Estrogen receptor-α directly regulates the hypoxia-inducible factor 1 pathway associated with antiestrogen response in breast cancer

Significance

About 1.7 million new cases of breast cancer occur every year, 70% of which are estrogen receptor-α (ERα) positive. Antiestrogen therapy to block ERα function is the most important approach in treatment of ERα⁺ patients. However, resistance eventually will develop for various reasons. Here we demonstrate that hypoxia-inducible factor 1α (HIF-1α) is a direct transcriptional target of ERα, which may compensate for ERα function loss because many other ERα targets are also HIF-1α targets. We further show that HIF-1α is able to confer cancer cell resistance to ERα antagonists tamoxifen and fulvestrant, and the expression of HIF-1α is associated with poor survival to endocrine therapy in ERα⁺ patients. Our findings thus have revealed a previously unidentified mechanism for antiestrogen resistance.

Abstract

A majority of breast cancers are driven by estrogen via estrogen receptor-α (ERα). Our previous studies indicate that hypoxia-inducible factor 1α (HIF-1α) cooperates with ERα in breast cancer cells. However, whether ERα is implicated in the direct regulation of HIF-1α and the role of HIF-1α in endocrine therapy response are unknown. In this study we found that a subpopulation of HIF-1α targets, many of them bearing both hypoxia response elements and estrogen response elements, are regulated by ERα in normoxia and hypoxia. Interestingly, the HIF-1α gene itself also bears an estrogen response element, and its expression is directly regulated by ERα. Clinical data revealed that expression of the HIF-1α gene or a hypoxia metagene signature is associated with a poor outcome to endocrine treatment in ERα⁺ breast cancer. HIF-1α was able to confer endocrine therapy resistance to ERα⁺ breast cancer cells. Our findings define, for the first time to our knowledge, a direct regulatory pathway between ERα and HIF-1α, which might modulate hormone response in treatment.

Estrogen receptor-α (ERα) is an estrogen-dependent nuclear transcription factor that is critical for mammary epithelial cell division and breast cancer progression (1, 2). ERα is expressed in ∼70% of breast tumors (3), the majority of which depend on estrogen signaling, thereby providing the rationale for using antiestrogens as adjuvant therapy to treat breast cancer (4). Tamoxifen is a first generation selective ER modulator and has been widely used in breast cancer prevention and treatment (4). Although now replaced by aromatase inhibitors as first-line treatment in postmenopausal women, it still remains important in premenopausal breast cancer and after failure of aromatase inhibitors. Tamoxifen acts as an antagonist in breast cancer cells by competing with estrogen for the ER. Tamoxifen-bound ER recruits the nuclear receptor corepressor and histone deacetylase (HDAC), as opposed to coactivators, leading to transcriptional repression (5). Although hundreds of thousands of patients have benefited from tamoxifen treatment, its efficacy is limited to an average time of 15 mo in patients with metastatic disease (6), as resistance often develops (7). Many mechanisms have been proposed to account for tamoxifen resistance (8), including loss of ERα expression or expression of truncated ER isoforms, posttranslational modification of ERα, deregulation of ERα coactivators, and increased receptor tyrosine kinase signaling. Recent studies further indicate that somatic ERα mutation (9, 10), as well as genomic amplification of distant ER response elements (11), could contribute to hormone therapy resistance.

Our clinical studies suggest that the in vivo tumor environment may play a role in tamoxifen resistance, as hypoxia-inducible factor 1α (HIF-1α) protein expression was associated with tamoxifen resistance in neoadjuvant, primary therapy of ERα⁺ breast cancers (12), as well as resistance to chemoendocrine therapy (13).

HIF-1α is a master regulator of oxygen homeostasis, which is rapidly degraded in normoxia by the tumor suppressor, von Hippel–Lindau protein (VHL), but is stabilized in hypoxia (14). This process is mainly determined by the hydroxylation of HIF-1α catalyzed by prolyl hydroxylases. HIF-1α has been associated with an aggressive phenotype of breast cancer: that is, large tumor size, high grade, high proliferation, and lymph node metastasis (15). Increased HIF-1α is also associated with ERα-positivity (15), whereas HIF-1β, the partner of HIF-1α, has been shown to function as a potent coactivator of ER-dependent transcription (16). Further studies revealed that in ER⁺ T47D breast cancer cells, combined hypoxia and E2 treatment had additive effects on expression of some genes (17), although the mechanism is not clear. We have previously shown that HIF-1α and ERα can coordinate expression of genes, such as lysine-specific demethylase 4B/Jumonji domain-containing 2B (KDM4B/JMJD2B), an H3K9me3/me2 histone demethylase, which is targeted by both ERα and HIF-1α and epigenetically regulates cell cycle progression (18). The genomic locus of KDM4B bears both HIF-1α and ERα binding elements (18, 19). These data collectively suggest that HIF-1α and ERα are functionally associated. However, how these two important oncogenic pathways interact has not yet been defined. In addition, whether HIF-1α plays an autonomous role in modulating endocrine therapy efficacy, such as tamoxifen resistance, is unknown. In this study, we investigated the role of ERα in the regulation of HIF-1 signaling and how HIF-1 signaling is involved in endocrine drug response.

Results

ERα Signaling Regulates Hypoxia/HIF-1α Pathway.

