Selective resistance to the PARP inhibitor olaparib in a mouse model for BRCA1-deficient metaplastic breast cancer

GEMMs have the potential to capture the cell-intrinsic and cell-extrinsic factors that drive de novo tumor formation, making them exceptionally valuable for cancer research (11). To speed up the development of novel multiallele GEMMs of human cancer, we previously developed a GEMM-ES cell (ESC) strategy, which permits rapid introduction of mutant alleles into ESCs derived from existing, well-characterized GEMMs and subsequent production of experimental cohorts of chimeric mice that can be directly used for preclinical studies (12, 13). To validate candidate oncogenes and their contribution to therapy resistance in BRCA1-associated breast cancer, we set out to develop a female GEMM-ESC pipeline, based on our K14cre;Brca1^F/F;Trp53^F/F (KB1P) mouse mammary tumor model, in which Cre-mediated stochastic inactivation of BRCA1 and p53 leads to mammary tumors and, to a lesser extent, skin tumors (14). We derived clonal ESC lines from male and female KB1P blastocysts and validated their quality and potency by producing chimeric mice via blastocyst injection (Fig. 1A). As expected, sex conversion was evident for the male KB1P clone C3. Clone C3 and female clone A2 contributed well to chimeras, with the majority of animals scoring greater than 40% chimerism, and displayed germ-line transmission. For a second female KB1P clone (C7), more than half of the chimeras showed low chimerism and did not display germ-line transmission, indicating that the quality of this clone was suboptimal (Fig. 1B).

Female ESCs tend to undergo XO conversion during in vitro culture. Indeed, copy number variation (CNV) sequencing of female KB1P ESC lines showed loss of one X-chromosome, which was confirmed by whole chromosome painting (Fig. S1 A and B). Previous studies reported that XO mice are viable and fertile but can exhibit some degree of growth retardation, high-frequency hearing loss, reduced thyroid activity, reduced body temperature, and behavioral abnormalities (15, 16). Chimeric mice derived from female KB1P ESCs did not display any abnormalities, were fertile, and gave germ-line transmission, indicating that our female KB1P ESCs contribute to the development of healthy chimeric offspring regardless of their XO status.

Next, we monitored female KB1P chimeras for development of mammary tumors and compared the mammary tumor latency to that of the original KB1P GEMM (Fig. 1C). Chimeras of KB1P clone A2 developed tumors with a significantly increased median latency of 266 d vs. 209 d for the original KB1P GEMM (P < 0.0001), most likely because of the smaller number of mammary and skin epithelial cells that undergo Cre-mediated loss of BRCA1 and p53 compared with the original model (Fig. 1C). There was no significant difference in tumor latency for chimeras of KB1P clone C7, with a median tumor latency time (t₅₀) of 237 d.

Histologically, mammary tumors from KB1P chimeras were estrogen and progesterone receptor-negative and were classified as solid carcinomas, thus showing morphological features comparable to KB1P GEMM tumors (Fig. 1 D and E). To assess inactivation of BRCA1 and p53 in tumors from KB1P chimeras, we performed Southern blot analysis, which confirmed recombination of the conditional Brca1F and Trp53F alleles in all cases (Fig. S2). Given that BRCA1 loss leads to genomic instability, we assessed DNA CNVs in mammary tumors from KB1P chimeras by CNV sequencing, which confirmed that KB1P chimeric tumors display genomic instability to the same extent as KB1P GEMM tumors (Fig. 1F). Thus, our female GEMM-ESC approach yields healthy chimeras that develop mammary tumors that closely resemble tumors from the original mouse model, and is therefore a novel tool to study breast cancer.

