RAD51 Mediates Resistance of Cancer Stem Cells to PARP Inhibition in Triple-Negative Breast Cancer

Triple-negative breast cancer (TNBC) is an aggressive subtype of breast cancer that accounts for 10% to 17% of all breast cancers (1). A subset of TNBCs develops in BRCA1 mutation carriers or in breast cells that lose functional BRCA1 activity. Because BRCA1 functions in DNA repair, synthetic lethality strategies involving the use of DNA repair inhibitors have been developed. These include inhibitors of PARP, an enzyme that detects single-strand breaks and recruits DNA repair molecules. Inhibition of PARP results in accumulation of DNA breaks, which are recognized and repaired by the DNA double-strand break pathway; thus in patients with loss of BRCA activity, cells are subjected to cell death due to excessive DNA damage. Successful implementation of a synthetic lethality therapeutic strategy was demonstrated in preclinical models where PARP inhibition was shown to selectively target breast cancer cells lacking functional BRCA1 or BRCA2 (2–4). Olaparib is a potent inhibitor of PARP1 and PARP2 and has been shown to potentiate the effects of DNA-damaging agents (5). Early clinical studies suggest that olaparib has considerable activity in patients with BRCA-deficient breast tumors (4, 6–9). However, the durability of this response and long-term clinical benefits of this approach have yet to be demonstrated.

There is increasing evidence that resistance to chemotherapeutic agents is mediated by a cellular subset that displays stem cell properties. These cancer stem cells (CSC) are functionally defined as cells that have self-renewal capacity, as well as the capacity to regenerate a phenotypically similar bulk tumor. CSCs mediate tumor metastasis and contribute to chemotherapy and radiotherapy resistance. Furthermore, these therapies often increase levels of CSCs in tumors. However, it is unknown whether the same effect occurs with targeted therapies, including PARP inhibitors (PARPi). In previous studies, we have reported that breast CSCs display increased expression of some DNA repair genes compared with bulk tumor cells (10). We identified RAD51 as a potential mediator of CSC resistance to PARPi. RAD51 forms a complex with BRCA1, and this complex mediates homologous recombination during DNA damage repair. This process is highly conserved throughout the phylogenetic tree. The association between elevated RAD51 levels and therapy resistance in cancer patients has been shown in a wide range of cancer types including breast cancer (11). CSCs in several tumor types have been reported to display increased efficiency in DNA damage repair (12–15), suggesting that CSCs may be insensitive to PARP inhibition. In the present study, we utilized in vitro and in vivo models of BRCA1-mutant and –wild-type TNBC to study the sensitivity of CSCs to PARPi and the role of RAD51 in mediating this response.

TNBC lines SUM149 and SUM159 (a gift from Dr. Stephen Ethier, Karmanos Cancer Institute) were maintained in Ham's F-12 (Invitrogen) supplemented with 5% FBS (Hyclone), 1X antibiotic-antimycotic, 10 μg/mL gentamicin (both from Life Technologies), 5 μg/mL insulin, and 1 μg/mL hydrocortisone (both from Sigma-Aldrich). MDA-MB-231 and HCC1937 were purchased from the ATCC and grown in DMEM and RPMI 1640 medium (Invitrogen), respectively, with 10% FBS (Hyclone), 1% antibiotic-antimycotic, and 10 μg/mL gentamicin (both from Life Technologies). All cells were incubated at 37°C with 5% CO₂.

The inducible TRIPZ-shRNAs were purchased from Dharmacon Open Biosystems (http://dharmacon.gelifesciences.com/openbiosystems/) to produce Tet-inducible RAD51 knockdown (KD) lentiviruses by transfecting 293T cells in the University of Michigan Vector Core Facility. The TNBC cells were infected with the presence of polybrene (8 μg/mL; Millipore) overnight, and the next day, medium containing viruses was washed and replaced with fresh medium. Puromycin (Invitrogen) selection was performed for 7 days.

Total RNA was extracted using an RNeasy Mini kit (Qiagen), and 1 μg of RNA was used for making cDNA with the High-Capacity cDNA Reverse Transcription Kit (Life Technologies). cDNA was analyzed using real-time quantitative RT-PCR assays in an StepOne Real-Time PCR system (Applied Biosystems). RAD51 and GAPDH primers were obtained from Applied Biosystems. For the DNA repair PCR array, RNA was extracted using an RNeasy Mini kit (Qiagen), and total RNA was quantified by Nanodrop (Thermo Fisher Scientific). RNA quality and integrity were assessed on the Agilent 2100 Bioanalyzer. cDNA was synthesized from 400 ng total RNA, loaded to the RT² Profiler qPCR Human DNA Repair Array (Qiagen), and amplified on a 7900HT Real-Time PCR system (Applied Biosystems) according to the manufacturer's instructions (Qiagen). Relative expression levels were calculated based on the DDCt method as described in the RT² Profiler PCR Array Handbook (06/2013; Qiagen).

