Introduction
Endocrine resistance in breast cancer is a process that appears to result from upregulation of growth factor and protein kinase signaling pathways that provide an alternate mechanism in support of tumor cell proliferation and survival [
1]-[
4]. Tamoxifen (TAM) has proven to be one of the most successful agents in the management of estrogen receptor-positive (ER+) breast cancers. When effective, it suppresses tumor growth and reduces the risk of relapse. Unfortunately, with time, about 50% of patients with ER+ breast cancer stop benefiting from TAM treatment and acquire resistance, leading to disease progression. Despite significant advances in defining some of the factors involved [
5]-[
8], the mechanisms underlying endocrine resistance are complex and not fully understood. Therefore, we have been interested in identifying and targeting, by inhibition or downregulation, key players that mediate endocrine resistance in ER+ breast cancer.
Many cancers are maintained in a hierarchical organization of rare cancer stem cells (CSCs) and more plentiful differentiated tumor cells. CSCs that are resistant to treatment not only have the capacity to give rise to differentiated tumor cells but also can lead to recurrence, metastasis and disease progression [
9]-[
11]. Therefore, endocrine resistance might be associated with the outgrowth of CSCs by promoting expansion of the CSC population or augmenting the production of key factors that regulate the CSC phenotype.
In our previous studies, we reported a correlation between overexpression of the protein 14-3-3ζ and early onset of recurrence in breast cancer patients [
12]. We also uncovered a previously unknown relationship between 14-3-3ζ and FOXM1 in TAM resistance in breast cancer, with 14-3-3ζ acting upstream of FOXM1 to enhance the expression of FOXM1-regulated genes [
13].
FOXM1 is a forkhead transcription factor that binds to chromatin and plays an important role in ERα signaling pathways [
14]. FOXM1 is a key regulator of the cell cycle and is essential for formation of the mitotic spindle and correct chromosome segregation [
15]. Its expression is very low in normal tissues, but elevated in many types of cancers [
16]-[
18]. High expression of FOXM1 is associated with a poor prognosis [
19]-[
22]. In addition to its role in mitosis and cytokinesis, this transcription factor regulates genes that control critical aspects of cancer, including differentiation [
23], angiogenesis [
24] and metastasis [
16],[
20].
In this study, we show that TAM-resistant (TamR) cells contain higher levels of FOXM1 than do parental cells sensitive to growth inhibition by TAM and that this is correlated with the presence of a larger CSC population. Further, in large cohorts of patient breast tumors that we examined, high FOXM1 RNA and protein levels were found to correlate with a significantly faster onset of tumor recurrence and reduced overall survival. In cultured cells, FOXM1 promoted breast cancer aggressiveness and therapy resistance which could be reversed by FOXM1 inhibition or knockdown. Our genome-wide analyses using chromatin immunoprecipitation followed by high-throughput sequencing (ChIP-seq) revealed that TAM-specific FOXM1 binding sites are associated with genes encoding markers of CSCs and invasiveness and that overexpression of FOXM1 increases the proportion of CSCs and directly regulates the production of factors that promote aggressiveness and therapy resistance in breast cancer.
Methods
Cell culture, small interfering RNA, overexpression and ligand treatments
MCF-7 and T47D cells were obtained from the American Type Culture Collection (Manassas, VA, USA) and TamR MCF-7 cells (TamR cells) described previously [
25] were cultured in minimal essential medium (MEM; Sigma-Aldrich, St Louis, MO, USA) supplemented with 5% calf serum (HyClone Laboratories, Logan, UT, USA), 100 μg/ml penicillin-streptomycin (Invitrogen, Carlsbad, CA, USA) and 25 μg/ml gentamicin (Invitrogen). Four days before control vehicle or ligand treatment, cells were seeded in phenol red-free MEM containing 5% charcoal-dextran-treated calf serum. Medium was changed on days 2 and 4 of culture before treatment. For three-dimensional cultures, 100 μl of Matrigel was spread in each well of a 12-well plate, and 8,000 cells were seeded and grown for 6 to 10 days. Spheroids were stained with Giemsa-Wright stain for 15 minutes at room temperature and washed twice with 1× phosphate-buffered saline (PBS) for 5 minutes each. Small interfering RNA (siRNA) experiments were carried out by transfecting 50 nM of siCtrl, siFOXM1 or siABCG2 from DharmaFECT reagent (Dharmacon, Lafayette, CO, USA) for 72 hours. Overexpression was performed as previously reported [
12].
