Background
Breast cancer is the second leading cause of cancer death for women. Most patients present with early disease and are treated with surgery, often followed by adjuvant radiotherapy and chemotherapy with or without endocrine therapy or trastuzumab given with curative intent. Nevertheless, 40–50 % of high-risk patients treated with adjuvant chemotherapy ultimately relapse as a result of having resistant disease [
1]. Despite the advent of targeted therapies, chemotherapy is also central to the treatment of women with metastatic disease, who often respond to palliative chemotherapy but in due course relapse due to drug resistance, including cross-resistance to structurally unrelated anti-cancer drugs [
2].
The taxanes and anthracyclines are widely used as adjuvant therapy as well as in metastatic cancer. Both target rapidly proliferating cancer cells. The taxanes interfere with microtubule depolymerisation, causing cell-cycle arrest [
3,
4], whereas anthracyclines introduce DNA breaks, form free radicals and covalently bind type II topoisomerase (Topo II)–DNA complexes [
5]. The taxanes and anthracyclines are both natural products and susceptible to resistance mediated by over-expression of the multidrug transporter P-glycoprotein. A well-established in vitro mechanism of resistance involves activity of multidrug resistance genes 1 and 2/3 (
MDR1 and
MDR2/3, respectively), which bind non-specifically to multiple drugs and actively export them across the cellular membrane [
6,
7]. Although this results in decreased intra-cellular drug concentrations and cytotoxicity, the clinical relevance of MDR genes remains to be determined. Other mechanisms include reduced Topo activity [
8,
9], reduced Fas ligand expression [
10] and downregulation of
TP53 expression [
11]. However, the molecular drivers of clinical anthracycline resistance remain largely unknown. We previously identified duplication of centromeric region on chromosome 17 (CEP17), a surrogate marker of chromosomal instability, as a predictive marker of clinical anthracycline sensitivity [
12‐
14]. However, identifying pathways that could be targeted in the clinic to eliminate anthracycline-resistant breast cancer remains a major challenge.
The aim of this study was to establish anthracycline-resistant breast cancer cell lines to (1) identify pathways driving resistance that are common to all breast cancers, regardless of their oestrogen receptor (ER) and human epidermal growth factor receptor 2 (HER2) status; (2) discover a predictive biomarker of anthracycline benefit; and (3) investigate alternative treatment options for patient groups that are not expected to respond to anthracycline regimens. Cell lines were chosen to reflect four major breast cancer subtypes [
15,
16]: MCF7 (ER+/HER2−, luminal A), ZR-75-1 (ER+/HER2+, luminal B), SKBR3 (ER−/HER2+, HER2-amplified) and MDA-MB-231 (ER−/progesterone receptor–negative [PR−]/HER2−, triple-negative), and they were exposed to increasing concentrations of epirubicin until resistant cells were generated. To identify mechanisms driving epirubicin resistance, we used complementary approaches, including gene expression analyses to identify signalling pathways involved in resistance and small-molecule inhibitors to reverse resistance. We demonstrated that a histone H2A- and H2B-containing module was associated with epirubicin resistance and that small-molecule inhibitors targeting histone pathways induced cytotoxicity in all epirubicin-resistant cell lines. Most importantly, the identified mechanism of resistance was recapitulated in the BR9601 clinical trial, where the patients with low expression of the histone module benefited from anthracycline treatment compared with patients with high expression of the same module (hazard ratio [HR] 0.35, 95 % confidence interval [CI] 0.13–0.96,
p = 0.042). Thus, in our study, we identified that chromatin remodelling represents an important mechanism of anthracycline resistance in breast cancer and established a reliable in vitro model system for investigating anthracycline resistance in all four breast cancer subtypes. As the histone modification can be targeted with small-molecule inhibitors, it presents a possible means of reversing clinical anthracycline resistance.