We next determined if the female GEMM-ESC strategy would be suitable to study the contribution of candidate oncogenes to mammary tumor formation. To allow rapid introduction of additional modifications, we made use of the Flp/FRT recombinase-mediated cassette exchange strategy (13, 17) (Fig. 2A). Female KB1P ESC clone A2 was targeted with an FRT homing cassette into the 3′UTR of the Col1a1 locus to produce K14cre;Brca1^F/F;Trp53^F/F;Col1a1^FRT (KB1P-Col1a1) ESCs (Fig. S3A). Next, a Cre-inducible allele of the MET receptor tyrosine kinase, which is associated with EMT and therapy resistance (18, 19), and expressed at high levels in breast cancer (20–22), was introduced in KB1P-Col1a1 ESC clone A2.1 to produce K14cre;Brca1^F/F;Trp53^F/F;Col1a1^{inv-CAG-Met-Luc} (KB1P-MET) ESCs (Fig. S3B). To allow for in vivo monitoring of MET expression by bioluminescence imaging, the Met cDNA was fused to a codon-optimized firefly luciferase (Luc) gene via an internal ribosomal entry site (Fig. S3C). Successful targeting of the FRT homing cassette and subsequent shuttling of the Cre-inducible invCAG-Met-Luc construct was confirmed by Southern blot analysis (Fig. S3 D and E). Luciferase expression was confirmed by bioluminescence imaging of Adeno-Cre–treated KB1P-MET ESCs (Fig. S3F). KB1P-Col1a1 and KB1P-MET ESCs produced high-grade female chimeras, which were monitored for mammary tumor development (Fig. 2B and Fig. S4A). Whereas KB1P and KB1P-Col1a1 chimeras developed mammary tumors with comparable latency, KB1P-MET chimeras developed mammary tumors with a significantly reduced latency of 176 d (P < 0.001; Fig. 2C), demonstrating that MET overexpression accelerates BRCA1-associated tumor development.

Histopathological analysis revealed that all mammary tumors from KB1P-Col1a1 chimeras were classified as carcinomas. In contrast, more than half of KB1P-MET chimeras developed carcinosarcomas, based on the presence of spindle cells that show E-cadherin loss and vimentin expression, indicating that MET expression promotes the development of metaplastic tumors with an EMT-like phenotype (Fig. 2 D and E). KB1P-MET carcinosarcomas also showed a claudin-low gene expression pattern, which is a key characteristic of breast cancers with an EMT phenotype (Fig. 2F) (8). In contrast to KB1P-MET carcinomas, mice bearing KB1P-MET carcinosarcomas developed tumors with a shorter latency (Table S1), indicating that these metaplastic tumors are more aggressive than carcinomas. Southern blot analysis confirmed Cre-mediated recombination of Brca1 and Trp53 in all tumors (Fig. S4B).

To confirm that KB1P-MET carcinosarcomas had undergone an EMT, we performed RNA-sequencing and determined the expression level of a panel of EMT signature genes (Fig. S5). KB1P and KB1P-MET tumors clustered perfectly into three groups. KB1P tumors express high levels of epithelial genes; KB1P-MET carcinomas express high levels of epithelial and mesenchymal genes, whereas KB1P-MET carcinosarcomas express high levels of mesenchymal genes. These results show that KB1P-MET mouse carcinosarcomas display an EMT phenotype and, in this respect, resemble BRCA1-deficient MBC in humans (23).

Mammary tumors from GEMMs can be transplanted orthotopically into WT syngeneic female mice while retaining their morphological features, molecular characteristics, and drug-sensitivity profile (24). To confirm that mammary tumors from chimeric mice can be used for transplantation and intervention studies, we orthotopically transplanted 1-mm³ tumor fragments from KB1P-Col1a1 and KB1P-MET chimeras into the fourth mammary gland of syngeneic WT female recipient mice. Mammary tumor growth was evident after 4–6 wk and histopathological features of the transplanted tumor outgrowths were preserved, demonstrating that chimera-derived mammary tumors are suitable for allografting.