We have previously shown that knockdown of ERα significantly down-regulated histone demethylase KDM4B expression (18), a HIF-1α transcriptional target, suggesting that HIF-1α function is compromised by loss of ERα even in hypoxia. To study whether ERα signaling is involved in the regulation of the hypoxia/HIF pathway, we used a chemical genetics approach in which the ERα⁺ breast cancer cell line MCF7 was treated with ICI182780 (fulvestrant) in normoxia and hypoxia to perform a global gene-expression profile analysis (Fig. 1A). ICI182780 is an ER antagonist with no agonist effects, which works by downregulating the ERα expression. Clinically, ICI182780 has been used in hormone receptor-positive metastatic breast cancer in postmenopausal women with disease progression following antiestrogen therapy. The gene-expression profiling results showed that a cluster of genes, such as stanniocalcin 2 (STC2), stanniocalcin 1 (STC1), solute carrier family 2 (Facilitated glucose transporter), Member 1 (SLC2A1) (also known as Glut-1), and lysyl oxidase (LOX) that were induced in hypoxia were down-regulated by ICI182780 in both normoxia and hypoxia (Fig. 1A). We then queried the iLINCS (integrative LINCS) genomics data portal to search compounds that regulate a similar gene pattern as ICI182780 induced in MCF7 cells. Among the top 15 hits, most are HIF-1α inducers or ERα modulators, including tamoxifen and ICI182780 (Fig. 1B). We analyzed these genes through ChIP enrichment analysis of transcription factors (ChEA) through the LINCS canvas browser II to examine what transcriptional factors directly regulate their expression. The top hits were ERα and HIF-1α (Fig. 1C). Thus, these data indicate that a subgroup of genes that are targeted by hypoxia/HIF-1α is also regulated by ERα signaling, which is dual responsive to hormone and oxygen. Interestingly, some of these genes can be bound by p53 in murine embryonic stem cells although the biological function is unclear (20). Cancer genomic sequencing reveals that p53 is more commonly mutated in triple-negative breast cancers than ERα⁺ patients (21), whereas MCF7 is p53 wild-type. ZNF263 is a transcription factor that regulates FoxA1 expression (22). FoxA1 is a pioneer factor that facilitates ERα for genomic binding (23), which further suggests that the estrogen-ER signaling pathway is involved in hypoxia/HIF response.

Fig. 1.

ERα signaling regulates hypoxia/HIF pathway. (A) MCF7 cells were treated with 1 μM of ICI182780 in normoxia and hypoxia (1% O₂) for 24 h. Extracted RNA from duplicated biological samples was subject to microarray analysis. Heatmap shows a subgroup of genes that are dual responsive to hormone and oxygen. (B) The dually responsive genes were queried with the LINCS program to search Connectivity MAP (CMAP) for compounds that induced a similar pattern to ICI182780. (C) ChEA of transcription factor binding to the dual responsive genes. (D) Venn diagram shows the common gene bound by both ERα and HIF-1α. (E) Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway annotation of the common genes from D.

To further confirm that ERα and HIF-1α directly bind their response elements in a subgroup of genes, we reanalyzed published ChIP sequencing data (24, 25). We found that among the 356 genes bound by HIF-1α, 202 (57%) of them were identified as the common genes bound by ERα as well (Fig. 1D and Table S1). KDM4B was one of the targets of both ERα and HIF-1α (Table S1), consistent with our previous studies (18, 26). Pathway analysis reveals that these common genes are involved in metabolism, cancer, and important signaling pathways, including Notch, MAPK, and insulin pathways (Fig. 1E).

ERα Signaling Regulates HIF-1α Expression.

Some genes, such as STC2, VEGFA, and KDM4B, bear both ERα and HIF-1α binding elements (18, 19, 27–30), and thus we initially hypothesized that blockade of the ER pathway might disrupt HIF-1α binding to its target genes. Surprisingly, we found that ERα signaling actually directly regulates HIF-1α expression. When MCF7 cells were grown without estrogen for 4 d and then placed in hypoxia or treated with the hypoxia mimetic deferoximine, 17-β-estradiol (E2) greatly enhanced HIF-1α expression (Fig. 2A). However, ICI182780 significantly reduced HIF-1α and ERα expression in hypoxia (Fig. 2B), consistent with its biological function in suppressing HIF-1α targets (Fig. 1). In addition, ICI1820780 treatment did not affect HIF-1α expression in MDA-MD-231 cells (Fig. 2B), an ERα⁻ cell line. Knockdown of ERα with two separate siRNA oligos or a SMARTpool of siRNA greatly reduced HIF-1α induction under hypoxia or hypoxic mimetic dimethyloxalylglycine (DMOG) treatment (Fig. 2 C and D). The proteasome inhibitor MG132, which stabilizes HIF-1α protein expression by preventing protein degradation, failed to rescue ERα knockdown-mediated down-regulation of HIF-1α (Fig. 2E). These data clearly show that the ERα signaling pathway directly regulates HIF-1α expression.

Fig. 2.

ERα signaling regulates HIF-1α expression. (A) After 4-d hormonal starvation, MCF7 cells were treated with 500 μM of DFO or 1% O₂ for 6 h in the presence or absence of 100 nM of E2. The whole-cell lysates were subject to Western blotting with indicated antibodies. V, Vehicle. (B) MCF7 and MDA-MB-231 cells were treated with 1 μM of ICI182780 in hypoxia (1% O₂) for 24 h. The whole-cell lysates were subject to Western blotting with indicated antibodies. (C and D) ERα in MCF7 cells was knockdown with two different siRNAs (C) or SMARTpool (D), which were treated with 200 µM of DMOG (C) or hypoxia (D). The whole-cell lysates were subject to Western blotting with indicated antibodies. (E) MCF7 cells were treated with siRNA SMARTpool against ERα, treated with 10 µM G132 for 16 h. The whole-cell lysates were subject to Western blotting with indicated antibodies.

