PARP Inhibition Induces Enrichment of DNA Repair–Proficient CD133 and CD117 Positive Ovarian Cancer Stem Cells

Ovarian cancer is the fifth leading cause of cancer-related death in women and the most lethal gynecologic cancer in the United States (1). In 2018, approximately 22,240 women will be diagnosed with ovarian cancer, and about 14,070 women will succumb to their disease (2). Although the majority of women with advanced-stage disease initially respond to platinum-based chemotherapy, tumor resurgence is observed in a majority of women rendering their disease difficult to control durably (3). Novel therapies to both prevent recurrence and actively treat recurrent disease are currently being investigated for women with ovarian cancer in order to prolong survival (4). Whole-genomic analyses have demonstrated that deficient homologous repair (HR) is a prevalent signature in high-grade serous ovarian carcinoma among women with and without germline breast cancer (BRCA) gene mutations (5). Notably, recent clinical trials have demonstrated that PARP inhibitors (PARPi) significantly prolong the time to recurrence and progression in all women with ovarian cancer, with the most robust antitumor activity observed in those women harboring a somatic or germline BRCA mutation (6). These clinical data have led to FDA approval of multiple PARPi as both maintenance and third-line therapies.

Currently, Olaparib (Lynparza, AstraZeneca), Niraparib (Junuvia, Tesaro), and Rucaparib (Rubraca, Clovis Oncology) are FDA-approved PARPi treatments for advanced ovarian cancer patients with and without deleterious germline and somatic BRCA mutations (7–9). Phase II and III clinical trials with PARPi in ovarian cancer have resulted in notable antitumor activity and increased progression-free survival. Although PARPi represent effective therapies, acquired resistance occurs in the overwhelming majority of patients and represents a major clinical hurdle (10, 11). This acquired resistance in ovarian cancer patients has prompted investigators to explore why patients, even those with somatic or germline BRCA mutation, develop resistance to this once effective therapy. Chemoresistance has been suggested as a potential mechanism by which the cancer cells are eluding this double shut down of the DNA repair pathway. A subset of chemoresistance has been attributed to secondary genetic and epigenetic events that restore functional HR pathway in HR-deficient tumors or activate other DNA repair pathways or increased expression of p-glycoprotein efflux transporter mediating multidrug resistance (12).

Alternatively, ovarian tumors may be composed of several cell populations that respond to PARPi in different ways due to fluctuating innate levels of HR efficiency. Given the highly heterogeneous nature of ovarian cancer, it is possible that a subset of the bulk tumor cells is less responsive to this treatment strategy. Cancer researchers in multiple solid and liquid tumor disease sites have long speculated that small subpopulations with stem-like features—often called cancer stem cells (CSC)—can repopulate tumors (13). Such rare cell populations are thought to turn over less frequently and are therefore more resistant to conventional therapies targeting the rapidly dividing cells. From a clinical perspective, CSCs are hypothesized to persist in tumors as a distinct population following standard cytotoxic therapy thereby contributing to disease relapse and distant metastasis (14).

In this study, we hypothesized that differential antitumor activity of PARPi in ovarian CSC and non-CSC populations contributes significantly to the development of PARPi resistance, independent of BRCA mutational status. We suspected that the antitumor activity of PARPi is due to the preferential targeting of a non-CSC population of cells. Although the CSC population manifests relative resistance leading to repopulation of tumor cells with the innate ability to overcome synthetic lethality, we sought to utilize BRCA-mutated and wild-type models to test this hypothesis and elucidate specific mechanisms that allow CSCs to manifest resistance to PARPi. We provide evidence that PARPi treatment resulted in the induction of an enrichment of cell populations expressing antigens linked to stem phenotype in ovarian cancer including CD133, CD117, and aldehyde dehydrogenase (ALDH). Furthermore, our results suggest that ovarian CSCs and non-CSCs respond differentially to DNA stress, and that CSCs may have more efficient DNA repair pathways due to an activation of embryonic repair mechanisms that may confer a survival advantage, contributing to treatment resistance and eventual recurrence.