As MBCs have been associated with therapy resistance and poor outcome (25), we set out to assess if the EMT status of KB1P-MET tumors would affect their response to therapy. When allografted tumors reached a volume of 200 mm³, tumor-bearing mice were treated with cisplatin (6 mg/kg on days 0 and 14) or olaparib (50 mg/kg daily) for a period of 28 d or until the tumor reached the maximum allowed size of 1,500 mm³. In line with previous studies (24, 26), KB1P-Col1a1 carcinomas regressed upon cisplatin and olaparib treatment (Fig. 3). Likewise, KB1P-MET tumors shrank upon cisplatin treatment, regardless of their morphological status (Fig. 3A). In contrast, whereas KB1P-MET carcinomas regressed upon olaparib treatment, KB1P-MET carcinosarcomas were intrinsically resistant to olaparib (Fig. 3B). This suggests that olaparib resistance is not induced by MET expression per se, but rather by MET-associated EMT.

Next, we determined if KB1P-MET carcinosarcomas display increased expression of drug efflux transporters, which were previously implicated in olaparib resistance (26). Indeed, we detected increased mRNA expression of the Abcb1a and Abcb1b genes, both encoding Pgp, in KB1P-MET carcinosarcomas, indicating that these tumors may display resistance by increased efflux of olaparib (Fig. 4A). Moreover, Abcb1a and Abcb1b mRNA levels were strongly correlated to the EMT status of the KB1P-MET tumors (Fig. 4B).

To demonstrate that Pgp mediates olaparib resistance of KB1P-MET carcinosarcomas, we determined the response of KB1P-MET carcinosarcomas to olaparib combined with the Pgp inhibitor tariquidar (XR9576) (27). A KB1P-MET carcinosarcoma with high Pgp expression (donor 2) responded well to the combination therapy, confirming that olaparib resistance is mediated by increased drug efflux by Pgp in this tumor (Fig. 4C). Olaparib resistance of a KB1P-MET carcinosarcoma with low Pgp expression (donor 3) could not be bypassed by coadministration of tariquidar, indicating that resistance of this carcinosarcoma was driven by a different mechanism.

Next, we determined the response of KB1P-MET carcinosarcomas to AZD2461, a PARP inhibitor with strongly reduced affinity to Pgp (28). KB1P-MET carcinosarcomas with high Pgp expression (donors 1 and 2) responded well to AZD2461, confirming Pgp-mediated resistance to olaparib (Fig. 4D).

To assess whether also human TNBCs with EMT features have elevated Pgp levels, we evaluated ABCB1 gene expression in several human TNBC subtypes. Similar to mouse KB1P-MET carcinosarcomas and in contrast to basal-like TNBCs with epithelial morphology, the majority of MBCs have a claudin-low phenotype (29, 30). We therefore determined ABCB1 mRNA levels in claudin-low TNBCs and compared them vs. basal-like TNBCs. As expected, claudin-low TNBCs showed a significantly higher EMT score than basal-like TNBCs (Fig. 5A). In addition, claudin-low TNBCs showed significantly increased ABCB1 mRNA levels compared with basal-like TNBCs (Fig. 5B), suggesting that also human MBCs may show intrinsic resistance to olaparib as a result of elevated Pgp-mediated drug extrusion.

All antibodies and their dilutions are summarized in Table S2.

ESCs were derived from KB1P mice and cultured as described previously (13). Male and female ESCs were injected into C57BL/6N blastocysts and implanted into pseudopregnant B6CBAF1/Ola foster mice. The percentage of chimerism of the resulting chimeras was scored by the absence of host-derived coat color. Germ-line transmission was determined by crossing chimeric mice to KB1P (FVB) mice, and the offspring were scored for coat color transmission.

Female KB1P ESCs were targeted with the Col1a1-frt targeting plasmid followed by Flp-mediated integration of the frt-invCAG-Met-Luc vector as described previously (13, 17). Further details are described in SI Methods.

