FOXA1 upregulation promotes enhancer and transcriptional reprogramming in endocrine-resistant breast cancer

Significance

FOXA1 augmentation, including by genetic aberrations, drives aggressive phenotypes of estrogen receptor-positive (ER⁺) breast cancer (BC). Here, we show that FOXA1 upregulation induces genome-wide enhancer reprogramming and adopts a superenhancer mechanism to activate the master transcription factor HIF-2α and a prometastatic transcriptional program. The hyperactive FOXA1/HIF-2α transcriptional axis is observed to be largely nonconcurrent with the ESR1 mutations in clinical ER⁺/HER2⁻ metastatic BC datasets, suggesting different mechanisms of resistance. Furthermore, a selective HIF-2α inhibitor, currently in clinical trials for advanced renal cell carcinoma and glioblastoma, inhibits the clonogenicity, migration, and invasion of endocrine-resistant BC cells. These findings demonstrate the role of FOXA1 upregulation in enhancer reprogramming and a therapeutic approach of targeting deregulated transcriptional programs to circumvent endocrine-resistant metastatic BC.

Abstract

Forkhead box A1 (FOXA1) is a pioneer factor that facilitates chromatin binding and function of lineage-specific and oncogenic transcription factors. Hyperactive FOXA1 signaling due to gene amplification or overexpression has been reported in estrogen receptor-positive (ER⁺) endocrine-resistant metastatic breast cancer. However, the molecular mechanisms by which FOXA1 up-regulation promotes these processes and the key downstream targets of the FOXA1 oncogenic network remain elusive. Here, we demonstrate that FOXA1 overexpression in ER⁺ breast cancer cells drives genome-wide enhancer reprogramming to activate prometastatic transcriptional programs. Up-regulated FOXA1 employs superenhancers (SEs) to synchronize transcriptional reprogramming in endocrine-resistant breast cancer cells, reflecting an early embryonic development process. We identify the hypoxia-inducible transcription factor hypoxia-inducible factor-2α (HIF-2α) as the top high FOXA1-induced SE target, mediating the impact of high FOXA1 in activating prometastatic gene sets and pathways associated with poor clinical outcome. Using clinical ER⁺/HER2⁻ metastatic breast cancer datasets, we show that the aberrant FOXA1/HIF-2α transcriptional axis is largely nonconcurrent with the ESR1 mutations, suggesting different mechanisms of endocrine resistance and treatment strategies. We further demonstrate the selective efficacy of an HIF-2α antagonist, currently in clinical trials for advanced kidney cancer and recurrent glioblastoma, in reducing the clonogenicity, migration, and invasion of endocrine-resistant breast cancer cells expressing high FOXA1. Our study has uncovered high FOXA1-induced enhancer reprogramming and HIF-2α–dependent transcriptional programs as vulnerable targets for treating endocrine-resistant and metastatic breast cancer.

Resistance to endocrine therapy in estrogen receptor-positive (ER⁺) breast cancer (BC) is common, and leads to poor clinical outcome (1). It has been shown that when ER is inhibited, tumors may activate growth factor receptor (GFR)-related pathways to drive endocrine resistance (2–4). However, with the exception of the mammalian target of rapamycin inhibitor everolimus (5), CDK4/6 inhibitors (6), and the PI3K-α isoform-specific inhibitor alpelisib (7), results of clinical trials using kinase inhibitors targeting the GFR-related pathways, especially as single agents, are mostly disappointing. Recurrent ESR1 mutations, observed in ∼30% of ER⁺ metastatic BCs (MBCs), especially those treated with aromatase inhibitors (AIs), are recognized as an important mechanism of endocrine resistance, but only in a subset of ER⁺ tumors (8). Other molecular mechanisms underlying endocrine resistance in the metastatic disease are still poorly understood.

FOXA1 is a transcription factor (TF) of the Forkhead box (FOX) protein family. It functions as a pioneer factor that binds to condensed chromatin to facilitate subsequent binding of ER (9) and other lineage-specific TFs. By characterizing multiple endocrine-resistant preclinical BC cell models, we have recently shown that high FOXA1 (H-FOXA1), via gene amplification and overexpression (OE), plays a key role in promoting endocrine-resistant cell growth and invasiveness by reprogramming the ER-dependent transcriptome (10). In addition, several clinical sequencing studies of ER⁺ disease reported that about 6% of primary and 10% of metastatic tumors harbor FOXA1 genetic aberrations, including gene amplification and missense mutations associated with FOXA1 activation (11, 12). A recent deep-sequencing study of the regulatory regions in primary breast tumors further revealed recurrent mutations at the FOXA1 promoter, resulting in high binding affinity for the E2F TF and increased FOXA1 gene transcription (13). FOXA1 up-regulation, due to increased expression and activity, has been reported in other metastatic tumors—including esophagus, lung, thyroid, and prostate (14–16)—suggesting potential shared transcriptional programs imposed by H-FOXA1 during malignant disease progression.