Human ovarian cancer cell lines (Supplementary Table SA) UWB1.289 MUT (BRCA1 mutant, 2594delC germline mutation in exon 11 and deletion of the wild-type allele; RRID: CVCL_B079) and UWB1.289 WT (RRID: CVCL_B078) were purchased from the American Type Culture Collection (ATCC). Human ovarian cancer cell line PEO1 (BRCA2-mutated, homozygous mutation 5193C>G; RRID: CVCL_2686) was purchased from Sigma Aldrich. OVCAR4 (RRID: CVCL_1672) and OVCAR3 (RRID: CVCL_0465) were purchased from National Cancer Institute-Developmental Therapeutics Program. OVCAR3 All cell lines were routinely tested for Mycoplasma, cultivated 37°C in 5% CO₂ humidity, and passaged until passage 12. UWB1.289 MUT and UWB1.289 WT were maintained in 50% RPMI 1640 (GIBCO, Life Technologies), 50% MEGM (Mammary Epithelial Growth Medium, MEGM Bullet Kit CC-3150; Lonza), 10% FBS (GIBCO, Life Technologies), and 1% Pen/Strep (Thermo Fisher Scientific). UWB1.289 WT is a stable cell line derived from UWB1.289 MUT (ATCC CRL-2945), in which BRCA1 function is restored through transfection with a plasmid carrying the wild-type BRCA1 gene. Selection is maintained by culturing the cells in 200 μg/mL of G-418 (Life Technologies). OVCAR3 cells were maintained in RPMI 1640 (GIBCO), 10% FBS (GIBCO), 1% Pen/Strep (Thermo Fisher Scientific), and 0.01 mg/mL of bovine insulin (Sigma-Aldrich). OVCAR4 cells were maintained in RPMI 1640 (GIBCO), 10% FBS (GIBCO), and 1% Pen/Strep (Thermo Fisher Scientific). PEO1 cells were maintained in RPMI 1640 (GIBCO), 10% FBS (GIBCO), 1% Pen/Strep (Thermo Fisher Scientific), and 2 mmol/L Sodium Pyruvate (GIBCO).

Olaparib and rucaparib were purchased from Selleckchem, and carboplatin was obtained from Sigma. IC₅₀ values (the concentration required to kill and/or inhibit growth of cells by 50% as compared with untreated control wells) of these drugs were estimated from concentration–response curves by using MTT colorimetric assay [MTT: 3(4,5-dimethylthiazol-2-yl)-2,5-dyphenyltetrazolium bromide] analysis. All the cell lines were plated in 96-well plates (10,000 cells/well in 150 μL media). After overnight incubation, olaparib (0.1, 1, and 10 μmol/L), rucaparib (5, 10, and 50 μmol/L), and carboplatin (10, 25, and 50 μmol/L) were added in six replicates to each population. Cell viability was measured after 72 hours (carboplatin) or 7 days (olaparib and rucaparib) using the MTT assay. A calibration curve was prepared using the data obtained from wells that contained a known number of cells. Cell proliferation was assessed by trypan blue (Thermo Fisher Scientific) staining and cell counting following treatment with the indicated concentrations of olaparib or rucaparib. Olaparib and rucaparib treatment was 7 days with retreatment every 2 days due to the half-lives of the drugs.

