HOXC10 Expression Supports the Development of Chemotherapy Resistance by Fine Tuning DNA Repair in Breast Cancer Cells Free

Many chemotherapeutic agents induce DNA damage, triggering cells to activate their DNA damage repair machinery. Nucleotide excision repair (NER) is a major DNA repair pathway that mediates repair of DNA adducts that cause inter- or intrastrand cross links (ICL) and the formation of pyrimidine dimers. Alkylating- and platinum-based agents often induce bulky DNA damage that is repaired via NER.

Large scale interaction screens as well as reciprocal affinity purifications have been instrumental in identifying several HOX proteins that interact with proteins critical for DNA damage repair processes (reviewed in ref. 1). HOXB7 associates with Ku70, Ku80, and DNA-Pkcs (2), and with PARP1 (3), whose polyADP-ribosylation activity enhances the kinase activity of DNA-PK at the initiation of non-homologous end joining (NHEJ). Recent evidence supports HOX protein participation in the cell replication machinery by associations at DNA replication origins (4). HOXC10 is one such example (5, 6); HOXC10 also has important functions in tissue regeneration (7). Recently, loss of HOXC10 expression was implicated in the development of resistance to estrogen response modulators in estrogen receptor (ER)–positive breast cancer (8). At the present time, the function of HOXC10 in breast cancer remains poorly understood.

Here, we report that HOXC10 expression is frequently higher in ER-negative breast carcinomas that are also chemotherapy resistant. We determined that HOXC10 contributes to chemoresistance through suppression of apoptosis and enhanced DNA damage repair, mediated through direct interaction with, and activation of CDK7. We also show that CDK7 inhibition can reverse chemotherapy resistance.

All cell lines were purchased from the ATCC and passages used were within 6 months of purchase. Stable MCF10A-Ras cells were established by transfection of LXSN-K-Ras vector into MCF10A cells; MCF7 wt and drug-resistant sublines were from Dr. A.M. Parissenti, Laurentian University, Sudbury, ON, Canada (9). DR-95 is a human fibroblast cell line stably expressing a pDR-GFP plasmid containing a mutated GFP gene with an 18 bp I-SceI endonuclease cleavage site and in-frame termination codon (10). Retrovirus and lentivirus were produced in HEK 293T cells. Human HOXC10 cDNA (Thermo Scientific) was cloned into the EcoRI and ClaI sites of pLPCX for retroviral production for creating HOXC10-expressing cell lines. Mutated HOXC10 constructs were generated with full-length myc-tagged HOXC10 cloned in pCDNA3.1-Neo at EcoRI and XbaI sites. The following reagents were used: doxorubicin (Sigma), paclitaxel, and gemcitabine (Tocris Bioscience) and carboplatin, BS-181 and SNS-032 (gifted by Selleckchem; ref. 11), TRC lentiviral Human HOXC10 shRNA (set of 4; Thermo Scientific); FlexiTube siRNA for HOXC10 (SI04296621), E2F1 (SI00300083), or CDK7 (SI02664795) (Qiagen). All other reagents were from Sigma.

Fresh frozen primary human tissues were used following approval of the Johns Hopkins Institutional Review Board.

Kaplan–Meier analyses were performed using 4,117 breast cancer patients. http://kmplot.com/analysis (12).

qRT-PCR was conducted using the MaximaSYBR Green/ROX Master Mix (Fermentas) per manufacturer protocol. The ΔΔC_t method was used, with GAPDH expression for normalization (13).

Six-well plates containing a 0.6% agar layer were overlaid with 3 × 10³ cells i n 0.3% agar layer. Colonies were counted after 7 days. Cell invasion assay was performed using The BD BioCoat Matrigel Invasion Chamber assay system (13).

Approved by Johns Hopkins Institutional Animal Care and Use Committee (IACUC), BALB/c nu/nu athymic mice (Sprague–Dawley–Harlan) received subcutaneous injections of 3 million cells/100 μL PBS/Matrigel (1:1), and were treated with doxorubicin (4 mg/kg BW/i.v./weekly) or paclitaxel (10 mg/kg BW/i.p./weekly).

Cells were grown in 12-well plates (2,000 cells/well), fixed with formalin, and stained with crystal violet. To quantitate growth, the dye was solubilized by 10% acetic acid, and absorbance was measured at 560 nmol/L.