FOXA1 binds to enhancers enriched in histone H3 lysine 4 mono/di-methylation (H3K4me1/me2) (17), where FOXA1 can further recruit histone methyltransferase (18) and enhance the hormone-driven ER activity in ER⁺ BC cells (19). This chromatin remodeling activity of FOXA1 is reminiscent of its role in inducing tissue-specific gene expression during the development and differentiation of liver, lung, kidney, pancreas, prostate, and mammary gland (20). A recent study of the ER⁺ BC epigenome revealed that the enhancer expansion in metastatic tumors is linked to FOXA1 and its network activation (21). However, the mechanism by which H-FOXA1 impacts global enhancers to promote an endocrine-resistant and metastatic phenotype is not clear. Furthermore, the key downstream mediators of H-FOXA1 signaling that could serve as therapeutic targets remain to be identified.

In this study, we integrated FOXA1 cistrome, epigenetic histone marks, and transcriptomic data from ER⁺ BC cell models expressing H-FOXA1 at the time of acquired or induced endocrine resistance, and characterized its role in driving BC enhancer reprogramming to activate prometastatic transcriptional programs. We identified the hypoxia-inducible transcription factor-2α (HIF-2α) as the top H-FOXA1–driven superenhancer (SE) target mediating H-FOXA1–induced transcriptional reprogramming in endocrine-resistant BC models. Importantly, we show that targeting HIF-2α by a selective small-molecule inhibitor, currently in clinical trial for advanced renal cell carcinoma, leads to significant reduction in clonogenicity, migration, and invasion of endocrine-resistant cells. Our study proposes a therapeutic approach, via blockade of SE-targeted TFs and aberrant transcriptional programs, to circumvent endocrine-resistant metastatic BC and possibly other types of aggressive cancers expressing H-FOXA1.

Results

FOXA1 OE Induces Enhancer Reprogramming in ER⁺ BC Cells.

To elucidate the impact of H-FOXA1 on genome-wide enhancers in ER⁺ BC, we first used a doxycycline (Dox)-inducible OE system in parental (P) endocrine-sensitive MCF7L cells, to achieve FOXA1 OE comparable to the high levels found in the MCF7L tamoxifen-resistant (TamR) derivative due to endogenous FOXA1 gene amplification (SI Appendix, Fig. S1A). We hypothesized that ectopic FOXA1 OE in P cells alters the enhancer landscape to activate transcriptional programs associated with multiple oncogenic pathways, such as GFR and integrin-related signaling (10), thereby promoting tumor progression. Upon FOXA1 induction, FOXA1 chromatin-immunoprecipitation and sequencing (ChIP-seq) revealed a substantial increase of 60,543 FOXA1 binding sites, while 12,721 preexisting sites remained but 15,210 were lost (Fig. 1A). The increased FOXA1 binding mainly occurred at intronic and intergenic regions (SI Appendix, Fig. S1B), similar to that observed for endogenous FOXA1 binding in both MCF7-P and TamR cells (10, 17, 22).

Fig. 1.

H-FOXA1 reprograms genome-wide enhancers in ER⁺ BC cells. (A) Heatmap representation of FOXA1 binding intensity based on ChIP-seq reads in MCF7L-P cells with FOXA1 OE (+Dox, duplicates) vs. non-OE (−Dox). Signals within 5 kb around the center of binding peaks are displayed in a descending order for each clustered FOXA1 binding event (i.e., lost, shared, and gained upon +Dox vs. −Dox). (Right) Plots of average signal of FOXA1 binding at the clustered regions. (B) Volcano plot of H3K27ac ChIP-seq of MCF7L-P cells with or without FOXA1 OE. The orange and green dots correspond to the regions with GAIN and LOSS H3K27ac in +Dox vs. −Dox cells, respectively. (C) Volcano plot of H3K4me1 ChIP-seq of MCF7L-P cells with or without FOXA1 OE. The purple and cyan dots correspond to the regions with GAIN and LOSS H3K4me1 in +Dox vs. −Dox cells, respectively. The threshold calling for GAIN/LOSS is set as fold-change > 4 and P < 1e-10. (D) Heatmap of FOXA1 binding signal over 5 kb around the center of H3K27ac GAIN and LOSS regions in +Dox vs. −Dox cells. (Right) Plots of average signal intensity of FOXA1 binding at these regions. (E) Same data representation for FOXA1 binding at H3K4me1 GAIN and LOSS regions as in D.