Cells were stained with Live-Dead (Pacific Blue, 1:600; Invitrogen) to discriminate living cells. Apoptotic and necrotic cells were discriminated using the Annexin/PI staining Kit (Roche) followed by flow cytometry. The following anti-human monoclonal antibodies were used to discriminate CD133⁺ and CD117⁺ cells: anti-CD133 [CD133/2 clone 293C3, 1:10, phycoerythrin (PE)-conjugated; Miltenyi Biotec; RRID: AB_2661207] and anti-CD117 [clone A3C6E2, 1:10, allophycocyanin (APC)-conjugated; Miltenyi Biotec; RRID: AB_2660103]. Following incubation with FcR-blocking reagent (Miltenyi Biotec), cells were resuspended in PBS buffer (PBS, 2% FBS, and 1 mmol/L EDTA) and stained with the relevant antibodies. Respective IgG isotype antibodies were included as negative controls in the first analysis and antibody titration analyses. In subsequent experiments, unstained cells were included as a control for background fluorescence. Cells were analyzed using a Fluorescent Activated Cell Sorting (FACS) LSRII cytofluorimeter (BD Biosciences). Data were collected from at least 1 × 10⁵ live cells/sample and analyzed with FlowJo 10.1 version (TreeStar). ALDH activity was evaluated by using the ALDEFLUOR Kit (Stemcell Technologies). Briefly, cells were harvested, washed, counted, and suspended in ALDEFLUOR buffer at a density of 1 × 10⁶ cells/mL. Cells were divided into two aliquots and incubated with 1.5 μmol/L ALDEFLUOR substrate in the presence or absences of the ALDH inhibitor DEAB. Following a 45-minute incubation at 37°C, the samples were analyzed using the FACS LSRII cytofluorimeter. For cell-cycle analysis cells were treated with either vehicle or 10 μmol/L. The cells were then fixed in 70% ethanol 72 hours after treatment, incubated with 10 μg/mL RNase A (Sigma-Aldrich) at room temperature, and stained with 1 μg/mL DAPI (Thermo Fisher). The DNA content of the cells (100,000 cells per experimental group) was quantified using a FACS LSRII cytofluorimeter and analyzed using cell-cycle software of FlowJo 10.2 version. For cell separation, cell suspensions stained with human anti-CD133 and anti-CD117 antibodies (Miltenyi Biotec) were separated using a FACSAria sorter (BD Biosciences) to obtain the double-negative CD133⁻ CD117⁻ cell populations, and live cells were discriminated with propidium iodide (PI; 1 mg/mL, Thermo Fisher). To acquire a pure CD133⁺ population, the AutoMacs cell separator (Miltenyi Biotec) was used. Specifically, cells stained with PE-conjugated anti-CD133 antibody (Miltenyi Biotec) were magnetically labeled with anti-PE MicroBeads UltraPure (Miltenyi Biotec) and loaded on a Macs column which was placed in the magnetic field of an AutoMacs separator. In all experiments, post-sort analysis using a FACS LSRII cytofluorimeter confirmed a population purity >90%. In the analysis where it was possible to discriminate the different cell populations by flow cytometry expression-dependent gates, CD133⁺CD117⁺ and CD133⁻CD117⁻ cells were compared (cell viability with Live-Dead staining and cell-cycle analysis). Conversely, in the analysis where it was important to obtain a relatively pure population of the different cell subtypes with limited DNA damage, the cells were sorted with the AutoMacs bead system and CD133⁺ and CD133⁻ cells were compared. Specifically, the bead separation system was favored instead of flow cytometric cell sorting for the analysis of the DNA damage induced by PARPi treatment (immunofluorescence and comet assay analysis). This bead system reduced background of DNA damage induced by flow cytometric cell sorting. It was important to obtain the highest purity >98% of the CD133⁺ and CD133⁻CD117⁻ cells for the gene card analysis. Although it was not possible to obtain this level of purity in the CD133⁺CD117⁺–specific sorted population in sufficient numbers.

The extreme limiting dilution assay (ELDA) was performed to analyze the sphere-forming efficiency after olaparib and rucaparib treatment in vitro. Cells were treated under monolayer culture conditions, harvested after 7 days, and plated in ultra-low-attachment 96-wells plates (Corning, Inc.) at different cell concentrations (24 wells with 1 cell/well, 24 wells with 10 cells/well, 24 wells with 50 cells/well, and 24 wells with 100 cells/well). Cells were maintained for 10 days in DMEM-F12 without serum, and bFGF and EGF growth factors were added to the culture medium every 3 days. Counting of spheres was performed after 10 days, and statistical analysis was performed using the ELDA software (http://bioinf.wehi.edu.au/software/elda/).

Using an Institutional Review Board–approved protocol (protocol number: 2005N000273) and in compliance with the Institutional Care and Use Committee, the animal experiments were conducted at our institution Massachusetts General Hospital. Twelve-week-old NOD/SCID mice at 12 weeks of age were s.c. injected with either 3 × 10⁶ OVCAR3 cells or 5 × 10⁶ PEO1 cells in 1:1L PBS:Matrigel (Corning Matrigel, BD Biosciences). After injection of tumor cells, the sizes of the resulting tumors were measured every other day using a caliper; the body weight of each mouse was also determined twice per week. The tumor volume was calculated using the following formula: (width² × height)/2. When the tumor volume reached 150 to 200 mm³, the mice were randomly divided into 2 groups (15 mice per group in OVCAR3 experiment, 7 mice per group in PEO1 experiment). Treatment regimen for olaparib included chronic daily administration with 50 mg/kg of olaparib for 14 consecutive days via i.p. injection. The control group was treated with a 4% DMSO/30% polyethylene glycol/ddH₂O solution alone. In OVCAR3 experiment, an acute treatment of 6 hours was performed, and tumors were formaldehyde-fixed and paraffin-embedded or frozen for further analysis. The other mice were euthanized with CO₂ when the tumor volume reached 900 mm³. Tumor volume was measured every 3 days. At the end of the treatment study, mice were euthanized and xenografts were harvested. Portions of each xenograft were snap frozen as well as formaldehyde-fixed and paraffin-embedded for further analyses. Tumors were processed following a previously described protocol (15), and H-2K^d+ mouse cells were removed using a FITC-conjugated antibody (BD Biosciences) and Macs LD columns (Miltenyi Biotec) as per the manufacturers' recommendations. H-2K^d− cells were stained with Live-Dead (Pacific Blue, 1:600; Invitrogen), anti-CD133 (CD133/2 clone 293C3, 1:10, PE-conjugated; Miltenyi Biotec), and anti-CD117 (clone A3C6E2, 1:10, APC-conjugated; Miltenyi Biotec) and analyzed using FACS LSRII cytofluorimeter (BD Biosciences). Data were collected from at least 1 × 10⁵ live cells/sample and analyzed with FlowJo 10.1 version (TreeStar).

