acetate, enzalutamide, and darolutamide (ODM-201)] and of two Chk1 inhibitors (GDC-0575 and AZD7762) used alone on the viability of two AR-positive TNBC models, the MDA-MB-453 and SUM185PE cell lines (Fig. 3A; Supplementary Fig. S3). With an IC₅₀ of about 5 μmol/L on the two cell lines, abiraterone acetate was the most potent antiandrogen. Chk1 inhibitors have low IC₅₀s of about 50 nmol/L. Activity of the two Chk1 inhibitors was verified by detecting decrease in the ratio of phospho-Chk1 S296 on total Chk1 expression by Western blotting (not shown).

We then treated cells with the three antiandrogens combined with the two Chk1 inhibitors to determine their CI in the two cell lines. Results showed that abiraterone acetate plus Chk1 inhibitor combinations led to an effect that is at least additive (CI ranging from 0.85 to 1.1) in both cell lines for GDC-0575 and AZD7762 (Fig. 3B). This was not observed with the two other antiandrogens, enzalutamide and ODM-201, wherein the CIs rather pointed to antagonist effects (CI >1). Thus, we decided not to perform further studies with these two antiandrogens.

Abiraterone acetate used alone (at 5 μmol/L) had no effect on the cell cycle. Both Chk1 inhibitors (at 0.1 μmol/L) led to a reduction in G₁ and an accumulation in S-phase (Fig. 3C). There was a slight increase in this effect when abiraterone acetate was combined both with GDC0575 and AZD7762 in both cell lines.

Abiraterone acetate used alone had a weak effect on the percentage of apoptotic (AnnexinV-positive) and dead cells (PI-positive). Both Chk1 inhibitors induced at least a doubling of that percentage, reaching statistical significance in MDA-MB-453. The addition of Chk1 inhibitors to abiraterone acetate increased the percentage of apoptotic cells in MDA-MB-453 cell line but not in SUM185PE cell line (Fig. 3D). This effect was only statistically significant for AZD7762 in MDA-MB-453 cell line.

DNA damage and DNA replication stress were assessed using γH2AX-labeling at serine 139 phosphorylation (Fig. 4A and B). Abiraterone acetate used alone has no effect on γH2AX labeling. As expected, both GDC-0575 and AZD7762 used alone induced a sharp and significant increase in γH2AX labeling in the two AR-positive TNBC cell lines (Fig. 4A and B). When abiraterone acetate was added to GDC-0575, the effect of GDC-0575 was significantly increased in MDA-MB-453 cell line, (P < 0.001; Fig. 4A). The addition of abiraterone acetate to AZD7762 did not increase AZD7762 effect in both cell lines.

Taken together, these in vitro observations suggest that Chk1 inhibitor potentiates the effects of abiraterone acetate, especially in the MDA-MB-453 cell line. The abiraterone acetate–GDC-0575 combination was significantly effective on the cell viability in both cell lines and strongly increased DNA damage in MBA-MB-453 cell line. On the basis of these results, we decided to test this combination on MDA-MB-453 cells growth in vivo.

MDA-MB-453 cells expressing luciferase were injected orthotopically through the nipple into the mammary gland ducts of NSG mice. Treatment with abiraterone acetate significantly reduced tumor growth, measured by luciferase activities during the first 7 days (P < 0.001 at day 7), but its effect did not last after (Fig. 5A). Compared with vehicle or abiraterone acetate, GDC-0575 induced a clear reduction in tumor growth, from day 4 of treatment to the end of the experiment (Fig. 5A). The effect of abiraterone acetate combined with GDC-0575 was more pronounced than abiraterone acetate alone (but not than GDC-0575 alone), and reached statistical significance (P ≤ 0.01). Regarding the tumor doubling times (Fig. 5B), the combination treatment significantly delayed the tumor development compared with either abiraterone acetate (P < 0.001) or GDC05-75 (P < 0.01) alone.

The engrafted mammary glands were dissected out, fixed, and tumor development was assessed by identifying MDA-MB-453 cells in sections using an anti-CK7 antibody. Two types of tumoral components were identified within the mammary gland: in situ carcinoma and massive infiltrating carcinomas. The effect of the treatments on the measured ratio (infiltrating carcinoma/whole gland) is shown in Fig. 5C. These are consistent with the luciferase activity measurements made on live animals (Fig. 5A). Again, the combination treatment was the most effective in delaying tumor development.

This translational research study comprised of two aspects. First, we aimed to identify predictive markers of response to abiraterone acetate and new druggable cotarget(s) of the AR pathway using tumor samples from patients included in UCBG 2012-1 trial (7). Second, we went back to preclinical models to validate our findings.

Regarding the first part, we searched for predictive markers and new targets by analyzing tumor samples at the protein, DNA, and RNA levels. At the protein level, a modest PPV of AR has been demonstrated in the enzalutamide cohort (14). We show that AR expression did not accurately segregate responders from nonresponders to abiraterone acetate. At the mRNA level, there was no association between AR expression and response to abiraterone acetate. The lack of correlation between AR IHC expression and its RNA expression has previously been reported (33). In addition to AR, we analyzed 14 IHC markers including those previously identified in AR-positive TNBC such as FOXA1, GCDFP15 (33), or GGT1 (32). We identified a score with four IHC markers (AR, GCDFP15, FOXA1, and GGT1) which allowed discriminating responders from nonresponders. This score could be improved by morphologic apocrine characteristics that show high negative predictive value. However, the number of responders in this study was very small and this score needs to be tested in an independent data set of AR-positive TNBCs.