Genomic DNA of ESCs and tumor samples was extracted by proteinase K lysis and organic extraction with phenol-chloroform. For DNA sequencing, DNA libraries for Illumina sequencing were prepared by using the KAPA HTP Library Preparation Kit (KAPA Biosystems) according to the manufacturer’s instructions. The amount of dsDNA in the genomic samples was quantified by using the Qubit dsDNA HS Assay Kit (Invitrogen). As much as 250 ng of double-stranded genomic DNA was fragmented by Covaris shearing to obtain fragment sizes of 160–180 bp. Samples are purified using 1.8× Agencourt AMPure XP PCR Purification beads according to the manufacturer’s instructions (Beckman Coulter). The sheared DNA samples were quantified and qualified on the Agilent Technologies 2100 Bioanalyzer instrument by using the HS assay kit (Agilent Technologies). DNA library preparation for Illumina sequencing was performed by using the KAPA HTP Library Preparation Kit (KAPA Biosystems). During library enrichment, 10 PCR cycles were used to obtain enough yield for sequencing. After library preparation the libraries were cleaned up by using 1× AMPure XP beads. All DNA libraries were analyzed on a BioAnalyzer system by using DNA7500 chips (Agilent Technologies) to determine the molarity. As many as 10 uniquely indexed samples were mixed together by equimolar pooling, in a final concentration of 10 nM, and subjected to sequencing on an Illlumina HiSeq2000 machine in one lane of a single-read 50-bp run, according to the manufacturer’s instructions. Reads were aligned to the reference genome (hg19) by using the Burrows-Wheeler Alignment (BWA) backtrack algorithm (32) and counted in 20-kb nonoverlapping bins. These bin counts were corrected for GC bias using a loose fit. The “mappability” value of each bin is precomputed by summarizing the alignment results of all possible 51-mers from the reference sequence. A linear model intercepting 0 was used to fit the loess-corrected count data to the mappability values. The slope of this fit, multiplied with the mappability value for each bin, provides the bin’s reference value that is used to calculate the final log2 copy number ratios. Encyclopedia of DNA Elements (ENCODE) (33) blacklisted regions and bins with a mappability < 0.2 were excluded from the final dataset.

Mouse whole-chromosome paint X-Cy3 and mouse whole chromosome paint Y-rhodamine green were used as described previously (34).

Small fragments (1–2 mm in diameter) of mammary tumors from chimeric KB1P-Col1a1 and KB1P-MET mice were transplanted orthotopically into the fourth right mammary fat pad of WT FVB/N mice as described previously (24). Treatment was initiated when tumors reached a size of ∼200 mm³ (formula for tumor volume: 0.5 × length × width²). Cisplatin (6 mg/kg i.v.; Mayne Pharma) was administered at day 0 and 14. Olaparib (50 mg/kg i.p.) and AZD2461 (100 mg/kg per os) were administered for 28 consecutive days. Tariquidar (2 mg/kg i.p.; Avaant Pharmaceuticals) was administered 15 min before the olaparib injection for 28 consecutive days. Mice were killed after 28 d of treatment or when the tumor volume exceeded 1,500 mm³.

FFPE sections were stained with H&E. For immunohistochemical stainings, antigen retrieval was performed with citrate buffer, pH 6.0 (cytokeratin 5 and 8, progesterone receptor, cMet, and 53BP1) or Tris⋅EDTA, pH 9.0 (E-cadherin, vimentin, estrogen receptor-α, and phospho-Met). Subsequently, endogenous peroxidases were blocked with 3% (vol/vol) H₂O₂. Slides were preincubated with 10% (wt/vol) nonfat dry milk, 4% (wt/vol) BSA/5% (vol/vol) goat serum, or 2.5% (wt/vol) BSA/5% (vol/vol) rabbit serum in PBS solution. The first antibody was incubated overnight. Next, slides were incubated with HRP-conjugated Envision (Dako) or stained with biotin-conjugated secondary antibodies and incubated with HRP-conjugated streptavidin–biotin complex (Dako). Following detection with 3,3-diaminobenzidine-tetrahydrochloride (A-6926; Sigma), slides were counterstained with hematoxylin and dehydrated.