UWB1.289 WT and UWB1.289 MUT cell lines were treated either with 10 μmol/L olaparib or with vehicle for 72 hours. Following the treatment, the media were changed, and the cells were allowed to recover for an additional 24 hours without treatment. CD133⁻ and CD133⁺ fractions were isolated following each timepoint using the AutoMacs Microbead system (Miltenyi Biotec). After cell separation, CD133⁻ and CD133⁺ were subjected to an alkaline comet assay using the Trevigen Comet Assay Kit (Trevigen) following the manufacturer's protocol. Results from three experiments were statistically quantified by counting the number of cells with and without comet tails in a minimum of 100 cells per sample. DNA damage induced by olaparib was determined following comparisons between comet tail frequency in BRCA1-wild-type– and mutant-derived CD133⁺ and CD113⁻ cell populations 72 hours after vehicle or olaparib treatment. Parallel analyses after the 24-hour recovery period determined the relative DNA repair efficiency of CSCs versus non-CSCs.

UWB1.289 MUT, UWB1.289 WT, and OVCAR4 cell lines were stably transfected with the plasmid pDR-GFP (a gift from Maria Jasin; Addgene plasmid 26475) using FugeneHD (Roche). The HR reporter assay was performed as previously described (16). Briefly, the plasmid pCBA-SceI (Addgene plasmid 26477) was used to exogenously express and induce a double-strand break (DSB). Cells were then analyzed 48 hours later for GFP fluorescence using a LSRII flow cytometer. The results were analyzed using FlowJo software.

Data were analyzed with GraphPad Prism 6.0 (GraphPad Software). Data from replicate experiments are shown as mean values ± standard error. Comparisons between groups were analyzed by a two-tailed Student t test or two-way ANOVA, as appropriate. A P value < 0.05 was considered statistically significant.

We hypothesized that PARPi induces an enrichment of CD133⁺, CD117⁺, and ALDH⁺ cell populations that have been shown to manifest CSC-like properties (17). To test this hypothesis, we used five different ovarian cancer cell lines with varying BRCA1/2 mutation status (see Supplementary Table SA): (1) UWB1.289 MUT (BRCA1 mutant); (2) UWB1.289 WT in which BRCA1 function has been restored via stable transfection; (3) PEO1 (BRCA2 mutant); (4) OVCAR4 (BRCA1/2 wild type); and (5) OVCAR3 (BRCA1/2 wild type). Cells were treated with either olaparib or rucaparib, and CD133/CD117 expression levels and cell viability were measured 7 days after treatment by flow cytometry. The dose-response curves were similar for olaparib across all cell lines, whereas the response to rucaparib was cell line dependent (Supplementary Fig. SB1). Both PARPi showed specificity in reducing PAR levels (Fig. 1A). Olaparib treatment in all the cell lines was associated with a decrease of cell viability by Live Dead or Annexin/PI staining (Fig. 1B, left; Supplementary Fig. SB2) when compared with the vehicle-treated cells. The efficacy of olaparib at the lower concentrations was more pronounced in the BRCA1-mutant UWB1.289 cell line compared with the BRCA1 restored wild-type line; however, at the highest concentration, there was no difference between the two cell lines. Olaparib treatment at 10 μmol/L concentration showed an enrichment of CD133⁺ and CD117⁺ cells (Fig. 1B, P value ≤ 0.001) in all the cell lines, with the exception of PEO1 cell line in which 90% of the cells expressed CD133 before the treatment. RT-PCR analysis confirmed the increase of PROM1 (CD133) and KIT (CD117) expression in all cell lines as well as increased expression of the stem-related genes SOX2 and POU5F1 (OCT4; Supplementary Fig. SC and Supplementary Table S1). We also investigated the activity of ALDH, a detoxifying enzyme recognized as CSC marker. Olaparib treatment of all cell lines was associated with an enrichment of cells with elevated ALDH activity, further supporting its role in chemoresistance and survival mechanisms in CSCs (Supplementary Fig. SD). Enrichment of CSC-related marker expression was drug dose-dependent and not specific to one PARPi. Rucaparib showed a similar enrichment of CD133⁺ and CD117⁺ cells (Supplementary Fig. SE2, P value ≤ 0.001). To demonstrate that olaparib specifically targeted the CD133⁻CD117⁻ cells, we performed posttreatment flow cytometric analysis to determine the cell viability of CD133⁺, CD117⁺, CD133⁺CD117⁺, and CD133⁻CD117⁻ cell fractions (Fig. 1C). The graphs reveal that olaparib treatment at 10 μmol/L had a negative effect on the survival of CD133⁻CD117⁻ cells, with a >50% decrease in viability 7 days after treatment. In contrast, CD133⁺, CD117⁺, and CD133⁺CD117⁺ cell populations did not show any change in viability, confirming their ability to survive the PARPi treatment. To corroborate these data, we confirmed the CSC properties in the cells enriched with olaparib and rucaparib treatment. Their sphere-forming efficiency in vitro was tested in a limiting-dilution assay and analyzed by the ELDA software. After olaparib and rucaparib treatment, the enriched CD133⁺ and CD117⁺ cell fractions demonstrated a corresponding increase in their sphere-forming capacity (Fig. 2A; Supplementary Table SE4). Interestingly, the enrichment of CD133⁺ and CD117⁺ cells in response to olaparib or rucaparib in vitro was not different when comparing BRCA-mutated cell lines (UWB1.289 MUT and PEO1) with BRCA-wild-type lines (UWB1.289, OVCAR3, and OVCAR4), suggesting a potential resistance mechanism to PARPi therapies independent of BRCA mutation status.