Cells at 70% to 80% confluence or after double thymidine synchronization were collected, permeabilized, pelleted, and resuspended in an isotonic-buffered propidium iodide (PI) staining solution containing RNase A (0.1 mg/mL) and PI (20 μg/mL). Samples were run on the BD FACSCalibur system (Becton Dickinson), and data analyzed using WinMDI 2.9 software.

A total of 1 × 10⁵ cells were seeded in 12-well cell culture plates, and transfected with a total of 1.6 μg of plasmids, including reporter, expression, and pCMV-β-galactosidase plasmids using Lipofectamine 2000 (14). NF-κB-luc (Igκ2-IFN-luc wt or mut) was a kind gift from Dr. Joel L. Pomerantz, Johns Hopkins University School of Medicine, Baltimore, MD (15). The TOPFlash plasmid (16) was purchased from Addgene.

Exponentially growing cells were exposed to drugs or ultraviolet light (9). Twenty-four hours later, cells were reseeded (1–2 × 10³/well) in triplicate in 6-well plates. Viable colonies were fixed, stained with crystal violet, and counted 1 to 2 weeks later.

A total of 2.5 × 10³ cells per well were plated in 96-well plates in triplicate and treated with drugs alone or in combination, and MTT assay was performed (17). Values are expressed as percent survival of the vehicle-treated control.

Caspase-3 and -7 activities were measured with the Caspase- Glo Assay Kit (Promega) according to the manufacturer's instructions. Values were expressed as the percentage of activity of the vehicle-treated control.

Cells were lysed with RIPA buffer and processed as described previously (17). The following antibodies were used: anti-CDK7 and anti-XPD (Santa Cruz Biotechnology) and BCL-xL and BIRC3 (Abcam). All other antibodies were from Cell Signaling Technology.

The assay was performed (18) with some modifications. pGL3-basic (Promega) was exposed to UV light [Stratalinker UV-Crosslinker 1800 (Stratagene)] or treated with 100 to 1,000 nmol/L cisplatin. To measure HR-mediated repair, 1 μg of vector digested with HindIII was transfected into cells with pCMV-β-galactosidase. Luciferase and β-galactosidase activities were measured after 48 hours.

The alkaline Comet assay to measure DNA strand breaks in single cells was performed according to the manufacturer's protocol (CometAssay Kit; Trevigen). Comet tail moments of ≥100 cells were measured and quantified using the CometScore software.

Chromosomal aberration analysis was performed as described previously (18, 19). Exponentially growing cells were gamma irradiated (3 G) and analyzed for metaphase aberrations after 12 hours (20). Cisplatin-induced chromosome aberrations were analyzed as described previously (1 9).

Cells growing in chamber slides (Nunc Lab-Tek II) were treated with Dox (200 nmol/L) or gemcitabine (50 nmol/L) for 24 hours. After fixation and permeabilization, cells were probed with an antibody against phosphorylated H2AX-Ser¹³⁹ (Upstate Biotechnology); H2AX foci were visualized using a Zeiss Axio Scope fluorescent microscope and scored with the ImageJ software (v1.47, NIH, Bethesda, MD). One-hundred cells or more were evaluated.

DR-95 cells were transfected with pI-Sce1, pEGFP, or pCMV (21, 22), harvested after 72 hours, and %GFP-expressing cells was measured by flow cytometry. Frequency of recombination events = mean %GFP-positive cells transfected with pI-Sce1/Mean %GFP-positive cells transfected with pEGFP.

DR95 cells were electroporated with I-pCBASce. ChIP was done using antibodies to Rad51 (Abcam); BRCA1 (Cell Signaling Technology); KU80 (Cell Signaling Technology; ref. 20). Quantification of ChIP DNA was carried out by real-time PCR in triplicate using the LightCycler Fast Start DNA Master SYBR Green I (Roche Applied Sciences).

DNA fiber spreads were prepared (19) with minor modifications. Cells in exponential phase were labeled for sites of ongoing replication with 5-iododeoxyuridine (IdUrd; 50 μmol/L) followed by exposure to hydroxyurea (4 mmol/L), washed, and labeled with 5-chlorodeoxyuridine (CldUrd; 50 μmol/L). Fibers were analyzed using ImageJ software.