HIF-1α Gene Bears Estrogen Response Element and ERα Enhances HIF-1α Transcription.

HIF-1α is labile and subject to rapid proteasomal degradation. Most studies focus on the regulation of protein stability of HIF-1α but how HIF-1α is transcriptionally regulated is not well understood. The down-regulation of HIF-1α by knockdown of ERα was not rescued by the proteasome inhibitor MG132, indicating that ERα regulates HIF-1α expression in a protein stability-independent manner (Fig. 2E). Supporting this hypothesis, the ERα antagonist tamoxifen treatment resulted in down-regulation of HIF-1α expression in MCF7 but not MDA-MB-231 cells, although tamoxifen caused ERα protein up-regulation (Fig. S1A), suggesting impaired ERα transcriptional function is involved. To determine whether ERα signaling directly regulates HIF-1α gene expression, we treated hormone-starved MCF7 and MDA-MB-231 cells with estrogen. RT-PCR data revealed that HIF-1α, KDM4B, and STC2 mRNA was significantly up-regulated in MCF7 but not MDA-MB-231 cells (Fig. 3A and Fig. S1 B and C). RT-PCR results further showed that HIF-1α mRNA level was significantly reduced after ICI182780 treatment, indicating an ER-mediated transcriptional mechanism regulates HIF-1α transcription.

Fig. S1.

(A) MCF7 and MDA-MB-231 cells were treated with 1 μM of tamoxifen in hypoxia (1% O₂) for 24 h. The whole-cell lysates were subject to Western blotting with indicated antibodies. (B and C) After 3-d hormonal starvation, MCF7 cells were treated with 100 nM of E2 for 24 h. Real-time PCR was performed to assess KDM4B and STC2 gene expression. Experiments were performed as triplicates. **P < 0.01; NS, not significant.

Fig. 3.

ERα directly binds EREs on the HIF-1α gene to enhance HIF-1α transcription. (A) After 3-d hormonal starvation, MCF7 cells were treated with 100 nM of E2 for 24 h. Real-time PCR was performed to assess HIF-1α gene expression. (B) MCF7 cells were treated with 1 µM of ICI182780 for 24 h. Real-time PCR was performed to assess HIF-1α gene expression. (C) The gene structure of HIF-1α and the sequence that bears ERα binding sites (ERE) and FOXA1 binding sites, as indicated. The black bar represents each exon and the black line represents each intron of HIF-1α. NB1 and NB2 represent the primer location for ChIP PCR to assess ERα binding as negative controls, as shown in D. (D) After 3-d hormonal starvation, MCF7 cells were treated with 100 nM of E2 or tamoxifen for 24 h. ChIP PCR was performed to assess ERα binding among the regions as indicated in C. (E) The ERE region and the mutant were cloned into pGL3 luciferase reporter. The plasmids and the pRL Renilla Luciferase control vector (to normalize the ERE reporter) were transfected into MCF7 cells. Dual luciferase assay was performed. (F) MCF7 cells were incubated in hypoxia for 8 h. ChIP PCR was performed to assess ERα binding at ERE of HIF-1α. **P < 0.01.

By analyzing the HIF-1α genomic sequence that bears 15 exons and 14 introns (Fig. 3C), we found that there is a canonical estrogen response element (ERE) located in the first intron. Interestingly, there is also a FOXA1 binding site that is 64 nucleotides downstream of the ERE (Fig. 3C), further supporting it as a bona fide ERα binding element because FOXA1 is a pioneer factor that facilitates ERα recruitment (23). In fact, all ER-chromatin interactions and gene-expression changes depend on the presence of FOXA1 (23). To test whether ERα is able to bind the ERE of the HIF-1α gene, we treated MCF7 cells with E2 or tamoxifen and then performed ChIP and real-time PCR assays. The results clearly showed that ERα bound to this ERE compared with the surrounding negative regions, and E2 and tamoxifen treatment further enhanced ERα binding (Fig. 3D). Next, we cloned the ERE sequences (Fig. S2 A and B) into a luciferase reporter and performed a luciferase assay to assess the regulatory function of ERα. Indeed, the luciferase activity was significantly high, but the ERE mutant abrogated the activity in MCF7 cells (Fig. 3E). We also tested the binding of ERα in hypoxia condition. The results showed that ERα still bound at the ERE of HIF-1α under hypoxia (Fig. 3F). Thus, these data demonstrate that we have identified HIF-1α as a direct target of ERα, potentially explaining why HIF-1α is associated with ER-positivity in breast cancer samples.

Fig. S2.

ERα binding region in the genomic locus of HIF-1α (A) and the corresponding mutant of HIF-1α gene (B) were sequenced by Sanger sequencing. The sequences were then BLAST with the National Center for Biotechnology Information database. The blue nucleotides indicate sequenced ER binding region and the black nucleotides indicate the genomic sequence of HIF-1α gene. The red nucleotides indicate the ERα element.