The enrichment of CD133⁺ and CD117⁺ cells after olaparib treatment in vitro led us to investigate whether we would observe a similar effect in vivo. OVCAR3 and PEO1 cells were injected s.c. in NOD/SCID mice, and a 2-week course of daily olaparib treatment began when tumor volume reached approximately 150 to 200 mm³. Olaparib alone blunted tumor growth (P value < 0.001) compared with vehicle control in both models (Fig. 2B, top plot). In both in vivo experiments, flow cytometric analysis of tumors harvested at the end of the treatment period revealed an increased frequency of CD133⁺ and CD117⁺ cells in olaparib-treated tumors compared with vehicle-treated controls (Fig. 2B, bottom plot, P value < 0.01). This enrichment of CD133⁺ and CD117⁺ cell populations is similar to what we observed in our in vitro analyses and confirms PARPi-mediated enrichment of CSCs.

Many studies suggest that normal stem cells respond differently to genotoxic injury by activating a more efficient DNA repair mechanism as compared with differentiated cells (18–20). Growing evidence suggests that CSCs may also utilize these same mechanisms of DNA repair to mediate resistance to both chemotherapy and radiotherapy (21–23). To understand better how ovarian CSCs may escape the negative effects of PARPi, we assessed cell-cycle progression, DNA damage, and DNA repair in the CSC and non-CSC fractions. Analysis of cell-cycle distribution in purified fractions after olaparib treatment showed a greater percentage of cells in G₂–M in CD133⁺CD117⁺ cells compared with CD133⁻CD117⁻ cells in all cell lines, suggesting that olaparib induces a G₂–M cell-cycle arrest in CSCs (Fig. 3, P value < 0.0001). Cells arrest in G₂–M phase to repair DNA DSBs through HR (24). Histone H2AX is phosphorylated in response to DNA damage and is an early driver of the recruitment and localization of DNA repair proteins (25). Nuclear γH2AX (phosphorylated H2AX) and RAD51 foci formation was monitored by immunofluorescence in the UWB1.289 BRCA1-wild-type and BRCA1-mutant cell lines following vehicle or olaparib treatment (Fig. 4A). Cells treated with carboplatin were included as a control. Confocal imaging confirmed that both γH2AX and RAD51 foci accumulated in the nuclei after olaparib treatment of the BRCA1-wild-type and -mutant cell lines, and this increase in RAD51 foci was statistically significant (Fig. 4A, P value < 0.0001). The levels of γH2AX and RAD51 were assessed by Western blotting following treatment with vehicle or PARPi in UWB1.289 WT and MUT cell lines (Supplementary Fig. SF). γH2AX and RAD51 protein levels were similarly increased after olaparib treatment compared with vehicle, regardless of BRCA gene status.

To explore HR pathway activity in CSC and non-CSC cell populations, CD133⁺ and CD133⁻ subfractions were isolated from UWB1.289 BRCA1-wild-type and -mutant cell lines, and treated with either vehicle or olaparib. Confocal imaging of RAD51 and γH2AX foci formation demonstrated an increase in the number of RAD51 foci in CD133⁺ cells compared with CD133⁻ cells after olaparib treatment. For this experiment, we chose to minimize any potential background damage from sorting. Therefore, the cells were sorted with the AutoMacs bead system (see Materials and Methods). Quantitative comparison of 100 cells showed an increase of RAD51 expression in the nuclei of CD133⁺ cells (Fig. 4B, bottom plot) in both BRCA1-mutant and BRCA1-wild-type cells (P value < 0.0001), suggesting a different olaparib-induced DNA damage response in CSCs that is independent of BRCA1 gene status.