One milligram of protein lysate was subjected to immunoprecipitation overnight at 4°C with CDK7 antibody (C-19; Santa Cruz Biotechnology) or normal rabbit IgG control (sc-2027; Santa Cruz Biotechnology); immunoblotting was performed using: Myc-Tag (9B11; Cell Signaling Technology), RNA Polymerase II (CTD4H8; Millipore), and XPD [sc-101174, Santa Cruz Biotechnology; ref. 14).

Three hundred micrograms of protein extract from drug-treated cells was used for immunoprecipitation with CDK7 antibody. The immunoprecipitate was resuspended in 40 μL of kinase buffer (50 mmol/L HEPES, pH 7.5, 10 mmol/L MgCl₂, 250 μmol/L EGTA, 10 mmol/L β- glycerophosphate, 1 mmol/L DTT), 10 μL of 50 nmol/L ATP, and 20 ng GST-CDT peptide (P4016, Proteinone) as substrate, incubated for 1 hour at 30°C, added to an equal volume of kinase-GLO reagent (Promega), and incubated for 15 minutes at room temperature. Control reactions lacked the CTD peptide substrate. Luminescence was recorded and expressed as relative RLU to the untreated control cells.

Results were expressed as mean ± SEM of at least three independent experiments. Paired Student t test or ANOVA tests were performed for data analysis. All statistical analyses were performed using GraphPad Prism version 4.03 (GraphPad Software, Inc.). In all figures, *, P < 0.05; **, P < 0.01; and ***, P < 0.001.

Details of constructs, materials, and experimental procedures are provided in Supplementary Materials and Methods online.

By HOX tiling mRNA expression array, compared with normal breast, HOXC10 was among the top five overexpressed genes in breast cancer (23), and higher than all 39 HOX genes in distant metastasis (Fig. 1A). In Oncomine (24), HOXC10 was among the top 1% transcripts in breast compared with other carcinomas (Fig. 1B), and in TCGA (https://tcga-data.nci.nih.gov/), was equally prevalent in all breast carcinomas (Fig. 1C). qRT-PCR analysis of our primary tissue panel validated this finding; median HOXC10 expression was 10-fold higher in primary invasive carcinomas and 30- fold higher in distant metastatic tissues (Fig. 1D). No correlation of HOXC10 expression with relapse-free survival (RFS) or overall survival (OS) was observed in breast cancer patients in the GEO (http://www.ncbi.nlm.nih.gov/geo) and the METABRIC cohorts (Fig. 1E, a and F, a; ref. 25). In chemotherapy treated patients, high HOXC10 expression correlated with short RFS (Fig. 1E, b and Supplementary Fig. S1A and S1B) and short OS (Fig. 1F, b and Supplementary Fig. S1C), and even more consistently in the subset of ER-negative patients treated with chemotherapy (Fig. 1E, c and F, c). Cox multivariate regression analysis, taking all clinical parameters into account, revealed a highly significant (P = 0.00013) inverse correlation between HOXC10 expression and RFS.

As a first step to investigating the contribution of HOXC10 to drug resistance, we analyzed HOXC10 expression in chemotherapy-resistant MCF7-sublines (9) that displayed resistance (>250-fold) to epirubicin, paclitaxel and docetaxel. HOXC10 was expressed at 2- to 8-fold higher levels in MCF-7- R sublines (Fig. 2A) and in newly developed MDA-MB-231-R sublines that emerged (>30 days) following exposure to drugs (Supplementary Fig. S2A). Depleting HOXC10 levels in MCF-7-Tax-R subline restored its response to paclitaxel (Fig. 2B). Further studies were carried out in eight breast cancer cell lines with varying levels of HOXC10 mRNA (Supplementary Fig. S2B). On exposure of cells in culture to drugs, induction of HOXC10 expression was rapid, starting at day 1 in MDA-MB-231 (Fig. 2C), SUM159 (Fig. 2D), and SUM149 (Supplementary Fig. S2C). Basal levels of HOXC10 expression determined response to Dox. Mouse xenografts of low HOXC10-expressing MDA-MB-231 (Fig. 2E) responded to Dox significantly better than high HOXC10-expressing SUM159 (Fig. 2F) or HCC1954 (Supplementary Fig. S2D). By qRT-PCR, MDA-MB-231 (Fig. 2G) and SUM149 (Supplementary Fig. S2E) tumors that resumed growth during treatment showed significantly higher HOXC10 expression compared with SUM159 (Fig. 2H) and HCC1954 (Supplementary Fig. S2F) supporting the argument that upregulated HOXC10 expression correlated with resistance to chemotherapy.