Methods
BR9601 trial
The BR9601 trial (ClinicalTrials.gov identifier NCT0003012) investigators recruited 374 pre- and post-menopausal women with completely excised, histologically confirmed breast tumours and a clear indication for adjuvant chemotherapy. Patients were randomised between 8 cycles of CMF (intravenous cyclophosphamide 750 mg/m
2, methotrexate 50 mg/m
2 and 5-fluorouracil 600 mg/m
2) every 21 days, and E-CMF (4 cycles of epirubicin 100 mg/m
2 every 21 days followed by 4 cycles of the same CMF regimen) [
17] (Additional file
1: Figure S1). The protocol was approved by central and local ethics committees, and each patient provided written informed consent before randomisation. For the present analysis, tissue blocks were retrieved and RNA was extracted. The primary outcomes of the BR9601 study were relapse-free survival and OS, although distant relapse-free survival (DRFS) was also reported [
17].
Cell culture
Breast cancer cell lines (MDA-MB-231, MCF7, ZR-75-1, SKBR3) were purchased from the American Type Culture Collection (Manassas, VA, USA) and cultured in Dulbecco’s modified Eagle’s medium (except SKBR3, cultured in RPMI 1640 medium) supplemented with 10 % heat-inactivated foetal bovine serum and Gibco 1 % l-glutamine (Thermo Scientific, Burlington, ON, Canada). Epirubicin-resistant cell lines were generated by exposing native cells to increasing concentrations of epirubicin with an initial concentration set at 0.5 nM. Resistance was defined when the half-maximal inhibitory concentration (IC50) value superseded the IC50 value of the corresponding native cell line and resistant cells could not tolerate further increase in drug concentration. Drug resistance and cross-resistance were determined by exposing cells to drug concentrations ranging from 0.3 to 3000 nM for 72 h. Cell viability was determined using the Cell Counting Kit-8 (CCK-8; Dojindo Molecular Technologies/Cedarlane Laboratories, Burlington, ON, Canada). IC50 values were calculated using Prism 5 software (GraphPad Software, La Jolla, CA, USA).
Flow cytometry
For cell-cycle analysis, cells were synchronised by the double-thymidine block [
18] and incubated with dimethyl sulphoxide (DMSO) or epirubicin doses established for each cell line: 25 nM for MDA-MB-231, 30 nM for MCF7, 15 nM for SKBR3 and 10 nM for ZR-75-1. Cells were collected at 48 h, fixed with 80 % ethanol and incubated with 2 mg/ml RNase A and 0.1 mg/ml propidium iodide (Sigma-Aldrich, Oakville, ON, Canada) before analysis. For apoptosis experiments, cells were treated with DMSO or epirubicin at the concentrations described above and collected at 72 h for staining with annexin V apoptosis detection eFluor 450 (eBioscience, San Diego, CA, USA). Data were collected using a FACSCanto II flow cytometer and FACSDiva software (BD Biosciences, Mississauga, ON, Canada) and analysed using FlowJo software (Treestar, Ashland, OR, USA).
Cell proliferation
Cells were cultured in the presence or absence of epirubicin for up to 96 h (see
Flow cytometry section above for epirubicin concentrations). Cells were collected at 24, 48, 72 and 96 h and counted using a Vi-CELL Cell Viability Analyzer (Beckman Coulter, Mississauga, ON, Canada). Data were analysed using GraphPad Prism 5 software.
Microarray
Illumina HumanHT-12 v4 BeadChips (Illumina, San Diego, CA, USA) were used for the whole genome microarray analysis by the UHN Microarray Centre, Toronto, ON, Canada. Total RNA was extracted with the RNeasy Mini Kit (Qiagen, Toronto, ON, Canada) and used for profiling gene expression changes. Raw data (Gene Expression Omnibus accession number [GEO:GSE54326]) were normalised with the R3.0.0 lumi package using simple scaling normalisation; the 10 % most variable probes were retained for differential analysis using the genefilter package. Differentially expressed probes were identified using limma with a Benjamini–Hochberg corrected p value cut-off of 0.05.