ChIP and ChIP-reChIP assays
Cells were treated with 0.1% EtOH (vehicle) or 1 μM 4-hydroxytamoxifen (4-OH-TAM) for 45 minutes after pretreatment for 1 hour with the FOXM1-selective alternate reading frame (ARF) peptide inhibitor or mutant ARF control peptide [
26] or with extracellular signal-regulated kinase kinase 1 (MEK1) inhibitor (AZD6244; Sellek Chemical, Houston, TX, USA) or control vehicle. After treatment, chromatin was cross-linked using 1% formaldehyde for 15 minutes at room temperature. Cells were washed with PBS, harvested and sonicated three times for 10 seconds in ChIP lysis buffer. Lysates were centrifuged for 10 minutes at 4°C. For immunoprecipitation of DNA-protein complexes, lysates were incubated overnight with antibodies to FOXM1 (GeneTex, Irvine, CA, USA) or extracellular signal-regulated kinase 2 (ERK2; Santa Cruz Biotechnology, Santa Cruz, CA, USA). Complexes were washed three times with radioimmunoprecipitation assay (RIPA) buffer (three times) and two times with Tris-EDTA (ethylenediaminetetraacetic acid). Following the overnight incubation at 65°C, ChIP DNA was isolated using a QIAGEN PCR purification kit (QIAGEN, Valencia, CA, USA) as per the manufacturer's suggestions. The DNA was used for ChIP-seq analysis and quantitative real-time PCR.
Sequential chromatin immunoprecipitation (ChIP-reChIP) experiments were done following the same ChIP protocol. After the first pull-down, immunoprecipitated material was recovered with 10 mM dithiothreitol in immunoprecipitation buffer at 37°C for 30 minutes, diluted and subjected to a second round of immunoprecipitation. Quantitative RT-PCR (qRT-PCR) was used to calculate recruitment to the regions studied, as described elsewhere [
27].
ChIP-seq and clustering analysis
For characterization of the FOXM1 and ERK2 cistromes from cells treated with 4-OH-TAM, the ChIP DNA was prepared into libraries according to Illumina Solexa ChIP-seq sample-processing methods (San Diego, CA, USA), and single-read sequencing was performed using the Illumina Solexa Genomic Analyzer using methods detailed previously [
28]. Sequences generated were mapped uniquely onto the human genome (hg19) by Bowtie2 [
29] with the default settings. A model-based analysis of ChIP-Seq algorithm [
30] was used to identify enriched peak regions (default settings) with a
P-value cutoff of 6.0E-7 and false discovery rate of 0.01. ChIP-seq data for FOXM1 and ERK2 binding sites are given as BED files in Additional file
1: Table S1. Cistrome data for ERα in MCF-7 cells treated with Tam are derived from a previous study [
31].
The seqMINER density array method with a 300-bp window in both directions was used for the generation of clusters (that is, groups of loci having similar compositional features) [
32]. This ChIP-seq data interpretation platform allows the comparison and integration of multiple ChIP-seq data sets and their extraction and visualization of specific patterns as described previously [
28]. BED files for each cluster were used for further analysis with Galaxy Cistrome integrative analysis tools (Venn diagram, conservation, Cis-regulatory Element Annotation System (CEAS)) [
33].