The accumulation of RAD51 foci in the nucleus suggested to analyze the DNA damage and repair rate using the comet assay, which can discriminate between a cell with DNA damage (as evidenced by a comet tail) and a cell without DNA damage (no comet tail). UWB1.289 BRCA1 WT and MUT-derived CD133⁺ and CD133⁻ cells were isolated using the AutoMacs bead system to avoid FACS-induced DNA damage. The comet assay was performed after 72 hours after olaparib treatment to confirm the induction of DNA damage. At that time, olaparib was removed from the culture medium, and cells were allowed to recover for an additional 24 hours. A second comet assay was then assessed to determine the relative DNA repair efficiency following recovery from the olaparib treatment. Confocal imaging showed an increase of cells with comet tails after olaparib treatment (Supplementary Fig. SG; P value < 0.001), suggesting DNA damage in the treated cells and consistent with the observed increase in γH2AX foci formation in both CD133⁺ and CD133⁻ cells (Fig. 4B). After the 24-hour recovery period, confocal imaging showed an increased number of CD133⁺ cells (P value < 0.0001) without a comet tail relative to that of CD133⁻ cells (P value < 0.01; Fig. 5A), suggesting that the CSC population is better able to repair DNA damage within 24 hours after cessation of olaparib treatment. Quantification of these data (Fig. 5B) showed that although both cell populations have a statistically significant increase in cells undergoing active DNA repair, a greater percentage of the CD133⁺ CSC population manifested a more efficient DNA repair rate, which was comparable with that of untreated cells. This differential DNA repair activity in CSCs was not correlated with BRCA1 mutation status.

Expression analysis of a subset of genes involved in DSB repair was performed to identify the specific pathways associated with PARPi resistance in CSCs. Although previous work demonstrated that drug inhibition or gene silencing of the RAD51 family of proteins affected response to PARPi treatment (26), there is no information regarding the relative efficiency of DNA DSB repair in CSCs. This lack of data stresses the importance of the HR pathway in DNA repair and PARPi resistance. We utilized custom PCR arrays comprising genes associated with DSB repair and BRCA-ness to analyze the relative expression of each gene in UWB1.289 WT- and MUT-derived CD133⁺ and CD133⁻ CD117⁻ cell populations. Olaparib treatment induced an increase in the expression of the RAD51 gene in both CD133⁺ (Fig. 6A; Supplementary Table S2) and CD133⁻CD117⁻ cell populations (Fig. 6B; Supplementary Table S2), confirming the important role of RAD51 proteins in the DNA damage response. Olaparib treatment was also associated with marked alteration in the expression of the DMC1 gene which encodes an embryonic RAD51 homolog that is not expressed in somatic tissue. DMC1 expression in CD133⁺ cells derived from both UWB1.289 WT and MUT cell lines was increased more than 2 times when compared with expression in vehicle-treated cells (Fig. 6A; P value < 0.0001). An olaparib-induced increase in relative DMC1 expression was also observed in the CD133⁻ CD117⁻ cell populations but was much less robust than the one detected in CD133⁺ cells (Fig. 6C). Appreciating the challenge of obtaining sufficient numbers of CD133⁺-sorted cells to analyze DMC1 protein expression by Western blot, we chose to use the immunofluorescence approach to confirm changes nuclear DMC1 protein expression. CD133⁺ and CD133⁻ cell fractions were isolated from UWB1.289 WT and MUT cells by AutoMacs bead system to reduce cell sorting–correlated DNA damage. DMC1 foci formation was analyzed in the samples following treatment with either vehicle or olaparib treatment. Confocal imaging revealed specific DMC1 foci formation in CD133⁺ cells compared with CD133⁻ cells (Fig. 6D) after olaparib treatment. Quantitative comparison of 100 cells revealed an increase of DMC1 expression in the nuclei of CD133⁺ cells (Fig. 6D, bottom plot) derived from both the BRCA1-wild-type and the BRCA1-mutant cell lines (P value < 0.0001). Specifically, DMC1 foci accumulation in CD133⁺ cells was PARPi treatment-dependent and not induced by carboplatin, a more traditional cytotoxic drug (Fig. 6E; Supplementary Fig. SH). These data suggest that the differential DNA repair mechanism in CSCs involves the DMC1 recombinase. RAD51 and DMC1 are important regulators in the initiation of HR (27). The olaparib induced increase of RAD51 and DMC1 foci in the nucleus of CD133⁺ cell populations suggests that CSCs can circumvent PARPi treatment by manifesting inherent HR processes. To provide some proof of concept for the role of DMC1 recombinase as a potential contributor to DNA repair efficiency in the CSCs, we induced overexpression of DMC1 gene in the CD133⁻ sorted cells by plasmid transfection to observe whether the nonstem cells would display the differential phenotype observed in the CSCs. The DNA repair rate was analyzed after 24-hour recovery period after olaparib treatment. We confirmed the DMC1 protein overexpression by Western blotting after the transfection and before the 24-hour recovery period in both UWB1.289 WT and MUT cell lines (Fig. 7A). CD133⁺ and CD133⁻ cells were isolated after a 24-hour recovery period, and the DNA repair efficiency was quantified with the comet assay. DMC1 overexpression resulted in enhanced DNA repair in the CD133⁻ cells after olaparib treatment and provided a protective role from inherent DNA damage in the vehicle-treated cells, compared with the CTRL vector-transfected cells (Fig. 7B).