To characterize the effects of HOXC10 overexpression, we derived stable clones of MCF10A-Ras [stably expressing K-ras (Gly¹²-Val)] and MDA-MB-231–overexpressing myc-tagged HOXC10, and SUM159 and SUM149 cells depleted of HOXC10 by 50% to 70% (immunoblot in Fig. 3A) using specific shRNAs. Properties of anchorage-independent growth (Supplementary Fig. S3A and S3B), colony formation (Supplementary Fig. S3C), invasion through Matrigel (Supplementary Fig. S3D), proliferation (Supplementary Fig. SE), quantified in (Supplementary Fig. S3F), and tumor growth in mice (Fig. 3B and Supplementary Fig. S3G) were all significantly decreased by depleting HOXC10 and enhanced by overexpressing HOXC10 in tumor cells. By colony survival assays, Dox, gemcitabine, carboplatin, and Tax were more cytotoxic in SUM159-shC10 (Fig. 3C and Supplementary Fig. S3H), MCF7-shC10 and SUM149-shC10 (Supplementary Fig. S3H) than in MCF10A-Ras-C10 cells (Supplementary Fig. S3I). This body of data, supporting an increased oncogenicity and aggressiveness of HOXC10-expressing breast cancer cells, was further corroborated by MTT assays in SUM159-shC10 (Supplementary Fig. S3J) and in MCF10A-Ras-10 cells (Supplementary Fig. S3K).

The activation of caspase-3/7 is a reliable marker of cells undergoing apoptosis (26). Susceptibility to the drugs was reflected by an increase in caspase 3/7 activity in SUM159-shC10 cells (Fig. 3D) and a decrease in MCF10A-Ras-C10 cells (Fig. 3E). Apoptotic cells were identified on the basis of DNA content frequency histograms as cells with fractional "sub-G₁" DNA content (27). In response to Dox, there was a decrease in sub-G₁ population in SUM159shC10 cells (Fig. 3F) and an increase in MCF10A-Ras-C10 cells (Fig. 3G). Also, several pro- and antiapoptotic mRNAs were modulated in SUM159-shC10 and MCF10A-Ras (Supplementary Fig. S3L and S3M), and further intensified by drug treatment (Fig. 3H and Supplementary Fig. S3N). Western blot analysis of SUM159-shC10 cells showed that antiapoptotic genes, hsp27, Birc 3, PI3K, and BCL2 levels were decreased, whereas proapoptotic genes, p21, and BID levels were increased by depletion of HOXC10 (Fig. 3I). Dox-treated MCF10A-Ras-vec xenografts regressed by week 4, whereas MCF10A-Ras-C10 tumors grew after week 2 (Fig. 3J). Consistent with this observation, MCF10A-Ras-C10 tumors showed high mRNA expression of many antiapoptotic genes (Fig. 3K).

Given that the antiapoptotic genes examined are known direct targets of NF-κB (28), we measured NF-κB activity using the NF-κB–responsive reporter Igκ2-IFN-LUC, with and without Dox treatment. The basal activity of NF-κB was lower in SUM159-shC10 (Supplementary Fig. S3O) and high in MCF10A-Ras-C10 (Supplementary Fig. S3P) compared with their respective controls, and further enhanced by Dox treatment. These data are consistent with findings that the basal activity of NF-κB is much higher, and maintained as such, in Dox-resistant MCF7 cells (29). Thus, HOXC10 upregulates NF-κB to achieve both activation of antiapoptotic pathways, and increases cell survival following stress.

Several homeodomain-related proteins have been functionally related to DNA damage repair (1, 2, 30, 31). Overexpression of HOXA5 in S. cerevisiae upregulates components of the mismatch repair (MMR) system, important for the detection and repair of DNA damage (32). Hence, we investigated whether HOXC10 participates directly in DNA repair and if so, the type of damage and cell-cycle stage of repair.