Genomic DNA was isolated from tissue by proteinase K lysis and organic extraction with phenol-chloroform. Southern blot analysis was performed by using 10 µg genomic DNA, digested with EcoRI/StuI or BglII to determine the status of Brca1 or Trp53. Blotting and hybridization was performed as described previously (35). Probes were radioactively labeled by PCR. Primers for amplification of the 86-nt Brca1 exon 14 probe were as follows: forward primer, 5′-AGC CAA AAT TTG AAG AGC G -3′; and reverse primer, 5′-ATC CCT GAC TCG TCA TCT CC -3′. Hybridization of EcoRV/StuI digested DNA to the Brca1 exon 14 probe resulted in 7-kb (WT Brca1 allele) and 6-kb (deleted Brca1 allele) bands. Primers for amplification of the 700-nt Trp53 5′ XbaI probe were as follows: forward primer, 5′-CTA CCT GAA GAC CAA GAA GG-3′; and reverse primer, 5′-TGG AGG ATA TGG ACC CTA TG-3′. Hybridization of BglII digested DNA to the Trp53 5′ XbaI probe resulted in 18-kb (WT Trp53 allele) and 9.4-kb (deleted Trp53 allele) bands.

The Col1a1-frt targeting construct and the Flpe overexpressing plasmid were provided by J. Gribnau (Erasmus Medical Center Rotterdam, Rotterdam, The Netherlands) (17). Met cDNA was inserted as an FseI-PmeI fragment into the frt-invCAG-Luc vector, resulting in the frt-invCAG-Met-Luc vector. Targeting of the Col1a1 locus with the Col1a1-frt construct and Flp-mediated integration of the frt-invCAG-Met-Luc plasmid was performed as described previously (13). Briefly, female KB1P ESCs (5 × 10⁶) were electroporated with 10 µg XhoI digested Col1a1-frt targeting plasmid DNA in 0.4-cm Gene Pulser cuvettes (Bio-Rad) at 3 µF, 0.8 kV for 0.1 ms in a Bio-Rad Gene Pulser. Cells were plated on a 57-cm² tissue culture dish precoated with 0.1 mg/mL laminin (Sigma). After 24 h, Geneticin (200 ng/mL; Gibco) was added to the culture medium to select positive clones. The medium was refreshed every other day. After 10–14 d, individual clones could be picked, dissociated, and transferred to a 0.1% gelatin-coated 96-well plate. After expending neomycin-resistant ESC clones, genomic DNA was isolated and used for PCR and Southern blot analysis to screen for correctly targeted clones.

Flip-in of Met was achieved by cotransfecting the Flpe overexpressing plasmid (0.6 µg) and the frt-invCAG-Met-Luc vector (4.8 µg) in a Col1a1-frt targeted KB1P ESC clone (KB1P-Col1a1). KB1P-Col1a1 ESCs (1 × 10⁶) were plated on a laminin-coated 57-cm² dish. After 2 d, plasmid DNA was mixed with 9 µL Lipofectamine 2000 (Invitrogen) in 250 µL OptiMEM (Gibco) and added to the culture medium. After 6 h, the medium was refreshed, and, after 24 h, Hygromycin-B (150 µg/mL; Invitrogen) was added. Individual clones were picked after 10–14 d and expanded, and genomic DNA was isolated to screen for correct shuttling. Screening of Col1a1-frt targeted and frt-invCAG-Met-Luc Flp-in ESC clones was performed by PCR and Southern blot as described previously (13).

KB1P-MET ESCs were cultured as described (13) for 24 h in six-well plates and infected with viral Ad5-CMV-Cre particles (6 × 10⁷; Gene Transfer Vector Core, University of Iowa). Three days after infection, luciferase activity in ESCs was determined by bioluminescence imaging.