Motif and Gene Ontology category analysis
Overrepresented Gene Ontology (GO) biological processes were determined by utilizing the web-based DAVID Bioinformatics Resources database [
34],[
35], GeneSpring and web-based GREAT (Genomic Regions Enrichment of Annotations Tool) software [
36]. Motif enrichment analysis was done using SeqPos [
33]. Conservation of the binding sites was determined using web-based CEAS software of the Cistrome/Galaxy platform [
37]. Default parameters were used in all software.
RT-PCR and quantitative PCR
Total RNA was isolated from cells using TRIzol reagent (Invitrogen). RNA samples were reverse-transcribed using SuperScript II reverse transcriptase (Invitrogen), and RT-PCR was carried out on the ABI Prism 7900HT Sequence Detection System using SYBR Green PCR Master Mix (Applied Biosystems, Foster City, CA, USA) as described previously [
38]. Primer sequences for the genes studied were obtained from the Harvard Primer Bank [
39]. Sequences are available on their website.
Microarray gene expression data analysis and statistics
Total RNA was used to generate complementary RNA (cRNA), which was labeled with biotin according to protocols recommended by Affymetrix (Santa Clara, CA, USA). All analyses were done using three or more samples for each treatment. The biotin-labeled cRNA was hybridized to Affymetrix U133 plus 2.0 GeneChips, which contain oligonucleotide probe sets for over 47,000 transcripts. After being washed, the chips were scanned and analyzed using Affymetrix processing software. All microarray gene expression data have been deposited in the Gene Expression Omnibus database [GEO:GSE55204]. CEL files were processed using GeneSpring GX 11.0 software (Agilent Technologies, Santa Clara, CA, USA) to obtain fold changes and
P-values with the Benjamini and Hochberg multiple-test correction [
40] for each gene for TAM treatment relative to the vehicle control in each cell background. We considered genes with fold changes greater than two and
P-values <0.05 as statistically significant and differentially expressed. For analyses of microarray data sets from TAM- treated breast cancer patients, we used Frasor
et al. data [GEO:GSE1379] [
38] and Buffa
et al. data [GEO:GSE2221] [
41]. Multifactor analysis was computed in WinSTAT statistics add-in for Excel software (R. Fitch software). Differences between two groups were assessed using an unpaired t-test. Data involving more than two groups were assessed by analysis of variance with Dunnett's multiple-comparisons test. Differences were considered significant at
P < 0.05. Additional statistical analyses done are indicated in the figure legends.
Western blot analysis
Whole-cell extracts were prepared using 1× RIPA lysis buffer (Upstate/Chemicon, Billerica, MA, USA) supplemented with 1× cOmplete Protease Inhibitor Cocktail mixture (Roche Applied Science, Basel, Switzerland). Proteins were separated on 4% to 20% gradient SDS-PAGE gels and transferred to nitrocellulose membranes. For Western blot analysis, we used antibodies against FOXM1, ERK1 and ERK2 (Santa Cruz Biotechnology), β-actin (Sigma-Aldrich), phosphorylated mitogen-activated protein kinase (pMAPK) (Cell Signaling Technology, Danvers, MA, USA) and CD44 (BD Biosciences, San Jose, CA, USA).
Cell proliferation assay
A WST-1 assay (Roche Applied Science) was used to quantify cell viability. Absorbance was read at 450 nm on a PerkinElmer Victor X Multilabel Plate Reader (PerkinElmer, Waltham, MA, USA), and all assays were performed in triplicate as described elsewhere [
13],[
42].
Fluorescence-activated cell sorting and immunofluorescence
For fluorescence-activated cell sorting (FACS), cells were detached and then stained with antibodies for CD44, CD24, ABCG2 (BD Biosciences and Cell Signaling Technology) at 1:100 dilution in PBS containing 1% fetal calf serum. FACS-sorted cells were collected into cell medium for plating or into RNAlater™ buffer for RNA extraction. To test for ABCG2+ activity, 1 × 106 cells were incubated with 5 μM Hoechst 33258 dye at 37°C for 90 minutes. All samples were analyzed and sorted using a FACSAria III instrument (BD Biosciences).