A host–cell reactivation assay was used to assess cellular ability to repair exogenously damaged DNA. A pGL3 (luciferase-reporter) plasmid was exposed to UV or cisplatin, or digested with Hind III (to generate double-strand breaks, DSB) followed by transfection into SUM159-shC10 cells or MCF-10A-Ras-C10 cells and their respective controls. SUM159-shC10 cells displayed a significantly decreased ability to repair DNA damage (Fig. 4A), whereas the reverse was shown for UV-induced DNA damage in MCF-10A-Ras-C10 cells (Fig. 4B). Clonogenic cell survival assays in both MCF10A-Ras-C10 and SUM159-shC10 cells exposed to UV (Fig. 4C, quantified in S4A) supported these findings. However, repair of DNA DSBs caused by Hind III digestion (Fig. 4A) remained unchanged in SUM159-shC10 cells, suggesting no involvement of the NHEJ pathway. Furthermore, ionizing radiation (IR) induced G1-chromosome damage repair was not affected in SUM159-shC10 cells (Supplementary Fig. S4B), wherein DSB repair by NHEJ is predominant. On the other hand, SUM159-shC10 cells had a higher frequency of S-phase–specific IR-induced chromosomal aberrations (Supplementary Fig. S4C), suggesting that HOXC10 expression enhances repair, possibly by the homologous recombination (HR) pathway. MCF10A-Ras-C10 cells treated with cisplatin showed significantly reduced number of cells with chromosomal aberrations at metaphase (Supplementary Fig. S4D) compared with the vector control cells. Collectively, the data strongly support the hypothesis that HOXC10 is an important participant in DNA DSB repair by HR.

We had noted that HOXC10 expression had a dramatic effect on survival following exposure to UV light or the platinum drugs (Fig. 4A–C and Supplementary S4A). DNA damage induced by such agents is repaired by nucleotide excision repair (NER) and HR. Therefore, we measured the activity of the NER pathway by the alkaline comet assay (Fig. 4D), a single-cell gel electrophoresis method in which the intensity of the comet tail of the migrating cells relative to the head reflects the number of DNA breaks (33). Cells depleted of HOXC10 had significantly more residual UV-, Dox- and gemcitabine-induced DNA damage 24 hours after treatment (Fig. 4E). Furthermore, time-course analysis after UV treatment showed that DNA damage in SUM159-scr cells had returned to baseline levels in 24 hours, but was still 4-fold higher in SUM159-shC10 cells (Fig. 4E), revealing inefficient DNA repair in the absence of HOXC10. Improved response to DNA damage was evident in MCF10A-Ras-C10 cells compared with the vector-control cells (Fig. 4B, C, and F). The reduced survival of SUM159-shC10 cells in colony formation assays (Fig. 4C) also correlated with higher levels of residual DSBs.

DNA damage that results in DSBs is always followed by the phosphorylation of the histone, H2AX (34). Newly formed phosphorylated protein γ-H2AX is the initial step of damage detection to recruit repair proteins (34). We quantified the number of residual γ-H2AX foci 24 hours after treatment of SUM159-shC10 cells with Dox or gemcitabine. On average, 20% to 30% more γ-H2AX foci accumulated in SUM159-shC10 cells after drug treatment (Fig. 4G) with almost 50% to 60% more cells displaying ≥100 foci/nucleus (Supplementary Fig. S4E). A time-course analysis of γ-H2AX protein after UV or Dox treatment showed no differences in the initial induction of γ-H2AX (10 minutes to 8 hours; Supplementary Fig. S4F and S4G). Residual γ-H2AX measured 24 hours after treatment, and not the initial kinetics of γ-H2AX formation, is a better predictor of cell viability (35, 36). These results indicated that HOXC10 might be more involved in DNA damage processing rather than in damage sensing. No major difference was observed in the phosphorylation status of known DNA damage sensors of the ATR/ATM pathways after Dox or gemictabine treatment of SUM159-shC10 (Supplementary Fig. S4H), thus confirming that HOXC10 is not involved in DNA damage sensing.