Western blotting was performed to test the efficiency of RAD51 KD and overexpression. Infected cells with or without (control) doxycycline (Sigma-Aldrich) treatment were lysed in radioimmunoprecipitation assay buffer (Sigma-Aldrich) containing protease and phosphatase inhibitors (Thermo Scientific). Note that 40 μg extracts from each sample was electrophoresed on a 4% to 12% Bis-Tris gel and transferred to a polyvinylidene difluoride membrane (Life Technologies). RAD51, BRCA1 (Cell Signaling Technology), ALDH1A1 (LSBio), and horseradish peroxidase (HRP)–conjugated β-actin (Sigma-Aldrich) were applied overnight at 4°C or 2 hours at room temperature in 5% non-fat milk (Bio-Rad) followed by secondary antibody anti-mouse-HRP or anti–rabbit-HRP (Cell Signaling). The membrane was stripped between antibodies using Restore Western Blot Stripping Buffer, and the staining was detected by Super Signal West Pico Chemiluminescent Substrate (Pierce).

To study RAD51 activity, cells were irradiated with a single dose of 4 Gy with an IC-320 orthovoltage irradiator (Kimtron Medical) and then fixed at different time points after radiotherapy treatment. MTT assay was conducted to test the role of RAD51 KD in cellular survival with olaparib (Selleckchem) treatment. Cells were plated into 96-well plate with or without doxycycline at 1 μg/mL (Sigma-Aldrich). Olaparib was added 72 hours after doxycycline induction and then every 3 days for 7 days. A lower dose of doxycycline (10 ng/mL) was used to induce 2-fold KD in the SUM149KD cell line.

The mice were in-house bred and housed in pathogen-free rodent facilities at the University of Michigan. All supplies (cages, chow, and sterile water) were autoclaved, and all experiments were conducted according to standard by the University Committee on the Use and Care of Animals. SUM149 and SUM159 cells were injected into the mammary fat pads of 6- to 8-week-old female NOD/SCID mice. Olaparib was administrated 15 mg/kg daily via intraperitoneal injection, and doxycycline (2 mg/mL) was administrated in the water supply [5% sucrose (w/v); Sigma-Aldrich], and tumors were monitored weekly. Animals were euthanized by the end of treatment or when the tumors reached an average of 600 mm³. Tumors were minced and digested by 1X collagenase/hyaluronidase (Stem Cell Technologies) in medium 199 (Invitrogen), and filtered through a 40-μm nylon mesh. Live tumor cells were sorted by fluorescence-activated cell sorting using anti-human HLA-A, B, C (BioLegend) and DAPI (Sigma-Aldrich). Tumor cells from each treatment group were implanted at 5,000, 500, and 50 cells into each mammary fat pad of NOD/SCID mouse, and the mice were euthanized when tumors volume reached 600 mm³.

The ALDEFLUOR assay was performed according to the manufacturer's (Stem Cell Technologies) instruction. Dissociated cells were suspended in assay buffer containing ALEDEFLUOR substrate and incubated with or without aldehyde dehydrogenase inhibitor diethylaminobenzaldehyde. MoFlo Astrios (Beckman Coulter) and Summit software were used for data acquisition and analysis.

Flow-sorted cells grown on chamber slides were fixed in 4% paraformaldehyde for 10 minutes and permeabilized with 0.25% Triton-100. Primary antibodies RAD51 (Santa Cruz Biotechnology) and γH2AX (Millipore) were applied at room temperature for 1 hour followed by Alexa-fluor 488 and 546 (Invitrogen) for 30 minutes. Cells were then mounted with ProLong gold antifade reagent with DAPI (Invitrogen).

One- and two-way ANOVA test was applied for analyzing in vitro and in vivo comparisons, and correction for multiple comparisons was performed using Tukey or Sidak test as appropriate. Image J was used for Western blot quantification. StepOne software (Applied Biosystems) was used for analyzing real-time PCR data, and extreme limiting dilution analysis was performed for calculating CSC frequency (16).