To determine the inherent DNA repair capacity of CD133⁺ and CD117⁺ cells relative to that of their CD133⁻ and CD117⁻ counterparts, we utilized a reporter assay which assesses HR activity (Fig. 7C). UWB1.289 BRCA1 WT and MUT cells stably expressing the DR-GFP reporter were transiently transfected with a plasmid encoding the endonuclease I-SceI to generate a specific DSB in the GFP coding region. In this assay, functional GFP can only be restored if the DSB is repaired using the downstream GFP fragment as a template for HR (Fig. 7C). GFP expression detected after 48 hours of the endonuclease-induced DSB damage revealed a more efficient DNA repair capacity in the CD133⁺ and CD117⁺ cell populations, as evidenced by a statistically significant increase in GFP expression (Fig. 7D, P value < 0.05) when compared with the one in the CD133⁻ CD177⁻ population. These data suggest that CSCs harbor inherent DNA DSB repair mechanisms that allow them to adapt to selective pressure and manifest HR.

Recent clinical trials have confirmed the clear clinical benefits of PARPi in the treatment of women with ovarian cancer, yet investigators acknowledge its limited impact on the frequent acquired resistance in the overwhelming majority of patients (28). We hypothesized that treatment with a PARPi would result in an enrichment of CSC populations that persist beyond therapy and eventually serve to seed tumor resurgence. Utilizing in vitroand in vivo models, we demonstrated that olaparib treatment enriched ovarian cancer stem–like CD133⁺ and CD117⁺ cell populations, as well as inducing cell death in the CD133⁻ CD117⁻ non–stem-like population, regardless of BRCA mutation status. A similar response to the PARPi rucaparib suggested this result was not limited to a single PARPi but was more likely to be a class effect. The CD133⁺ cell fractions displayed an extended G₂–M phase which was more pronounced following PARPi treatment, suggesting that the CSC population may be undergoing DNA repair. Gene expression analyses utilizing arrays focused on HR and non-homologous end joining (NHEJ) genes revealed a preferential expression of DMC1 in the CD133⁺ fractions that again was also more pronounced in response to olaparib. Assessment of γH2AX along with RAD51 and DMC1 foci revealed that RAD51 and DMC1 foci were abundant in the CSC population when compared with the non-CSC population. RAD51 and DMC1 are both associated with HR repair (29). Further analysis using the comet assay and an HR reporter assay to better determine DNA repair efficiency confirmed the capacity of ovarian CSC to respond to and repair DNA damage. Collectively, these data reveal the enhanced DNA repair capability of stem-like ovarian cancer cells which likely enables them to escape the PARPi-induced disruption in their genomic integrity leading to their survival and driving of disease recurrence.

Maintenance of stem cell genomic integrity is critical for tissue and organ homeostasis (30). Consequently, it has long been suggested that adult stem cells have a greater capacity to withstand DNA damage when compared with their non–stem cell counterparts (31, 32). It has also been proposed that adult stem cells have the capacity to repair their DNA damage more efficiently when induced by intrinsic cellular metabolism or extrinsic factors like ionizing radiation or pharmacologic agents (33). Proposed mechanisms by which stem cells evade irreparable damage include the stem cell's relatively quiescent state. The infrequent turnover limits endogenous DNA damage which might occur as the result of normal cellular metabolism. Alternatively, stem cells can often be found in protective microenvironments (34). Lastly, stem cells may rely on more efficient DNA damage response and repair mechanisms.