These findings were further substantiated by the HR-DSB repair assay of lesions induced by I-Sce1 endonuclease performed in DR-95 cells engineered to stably express a pDR-GFP plasmid containing a mutated GFP gene with an 18 bp I-Sce1 endonuclease cleavage site and an in-frame termination codon (10, 20). Efficient repair of the Sce-1 cleavage site restores expression of GFP. DR-95 cells, with and without HOXC10 silencing (using siRNA) were transfected with the I-Sce1 expression plasmid. DR95-siHOXC10 showed reduced repair of I-Sce-1–induced lesions. Consequently, restoration of GFP expression occurred at a significantly lower efficiency than in control cells (Fig. 4H). Reduction of GFP expression was comparable with cells with reduced expression of BRCA-1, a known DNA repair protein, used as a positive control (Fig. 4H). Thus, repair of DSB by HR is impaired in cells lacking HOXC10 expression.

To investigate a role of HOXC10 in the recruitment of HR repair proteins at the site of a single DNA DSB, we performed chromatin immunoprecipitation (ChIP) in DR-95 cells (10, 20). We compared the levels of BRCA1, RAD51, KU80, and HOXC10 at different distances from an I-Sce1–induced DSB site using ChIP analysis with specific primers (10) in cells with and without expression of HOXC10. Depletion of HOXC10 using siRNAs did not decrease KU80 levels in close proximity to the DSB, whereas levels of RAD51 and BRCA1 at the break site were significantly reduced (Fig. 4I). This finding in HOXC10-depleted cells strongly supported the role of HOXC10 in facilitating the recruitment of proteins involved in HR. However, HOXC10 itself was undetectable at the I-Sce1 cleavage site, ruling out a direct role for HOXC10 in DSB repair.

We therefore considered the possibility that HOXC10 has a role in the initiation of new origins of DNA replication. Single-molecule examination of replication dynamics by DNA fiber analysis (19) after hydroxyurea (HU) treatment in SUM159-shC10 and MCF10A-Ras-C10 cells showed that HOXC10 is required for the initiation of new origins, but not for the accumulation of stalled forks (Fig. 4J and K).

Collectively, the data suggest that HOXC10, as a transcriptional factor, affects the chromatin status and sets the stage for HR by, (i) decreasing S-phase chromosome aberrations, (ii) enhancing resection as shown by γ-H2AX foci, (iii) increasing HR based on l-Scel endonuclease–induced DSB repair, and (iv) resolving stalled replication forks as shown by DNA fiber analysis, possibly leading to initiation of DNA replication following DNA damage.

We had thus far observed that HOXC10 enhances cell survival by affecting apoptosis, NF-kB activity, and damage repair, mainly by the HR pathways. One key protein that links all three pathways is cyclin-dependent kinase 7 (CDK7). In metazoans, CDK7 has essential roles in transcription as a component of the general transcription factor, TFIIH. CDK7 is an effector kinase, which also phosphorylates RNA Pol II and other proteins during transcription after the pausing of the TFIIH complex following DNA damage (37). Highly relevant to this study, HOXC10 was found to be the tightest binding protein linking the CAK (Cdk-activating kinase, CDK7) complex to TFIIH in a yeast 4-hybrid system (38). We therefore tested for an association of HOXC10 with CDK7 by coimmunoprecipitation. A weak interaction between the two proteins increased significantly upon treatment with Dox and gemcitabine (Fig. 5A). This interaction was reduced significantly upon addition of a pharmacologic inhibitor of CDK7 activity, SNS-032 (Fig. 5A), or by expressing the dominant negative, kinase-dead mutant CDK7 (D155A) in the cells (Fig. 5B). We concluded that CDK7 and HOXC10 formed a complex in breast cancer cells, which became more abundant upon exposure to drugs.