Network-based analysis
To identify functionally relevant modules, genes demonstrating consistent directionality of significant expression changes were analysed using the Cytoscape Reactome Functional Interaction (FI) plugin in Cytoscape 2.8.3. Symbols were loaded as a gene set and interactions from the FI network 2012 version, including FI annotations and linker genes. Network modules were identified using spectral clustering and pathway enrichment computed for each module using the Reactome FI plugin functions. Reactome pathways exhibiting false discovery rate (FDR) values less than 0.01 were considered enriched.
Pharmaceutical inhibitors
All inhibitors were provided by the drug discovery group at the Ontario Institute for Cancer Research (Toronto, ON, Canada). Cells were seeded at 1000–1500 cells/well into 384-well plates (Greiner Bio-One, Mississauga, ON, Canada). After 24 h, resistant cells were exposed to epirubicin at the selection doses established (see
Flow cytometry section above), then exposed to histone deacetylase (HDAC) inhibitors (HDACi) dissolved in DMSO in 12 concentrations ranging from 0.0026 to 10 μM using HP D300 digital compound dispenser (Tecan Systems, San Jose, CA, USA). The DMSO concentration did not exceed 0.5 % in the final drug solution. After 72 h, the effects of inhibitors were determined using CellTiter-Glo Luminescent Cell Viability Assay (Promega, Madison, WI, USA) and the Wallac EnVision 2104 Multilabel Reader (PerkinElmer, Woodbridge, ON, Canada). Raw data were normalised to negative (media) and positive (20 μM staurosporine) controls and analysed using GraphPad Prism 5.
Quantitative RT-PCR
RNA was isolated from cultured cell lines using the RNeasy Mini Kit (Qiagen, Toronto, ON, Canada). A total of 20 ng of RNA was analysed using TaqMan gene expression assays (HIST1H2BD, Hs00371070_m1; HIST1H2BK, Hs00955067_g1; HIST1H2AC, Hs00185909_m1) and EXPRESS One-Step Superscript qRT-PCR universal kit according to manufacturer’s protocol (Life Technologies, Burlington, ON, Canada). Reactions were run using Applied Biosystems ViiA 7 Real-Time PCR instrument and software (Life Technologies). Transcript levels were quantified from the standard curve generated from the control Universal Human Reference RNA samples (Agilent, Mississauga, ON, Canada). Statistical significance was determined using an unpaired t test.
Immunoblotting
Whole-cell lysates (WCL) were prepared in radioimmunoprecipitation assay (RIPA) buffer supplemented with cOmplete Mini Protease and PhosSTOP phosphatase inhibitors (Roche, Laval, QC, Canada). For cell line characterisation, 10–50 μg of total protein was run on 4–20 % Mini-PROTEAN TGX Precast Gels (Bio-Rad Laboratories, Mississauga, ON, Canada). For histones, cells were collected in 0.1 % Nonidet P-40 in phosphate-buffered saline to release nuclei. WCL were prepared by adding equal volumes of 2× RIPA buffer supplemented with 25 U of Benzonase Nuclease (Sigma-Aldrich) and cOmplete Mini Protease Inhibitor Cocktail (Roche), incubating on ice for 30 minutes and sonicating for 15 minutes with 30-second on-off intervals. Twenty micrograms of WCL were run on a 12 % gel. A list of primary antibodies used in immunoblotting is provided in Additional file
1: Table S5. Signals were developed with BM Chemiluminescence Blotting Substrate (POD) (Roche) and a ChemiDoc imaging system (Bio-Rad Laboratories).