To determine how the enhancer landscape evolves upon H-FOXA1 induction, we next performed ChIP-seq of the two enhancer marks, H3K27 acetylation (ac) and H3K4me1. Differential peak analysis identified more regions with increased (GAIN) than with decreased (LOSS) H3K27ac (5,010 vs. 1,722) (Fig. 1B). Additionally, substantially more GAIN than LOSS regions of H3K4me1 (7,516 vs. 68) were identified in +Dox vs. −Dox cells (Fig. 1C). These findings suggest that ectopic FOXA1 OE in ER⁺ BC cells promotes an overall chromatin state amenable for gene transcription. To further assess the occupancy of H-FOXA1 at enhancers, we overlaid the FOXA1 cistrome signal with the altered enhancer histone marks. This analysis showed concordant increase and decrease in FOXA1 binding at the H3K27ac GAIN and LOSS regions, respectively (Fig. 1D). H3K4me1 GAIN regions were also enriched for increased FOXA1 binding (Fig. 1E). Reciprocally, substantially increased enhancer marks were observed at the FOXA1 GAIN regions, especially those with high-intensity FOXA1 peaks (SI Appendix, Fig. S2A), of which 15% and 33% were marked with H3K27ac and H3K4me1, respectively (SI Appendix, Fig. S2 B and C). These data suggest that in this inducible FOXA1 OE cell model, only a subset of the gained FOXA1 binding sites is fully engaged in enhancer activation, which could possibly be attributed to the insufficiency in additional essential epigenetic modulators, beyond FOXA1.

H-FOXA1–Induced Enhancer Reprogramming Coordinates Gene-Expression Profiles Enriched for Developmental and Prometastatic Processes.

We focused our subsequent analysis on enhancer GAIN regions, which represent the majority of the events during enhancer reprogramming induced by H-FOXA1. More than 90% of GAIN regions of H3K27ac and H3K4me1 were outside promoter regions (Fig. 2 A and B). It has been shown that the chromatin state at enhancers, rather than at promoters, strongly correlates with cell-type–specific gene-expression programs (23). To assess the impact of H-FOXA1–induced enhancer reprogramming on gene expression, we integrated these histone ChIP-seq data with our previously obtained RNA sequencing (RNA-seq) data in this cell model (10). The overall density of H3K27ac, over 2 to 10 kb from a transcription start site (TSS) per gene, was coordinately increased or decreased in the up-regulated (UP) or down-regulated (DN) gene sets induced by FOXA1 OE, respectively (Fig. 2C). A similar correlation was also observed for H3K4me1 around the differentially expressed genes (Fig. 2D). The partial overlap of these 2 histone marks at GAIN enhancer regions upon FOXA1 OE (Fig. 2E) recalls the fact that, in developmental model systems, the most open enhancer regions are marked by H3K4me1, but only active enhancers carry both H3K4me1 and H3K27ac (24). We therefore annotated the differentially expressed genes upon FOXA1 OE with the distribution of the enhancers’ marks. As expected, the UP genes were enriched for GAIN enhancers, especially those marked with both H3K27ac and H3K4me1 (Fig. 2F). In contrast, compared to the genes with no change (NC) in expression, the DN genes upon FOXA1 OE were less enriched for the enhancers marked with H3K27ac. Enhancers marked with H3K4me1 alone were equally enriched in the DN and NC genes. It is possible that for those NC genes harboring H3K4me1-only enhancers, additional TFs are needed to fully activate the so-called poised or predetermined enhancers (24), similar to what is seen in embryonic gene transcription.

Fig. 2.

H-FOXA1–induced enhancer reprogramming coordinates gene expression toward developmental and prometastatic transcriptional programs. (A) Pie chart showing the genomic annotations of H-FOXA1–induced H3K27ac GAIN regions according to the location of a given peak. (B) The same pie chart as in A for H3K4me1 GAIN regions. (C) Violin plots showing the distribution of UP and DN genes (|G-fold| > 2) upon H-FOXA1 with average H3K27ac signal over 2 to 10 kb from the TSS of each gene in +Dox vs. −Dox cells. The H3K27ac signal was calculated over a 2-kb range centered on each peak and normalized by a total of 10 million reads. (D) Same violin plots as in C for the altered genes with differential H3K4me1. P value was calculated by a paired Student’s t test. (E) Venn diagram representing overlap between H-FOXA1–induced H3K27ac GAIN and H3K4me1 GAIN regions in +Dox vs. −Dox cells. (F) Bar charts showing the proportion of genes within each of the categories of H-FOXA1–induced altered (UP or DN) and NC genes that overlap with the genes near enhancers (within ±100 kb from TSS) demarcated with GAIN H3K27ac and/or H3K4me1. P value was calculated by Bonferroni-corrected multiple pairwise comparisons with a Fisher’s exact test. *P < 0.05, ***P < 0.001. (G) Bar charts showing enriched GO terms for the genes predicted from the GAIN enhancers marked by both H3K27ac and H3K4me1 (n = 1,024). (H) Bar charts showing enriched GO terms for the FOXA1-induced UP genes with GAIN enhancers marked by both H3K27ac and H3K4me1 (n = 253). Dashed gray line indicates P = 0.05.