Interestingly, we did not see positive correlation between HIF-1α and ERα at transcript levels from tumor cohorts; instead, we found a negative correlation between them (Fig. S3 A and B). This may be explained by the hypothesis that ER⁻ breast cancer has been epigenetically remodeled to express high HIF-1α transcript. This finding led us to further speculate that ERα loss need to be compensated by high HIF-1α activity.

Fig. S3.

The expression of HIF-1α in ER⁺ and ER⁻ breast cancers from the TCGA cohort (A) and Curtis cohort (B), downloaded from Oncomine (https://www.oncomine.com).

Tamoxifen-Bound ERα Inhibits HIF-1α Expression.

Tamoxifen is an antagonist of ERα, competing with estrogen for the ER binding. Tamoxifen-bound ERα recruits the corepressors nuclear receptor corepressor and HDAC to silence gene transcription of ERα targets (5). The discovery of enhancing HIF-1α by E2-bound ERα prompted us to examine the effect of tamoxifen-bound ERα. Tamoxifen treatment of MCF7 cells greatly increased HDAC1 binding on the ERE of HIF-1α. Interestingly, ICI182780 did not affect HDAC1 binding, but remarkably reduced the ERα binding (Fig. 4A). These data indicate that the two compounds inhibit ERα function through different mechanisms.

Fig. 4.

Tamoxifen-bound ERα inhibits HIF-1α expression. (A) MCF7 cells were cultured in standard DMEM and treated with 1,000 nM of tamoxifen or ICI182780 for 48 h. ChIP PCR was performed to assess HDAC1 and ERα binding the ERE of HIF-1α. (B) MCF7 and TamR-MCF7 cells were transfected with siRNA SMARTpool against ERα. After 36-h transfection, cells were incubated in normoxia and hypoxia for 16 h. Western blotting was used to assess the indicated proteins. (C) BT474 cells were treated as in B. (D) MCF7 and TamR-MCF7 cells were transfected with ERα expression plasmids. MCF7 cells were cultured in normoxia in standard DMEM while Tam-MCF7 cells were cultured in phenol-free DMEM with 100 nM of tamoxifen. After 48-h transfection, cells were harvested for Western blotting. The two HIF-1α blots were for short and long exposure, respectively. (E) The cartoon shows the E2-bound and tamoxifen-bound ERα exert opposite effect on HIF-1α transcription.

We then used an in vitro established tamoxifen-resistant MCF7 (TamR-MCF7) cell line and BT474 cells, which are intrinsically resistant to tamoxifen (Fig. 4 B and C). These cells were cultured in media containing tamoxifen for propagating. When ERα was depleted in these cells, HIF-1α expression was up-regulated (Fig. 4 B and C). Overexpression of ERα in TamR-MCF7 cells significantly reduced HIF-1α expression, but not for the parental MCF7 cells in normoxia (Fig. 4D). Longer exposure of the film showed that overexpression of ERα enhanced HIF-1α in parental cells (Fig. 4D). Interestingly, we noticed that the basal levels of HIF-1α expression in Tam-MCF7 cells is higher than in parental cells, indicating epigenetic effects are involved after cells acquire tamoxifen resistance. These data indicate that E2-bound ERα induces—but tamoxifen-bound ERα suppresses—HIF-1α expression (Fig. 4E).

HIF-1α Confers Tamoxifen Resistance on ERα⁺ Breast Cancer Cells.

Because HIF-1α was a downstream target of ERα and enhanced ERα target expression, we hypothesized that HIF-1α may modulate endocrine efficacy in ER⁺ breast cancers. We therefore assessed the role of HIF in regulation of breast cancer cell survival or proliferation in response to inhibitors of ERα. We first used retroviral vector-mediated transduction to stably introduce HIF-1α and HIF-2α into T47D and MCF7 cells (Fig. 5A and Fig. S4). HIF-2α expression significantly suppressed proliferation of both cell lines, whereas expression of HIF-1α did not. Thus, we were unable to generate stable cell lines with HIF-2α. Although both parental MCF7 and T47D and their derivative HIF-1α–expressing cells responded to tamoxifen treatment, the HIF-1α–expressing cells were at least twofold more resistant in normoxia (Fig. S4) and long-term treatment showed more remarkable effect (Fig. 5B), demonstrating that HIF-1α is able to confer tamoxifen resistance. We also treated the cells with ICI182780. Although the parental and HIF-1α expressing cells responded to ICI182780 similarly after 1 wk of treatment (Fig. S4B), long-term treatment (18 d or 4 wk) with 10 nM or 100 nM led to development of more resistant colonies (Fig. 5 B and C). Although 1,000 nM of ICI182780 efficiently suppressed both cell lines, HIF-1α–expressing cells gave rise to more large resistant colonies (Fig. 5C).

Fig. 5.

HIF-1α confers tamoxifen resistance to ER⁺ breast cancer cells. (A) Western blotting to assess MCF7 cells expressing HIF-1α. (B) The MCF7 control and MCF7–HIF-1α cells were treated with tamoxifen or ICI182780 with indicated concentration for 18 d. Cell colonies were stained with Crystal violet. (C) The MCF7 control and MCF7–HIF-1α cells were treated with ICI182780 for 4 wk. Cell colonies were stained with Crystal violet. (D) Tumorspheres generated from the parental (Upper) and HIF-1α expressing (Lower) MCF7 cells were treated for 5 d with 1 µM of tamoxifen, 1 µM of ICI182780, and 0.5 µM of RITA. Photos were taken under microscope (10×). (E) After counting 10 different 10× fields, the number of tumorsphere counts were averaged and compared with Student’s t test (two tailed, **P < 0.01). (F) Kaplan–Meier analyses for relapse free survival of the cohort of patients with ER-positivity, receiving tamoxifen treatment only without chemotherapy. Affymetrix ID for HIF-1α used was 200989_at. The cut-off value used in analysis was 3,043 and the expression range of the probe was 439–17,198. Patient number for low HIF-1α (black) and high HIF-1α (red) is presented under the following months. (G) Kaplan-Meier analyses for overall survival of the cohort of patients with ER-positivity, receiving tamoxifen treatment with chemotherapy. The cutoff value was 3035 and the expression range was 456–11726. Patient number for low HIF-1α (black) and high HIF-1α (red) are presented under the following months.