Different sets of proteins are phosphorylated by CDK7 depending on its function in cell division (i.e., CDK1, 2, 4, 6) or in transcription as a complex with TFIIH (37). We therefore investigated whether binding of HOXC10 to CDK7 altered its substrate preference. Also, elongating RNA Pol II can be arrested by endogenous and exogenous DNA lesions such as UV-induced pyrimidine dimers, adducts induced by anticancer drugs, and DSBs. These transcription-blocking lesions located on the transcribed strand are primarily repaired by transcription-coupled repair, the NER pathway (39). XPD helicase is a key member of the human TFIIH complex that anchors CAK kinase (cyclin H, MAT1, and CDK7) to TFIIH and unwinds DNA for transcription and for repair of duplex warping damage by nucleotide excision repair (NER), thereby maintaining genomic integrity (40). We therefore examined whether HOXC10 affects the composition of the TFIIH–XPD–CAK complex through its protein/protein interaction with CDK7. The binding of XPD to CDK7 was reduced in SUM159-shC10 cells (Fig. 5C) and was increased in MCF10A-Ras- C10 cells, compared with their respective control cells (Fig. 5D). This finding suggested that HOXC10 may have a role in anchoring CAK to XPD, thus maintaining the integrity of the holoenzyme TFIIH during response to DNA damage. This finding is also consistent with reports that in repair-deficient cells, the association of CAK kinase, but not of XPD, to damaged DNA was reduced (41).

Next, we investigated whether the association of CDK7 with the RNA Pol II is modulated by HOXC10. We found that the association between CDK7 and RNA Pol II in drug-treated cells was significantly stronger in the presence of HOXC10 in MCF10A-Ras-C10 (Fig. 5E) and weaker in the absence of HOXC10 in SUM150-shC10 (Fig. 5F). Furthermore, the kinase activity of CDK7 on the C-terminal domain (CTD) of RNA Pol II was significantly reduced in SUM159-shC10 cells after DNA damage (Fig. 5G). MCL1 is an antiapoptotic protein that is rapidly depleted upon inhibition of RNA Pol II, permitting measures of its mRNA to serve as a surrogate for CDK7 activity (42). The lack of CDK7 activity is likely to impede recovery of the transcriptional machinery, thereby promoting cell death after DNA damage as was reflected by MCL1 expression (Fig. 5H). Conversely, the enhanced interaction between Pol II and CDK7 in MCF10A-Ras-C10 cells (Fig. 5E) promotes cell survival following DNA damage (Fig. 5I). Consistent with this data, newly emerging drug-resistant MCF7 sublines in cell culture showed increases in both MCL1 and HOXC10 mRNA expression (Supplementary Fig. S5).

Inhibiting CDK7 is of special interest in cancer therapeutics as it affects multiple signaling pathways (43). Two known selective CDK7 inhibitors are BS-181 (44) and SNS-032 (11); the latter is in phase I studies (http://clinicaltrials.gov/). Combining SNS-032 or BS-181 at minimally cytotoxic doses (∼20%) with chemotherapy significantly improved the response of both MCF10-Kras-C10 cells (Fig. 6A and B) and SUM159-shC10 cells (Supplementary Fig. S6A and S6B). In taxol and epirubicin-resistant MCF7 sublines, Western blot analysis showed that CDK7 (pThr170) and CTD (pSer5) were activated (Fig. 6C) with no change in cell-cycle kinases, CDK1 and CDK2. qRT-PCR analysis showed an increase in both HOXC10 and MCL1 mRNAs (Fig. 6D). Treatment with taxol combined with BS-181 restored drug susceptibility of Tax-R-MCF-7 cells, as seen in colony formation assays (Fig. 6E), and caused cell kill in parental MCF-7 cells as shown by MTT assays (Fig. 6F). Similarly, BS-181 restored sensitivity of two MDA-MB-231-Tax-R sublines to taxol (Supplementary Fig. S6C and S6D), an effect not achieved using inhibitors to CDK1/2/9, AZD5438 (45) and CDK4/6, PD0332991 (Supplementary Fig. S6E and S6F; ref. 46). These data suggest a specific role for CDK7 in resistance to chemotherapeutic drugs.

In summary, our data support a model of HOXC10 action wherein it is intimately involved in multiple steps in the process of HR-mediated DNA repair following chemotherapy exposure of breast cancer cells and in the development of drug resistance in the long term. HOXC10 is an integral component of the CAK complex, which allows restart of transcription following DNA damage (Fig. 7).

Chemotherapy failure in breast cancer claims at least 80,000 lives each year worldwide. Here, we present evidence that HOXC10 overexpression is common and functionally important for onset of resistance to chemotherapy in ER-negative breast cancer and is also prognostic of poor outcome primarily in patients treated with chemotherapy.