Invasion assay
Breast cancer cells were seeded on precoated filters (8-μm pore size) after membrane rehydration (BD Biosciences). Following incubation for 48 hours at 37°C, cells were fixed in 10% formalin buffer and stained using crystal violet. Noninvasive cells on the surface of the filter were removed using a cotton swab. Invasion was quantified by determining the percentage of cells that had invaded the filter compared to the total number seeded as described previously [
13],[
42].
Breast tumor cohort and FOXMimmunohistochemistry and statistical analysis
A tissue microarray (TMA) from the Samsung Medical Center Breast Cancer Biomarker Study was utilized for the analysis of FOXM1 status. Detailed clinical features and molecular subtype classification have been reported elsewhere [
43],[
44]. Briefly, from among 815 tumors, 501 were assigned as ERα-positive and used for the immunohistochemical detection of FOXM1 expression. TMA sections were incubated for 1 hour at room temperature with mouse anti-human FOXM1 antibody (ab55006; Abcam, Cambridge, MA, USA) diluted 1:400. The detection system EnVision+ for mouse antibody (K4001; Dako, Glostrup, Denmark) was applied according to the manufacturer's instructions. Slides were stained with liquid diaminobenzidine tetrahydrochloride (DAB+), a high-sensitivity substrate chromogen system (K3468; Dako). Counterstaining was performed with Mayer's hematoxylin. FOXM1 expression was scored using a semiquantitative method based on the following four classes: score 0 (no staining or nuclei staining observed in <10% of the tumor cells), score 1+ (faint nuclear staining detectable in >10% of the tumor cells), score 2+ (weak to moderate nuclear staining observed in >10% of the tumor cells) and score 3+ (strong nuclear staining observed in >30% of the tumor cells). Representative photomicrographs of each of the scoring categories are shown in Additional file
2: Figure S4. Patients with tumor scores ranging from 0 to 1 were classified as FOXM1-negative/low expression, and those who had scores of 2+ and 3+ were classified as FOXM1-high expression group. Disease-free survival was defined as the time from the date of diagnosis to the date of documented relapse, including locoregional recurrence and distant metastasis. Survival curves were constructed using the Kaplan-Meier method, and the logrank test was used to compare the mean survival rates across the groups. The logrank test with Bonferroni's correction was used for the subgroup survival analysis.
Accession numbers and data availability
Gene expression data are available in the GEO database [GEO:GSE55204]. ChIP-Seq data files for FOXM1 and ERK2 binding sites in TAM-treated cells are given as BED files in Additional file
1: Table S1.
Discussion
Our findings reveal that TAM resistance is associated with upregulation of FOXM1 and with a FOXM1-dependent gene expression program that enhances cell proliferation and invasiveness and elicits an increase in the proportion of CSCs within the breast cancer cell population. These cells expressed many markers associated with stem cells and with decreased patient survival [
26], including CD44+ and CD24-/low markers, and elevated EMT markers and properties. They also showed high expression of ABC transporters that can result in tumor stem-like cells being resistant to conventional therapies due to drug efflux. These observations provide guidance for how one might optimally combine agents targeting specific characteristics of CSCs with conventional treatments that reduce tumor bulk, thereby effecting long-term benefits of ablating not only the overwhelming majority of the differentiated tumor cells but also removal of the more endocrine-resistant CSCs that can result in repopulation of the tumor [
53]-[
56]. Indeed, inherent drug resistance of CSCs is considered to be a crucial limitation to treatment effectiveness [
56].