HCC1937 and MDAMB231 cells were purchased from the ATCC, and their identity is routinely monitored by short tandem repeat (STR) profiling. SUM149 and SUM159 cells were gifts from Dr. Steven Ethier. All cell lines were cryopreserved within 10 passages, and no thawed cell aliquots were cultured for more than 6 months continuously. Cells were monitored regularly for mycoplasma contamination using the MycoAlert Mycoplasma Detection Kit (Lonza). No cell authentication was performed by the authors.

PARPi have been reported to effectively target BRCA1-mutant breast cancer cell lines and tumors (2, 3, 17). We therefore tested the cytotoxic effect of olaparib on four TNBC cell lines including both BRCA1-mutant and BRCA1–wild-type cell lines: SUM149, SUM159, HCC1937, and MDA-MB-231, and the status of BRCA1 deficiency was confirmed by Western blotting (Supplementary Fig. S1A). As expected, at clinically relevant concentrations, PARPi treatment resulted in fewer viable cells in BRCA1-mutant cell lines (SUM149 and HCC1937) compared with those with wild-type BRCA1 (SUM159 and MDAMB231; Fig. 1A and B). These data confirm the synthetic lethality of PARP inhibition and BRCA1 mutation in TNBC.

Accumulating evidence from a number of solid tumors suggests that CSCs are more resistant to chemotherapy and radiotherapy as compared with bulk tumor cells, but there is limited data on the effect of PARPi on CSC populations. We therefore assessed whether PARP inhibition targets the CSC population in TNBC cell lines using ALDEFLUOR assay to identify CSCs. Despite the dramatic proapoptotic effect of PARP inhibition on BRCA1-mutant bulk tumor cells, the proportion of CSCs, identified by high aldehyde dehydrogenase activity, was elevated after 7 days of PARPi treatment in BRCA1-mutant cell lines SUM149 and HCC1937 (Fig. 1C). More importantly, the absolute CSC number, which was determined by the total cell number multiplied by the percentage of ALDEFLUOR-positive cells, was not affected by PARP inhibition (Fig. 1D). PARP inhibition also had limited effect on the CSCs in BRCA1–wild-type SUM159 and MDA-MB-231 (Fig. 1C and D), consistent with the expectation that neither bulk tumor cells nor CSCs are affected by PARPi treatment because of the presence of functional BRCA1. Our data suggest that although PARPi effectively targets bulk populations in BRCA1-deficient tumor cells, the CSCs in these tumors are resistant to PARP inhibition.

Although the mechanisms of PARPi resistance in CSCs are not fully understood, we have previously demonstrated that the CSC population expresses elevated levels of several DNA repair proteins (10). We thus performed an exploratory PCR array of 84 key DNA repair genes. From this array, we identified RAD51, which assists in DNA double-strand break repair, as having higher expression in CSCs compared with bulk tumor cells in SUM149 cells (Supplementary Fig. S1B). To confirm the protein expression levels of RAD51 in CSC and non-CSC populations, SUM149 and SUM159 cells were sorted based on aldehyde dehydrogenase activity. Western blot revealed 1.76- and 1.13-fold increases, respectively, in RAD51 protein levels in the CSC populations of SUM149 and SUM159 compared with the non-stem cells (Fig. 2A and B). To assess RAD51 localization, sorted cells were seeded onto chamber slides and irradiated by 4 Gy single dose of X-ray radiation to induce DNA double-strand breaks, and the extent of RAD51 foci formation was determined at 0, 3, 12, and 24 hours after radiotherapy. ALDEFLUOR-positive SUM149 (BRCA1-mutant) cells displayed a significant increase in RAD51 foci staining at 3, 12, and 24 hours after radiotherapy compared with the non-CSC population. In contrast, in SUM159 (BRCA1–wild-type), RAD51 foci formation was similar in both CSC and non-CSC compartments (Fig. 2C and D; additional cell line data in Supplementary Fig. S2A). Consistent with previous studies showing impaired homologous recombinant DNA repair in cells with BRCA1 mutation (18–20), DNA damage induced much higher proportions of cells with RAD51 foci in the BRCA1–wild-type SUM159 cell line compared with SUM149 cells. In addition, long-term PARPi treatment generated cells with an increase in RAD51 expression in TNBCs (Supplementary Fig. S2B). Our data suggest a potential link between RAD51 and olaparib resistance found in the CSC population of BRCA1-mutant cells, thus targeting RAD51 may enhance the efficacy of PARPi.