In addition to an infrequent turnover rate, CSCs may employ alternative mechanisms of DNA repair to mediate resistance to chemotherapy and radiotherapy (35). In our study, an extended G₂–M phase was evident in the CSC fractions—independent of the treatment, which suggests that these populations were not in a fully quiescent state but more likely in a lag phase allowing for time to respond to DNA damage. The extended G₂–M phase was more prevalent post treatment with a PARPi, further suggesting that PARPi may affect CSCs even though these populations exhibited an improved ability to overcome the deleterious effects. Evidence of DNA damage in the CSCs was evident by the accumulation of γH2AX foci concurrent with the accumulation of RAD51 foci. Other investigators have demonstrated enrichment of CSCs after treating solid tumors with cytotoxics and other pharmacologic therapies targeting breast and ovarian cancer (36, 37).

Because olaparib extended the G₂–M phase in CSCs, we sought to determine whether there was differential gene expression related to HR and NHEJ pathways. The most striking difference observed was in the expression of DNA meiotic recombinase 1 (DMC1), which is thought to be a meiosis-specific recombinase. Normally, genetic recombination in meiosis contributes to genetic diversity through the induction of programmed DSBs, which are repaired through HR. Recent data have shown that malignant cells are able to adapt unique signaling pathways normally attributed to meiosis and/or embryonic development (38). DMC1 expression has been detected in glioblastoma cells and was induced by radiation (39). Depletion of DMC1 in glioblastoma multiforme (GBM) cells, together with radiation treatment, blunted the DNA damage response. Moreover, loss of DMC1 stabilized tumor growth and increased survival. Unlike GBM, in the ovarian cancer lines, the increase in DMC1 was only induced by PARPi (olaparib and rucaparib) as no evidence of DMC1 foci was observed in response to carboplatin. Although the GBM studies implied that DMC1 expression was ubiquitous, our analyses determined that it was primarily specific to ovarian cancer CSCs.

These data highlight DMC1 as an important mediator of HR preservation in CSCs treated with PARPi. Interestingly, although carboplatin treatment of ovarian cancer cells induced nuclear γH2AX and RAD51 foci in CD133⁻ and CD133⁺ cells, only PARPi induced an accumulation of γH2AX, RAD51, and DMC1 foci exclusively in the CD133⁺ population. This finding suggests an activation of alternative homologous recombination DNA repair pathways in stem-like cells. Given that two separate analyses provided evidence that CD133⁺ cells efficiently repair olaparib-induced DNA damage, we sought to assess whether DMC1 expression in CD133⁺ cells could augment DNA repair following olaparib treatment. Because DMC1 expression was induced by PARPi, the gene could not be knocked down. We therefore chose to stably transfect non-CSCs with DMC1 to determine if the transformed cells could mimic the DNA damage response observed in CSCs. Exogenous expression of DMC1 in the non-CSC population was sufficient to mount a DNA damage response to reduce the impact of PARPi. Collectively, these data highlight a potential DMC1-mediated resistance mechanism to PARPi therapy. Whether the enhanced DNA repair capability of ovarian cancer CSCs is mediated by DMC1 alone is yet to be determined. Regardless, our analyses strongly suggest that the enhanced DNA repair capability of CSCs contributes to the maintenance of their genomic integrity during PARPi treatment, thereby allowing them to survive and lead to the recurrence of disease.

P.A. Konstantinopoulos is a consultant/advisory board member for Pfizer, AstraZeneca, Vertex, and Merck. W.B. Growdon reports receiving a commercial research grant from, and is a consultant/advisory board member for Tesaro. No potential conflicts of interest were disclosed by the other authors.

Conception and design: C. Bellio, W.B. Growdon, B.R. Rueda

Development of methodology: C. Bellio, C. DiGloria, R. Foster, B.R. Rueda

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C. Bellio, C. DiGloria, B.R. Rueda

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C. Bellio, C. DiGloria, R. Foster, K. James, P.A. Konstantinopoulos, W.B. Growdon, B.R. Rueda

Writing, review, and/or revision of the manuscript: C. Bellio, C. DiGloria, R. Foster, K. James, P.A. Konstantinopoulos, W.B. Growdon, B.R. Rueda

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C. DiGloria, B.R. Rueda

Study supervision: B.R. Rueda

We are grateful to Dr. Unnati Pandya for her expertise in DNA transfection and sequencing.

This work was funded by the Advanced Medical Research Foundation (C. Bellio, C. DiGloria, R. Foster, and B.R. Rueda) and the Vincent Memorial Research Funds (B.R. Rueda).

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.