Beetle luciferin (Promega) was dissolved at 15 mg/mL in sterile PBS solution and stored at −20 °C. For in vitro studies, luciferin solution was diluted 10 times in culture medium. For in vivo studies, luciferin solution was injected i.p. (0.01 mL/g body weight) and animals were anesthetized with 2–3% isoflurane. Light emission was measured 15 min after luciferin administration by using a cooled CCD camera (IVIS; Xenogen), coupled to Living Image acquisition and analysis software over an integration time of 1 min. Signal intensity was quantified as the Flux (photons per second) measured over the region of interest.

Total RNA of tumor samples was isolated with TRIzol (Invitrogen), and the integrity of RNA was verified by denaturing gel electrophoresis. Reverse transcription, hybridization, ligation, PCR amplification, and fragment analysis by capillary electrophoresis were performed as described previously (24).

Total RNA was extracted using TRIzol reagent (Ambion Life Technologies) according to the manufacturer’s instructions. For RNA sequencing, strand-specific sequencing libraries were generated by using the TruSeq Stranded mRNA sample preparation guide (Illumina) according to the manufacturer’s protocol. Strand-specific sequencing libraries were generated by using the TruSeq Stranded mRNA sample preparation guide (Illumina) according to the manufacturer’s instructions. Briefly, polyadenylated RNA from intact total RNA was purified by using oligo-dT beads. Following purification, the RNA was fragmented, random primed, and reverse transcribed by using SuperScript II Reverse Transcriptase (Invitrogen) with the addition of actinomycin D. Second-strand synthesis was performed by using polymerase I and RNaseH with replacement of dTTP for dUTP. The generated cDNA fragments were 3′-end adenylated and ligated to Illumina paired-end sequencing adapters and subsequently amplified by 12 cycles of PCR. The libraries were analyzed on a 2100 Bioanalyzer (Agilent) using a 7500 chip (Agilent) and diluted to 10 nM. The 10-nM RNAseq libraries were denatured and diluted in hybridization buffer (Illumina) to 7 pM concentrations and spiked (0.5%) with a standard PhiX library. Cluster generation was achieved in a c-Bot instrument (Illumina), and deep sequencing was done with a HiSeq2000 machine (Illumina).

A published EMT signature (36) was converted from human to mouse, resulting in 239 epithelial and 224 mesenchymal genes, of which 235 epithelial and 223 mesenchymal genes could be mapped to Ensembl gene identifiers by using Ensembl Biomart (31, 37). Expression data were mean-centered per gene or probe for each data set. The EMT score of each tumor in all data sets is calculated by subtracting the mean log2 expression of the epithelial genes from the mean of the log2 expression of the mesenchymal genes. Consequently, tumors with a mesenchymal gene expression profile display a positive EMT score and epithelial tumors have a negative EMT score. Correlation between EMT score and Abcb1a, Abcb1b, and Abcg2 expression was calculated by using Spearman correlation.

To determine ABCB1 expression levels in human TNBCs, we used the previously described dataset GSE10885, which contains 58 basal-like samples and 17 claudin-low samples (8).

Small fragments of KP (K14cre;Trp53^F/F) carcinoma and KB1P-MET carcinosarcomas were transplanted orthotopically into the fourth right mammary fat pad of WT FVB/N mice. When tumors reached a size of ∼10 × 10 mm, the tumor was irradiated at a dose of 15 Gy by using a cone-beam irradiator. Mice were killed 2 h after irradiation, and tissue was processed. FFPE sections were deparaffinized and rehydrated, after which antigen retrieval was performed on the sections with Target Retrieval solution (DAKO). Tissue was permeabilized by using 0.2% Triton in PBS solution for 20 min at room temperature and subsequently incubated with DNase I (1,000 U/mL; Roche) for 1 h at 37 °C. Sections were incubated with RAD51 antibody (gift from R. Kanaar, Erasmus MC, Rotterdam, The Netherlands) and subsequently with goat anti-rabbit Alexa Fluor 568. Slides were mounted by using Vectashield with DAPI. A minimum of five different areas were imaged (z-stack) per tumor; images were analyzed by using ImageJ to quantify the percentage of cells with more foci than the threshold of five foci per nucleus.