The association between RAD51 and CSCs led us to further study this molecule. We infected SUM149 and SUM159 cells with a doxycycline-inducible RAD51 shRNA lentiviral system. The efficiency of RAD51 KD was confirmed by real-time PCR of sorted CSC and non-CSC cells to ensure the efficiency of silencing RAD51 in both populations. Western blot and immunofluorescence foci staining were also performed to ensure KD at protein level (Fig. 3A and B; Supplementary Fig. S3). RAD51 KD resulted in a 50% decrease (P < 0.05) in the viable cell count in SUM149 cells, likely due to G₀–G₁ phase cell-cycle arrest, as we demonstrated a 20% increase of SUM149 cells in G₀–G₁ phase compared with those with RAD51 KD (P < 0.01). However, RAD51 KD showed limited effect on SUM159 cell counts. In addition, we demonstrated that RAD51 KD did not affect CSC populations in both cell lines (Supplementary Fig. S4). To test the hypothesis that RAD51 facilitates resistance to PARP inhibition, RAD51 KD SUM149 and SUM159 cells were treated with olaparib. The MTT assay revealed that RAD51 KD increased sensitivity to the PARPi (P < 0.001) at low concentrations (10 nmol/L and 100 nmol/L) in SUM149 after 7 days of treatment (Fig. 3C). More importantly, a decrease in RAD51 sensitized the CSC population in SUM149 cells to the PARPi treatment, and the effect was dose-dependent (Fig. 3E). On the other hand, BRCA1-wild-type SUM159 with functional RAD51 did not respond to PARP inhibition at 10 nmol/L and 100 nmol/L, whereas a 60% decrease in viable cells was induced by PARPi after RAD51 KD (P < 0.001; Fig. 3D). The SUM159 CSC population also decreased with olaparib treatment in the RAD51 KD cells (P < 0.05 and <0.001; Fig. 3F). RAD51 KD also sensitizes HCC1937 (BRCA1-mutant) and MDAMB231 (BRCA1–wild-type) CSCs to PARPi (Supplementary Fig. S5). Two additional pTripZ-RAD51 shRNAs were also used to confirm the specificity of RAD51 KD on PARPi sensitivity (Supplementary Figs. S6 and S7). These results support our hypothesis that RAD51 mediates resistance of CSCs to PARPi in BRCA1-mutant cells as well as BRCA1–wild-type breast cancer cells.

Because the baseline difference in RAD51 expression between CSCs and bulk tumor cells is small, we verified our KD findings at more physiologic levels of RAD51. Using a lower dose of doxycycline to induce the RAD51 KD, we achieved approximately 2-fold downregulation at both the mRNA and protein levels. We demonstrated that downregulating RAD51 to levels similar to those found in bulk cells renders CSCs sensitive to PARPi as indicated by a decrease in absolute CSC numbers (Supplementary Fig. S8).

We next determined the effects of PARP inhibition and RAD51 KD in an in vivo model of TNBC. SUM149 cells infected with inducible RAD51 shRNA vector were injected into the mammary gland fat pads of NOD/SCID mice. The mice were divided into four treatment groups: vehicle control, olaparib, RAD51 KD (doxycycline water), and combined (olaparib plus RAD51 KD) for a period of 8 weeks. Each treatment group was further divided into two groups: one group received treatment immediately after cell implantation and the other group received delayed treatment that was started when the tumors reached an average of 15 mm³ in volume. Although olaparib treatment had substantial effects on cell viability in vitro, it only inhibited tumor growth in the early treatment scenario (P < 0.05 compared with vehicle control; Fig. 4A). In contrast, KD of RAD51 significantly reduced tumor growth compared with control treatment in early (P < 0.0001) and late treatment (P < 0.001) groups. The addition of PARP inhibition to RAD51 KD further reduced tumor size by approximately 60% and 75% in early and late treatment groups, respectively. Although tumor growth resumed when RAD51 expression was restored, the inhibitory effect of combined treatment was sustained even after treatment was terminated (Fig. 4A and B). Tumors from RAD51 KD and combination treatment groups were 60% and 92% smaller (by weight), respectively, when compared with mice with control and were 45% and 87% smaller (by weight), respectively, compared with PARP inhibition alone in early treatment (Fig. 4C). In the late treatment groups, RAD51 KD and combined treatment reduced the tumor weight by 60% and 80%, respectively, compared with mice receiving vehicle control and PARPi alone (Fig. 4D). Our in vivo results are consistent with in vitro data suggesting that knocking down RAD51 inhibits SUM149 tumor progression, and the addition of a PARPi enhances this effect.