Fig. S4.

(A) MCF7 control and MCF7-HIF-1α cells were treated with 100 nM of tamoxifen for 96 h. Cell were stained with Crystal violet and counted under microscope; cell number was normalized with control treatment. (B) T47D cells expressing HIF-1α were treated with 1 μM of tamoxifen or 1 μM of ICI182780 for 48 h. The indicated proteins were analyzed by immunoblotting. (C) T47D cells expressing HIF-1α were treated with 1 μM of tamoxifen or 1 μM of ICI182780 for 7 d. Cell colonies (here we define >20 cells as one colony) were stained with Crystal violet and counted under a microscope. **P < 0.01.

We further tested the capacity for tumorsphere formation of the HIF-1α–expressing and parental cells using a mammosphere 3D culture system. HIF-1α did not affect tumorsphere formation; however, HIF-1α conferred significant resistance to tamoxifen and ICI182780 compared with the parental control (Fig. 5 D and E) (P < 0.01). We also tested another drug, RITA, which inhibits HIF-1α expression and induces apoptosis of MCF7 cells (31). The results showed that RITA equally reduced the tumorsphere formation in both HIF-1α–expressing and parental cells (Fig. 5 D and E).

Hypoxia Metagene Signature and High HIF-1α Gene Expression Show a Poor Response to Tamoxifen Treatment in ERα⁺ Breast Cancer.

To determine whether hypoxia/HIF-1α is associated with tamoxifen effectiveness in patients with breast cancer, we first examined the hypoxia status of two groups of breast cancer patients by a hypoxia/HIF-1α metagene signature (32). We then compared relapse-free survival of ERα⁺ breast cancers classified as hypoxic or normoxic by their gene-expression profiles with this hypoxia/HIF-1α metagene signature. In those without adjuvant tamoxifen therapy, hypoxic cases had a significantly worse outcome than the normoxic breast cancers (P = 0.001) (Fig. S5A). A significant difference in outcome for those treated with tamoxifen remained (P = 0.03) (Fig. S5A). Then we analyzed whether the HIF-1α gene expression itself correlated with tamoxifen response in a large cohort of ER⁺ patients from public data (33). Kaplan–Meier analysis results showed that patients with high level of HIF-1α gene expression had a poorer relapse-free survival to endocrine therapy or tamoxifen treatment alone (P = 0.0093) (Fig. 5G and Fig. S5B), although overall survival was not significantly different (Fig. S5C). When chemotherapy was included for those patients who received tamoxifen, HIF-1α is also associated with poor overall survival (P = 0.017) (Fig. 5H). These data further indicate that HIF-1α may be directly involved in modulating tamoxifen response in ER⁺ patients. Interestingly, high HIF-2α is associated with better survival in ER⁺ breast cancer patients who received endocrine therapy (Fig. S5D). Although statistically not significant, HIF-2α tends to be associated with better survival in tamoxifen-treated patients (Fig. S5E). These results were consistent with our findings that HIF-2α overexpression was harmful for ERα⁺ cancer cells and the clinical data that HIF-2α is significantly lower in breast cancer tissues than the normal breast tissue (Fig. S6).

Fig. S5.

(A) Kaplan–Meier analyses of relapse-free and overall survival of ER⁺ breast cancer patients without tamoxifen treatment. These ER⁺ breast cancers are classified into hypoxic and normoxic groups according to expression of a hypoxia metagene signature. (B) Kaplan–Meier analyses for relapse free survival of the cohort of patients with ER-positivity, receiving endocrine therapy only without chemotherapy. Affymetrix ID for HIF-1α used was 200989_at. The cut-off value used in analysis was 3,212 and the expression range of the probe was 299–17,198. Patient number for low HIF-1α (black) and high HIF-1α (red) is presented under the following months. (C) Kaplan–Meier analyses for overall survival of the cohort of patients with ER-positivity, receiving endocrine therapy only without chemotherapy. Affymetrix ID for HIF-1α used was 200989_at. The cut-off value used in analysis was 3,243 and the expression range of the probe was 456–9,223. Patient number for low HIF-1α (black) and high HIF-1α (red) is presented under the following months. (D) Kaplan–Meier analyses for relapse-free survival of the cohort of patients with ER-positivity, receiving endocrine therapy only without chemotherapy. Affymetrix ID for HIF-2α (EPAS1) used was 200878_at. The cut-off value used in analysis was 3,210 and the expression range of the probe was 394–13,085. Patient number for low HIF-2α (black) and high HIF-2 (red) is presented under the following months. (E) Kaplan–Meier analyses for relapse-free survival of the cohort of patients with ER-positivity, receiving tamoxifen treatment only without chemotherapy. Affymetrix ID for HIF-2α (EPAS1) used was 200878_at. The cut-off value used in analysis was 2,472 and the expression range of the probe was 394–13,085. Patient number for low HIF-2α (black) and high HIF-2 (red) is presented under the following months.