At the DNA level, on the 91 cancer-associated sequenced genes, only TP53 segregated responders from nonresponders as it was mutated in 17% and 61%, respectively; however, the number of responders was small. Interestingly, NGS analyses identified a targetable alteration in 71% of patients. A majority of these alterations are related to the PIK3CA pathway supporting previous findings of high rate of PIK3CA pathway mutation among AR-positive TNBC (11, 35, 36). Preclinical data of antiandrogen plus PI3K inhibitor combination have also been published (11) on the basis of those findings and clinical trials are ongoing.

Gene expression analysis revealed that several genes linked to DDR, including CHEK1, are overexpressed in nonresponders compared with responders. CHEK1 activation transiently delays cell-cycle progression through the S and G₂–M-phases, allowing DNA breaks to be efficiently repaired (37). In breast cancer, Chk1 expression has been linked to early local recurrence (38). Interestingly, nonresponders to abiraterone acetate present a high rate of TP53 mutation, while P53 inactivation could be a predictive marker of Chk1 inhibitor efficacy (39). In addition, AR mediates regulation of DDR in classic AR-driven cancer model: prostate cancer (15). Blocking Chk1 using the small-molecule inhibitor AZD7762 increased DNA damage and apoptosis, when used in combination with enzalutamide (40), and therefore, this combinatorial therapeutic strategy is currently under clinical investigation (41).

These results and data encouraged us to go back to preclinical AR-positive TNBC models and to test for the efficacy of combinations of Chk1 inhibitors and antiandrogens. In vitro, the two Chk1 inhibitors (GDC-0575 or AZD7762) had very potent effects on the AR-positive cell line viability, with IC₅₀ in the nanomolar range. This is in concordance with Bryant and colleagues study (41), who reported an IC₅₀ of 100 nmol/L for AZD7762 in MDA-MB-453 cells. Moreover, we show that Chk1 inhibitors had additive effects when used in combination with abiraterone acetate on cell viability, cell-cycle progression through S or G₂–M-phases, accumulation of apoptotic/necrotic cells, and DNA breaks. We further validated the efficacy of abiraterone acetate plus Chk1 inhibition combination using a previously described orthotopically engrafted mice model (42, 43). Direct injection in the mammary gland allowed growing cells in their natural location and microenvironment. Abiraterone acetate–GDC-0575 combination showed a substantial delay in in vivo tumor development compared with monotherapy confirming our in vitro findings.

The lack of additive or synergistic effect of Chk1 inhibitors with the two AR antagonists tested (enzalutamide and ODM-201) could be explained by a different mechanism of action compared with abiraterone acetate. Abiraterone acetate acts as a CYP17A1 inhibitor. One could argue on the interest to use it on in vitro assays, related to its enzymatic effect even if intracrine production of androgens has been demonstrated in advanced prostate cancer (44). However, abiraterone acetate activity in prostate preclinical model was demonstrated to be also linked to a competitive binding to the AR (45, 46). Although the additive effects we measured in vitro were not as pronounced as the ones described using chemotherapy combinations, they were in-line with the effects of combinations using antiandrogens reported in MDA-MB-453 cells (11, 47).

In conclusion, this study identified Chk1 as a potential target in AR-positive TNBCs. The sensitization of cells with a Chk1 inhibitor might be an innovative therapeutic strategy. A clinical trial in patients with AR-positive TNBC evaluating abiraterone acetate combined with a Chk1 inhibitor should be considered.

T. Grellety is a consultant/advisory board member for Novartis. H. Bonnefoi is a consultant/advisory board member for Bayer and Pfizer. No potential conflicts of interest were disclosed by the other authors.

Conception and design: T. Grellety, C. Callens, M. Pulido, H. Bonnefoi, B. Cardinaud

Development of methodology: T. Grellety, C. Callens, E. Richard, M. Pulido, H. Bonnefoi

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T. Grellety, E. Richard, A. Gonçalves, G. MacGrogan, H. Bonnefoi

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): T. Grellety, C. Callens, A. Briaux, M. Pulido, P. Gestraud, H. Bonnefoi

Writing, review, and/or revision of the manuscript: T. Grellety, C. Callens, V. Vélasco, M. Pulido, A. Gonçalves, P. Gestraud, G. MacGrogan, H. Bonnefoi, B. Cardinaud

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): T. Grellety, C. Callens, E. Richard, V. Vélasco

Study supervision: T. Grellety, H. Bonnefoi, B. Cardinaud

Janssen-Cilag contributed to UCBG 2012-1 study with an educational grant to Unicancer and by providing abiraterone acetate. The sponsor of UCBG 2012-1 trial (Unicancer) designed and coordinated the trial. We thank the SIRIC BRIO (Site de Recherche Intégrée sur le Cancer-Bordeaux Recherche Intégrée Oncologie) and GIRCI-SOOM (Groupement Interrégional de Recherche Clinique et d'Innovation Sud-Ouest Outre-Mer) for their financial support (grant API-K, GIRCI-SOOM 2015). We would like to thank Audrey Laroche for her helpful advices on the Chk1 inhibitors assays, Benoit Rousseau and Julien Izotte (Animal facility, University of Bordeaux) for supplying and caring of the mice. We thank Richard Iggo for his logistical support. We thank Genentech for kindly providing GDC-0575. We thank Ravi Nookala of Institut Bergonié for the medical writing service.

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.