Small interfering RNA transfection of ZR75-1- and MDA-MB-231-resistant cells
For CCK-8 assays, a total of 7 × 104 ZR75-1 epirubicin-resistant cells and MDA-MB-231 epirubicin-resistant cells were transfected with Lipofectamine RNAiMAX (Life Technologies) and 10 nM Dharmacon ON-TARGETplus siRNA reagent human SMARTpool (GE Healthcare, Lafayette, CO, USA) targeting HIST1H2AC (L-011435-01-0005), HIST1H2BK (L-013323-02-0005) or both according to the manufacturer’s instructions. Negative controls included media only, Lipofectamine only or mock transfection with non-targeting small interfering RNA (siRNA; D-001810-10-05). Cells were exposed to 0.3–3000 nM epirubicin for 72 h before their viability was determined using the CCK-8 kit. For flow cytometric analyses, 2 × 105 cells were plated in 6-well plates and transfected with 10 nM siRNA or control as described above. Samples were collected at 72 h for quantitative real-time polymerase chain reaction (qRT-PCR) and flow cytometric analyses.
nCounter CodeSet and data pre-processing
The nCounter gene expression CodeSet (NanoString Technologies, Seattle, WA, USA) included 7 genes within the histone module and 11 additional genes that were identified in the KEGG PATHWAY database [
19] as being important for histone function (Additional file
1: Table S6).
HIST1H2AC was excluded from the CodeSet because probes cross-hybridised to other genes. All 18 genes were functionally related (Additional file
1: Figure S6). Messenger RNA (mRNA) CodeSets were processed on nCounter according to the manufacturer’s instructions. Raw mRNA abundance data were pre-processed using the NanoStringNorm R package (v1.1.19; Additional file
2: Methods). A range of pre-processing schemes was assessed to optimise normalisation parameters as previously described (Haider S., Yao C. Q., Sabine V. S., Grzadkowski M., Starmans M. H. W., Wang J., Nguyen F., Moon N. C., Lin X., Drake C., Crozier C. A., Brookes C. L., van de Velde C. J. H., Hasenburg A., Kieback D. G., Markopoulos C. J., Dirix L. Y., Seynaeve C., Rea D. W., Kasprzyk A., Lambin P., Lio P., Bartlett J. M. S., Boutros P. C.: Using pathways for cross-disease biomarker discovery, in preparation).
Survival modelling
To assess whether individual genes are prognostic of survival, each gene was median dichotomised into low- and high-expression groups and univariate Cox proportional hazards models were fit (Additional file
1: Figure S7). Survival analysis of clinical variables modelled age as binary variable (dichotomised at age >50 years), while nodal status, pathological grade, ER status and HER2 status were modelled as ordinal variables (Additional file
1: Figure S1B). Tumour size was treated as a continuous variable.
mRNA network analysis
We hypothesised that integrating molecular modules could improve residual risk prediction relative to DRFS and OS. For each module, we calculated a module dysregulation score (Additional file
2: Methods), which we used in a univariate Cox proportional hazards model. A stratified fivefold cross-validation approach was applied, and models were trained in the training cohort and validated in the
kth testing cohort using 10-year DRFS as an end point. All survival modelling was performed on DRFS and OS in the R statistical environment with the survival package (v2.37-7). Treatment by marker interaction term was calculated using Cox proportional hazards model.
Discussion
Anthracycline resistance is a major obstacle to the effective treatment of women with breast cancer. Although various mechanisms may contribute to anthracycline resistance, including activation of drug transporters, reduced activity of Topo IIα and inhibition of apoptosis, the majority of the molecular mechanisms involved in clinical drug resistance remain unknown. Using a panel of cell lines representative of the major molecular subtypes of breast cancer, we have shown that deregulation of histones involved in chromosome maintenance, epigenetic pathways, cell division and gene regulation is observed consistently in epirubicin-resistant cell lines. This observation was then validated clinically in the BR9601 adjuvant clinical trial.