Fig. S6.

The expression of HIF-2α in normal breast and breast cancers from the TCGA cohort (A) and Curtis cohort (B), downloaded from Oncomine. **P < 0.001.

HIF-1α Overexpression Confers Advantage of Tumor Growth and Resistance to Tamoxifen Treatment.

To further investigate the effect of HIF-1α on tamoxifen treatment in vivo, orthotopic xenograft tumor was established using MCF7 cells overexpressing HIF-1α. Compared with the control, HIF-1α overexpression resulted in rapid tumor growth in NOD SCID-γ (NSG) mice that had to be killed before the control reached comparable sizes of tumors (Fig. 6A). Tamoxifen treatment only modestly delayed tumor growth with HIF-1α overexpression (P > 0.05), similar to the in vitro data (Fig. 5), indicating that HIF-1α confers tamoxifen resistance.

Fig. 6.

The working model between ERα and HIF-1 pathway. (A) Mammary fat pad tumor growth assessment of HIF-1α–overexpressing MCF7 cells compared with parental MCF7 cells implanted with tamoxifen citrate (5 mg; TAM) or placebo pellet (CTL) when HIF-1α tumors reached 150 mm³ (indicated by red arrow) in NSG female mice with estrogen pellets implanted (0.72 mg; n = 5 per group). NS, nonsignificant; ***P < 0.001. (B) HIF-1α gene bears ERα binding site whose transcription is directly regulated by ERα signaling pathway. Under hypoxia, produced HIF-1α protein is stabilized. Dual responsive genes bear both ERα and HIF-1α binding sites, which can be regulated by two signaling pathways. ERα directly regulates the HIF-1α pathway associated with antiestrogen response in breast cancer.

Discussion

To develop novel therapeutics to treat breast cancer, a deeper understanding of the molecular mechanism of ER-driven cancer is important, as the most common type of metastatic breast cancer is endocrine receptor-positive. The association of ERα-positivity and HIF-1α from clinical studies (12, 15, 34) supports our findings that these two pathways may act in cooperation to promote breast cancer progression. However, the basis of these previous clinical observations was unclear because ER⁻ tumors have a greater proliferation rate and are more hypoxic than ER⁺ tumors. Hence, it was unclear why HIF-1α should be more highly expressed in ER⁺ tumors (15, 34). Our data, which demonstrate that ERα regulates HIF-1α expression, provide a mechanism for these clinical observations. HIF-1α is a labile protein that is rapidly degraded by VHL-mediated proteasomal degradation in normoxia but stabilized in hypoxia. However, the transcriptional regulation of HIF-1α is not well studied, except in one report showing that the NF-κB pathway regulates HIF-1α gene expression (35). Here we show that the HIF-1α gene bears a canonical ER-binding element that responds to estrogen signaling, demonstrating a direct regulatory link between the ERα and HIF-1α pathways in breast cancer (Fig. 6B).

Interestingly, we found a subgroup of genes that are dually responsive to hormone and oxygen (Fig. 1). These genes were up-regulated by hypoxia but the ERα antagonist ICI182780 significantly reduced their expression. Previous studies also show that some genes, such as KDM4B, STC2, and VEGFA, bear both a hypoxia response element and ERE (18, 19, 26–30). This finding may indicate that the ER signaling pathway can be enhanced by HIF-1α induction and vice versa (Fig. 6B). The physiological significance of the cross-talk between these two pathways in breast or ovary development warrants further study. Nevertheless, we envisage that both pathways form a positive feedback loop to enhance the common downstream target gene expression. But is HIF-1 required for ERα activity? We previously showed that depletion of HIF-1α only partially affected KDM4B expression in hypoxia, whereas depletion of ERα nearly abrogated KDM4B expression (18), indicating that HIF-1 might not be required for ERα activity but synergizes with ERα. Here we further tested this by knocking down HIF-1α in tamoxifen-resistant MCF7 and BT474 cells and obtained a similar result that KDM4B expression was only partially reduced (Fig. S7), further suggesting that ERα function may not rely on HIF-1α. However, we cannot exclude the possibility that ERα may require HIF-1α function for activation of certain specific genes or under specific conditions.

Fig. S7.

After 48-h knockdown of HIF-1α and HIF-2α, tamoxifen resistant MCF7 (A) or BT474 (B) cells were incubated in 1% O₂ for 24 h. The indicated proteins were assessed by Western blotting.

Another important conclusion of our findings is that overactive HIF-1α function may partially compensate for estrogen signaling when ERα function is compromised, such as under the circumstances of hormone therapy. When ERα⁺ breast cancer cells were transduced with HIF-1α, the cancer cells became much more resistant to tamoxifen and ICI182780 treatment (Fig. 5). This finding is consistent with the fact that the HIF-1α and hypoxia gene signature were correlated with poorer survival in response to hormone therapy. Although the molecular mechanism by which HIF-1α confers tamoxifen resistance needs to be further defined, our recent findings indicate that the histone demethylase KDM4B is important in coordinating HIF-1α and ERα. KDM4B is a direct target of both HIF-1α and ERα and regulates expression of many genes in normoxia and hypoxia and cell cycle progression (18). Thus, HIF-1α may drive gene expression of KDM4B and other genes to compensate for tamoxifen inhibition of ERα signaling. This may be more important in vivo as the common factors are secreted extracellular signaling molecules. These results complement recent data showing the importance of HIF-1α in triple-negative breast cancer that is driven by XBP1 in response to unfolded protein or endoplasmic reticulum stress (36). Interestingly, XBP1 has been shown to confer both estrogen independence and antiestrogen resistance in breast cancer cell lines (37). Thus, it is also possible that cellular stress-induced XBP1 might interact with HIF-1α to confer antiestrogen resistance. Therefore, our data suggest that targeting the HIF-1α signaling pathway might increase efficacy of endocrine therapy in breast cancers.