We found that the CSCs represent only a small proportion of the MCF-7 cell population, but that this fraction is increased fivefold in TamR cells. Our observations uncover a novel role for FOXM1 in inducing expansion of the CSC-like population and in promoting an aggressive and endocrine-resistant phenotype. These effects of FOXM1 likely underlie the strong association we have observed between high tumor FOXM1 and poor clinical outcome for patients with ER+ breast cancers. Our examination of several large data sets cumulatively representing about 1,000 ER+ breast tumors indicates that high FOXM1 expression occurs in about 20% of ER+ breast cancers. Of note, the authors of a recent report showed that FOXM1 and its regulated target genes
AURKA, AURKB and
BIRC5/survivin display the greatest prognostic discrimination among a panel of genes analyzed for overall survival of patients with ER+ breast cancer and an intermediate Oncotype DX 21-gene recurrence score. High expression of these genes predicts a poorer outcome and suggests more aggressive selection of adjuvant chemotherapy for these patients [
57].
As schematized in the model (Figure
6H), we show that FOXM1 is elevated by TAM in a time-dependent manner and that its expression is associated with markers of TAM resistance. In previous studies, we identified the association between FOXM1 and 14-3-3ζ, a protein also found to be upregulated by TAM and elevated in TAM-resistant tumors [
13] via deregulation of miR-451 that targets 14-3-3ζ [
42] (Figure
6H). Our data now reveal that FOXM1, a member of the family of forkhead transcription factors, fosters the enrichment of CSCs expressing stem cell markers (for example, ABCG2, NF-YA, NF-YB and NF-YC), mitosis-related genes and genes fostering invasiveness and motility (Rho-GTPases).
By ChIP-Seq and ChIP-reChIP, we show that TAM induced recruitment of FOXM1 to the promoter regions of cell-cycle mitosis-related genes and genes encoding stem cell markers in MCF-7 and TamR cells, supporting our hypothesis that FOXM1 promotes the expansion of a highly proliferative CSC-like progenitor population that is capable of self-renewal and can give rise to differentiated progeny. We also observed FOXM1 upregulation of EMT markers.
We focused much of this study on our novel finding of the regulation by FOXM1 of stem cell–related genes that were found by seqMINER analysis to be enriched in the C1 cluster. Moreover, this cluster was also enriched for targets of miR-34a, recently reported to be important in regulating the expression of self-renewal genes [
58]. Interestingly, the FOXM1 C1 cluster binding sites are co-occupied by ERK2, suggesting a sophisticated mechanism by which FOXM1 and MAPK signaling may participate in the development of endocrine resistance. Indeed, resistance to endocrine therapies is known to be associated with enhanced signaling through MAPK [
1]-[
3],[
5],[
8],[
45]. We show in this study that the inhibition of MAPK activation with MEK1 inhibitor, or alteration of FOXM1 expression by the specific inhibitor ARF, impaired the recruitment of these factors to chromatin, indicating that these two factors control each other's binding to C1 genomic regions and that their copresence is essential for the activation of transcription of C1 genes. Of note, it has been shown that pMAPK induces phosphorylation of FOXM1, enabling its translocation to the nucleus and binding to genomic elements [
59].
Further, our study reveals the interdependence of FOXM1 and MAPKs, with FOXM1 regulating the expression of MAPK and FOXM1-knockdown decreasing the level of MAPK. Of interest, the binding sites co-occupied by FOXM1 and MAPK are highly conserved among species, which suggests an evolutionarily conserved function for these genomic locations in different organisms. Moreover, our bioinformatics analysis of cluster C1 FOXM1-regulated genes was predictive of clinical outcome in women with TAM-treated tumors. Among these genes, we found well-described FOXM1 target genes such as
B-Myb, c-Jun and
c-Fos, as well as important genes involved in stem cell maintenance. Among the genes classified as CSC markers, we found multidrug resistance proteins (
MDR1, ABCG5 and
ABCG2), the nuclear transcription factor
NF-YA/B/C[
60] and
SIRT1[
61].