Histone
H2A and
H2B variants are emerging as mediators of drug sensitivity and resistance in cancer [
22,
23]. We have shown that the dysregulation of histones is associated with increased cell-cycle progression, specifically the release of a G
2/M cell-cycle block in the presence of epirubicin, and with a reduction in apoptotic cell death. Interestingly, transcriptional knockdown of the two histone variants contributing to the dysregulation signature did not completely sensitise cells to anthracycline, possibly for a few reasons. First, although the transcript levels were reduced by 6–53 %, it is possible that the protein levels remained unchanged during our experimental window. We were not able to assess protein expression of each specific variant, because antibodies are not yet commercially available. Second, even if the protein levels were sufficiently diminished, it is still possible that other histone variants functionally substituted for HIST1H2AC and HIST1H2BK because there are 9
H2A and 11
H2B non-allelic histone variants [
24]. Third, the module contains 16 other genes that perform together with the histone genes in this functional module. This notion is shifting away from the previous efforts that were focused on discovering single genes as biomarkers by using fold-change differences in gene expression as the means of selecting promising biomarker candidates. Instead, the FI network approach relies on the strength of the gene-to-gene interactions and is based on how closely the genes are functionally related. This entire module was identified to be a predictive biomarker of anthracycline benefit, which allowed us to focus our efforts on identifying a drug that could target the function of an entire module rather than one of its components. Indeed, using a small-molecule inhibitor screen, we have shown that drugs directly targeting histone function (HDACi as well as cell-cycle inhibitors; data not shown) are cytotoxic to epirubicin-resistant cells and could be considered as an alternative treatment option for patients who do not respond to epirubicin (Fig.
7b). Collectively, these data suggest that modification of histone-regulated pathways represents a key “druggable” target in patients with epirubicin-resistant breast cancers.
Epirubicin-resistant cell lines were generated by exposing native, non-resistant cell lines to increasing concentrations of epirubicin. Interestingly, only a single cell line, SKBR3, upregulated drug transporters, and this was associated with cross-resistance to taxanes. Previously, Hembruff et al. [
25] developed epirubicin-resistant MCF7 cells and established that a specific selection dose must be surpassed to activate drug transporters. For MCF7, this critical threshold concentration was around 30 nM [
19]. Although this concentration is identical to the selection dose of our resistant MCF7 cells, MDR was not upregulated, suggesting a stochastic nature of molecular events that take place en route to drug resistance. Importantly, it indicates that there exists a previously unappreciated MDR-independent mechanism of resistance that should be evaluated for clinical relevance.
Our study revealed that one of those mechanisms involves upregulation of
H2A and
H2B genes and several pathways, including epigenetic and cell-cycle pathways. H2A and H2B histones form octamers with H3 and H4 histones, which participate in packaging of DNA into nucleosomes [
26]. These histones are replication-dependent and cell-cycle–regulated, increasing 35-fold in S phase during DNA replication [
27]. Thus, elevated histone transcript levels may be a consequence of a stalled cell cycle as cells struggle to repair epirubicin-induced DNA damage. However, because resistant cells did not stall, we eliminated the possibility that upregulated histone transcripts were a mere reflection of accumulated mRNA.
An alternative explanation, supported by the ability of HDACi to sensitise resistant cells to epirubicin, is that upregulation of histones contributed to (1) activation of resistance pathways, (2) silencing of molecular pathways that sensitise cells to anthracyclines, and/or (3) decreased accessibility of epirubicin to DNA. H3 and H4 histone modification patterns strongly associate with either active or repressed gene transcriptional status. Current understanding of H2A and H2B histone modifications is based on studies in yeast and few tumour cell lines; nonetheless, a few important features of H2A and H2B histone modifications have been revealed. First, modified sites are acetylated, phosphorylated and ubiquitinated, but not methylated [
28‐
30], a modification most commonly observed with H3 and H4 histones. This highlights the appropriate use of HDACi in our study and their potency due to numerous acetylation sites, although this does not eliminate the possibility that the inhibitors were acting on H3 and H4 histones as well. Because acetylated sites on H2A and H2B are associated with transcriptional activation [
28,
29], modifying the acetylation pattern may have activated transcriptional repressors and pro-apoptotic genes outlined in our model (Fig.