Upregulation of HOXC10 is likely to be an important cancer adaptation mechanism. Ectopic expression of HOXC10 leads to decreased drug susceptibility, while decreasing its levels enhances the cytotoxic effects of chemotherapy in both in vitro and in vivo models of breast cancer. Mechanistically, HOXC10 is involved in activating different survival pathways, including driving checkpoint recovery after DNA damage, a key pathway that allows cancer cells to overcome damage response arrest (Fig. 4 and Supplementary Fig. S4; ref. 47). HOXC10 does not affect the sensing of DDR but is required at later stages of the DNA damage repair by NER and HR (Fig. 4 and Supplementary Fig. S4). Cells expressing high levels of HOXC10 repaired DNA damage more efficiently and resumed transcription and growth; whereas their low-HOXC10–expressing counterparts eventually committed to apoptosis. Part of this response, we propose, may be attributed to the observed association of HOXC10 with the CAK complex.

The CAK complex along with TFIIH can participate in diverse functions, including transcription, DNA repair (NER), and cell-cycle regulation. We confirmed that HOXC10 binding to CDK7 enhances its kinase activity toward the CTD domain of RNAPII after DNA damage. We surmise that HOXC10 plays a role in bridging the gap between these 2 complexes, enhancing the recovery process after DNA damage. Indeed, treatment of drug-resistant MCF-7 cells with CDK7 inhibitors restored their susceptibility to chemotherapy (Fig. 6). The importance of current findings to breast cancer therapy stems from the recent attention given to CDK inhibition, including CDK7, in clinical trials either as single agents or in combination with chemo- or targeted therapies to overcome resistance (43, 48).

The second survival pathway that is activated by HOXC10 is shifting the balance between growth and apoptosis to allow continuous proliferation and survival under adverse conditions. Our data showed that HOXC10 facilitates the transition from G₁ to S phase and progression through the S-phase during the cell cycle (Fig. 3). As a consequence, cells with high HOXC10 levels have a growth advantage, especially under nonideal conditions, and restart their replication to overcome stressful conditions. At the same time, the increase in NF-κβ activity and the consequent increase in the levels of many antiapoptotic proteins likely decrease cell sensitivity to many stressors.

Because of HOXC10 involvement in survival and proliferation of cancer cells despite exposure to chemotherapy, HOXC10 is an attractive target to reverse chemotherapeutic resistance in breast cancer. Future studies could focus on developing direct inhibitors of HOXC10 or indirect modulators of its function by targeting its downstream effectors, such as CDK7.

H.Y. Chang has ownership interest (including patents) in Epinomics and is a consultant/advisory board member for RaNA Therapeutics. S. Sukumar reports receiving a commercial research grant from CEPHEID and is consultant/advisory board member for AVON Foundation. No potential conflicts of interest were disclosed by the other authors.

Conception and design: H. Sadik, R. Kumar, S. Park, N. Shah, H.Y. Chang, T.K. Pandita, S. Sukumar

Development of methodology: H. Sadik, R. Kumar, M. Hedayati, S. Park, T.G. Munoz, N. Shah, R.K. Pandita, J.C. Chang, H.Y. Chang, T.K. Pandita, S. Sukumar

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): H. Sadik, P. Korangath, N.K. Nguyen, M. Hedayati, W.W. Teo, N. Shah, J.C. Chang, T. DeWeese, H.Y. Chang, S. Sukumar

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): H. Sadik, B. Gyorffy, R. Kumar, M. Hedayati, H. Panday, T.G. Munoz, N. Shah, J.C. Chang, S. Sukumar

Writing, review, and/or revision of the manuscript: H. Sadik, B. Gyorffy, M. Hedayati, N. Shah, T.K. Pandita, S. Sukumar

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): H. Panday, O. Menyhart, R.K. Pandita, T.K. Pandita, S. Sukumar

Study supervision: T.K. Pandita, S. Sukumar

We thank Dr. Joel L. Pomerantz for providing us their responsive reporter plasmids NF-κB-luc, Dr. Amadeo M Parissenti for the drug-resistant MCF7 cell lines, and Dr. Alan Rein for reviewing the article.

This work was supported by funding from the DOD-BCRP BC093970 (H. Sadik), NCI-P50-CA88843 and P30-CA006973 (S. Sukumar) and NIH grants CA129537, CA154320, and GM109768 (T.K. Pandita).

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.