Gene ontology (GO) analysis (25) showed that the genes harboring the GAIN enhancers with both H3K27ac and H3K4me1 (n = 1,024) were enriched in proproliferation, antiapoptosis, and developmental signaling, such as “utero embryonic development” (P = 0.002), “TGF-β receptor signaling” (P = 0.003), and “neuron migration” (P = 0.024) (Fig. 2G). Most of the LOSS enhancers were marked with H3K27ac before FOXA1 OE (SI Appendix, Fig. S3A). The top GO term enriched for the genes harboring the LOSS enhancers was “regulation of Rho protein signal transduction” (SI Appendix, Fig. S3B). Interestingly, one of these GO term genes that was DN upon FOXA1 OE was the gene encoding DLC1 (SI Appendix, Fig. S3C), a Rho-GAP protein that was recently reported to play an estrogen-induced tumor-suppressor role (26). This finding is also in line with the reduced classic estrogen-regulated ER signaling in H-FOXA1–expressing cells, as we have previously reported (10). Focusing on the GAIN enhancer genes UP by FOXA1 OE (n = 253), GO analysis uncovered the enrichment of prometastatic processes, including “regulation of cell migration” (P = 0.002) and “membrane invagination” (P = 0.003) (Fig. 2H), which share the same characteristics of cellular morphogenesis during embryonic development. A recent study has shown that H-FOXA1 drives aberrant enhancers with increased H3K27ac to promote pancreatic cancer metastasis by activating an embryonic endoderm transcriptional program (27). Interestingly, the 253 H-FOXA1–induced genes harboring GAIN enhancers in our model were also highly enriched in the gene-expression profiles of embryonic foregut endoderm as well as the reported metastatic pancreatic cancer cell model expressing H-FOXA1 (27) (SI Appendix, Fig. S4 A and B). In addition, we analyzed the sequencing data from a large clinical MBC cohort (28) focusing mainly on the ER⁺ subtype. We found that the FOXA1 and ESR1 genetic aberrations (amplification and active mutations) are largely nonconcurrent in ER⁺/HER2⁻ metastatic tumors (n = 781) (SI Appendix, Fig. S4C), suggesting a mechanism of FOXA1 UP, different from that of the ESR1 mutations (29), in promoting ER⁺ BC endocrine resistance and metastasis. Overall, these findings support the role of H-FOXA1 and its induced enhancer reprogramming in activating prometastatic transcriptional programs associated with ER⁺ disease progression.

Endocrine Resistance Driven by FOXA1 Amplification Is Associated with a FOXA1 Cistrome That Correlates with Active Enhancers and Transcriptome.

The observation that ectopic H-FOXA1 in MCF7L-P cells induces enhancer reprogramming to activate prometastatic transcriptional programs suggests that H-FOXA1 might be the driving force for acquired endocrine resistance in ER⁺ disease progression. We therefore used our MCF7L-TamR cell model, harboring endogenous FOXA1 amplification and expressing H-FOXA1 (10), to further elucidate the correlation between FOXA1 cistrome, active enhancers, and associated gene expression. We performed H3K27ac ChIP-seq and identified twice as many H3K27ac gained (6,043) vs. lost (2,658) regions in TamR vs. P cells (Fig. 3A). Our previous FOXA1 cistrome data revealed an increased number of gained (31,766) vs. lost (15,778) binding sites in MCF7L-TamR vs. P cells (10). Plots of the H3K27ac signal over the 5-kb regions around the centers of the top 1,000 FOXA1 unique peaks in TamR vs. P cells showed a coordinated deposition of differential H3K27ac at these FOXA1 binding sites (Fig. 3B), in accordance with the overall enrichment of gained H3K27ac at the unique FOXA1 occupancy sites in TamR vs. P cells (SI Appendix, Fig. S5A). Conversely, FOXA1 unique binding in TamR and P cells was positively correlated with the TamR-gained and -lost H3K27ac, respectively (SI Appendix, Fig. S5B).

Fig. 3.

H-FOXA1 in acquired TamR cells driven by FOXA1 amplification coordinates active enhancers and an altered transcriptome. (A) Scatter plots showing relative enrichment of H3K27ac ChIP-seq signal in MCF7L-P vs. TamR cells. Peaks differentially enriched between P and TamR cells are shown in green and orange (fold-change > 4 and P < 1e-10). (B) Heatmap of FOXA1 binding and H3K27ac signal over 5 kb around the center of the top 1,000 FOXA1 binding sites with unique peaks observed in P or TamR cells. (Right) Plots of average H3K27ac signal at these regions. P value was calculated by the Wilcoxon matched-pairs signed rank test. (C) Pie chart showing the proportion of TamR-only FOXA1-binding sites without H3K27ac (gray), or with gained (orange) or lost/NC (green) H3K27ac signal. (D) Bar charts showing the proportion of genes within the altered (UP or DN) and NC groups (TamR vs. P) that overlap with the predicted genes with P-only or TamR-only FOXA1 binding, without (Left) or with overlaid H3K27ac (Right). P value was calculated by Bonferroni-corrected multiple pairwise comparisons with a Fisher’s exact test. (E) Enrichment for the AP-1 binding motif in the center (±500 bp) of TamR-only FOXA1 binding sites with gained vs. lost/NC H3K27ac. P value was calculated by the Wilcoxon matched-pairs signed rank test. (F and G) Enrichment for the FOXA1, AP-1, and ERE binding motifs around the center (±500 bp) of FOXA1/H3K27ac-gained enhancers with or without ER binding in TamR cells. Right panels show the top six enriched motifs at these regions. The occurrence of the motifs was normalized to the length and number of peaks. P value was calculated based on the cumulative distribution function of the hypergeometric distribution.