Materials and Methods

Cells were maintained in DMEM supplemented with 10% (vol/vol) FCS, 1% glutamine, and 1% penicillin-streptomycin. Western blot analysis, quantitative RT-PCR, tumorsphere formation, ChIP, retroviruses and plasmids, siRNA transfection, cell viability and colony formation assays, luciferase reporter assay, gene-expression profiling, data mining, patient details, and gene-expression profiling for patients sample, and in vivo xenograft experiment are described in SI Materials and Methods. Statistical analyses were two-tailed t tests, with P ≤ 0.05 considered statistically significant. All animal procedures were approved by University of Oxford ethical review and work was conducted in accordance with the UK Home Office guidelines, under the project licence PPL 30/2771.

SI Materials and Methods

Cell Culture and Reagents.

Breast cancer cell lines MCF7, T47D, MDA-MB231, and BT474 were purchased from ATCC. Tamoxifen-resistant MCF7 (Tam-MCF7) was provided by Ian R Nicholson (Tenovus Centre for Cancer Research, Welsh School of Pharmacy, Cardiff University, Cardiff, United Kingdom), which was maintained in phenol-red–free RPMI medium containing 5% charcoal-stripped steroid-depleted FBS, penicillin (50 IU/mL) and streptomycin sulfate (50 μg/mL), and 4-hydroxytamoxifen (100 nM in ethanol). Other cell lines were cultured in DMEM, supplemented with 10% FBS, penicillin (50 IU/mL), and streptomycin sulfate (50 μg/mL). For hypoxia incubation, cells were exposed to hypoxic conditions (1% O₂, 5% CO₂, and 94% N₂) in a Heto-Holten CellHouse 170 incubator (RS Biotech). 17β-estradiol (E2) was purchased from Calbiochem. ICI182780 was purchased from Tocris Bioscience. Desferrioxamine (DFO), 4-hydroxy-tamoxifen, and DMOG were purchased from Sigma. RITA was purchased from Selleck Chemicals. Anti-ERα antibody was purchased from Santa Cruz Biotechnology. Anti–HIF-1α was purchased from BD Transduction Laboratories or Novus Biologicals. Anti-Actin was purchased from Sigma.

Tumorsphere Formation Assay.

Tumorsphere formation assay was performed according to the instructions of the manufacturer (StemCell). Briefly, cells were trypsinized after washing twice with HBSS. Next, 50,000 single cells were cultured with MammoCult Human Medium containing Proliferation Supplement, 4 μg/mL of heparin, and 0.48 μg/mL of hydrocortisone in an ultra-low attachment six-well plate (Corning).

Retroviral Vector.

Full-length HIF-1α or HIF-2α cDNA was cloned into the pLZRS-IRES-GFP retroviral expression vector. HIF-1α, HIF-2α, and empty vector control retroviruses were produced by transfection of the retroviral vector into Phoenix amphotropic packaging cells. More than 90% of tumor cells expressed GFP after three continuous infections were achieved for all cell lines and the pooled cells were used for all experiments. Alternatively, HA-HIF-1α-pBabe-Puro (Addgene#19365) and HA-HIF-2α-pBabe-Puro (Addgene#26055) were packaged into retroviruses by the St. Jude Vector Laboratory. Cells were maintained in 1 µg of puromycin.

siRNA Transfection.

siRNA oligo sequences (HIF-1α, 5-UCAAGUUGCUGGUCAUCAG; HIF-2α, 5-ACUGCUAUCAAAGAUGCUG-3; ESR1#1, GAGAAGUAUUCAAGGACAU; ESR1#2, AAUGAUGAAAGGUGGGAUA; Luciferase, CUUACGCUGAGUACUUCGA) were synthesized by Dharmacon. The SMARTpool siRNA oligos for ERα and nontargeting control were purchased from Dharmacon. siRNAs were transfected into subconfluent cells using HiPerfect transfection reagent (Qiagen) according to the manufacturer’s instructions. Alternatively, reverse transfection with RNAiMAX (Invitrogen) was used according to the manufacturer’s instructions.

Plasmids and Transfection.

pHEO-ERα plasmid was a gift from P. Chambon, Institute for Genetics and Cellular and Molecular Biology, CNRS UMR, France. The pGL3-luciferase reporter vector was purchased from Promega. The ERE fragment and the corresponding mutant were cloned into pGL3-luciferase reporter vector at KpnI and XhoI sites and validated with Sanger sequencing by the St. Jude Hartwell Center (see validated sequence in Fig. S2). To transfect plasmids into cells, FuGene6 transfection reagent (Roche) or Lipofectamine 3000 (Life Technologies) was used according to the manufacturer’s protocol.

Western Blotting.

We performed Western blotting as described in ref. 18.

Cell Viability and Colony Formation Assays.