7c, point 1,
left). Second, the N-terminal ends of H2A and H2B histones possess a repression domain that inactivates gene transcription in approximately 10 % of the yeast genome [
28,
29], suggesting that these domains could have collaborated with acetylation patterns induced by HDACi to repress genes involved in resistance, such as those involved in cell cycle or apoptosis (Fig.
7c, point 2,
centre). Third, our model also recognises that resistance might have been reversed by an increased accessibility of epirubicin to DNA (Fig.
7c, point 3,
right).
Interestingly, although the cell lines were resistant, increasing epirubicin concentrations were ultimately cytotoxic to the cells, being indicative of partial drug resistance. This could be a consequence of dynamic changes in the population that is heterogeneous in terms of mutations [
31] and the ability to use different resistance mechanisms at increasing concentrations of epirubicin until toxic levels are reached. Alternatively, a non-mutation–driven model might have contributed to the partial resistance to epirubicin; that is, because the rate of cell kill is proportional to the rate of tumour growth [
32,
33], the effectiveness of epirubicin might have depended on the proportion of cells that were actively dividing vs. those that were in G
0 phase and unresponsive to the drug treatment until they entered the G
1 phase of the cell cycle [
31].
Regel et al. [
34] showed that the HDACi panobinostat sensitises gastric cancer cells to anthracyclines. Our findings are consistent with those of their study and show that multiple HDACi reverse anthracycline resistance in breast cancer cells. This is an important finding because many of the pharmacological inhibitors tested in our study are in use either as single agents or as combination therapies in phase II/III clinical trials [
35‐
37]. HDACi currently in clinical trials include panobinostat, quisinostat, givinostat, abexinostat, pracinostat, belinostat and mocetinostat (Additional file
1: Table S4). Because anthracycline resistance may lead to cross-resistance to taxanes [
2,
38], as it did in one of our resistant cell lines, it may be that taxanes, not anthracyclines, should be used in first-line treatment [
39]. Furthermore, the patients in this study received polychemotherapy as part of their standard of care and, as was appropriate at the time, received either CMF (standard of care at the time) or anthracyclines (experimental at the time). Given the focus of our research on identifying markers of anthracycline benefit, this trial design satisfies the requirement of Simon et al. for a biomarker validation study [
40]; however, further research in the context of taxane-based chemotherapy would be of value. As cancer cells could acquire resistance to HDACi [
36], sequential therapy involving HDACi, taxanes and anthracyclines will be an important aspect of clinical trial design and medical practice.
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
MB participated in the study design and coordination; performed flow cytometric analyses, siRNA experiments involving apoptosis and cell-cycle profiling and associated qRT-PCR, Western blotting for basal histone levels, and inhibitor screens; interpreted data; and drafted the manuscript. LL, NLy and KJT participated in generation of epirubicin-resistant cell lines. LL performed cell-counting and CCK-8 assays and HDACi screens and also characterised cells by Western blotting. NLy measured basal histone levels by qRT-PCR. NLo performed siRNA experiments involving epirubicin and CCK-8 assays that pertained to those experiments. PMK, IK and LDS performed network-based analysis of the microarray data. CQY and PCB normalised and processed gene expression data from the BR9601 trial bioinformatically. CJT provided BR9601 patient samples and participated in editing of the manuscript. RM participated in the small-inhibitor experimental design and interpretation of the HDACi results. JMSB conceived the study and helped draft the manuscript. MS conceived the study, participated in the generation of the cell lines, performed statistical analysis of the clinical data and helped draft the manuscript. All authors read, edited and approved the final manuscript.
MB has a doctoral degree in immunology with a focus on transcriptional regulatory networks. She is interested in elucidating mechanisms of anthracycline resistance and determining their clinical relevance. MS has a doctoral degree in veterinary virology. She has a strong background in breast cancer, in particular assay development and cell-based drug screening. Her research interests are in developing new diagnostic approaches to improve patient diagnosis and treatment using both preclinical and clinical models. The Ontario Institute for Cancer Research (OICR) is a translational research institute dedicated to research on the prevention, early detection, diagnosis and treatment of cancer.