Of the total gained FOXA1 binding sites in TamR vs. P cells, 29% carry gained and 10% carry lost/NC H3K27ac (Fig. 3C). Of note, the unique FOXA1 binding sites marked with H3K27ac upon FOXA1 OE in P cells shared more common regions with the unique FOXA1 binding in TamR vs. P cells (23% vs. 4%, respectively) (SI Appendix, Fig. S5C). Compared to the NC genes, the UP and DN genes in TamR vs. P cells were significantly enriched for the gained FOXA1-binding in TamR and P cells, respectively (Fig. 3D). The great enrichment observed by intersecting the altered transcriptome with the genes associated with FOXA1-bound active enhancers (Fig. 3 D, Right) further supports the enhancer-level activation of FOXA1 in regulating gene expression in endocrine resistance. We and others have previously shown that the AP-1 TF, mediating the alteration of the ER transcriptome, is a key determinant of endocrine resistance (30–32). We found that the AP-1 binding motif resides at a higher frequency at the FOXA1-bound enhancers with gained H3K27ac than those with lost/NC H3K27ac (Fig. 3E). Similar analysis of the unique FOXA1 binding upon FOXA1 OE in P cells also showed that the AP-1 motif was enriched at those regions overlapping with the unique FOXA1 binding in TamR vs. P cells (SI Appendix, Fig. S5D). Since ER remains expressed and functional in the MCF7L-TamR cells (10), we further integrated these FOXA1 enhancer-binding sites with our previously reported ER cistrome in the same model (33). Motif enrichment analysis showed that the AP-1 binding motif was the top enriched motif at the FOXA1/H3K27ac-gained enhancers, irrespective of ER binding (Fig. 3 F and G). Instead of the estrogen response elements (ERE), other motifs (e.g., SIX1, ELF3, and GRHL1) were enriched at the FOXA1/H3K27ac-gained enhancers without ER occupancy, suggesting an interplay between H-FOXA1 and AP-1–involved TF complexes, likely via an ER-independent mechanism, to promote enhancer reprogramming in endocrine resistance.

Activation of SE Formation Is Correlated with Reprogrammed FOXA1 Cistrome in Endocrine Resistance.

Recent studies have shown that aberrant SEs, a cluster of cis-regulatory elements controlling mammalian cell identity by activating lineage-specific transcription (34), underlie the deregulated gene network in several types of cancer (35). To further dissect the enhancer reprogramming in endocrine resistance, we identified SEs based on H3K27ac and annotated the TFs nearby these SEs in MCF7L-P and TamR cells (Fig. 4 A and B). More than 50% of these SEs in TamR cells were demarcated with gained H3K27ac (Fig. 4C). Importantly, 88% (575 of 653) of these SEs with gained H3K27ac in TamR cells were not shared by P cells (Fig. 4D), suggesting that the enhanced H3K27ac observed in TamR cells is associated with global SE reprogramming. Similarly, we also observed that the majority of the SEs with gained H3K27ac in the MCF7L-P cells with ectopic inducible FOXA1 OE were not shared by the P cells without FOXA1 OE (SI Appendix, Fig. S6 A–C). The top enriched motifs at the TamR H3K27ac-gained SEs include those bound by AP-1, the embryonic TFs (e.g., STAT5, SOX9, and SMAD3), and FOXA1 (Fig. 4 C, Right and Dataset S1). These findings are of particular interest because several AP-1 and SOX family members also harbor SEs at their own gene loci in TamR cells (Fig. 4B), potentially forming self-regulatory circuits as shown in embryonic stem cells (ESCs) (34). In agreement with our previous reports (30, 31), the essentiality of AP-1 activity at SEs for endocrine-resistant cell growth was further supported by the significantly higher sensitivity to gene knockdown (KD) of two AP-1 key components (JUN and FOS) in TamR vs. P cells (Fig. 4E).

Fig. 4.

SEs are engaged in H-FOXA1-induced enhancer reprogramming in endocrine resistance. (A and B) Hockey-stick plots showing ranked enhancers based on H3K27ac of MCF7L-P and TamR cells. A total of 1,036 and 1,144 SEs were annotated for the enhancers with a slope value of >1 in P and TamR cells, respectively. SE-targeted TFs are indicated with rank orders. (C) Venn diagram showing the overlap between gained (fold-change > 2) H3K27ac-demarcated enhancers and the SEs annotated in TamR cells. Motif enrichment analysis reveals the top enriched motifs within the overlapping regions of TamR H3K27ac-gained SEs (n = 653). (D) Venn diagram showing the overlap between TamR H3K27ac-gained SEs and the SEs annotated in P cells. (E) Bar charts showing the normalized cell growth rate of P and TamR cells with control (nonspecific, NS), c-Jun, or c-Fos siRNA KD. P value was calculated by an unpaired Student’s t test. *P < 0.05, **P < 0.01. (F) A χ² test of the contingency table counting the overlapping regions between the FOXA1 cistrome (unique and shared binding sites in TamR vs. P cells) and the SEs annotated in P and TamR cells. Pearson standardized residues for each intersection are color-scaled and the size of the circle is proportional to the amount of the contribution to the total χ² score. (G) Bar charts showing the proportion of the SEs that overlap with the unique and shared FOXA1 binding sites in TamR vs. P cells. P value was calculated by Bonferroni-corrected multiple pairwise comparisons with a Fisher’s exact test.