We concentrated in particular on studying the role of FOXM1 in regulating the expression of ABCG2 because ABCG2, also known as breast cancer resistance protein, belongs to the ATP-binding cassette family. A defining feature of CSCs is their ability to efflux Hoechst dye, leading to the identification of the SP that is associated with expression of the ABCG2 protein. Its expression has been found in several stem cell tissues, including lung and prostate cancer and glioblastoma [
62],[
63]. Breast cancer SP cells have a high drug efflux capacity owing to functional expression of ABC transporters such as ABCG2. Although the mechanism by which multidrug resistance genes work in inducing chemotherapy resistance has been described previously, a recent study has implicated multidrug resistance proteins in hormone resistance by showing that ABCG2 can efflux TAM [
64]. These reports support what we observed upon knockdown of either FOXM1 or ABCG2. With the reduction in cellular FOXM1 or ABCG2, or by inhibition of FOXM1 using ARF peptide, we were able to restore growth suppression by TAM to TamR cells, indicating that the levels of ABCG2 impact treatment response and that the upregulation of ABCG2 by FOXM1 could provide an explanation for the development of TAM therapy resistance.
In line with previous reports, our data show that our ABCG2+ SP had higher invasiveness potential compared to ABCG2- cells upon examination by three-dimensional Matrigel culture and invasion assays. We further determined that this phenomenon is associated with their elevated expression of CDC42 and RhoB genes, which harbor FOXM1 binding sites co-occupied by ERα.
Of note, we show that overexpression of FOXM1 induced a cell phenotype characterized by branching, extended chains of cells and cellular protrusions distinctive of a migratory phenotype and characterized by increased expression of
CDC42 and
RhoB and higher invasiveness. The GTP-binding proteins RhoB and CDC42 regulate the organization and turnover of the cytoskeleton and cell-matrix adhesions, which are a crucial feature in the acquisition of an invasive phenotype and the development of metastasis [
65],[
66]. Further, in line with what has been previously reported, our data confirm the binding of FOXM1 to matrix metalloproteinases and VEGF [
14] as well as the regulation of EMT markers, thereby associating FOXM1 at yet another level to the metastatic process [
20].
Conclusions
Collectively, our findings define FOXM1 as a master regulator of Rho-GTPase and stem cell marker expression and imply that reducing FOXM1 expression might be effective in blocking tumor progression in several critical ways: by decreasing the expression of mitosis-related genes, by reducing invasion potential and by diminishing the proportion of CSCs, thereby enhancing sensitivity to cancer therapeutic agents. Indeed, the authors of several recent reports have shown FOXM1 to be associated with resistance to chemotherapeutic agents [
67] and resistance to radiation treatment [
68].
Moreover, our functional work clearly shows that rendering the FOXM1 pathway inactive by RNAi knockdown or by use of the p19
ARF 26-44 peptide [
49], a selective FOXM1 peptide inhibitor called ARF, was highly effective in restoring endocrine sensitivity and suppressing breast cancer aggressiveness. This ARF inhibitor has already been shown to be effective in suppressing the development of hepatocellular carcinoma in a preclinical model [
69]. Taken together, our findings have clinical implications for breast cancer and potentially many other cancers where FOXM1/pMAPK signaling pathways are active, and make a case for the use of FOXM1 inhibitors in combination with current therapies, including protein kinase inhibitors, to improve effectiveness and long-term patient response to treatments.
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
AB conceived of and designed the project; performed the gene expression, genomic and cell biological assays; analyzed the assay data and clinical patient data sets; and wrote the manuscript. ZME conceived of and designed the project; performed the gene expression, genomic and cell biological assays; analyzed the assay data and clinical patient data sets; and contributed to the writing of the manuscript. YJK carried out the IHC experiments and analyses on large clinical TMAs, analyzed the data and contributed to data interpretation and the writing of the manuscript. YLC carried out the IHC experiments and analyses on large clinical TMAs, analyzed the data and contributed to data interpretation and the writing of the manuscript. HL assisted in designing the project, performed gene expression and cell biological assays and analyzed these data, and contributed to the writing of the manuscript. BSK conceived of, designed and directed the project; oversaw the experiments; analyzed all data; and wrote the manuscript. All authors read and approved the final manuscript.