In order to assess the effects of PARP inhibition on the CSC frequency, serial dilutions of SUM149 xenografts (control and treated) were re-injected into mouse mammary fat pads. Using this gold standard for validating the effect of compounds on CSC population, we were able to calculate the frequency of CSC from each treatment group and performed statistical comparisons using ELDA software as previously reported (16, 21, 22). Compared with control treatment, RAD51 KD and combined treatments reduced the CSC frequency by 3 and 10 times, respectively, statistical analysis revealed significant decrease in CSC frequency in combined groups compared to control (P < 0.01) and PARPi (P < 0.05) treatments. These results confirm our hypothesis that RAD51 mediates the sensitivity of CSCs to PARPi.

The sensitization to PARPi mediated by RAD51 KD that we observed in BRCA1–wild-type SUM159 cells in vitro led us to test whether we would observe the same effect in an in vivo mouse model. As described above, SUM159 cells were injected into the mammary fat pads of NOD/SCID mice, and treatments were started immediately (early treatment) or after the tumors reached around 15 mm³ (delayed treatment) and continued for 7 weeks. Both RAD51 KD and combination (PARPi and RAD51 KD) treatments slowed tumor progression compared with control (P < 0.05) in early treatment groups (Fig. 5A). This inhibitory effect was also seen when the treatment was given after the tumors were established (Fig. 5B); however, the decrease was not significant.

With regard to CSCs, re-implantation of the SUM159 xenograft cells from the combined treatment in both early and late treatment groups failed to form any tumor. This indicates that the strategy of combined RAD51 inhibition and PARPi is able to eradicate the tumor-initiating cell population. The CSC frequency of combined treatment decreased significantly compared with the control (P < 0.05), PARP inhibition (P < 0.001 and P < 0.01 in early and late treatment groups, respectively), and RAD51 KD (P < 0.01) groups (Fig. 5C and D). Together, these results indicate that our therapeutic regimen targets the BRCA1–wild-type SUM159 cells (both CSC and bulk-tumor populations), supporting our in vitro findings of the involvement of RAD51 in sensitivity of BRCA1–wild-type SUM159 cells to the PARPi olaparib, and importantly, it suggests an effective therapeutic intervention for targeting TNBC.

Despite the promising results of PARPi in treating TNBCs with BRCA deficiency, resistance is still a significant issue. The overwhelming evidence of the importance of CSCs to drug resistance led us to study the effect of PARP inhibition on CSCs. In this study, we used in vitro models and mouse xenografts to demonstrate the importance of RAD51 in PARPi resistance of CSCs in BRCA1-deficient and –wild-type TNBCs. Our study was limited to four cell lines due to the availability of BRCA-mutated breast cancer cells lines (23); however, we have also provided evidence that RAD51 plays a role in mediating resistance to PARPi in basal and claudin-low TNBC.

Sensitivity to PARP inhibition in cells that are deficient in components of the homologous repair pathway suggests a broader clinical potential for PARPi in treating triple-negative breast tumors (2). However, our results showing a relatively modest influence of PARPi on CSCs of both BRCA1-mutant and BRCA1–wild-type TNBC cells identify a potential mechanism for PARPi resistance. Although overexpression of PARP1 has been found to be associated with PARPi-resistant ALDEFLUOR-positive breast cancer cells (24), the molecular mechanisms mediating CSC resistance to PARP inhibition in TNBCs are not known. Here, we have shown that RAD51 is involved in this resistance.

The synthetic lethality of PARP inhibition in TNBCs is based on BRCA1 deficiency; however, the CSCs with higher RAD51 expression exhibited reduced sensitivity toward PARP inhibition. In addition, even among the TNBC cells with BRCA1 mutation, HCC1937 displayed less sensitivity toward the PARPi compared with SUM149, likely due to higher basal RAD51 level. Our results are consistent with previous findings showing that overexpression of RAD51 compensates for BRCA1 deficiency (25), therefore allowing the homologous recombinase pathway to repair DNA damage when PARP is inhibited. In addition, PARPi-resistant BRCA1-mutant TNBC cells, selected by long-term treatment with PARPi, exhibited elevated RAD51 activity compared with parental cells (26), supporting our observation of RAD51 in mediating sensitivity to PARPi. We have also examined the expression of RAD51 upstream regulators: BRCA1 and BRCA2 in CSC and non-CSC subpopulations and detected no significant differences in expression in these cell populations (data not shown). However, we cannot rule out involvement of other molecules upstream of RAD51 in mediating PARPi resistance.