We further examined the correlation of FOXA1 binding and these SEs using the distinct FOXA1 cistrome identified in the same model. FOXA1 unique binding in TamR vs. P cells was significantly enriched at the SEs of TamR vs. P cells (Fig. 4F). However, the observed elevated proportion of the SEs occupied by FOXA1 in TamR vs. P cells (22% vs. 6%) (Fig. 4G) suggests the contribution of H-FOXA1 in the endocrine resistance-associated SE formation. Similarly, gained FOXA1 binding was significantly enriched at the SEs in P cells with ectopic FOXA1 OE (SI Appendix, Fig. S6D). Notably, about 50% of the SEs identified in the P cells with ectopic FOXA1 OE overlapped with the SEs in TamR cells, and shared several enriched motifs, including SMAD3 and STAT5 (SI Appendix, Fig. S6E), signifying the role of H-FOXA1 in promoting SE formation.

HIF-2α Is the Top TF Target of H-FOXA1–Engaged SEs in TamR Cells.

To further determine the impact of H-FOXA1 on SE-induced gene transcription, we overlapped the TamR H3K27ac-gained SEs and the FOXA1 cistrome and identified the potential SE target genes (Fig. 5A). The majority of these genes harbor gained FOXA1 binding in TamR vs. P cells, which is in line with the fact that 74% of these FOXA1-engaged SE target genes were UP in TamR vs. P cells (SI Appendix, Fig. S7A). Importantly, changes in expression of these FOXA1-engaged SE target genes are highly concordant between the TamR cells with endogenous H-FOXA1 and MCF7L-P cells upon ectopic FOXA1 OE (Fig. 5B and SI Appendix, Fig. S7B), supporting the role of SEs in promoting transcriptional reprogramming even in P cells with ectopic FOXA1 OE.

Fig. 5.

HIF-2α is the top SE-targeted TF induced by H-FOXA1 in TamR cells and correlated with FOXA1 expression in ER⁺ tumors. (A) Waterfall plot showing the genes predicted from the TamR H3K27ac-gained SEs ranked by the differential binding signal of FOXA1 at these gained SEs in TamR vs. P cells. The lower panel shows the top genes with high FOXA1 occupancy that are differentially expressed in TamR vs. P (red vs. blue), and in MCF7L-P cells with (+Dox) vs. without (−Dox) FOXA1 OE (orange vs. green). (B) Scatter plot showing the correlation of expression changes of the TamR-gained SE genes from A in TamR vs. P cells and in P cells with (+Dox) vs. without (−Dox) FOXA1 OE. The extent of gained FOXA1 binding at each gene in TamR vs. P cells was color-scaled. P value was calculated using the Pearson correlation coefficient method. (C–E) qRT-PCR of HIF-2α and FOXA1 in three ER⁺ BC cell models and their endocrine-resistant derivatives made resistant to tamoxifen (TamR), estrogen-deprivation (EDR), or fulvestrant (FulR). Data represent mean ± SEM. *P < 0.05, **P < 0.01, ***P < 0.001, two-tailed t test. (F) H3K27ac and FOXA1 ChIP-seq profiles showing the gained SE overlapping with the gained FOXA1 binding at the first intron of the EPAS1 locus (encoding HIF-2α) in TamR vs. P cells. Three guide RNA target sites for the CRISPR/Cas9-based editing are indicated by pointers. (G) Bar plots of EPAS1/HIF-2α mRNA levels (normalized to GAPDH mRNA) in the MCF7L-TamR cells with CRISPR/Cas9-mediated intronic deletion (del.) at sites indicated in F. Cells edited by scrambled guide RNA sequences were used as the negative control. (H) ChIP-qPCR analysis of H3K27ac (shown as the percentage of input) in MCF7L-P/FOXA1 ±Dox cells at the three intronic sites indicated in F. Inset shows the EPAS1/HIF-2α mRNA levels upon FOXA1 OE. Data represent mean ± SEM. *P < 0.05, **P < 0.01, two-tailed t test. (I) Representative immunohistochemistry images of FOXA1 and HIF-2α in one of the ER⁺ tumors integrated in a tissue microarray. (Scale bar, 20 μm.) (J and K) Scatter plots showing the correlation between the Allred scores of FOXA1 and HIF-2α nuclear staining in ER⁺ (n = 89) and ER⁻ (n = 31) breast tumors. P value was calculated using a Spearman’s rank-order correlation test.