For cell viability assay, CellTiter 96 AQueous One Solution Cell Proliferation Assay (MTS) (Promega) or CellTiter Glo (Promega) was used according to the manufacturer’s instructions. For the colony formation assay, cells were incubated in the presence of 100 nM of 4-hydroxy-tamoxifen. The colonies were stained with 0.5% Crystal violet and scored and counted.

RNA Extraction and RT-PCR.

RNA was extracted using RNeasy Mini Kit from Qiagen and miRNA was extracted using miVana kit from Life Technologies. RT-PCR was performed using Applied Biosystems 7500 Real-Time PCR system. The results were analyzed using ΔΔCt methods. Primers sequences can be provided upon request.

ChIP.

MCF7 cells were starved for 3 d in phenol-free media with charcoal-stripped serum. Cells were then treated with 100 nM of tamoxifen or 17-β-estradiol for 8 h. Cells were fixed in 2% formaldehyde and performed with ChIP according to the manufacturer’s protocol (Millipore). ERα and HDAC1 antibody for ChIP was purchased from Santa Cruz Biotechnology. PCR primer sequences for ER element were: Forward, 5-ATTGAATGTTTGCTGGAACG -3; Reverse, 5-TTCCATAAGAGAATTCAGTTACTGTTC-3. Negative binding site 1 (NB1) were: Forward, 5-ATCGCTCAAGCCTTCTTTGT-3; Reverse, 5-TAGTCTGCAAAGGCCAGATG-3. Negative binding site 2 (NB2) were: Forward, 5-AGTAAATGGCAGAGCTTGGG-3; Reverse, 5-GGTCCTGCTGACAGAGTGAA-3.

Luciferase Reporter Assay.

MCF7 cells were seeded in antibiotics free DMEM media. Twenty-four hours later cells were transfected with 5 μg of pGL3-HIFER (wild-type) and pGL3-HIFER (mutant) together with 100 ng of pRL Renilla Luciferase Control Reporter Vector using lipofectamine 3000 (Life Technologies), according to the manufacturer’s protocol. Three days later after transfection, dual luciferase assay (Promega) was performed according to the manufacturer’s protocol.

Gene-Expression Profiling.

RNA was extracted from MCF7 cells using RNeasy mini kit (Qiagen). Biotinylated cRNA were generated following manufacturer’s protocols using the Illumina TotalPrep96 Amplification Kit (Ambion # 4393543). Biotin-labeled, amplified cRNA (1,500 ng per array) was hybridized to the Illumina HumanWG-6_V3 BeadChip according to the Manufacturer’s instructions (Illumina). Arrays were scanned with an Illumina BeadArray Reader using BeadScan 3.5.49.29917 software and direct hybridization assay, according to the Manufacturer’s instructions. Array data processing and analysis was performed using Illumina BeadStudio 3.2.6 software using the average chip normalization method. Data analysis was performed using R. The genes engaged in hypoxia and ICI182780 were analyzed by Cluster (38) and the heatmap was shown by Treeview. The GEO accession number for the microarray data is GSE61799.

Data Mining.

Kaplan–Meier analysis of breast cancer (kmplot.com/analysis/) was performed as described previously (39). For the LINCS query, on-line programs were performed by following the instructions at eh3.uc.edu/GenomicsPortals/viewiLincs.jsp and www.maayanlab.net/LINCS/LCB.

For ChIP-seq analysis, RefSeq gene annotation was downloaded from the University of California, Santa Cruz (hg18). The gene regions were extended upstream to include 10-kb promoter regions, and then overlap analysis was done with the 48,007 ERα peaks reported (GSE25021). A total of 11,525 RefSeq genes were bound by ERα either in the gene body or within 10-kb upstream of the transcription start site. These genes were further compared with the 356 genes bound by HIF-1α (GSE28352), and 202 (57%) genes were identified as the common genes bound by both HIF-1α and ERα. KEGG pathway analysis for the overlapping genes was done using DAVID Bioinformatics Resources (david.ncifcrf.gov/).

Patient Details and Gene-Expression Profiling for Patient Samples.

This patient series is part of a published series (40) and a detailed demographic table has been reported previously (41). It has complete 7 y of follow-up for all but four patients, and the median follow-up time for patients in the study alive and without a relapse was 12 y. Total RNA was isolated by TRIzol method (Invitrogen) according to the manufacturer’s instructions. mRNA expression was measured using Affymetrix U133 arrays. RNA was amplified using Ambion Illumina Amplification Kit. Methods for both protocols have been previously described (32). Expression data were preprocessed using gcrma and quantile normalized in Bioconductor (www.bioconductor.org). Log₂ data were used. A previously derived hypoxia signature (32) was mapped to these arrays using the original Affymetrix IDs; signature score in each sample was calculated as previously reported (32).

Xenograft and Tamoxifen Treatment.

For xenograft and tamoxifen treatment, 50 μL of matrigel containing 5 × 10⁶ MCF7 ± HIF-1αwere implanted into the mammary fat pad of NSG female mice, aged 5–6 wk, implanted with estrogen (0.72 mg; 90-d release; Innovative Research of America) or placebo pellet. When tumors reached 150 mm³ mice were randomized (n = 5 per group) and implanted with tamoxifen citrate (5 mg; 90-d release) or placebo pellet. Tumors were regularly measured until endpoint; mice were given pimonidazole (2 mg) 1.5 h intraperitioneally before culling.