Overexpression of RAD51 has been linked with therapy resistance, and CSCs are known to be more resistant to chemotherapy and radiotherapy compared with non-CSCs. In mouse embryonic stem cells, RAD51 protein is expressed at very high levels compared with differentiated cells (27); and human embryonic stem cells exhibit more efficient DNA damage repair compared with primary fibroblasts (28). In cancers, increased levels of RAD51 were found in invasive pancreatic cancer cells displaying CSC properties, and higher expression was also associated with aggressiveness of primary tumors (13). The CSCs of MDA-MB-231 repaired DNA damage more efficiently than the non-CSC cells after irradiation, as evidenced by a much higher expression of RAD51 foci (29). The factors regulating RAD51 expression in CSCs remain unknown. In addition to BRCA1/2, RAD51 is also linked to ERK1/2, miR-96, and hypoxia (30–32). The MEK/ERK signaling pathway has been found to facilitate tumor growth and angiogenesis of TNBC in mouse models (33); miR-96 enhanced breast cancer cell proliferation and anchorage-independent growth (34), while hypoxia was shown to promote CD44⁺/CD24⁻ expansion (35). Interestingly, miR-96 and hypoxia actually downregulate RAD51 expression, suggesting that CSCs of TNBC cells are RAD51 independent, which is consistent with our data that knocking down RAD51 alone did not reduce CSC frequency. However, RAD51 KD did slow down tumor progression, and this inhibitory effect is reversible when the expression was rescued, suggesting that RAD51 is involved in regulating CSC proliferation.

PARP inhibitions have shown promising results in subsets of TNBC patients that manifest deficiency in DNA damage repair. Our findings of the role of RAD51 in PARPi resistance in CSCs of BRCA1-mutated TNBCs and BRCA1–wild-type TNBCs suggest that resistance to PARPi may be overcome by targeting both CSCs and bulk-tumor cells. Furthermore, by targeting RAD51, it may be possible to greatly expand the sensitivity of TNBCs to PARPi, beyond those with defective BRCA1 proteins. This novel approach holds potential for significantly improved therapies for TNBC.

No potential conflicts of interest were disclosed.

Conception and design: Y. Liu, M.L. Burness, S. Bai, T.K. Luther, M.S. Wicha, S. Liu

Development of methodology: Y. Liu, M.L. Burness, R. Martin-Trevino, S. Bai, T.K. Luther, M.S. Wicha, S. Liu

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Liu, M.L. Burness, R. Martin-Trevino, J. Guy, R. Harouaka, M.D. Brooks, A. Fox, T.K. Luther, A. Davis, T.L. Baker, J. Colacino, M.S. Wicha, S. Liu

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y. Liu, M.L. Burness, J. Guy, R. Harouaka, A. Fox, J. Colacino, S.G. Clouthier, S. Liu

Writing, review, and/or revision of the manuscript: Y. Liu, M.L. Burness, T.K. Luther, S.G. Clouthier, M.S. Wicha, S. Liu

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y. Liu, M.L. Burness, R. Martin-Trevino, R. Harouaka, L. Shang, T.K. Luther, Z.-m. Shao, S. Liu

Study supervision: M.L. Burness, S.G. Clouthier, M.S. Wicha, S. Liu

The authors thank the University of Michigan Flow Core and Vector Core for excellent technical assistance. They also thank Drs. Felix Feng, Ming Luo, and Jill Granger for their comments during article preparation, and Drs. Guan-Tian Lang, Xin Hu, Dafydd Thomas, and Sofia Merajver for assistance in acquiring clinical specimens.

This work was supported by the DOD grant (W81XWH-12-1-0147), CAS stem cell grant XDA01040410, NSFC grants 81472741 and 81322033 (to S. Liu), The Dr. Frank Limpert Clinical Scholar Award and NSFC grant 81530075 for the tumor block cutting (to M.L. Burness), Susan G. Komen for the Cure Promise Award (KG120001), Breast Cancer Research Foundation Award (N015445), Senior Taubman Scholar Award, and Cis Maisel gift to support CSC research (to M.S. Wicha).

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.