We next focused on the EPAS1 gene as the top H-FOXA1–engaged SE target upon endocrine resistance. This gene encodes HIF-2α, and fits our selection criteria for TFs based on the facts that it was: 1) The top SE-targeted TF identified in TamR vs. P cells; 2) the top SE target carrying gained FOXA1 binding in TamR vs. P cells; 3) the most up-regulated SE-targeted gene in TamR vs. P cells; and 4) one of the top genes highly induced by FOXA1 OE in P cells (Fig. 5 A and B). Notably, the concordant EPAS1/HIF-2α induction upon FOXA1 OE in P cells, and reduction upon FOXA1 KD in TamR cells, were not observed for HIF1A, encoding another HIF family member, HIF-1α (SI Appendix, Fig. S7 C–F). In addition, compared to the P cells, HIF-2α was UP in other MCF7L endocrine-resistant derivatives (10, 36), in company with H-FOXA1 (Fig. 5C). Induction of HIF-2α, but not HIF-1α, was also observed in additional T47D and 600MPE endocrine-resistant cell models (10, 36, 37) expressing H-FOXA1 (Fig. 5 D and E and SI Appendix, Fig. S8 A–C). HIF-2α was slightly UP upon short-term tamoxifen treatment in MCF7L-P cells, but was robustly induced in TamR cells (SI Appendix, Fig. S8D). HIF-2α, but not HIF-1α, was also markedly UP in our previously reported MCF7L endocrine-resistant xenograft tumors, especially TamR and fulvestrant-resistant (FulR) (38) (SI Appendix, Fig. S8E). Kaplan–Meier meta-analysis (39) showed that high mRNA levels of HIF-2α, but not HIF-1α, predict poor distant metastasis-free survival of patients with ER⁺ tumors treated with endocrine therapy (SI Appendix, Fig. S9 A and B). In contrast, both HIF-1α and HIF-2α predict poor distant metastasis-free survival in patients with ER⁻ BC (SI Appendix, Fig. S9 C and D).

We then asked whether the associated SE is responsible for activating HIF-2α gene (EPAS1) transcription. Using CRISPR/Cas9-mediated gene editing, we deleted fragments of the SE harboring gained FOXA1 binding sites within the first intron of the EPAS1 gene in TamR cells (Fig. 5F). EPAS1 transcription was reduced by 60% and 30% in the cells with deleted proximal and distal intronic regions, respectively, adjacent to the EPAS1 TSS (Fig. 5G). Conversely, EPAS1/HIF-2α expression was highly UP upon FOXA1 OE in P cells, in line with the fact that H3K27ac was increased at 2 of the 3 EPAS1 SE regions (Fig. 5H). These data support the engagement of H-FOXA1-induced SE in activating EPAS1/HIF-2α gene transcription.

To determine the correlation of FOXA1 and HIF-2α expression in human samples, we performed immunohistochemistry for HIF-2α using tissue microarrays that we had previously stained for FOXA1 (10). We found that ∼78% of the ER⁺ tumors expressed HIF-2α (Fig. 5I). Importantly, the concordance between nuclear staining of FOXA1 and HIF-2α was high in ER⁺ but not in ER⁻ tumors (Fig. 5 J and K). The fact that HIF-2α expression also occurred in ∼26% of ER⁻ tumors with absent or low FOXA1 suggests additional regulatory mechanisms of HIF-2α activation, possibly via high GFR-related signaling (40), as also shown in neuroblastoma (41).

High HIF-2α Signaling Is Associated with Metastatic Traits and Poor Clinical Outcome of ER⁺ BC.

We have previously shown that ectopic FOXA1 OE elicits a wide spectrum of activated GFR-related signaling in multiple ER⁺ BC cell lines (10). We asked whether HIF-2α, as the top H-FOXA1–induced SE target, mediates the impact of H-FOXA1 on cellular signaling pathways. We first used reverse-phase protein arrays (RPPA) on MCF7L-P cells with Dox-inducible HIF-2α OE to identify differentially expressed proteins (Dataset S2). Intersection of the UP proteins upon HIF-2α OE and FOXA1 OE (10) revealed a significant overlap of 29 proteins corresponding to >70% of the UP proteins upon HIF-2α OE in P cells (Fig. 6A). Kyoto Encyclopedia of Genes and Genomes (KEGG) (42)-defined signaling analysis identified the top enriched oncogenic pathways of ERBB2, focal adhesion, and insulin (Fig. 6B), which noticeably were also the top enriched pathways upon FOXA1 OE in multiple ER⁺ BC cell models (10).

Fig. 6.

High HIF-2α signaling is associated with prometastatic traits and poor clinical outcome. (A) Venn diagram showing the overlap of UP proteins (measured by RPPA) between MCF7L-P cells with HIF-2α OE and P cells with FOXA1 OE. (B) Enriched KEGG pathways represented by the commonly UP proteins upon HIF-2α/FOXA1 OE. Significance in enrichment was calculated by a Fisher’s exact test (P = 0.05 is indicated by a dashed line). Inset showing the Western blot of HIF-2α upon ectopic OE. (C) Reads of HIF-2α and HIF-1α mRNA (fpkm) from RNA-seq of MCF7L-TamR cells with HIF-2α KD using two different siRNA sequences. (D) Hierarchical clustering of the differentially expressed genes (FDR < 0.05) in TamR cells with HIF-2α KD compared to the nonspecific (NS) KD control. (E) Bar charts showing enriched GO terms for the 138 DN genes (log₂R < −1) upon HIF-2α KD. Dashed gray line indicates P = 0.05. (F) Normalized enrichment score (NES) of the DN gene set from E in the TCGA ER⁺ breast tumors (n = 594) ranked by high vs. low HIF-2α mRNA levels. (G) Oncomine Concepts Map analysis was used to compare the high HIF-2α GS against the published BC GSs. Circles represent GSs with size proportional to number of genes and color based on enrichment significance (P < 1e-3, odds ratio > 2). Straight lines connect GSs with the line weight proportional to the degree of overlap. GSs sharing similar concepts but from independent studies are grouped together by dashed rectangles. (H) NES of the HIF-2α GS from G in the liver metastases of ER⁺ MBC (n = 70) expressing high vs. low HIF-2α mRNA levels. (I) NES of the HIF-2α GS in the same cohort of ER⁺ MBC stratified by the ESR1 mutant (mut, n = 22) vs. wild-type (wt, n = 48) status.

We next performed RNA-seq on MCF7L-TamR cells with HIF-2α KD using two validated small-interfering RNA (siRNA) sequences, showing no interference on HIF-1α gene expression (Fig. 6C). Overall, 917 and 1,107 genes were commonly DN and UP (false-discovery rate [FDR] < 0.05), respectively (Fig. 6D). Focusing on the top-ranked DN genes (n = 138), we found that the enriched GO terms were predominantly linked to the affected signaling pathways in P cells upon HIF-2α OE as shown above, including those associated with tumor metastatic traits, such as “extracellular matrix” (i.e., TNC, SERPINE1, and FN1), “focal adhesion” (i.e., CD44, ANXA1, and PLAUR), and “angiogenesis” (i.e., VEGFA, EPAS1, and KLF5) (Fig. 6E). Gene set enrichment analysis (GSEA) further showed that this HIF-2α–dependent gene set was significantly enriched in the transcriptomes of The Cancer Genome Atlas (TCGA) ER⁺ breast tumors (11) (Fig. 6F) expressing high HIF-2α, but not high HIF-1α (SI Appendix, Fig. S10A). Importantly, the transcriptional profiles UP in ER⁺ tumors expressing high HIF-2α were not enriched for the reported HIF-1α–dependent gene signature (GS) (43) (SI Appendix, Fig. S10B), suggesting that our HIF-2α–dependent gene set was not associated with HIF-1α signaling. We further defined the clinically relevant HIF-2α GS by selecting the top-ranked 40 genes with significant GSEA scores (Dataset S3). Oncomine Concepts Map analysis (44) revealed that this HIF-2α GS was significantly correlated with the gene sets overrepresented in breast tumors at advanced stages, with high GFR/ERBB2 signaling, or with metastasis (Fig. 6G and Dataset S4).

To further address the relevance of our HIF-2α GS especially in ER⁺ metastatic tumors, we evaluated the HIF-2α GS in an ongoing study of ER⁺ MBC with available transcriptomic profiles. Specifically, we utilized the currently available transcriptome of 70 biopsies exclusively obtained from liver metastases to minimize the divergence in gene expression of normal tissue from the site of the metastatic lesion (29). GSEA showed that our HIF-2α but not HIF-1α–dependent GS was significantly enriched in the ER⁺ liver metastases expressing high HIF-2α (Fig. 6H and SI Appendix, Fig. S10C). In line with the largely nonconcurrent FOXA1 and ESR1 genetic aberrations in MBC (SI Appendix, Fig. S4C), GSEA revealed an inverse correlation of the HIF-2α but not HIF-1α GS with the ESR1 mutations in these ER⁺ liver metastases (Fig. 6I and SI Appendix, Fig. S10D), suggesting potential different signaling exerted by high FOXA1/HIF-2α and ESR1 mutations in promoting endocrine-resistant MBC. Using additional clinical gene-expression profiling datasets (45, 46), we found that both FOXA1 and HIF-2α mRNA levels were significantly increased in cancer lesions from pancreatic metastases and prostate bone metastases, compared to their levels in the primary tumors (SI Appendix, Fig. S11). These findings suggest that high FOXA1/HIF-2α signaling identified in endocrine-resistant MBC is a common mechanism of disease progression, at least in a subset of tumors in other cancer types.

HIF-2α Inhibition Suppresses Aggressive Phenotypes of Endocrine-Resistant BC Cells.

To test whether HIF-2α is a viable druggable target to treat endocrine-resistant BC, we used a specific HIF-2α antagonist, the small-molecule inhibitor PT2385, with proven efficacy and a favorable safety profile, as shown in a recent phase I trial for advanced kidney cancer (47). PT2385 is a first-in-class antagonist that allosterically blocks the heterodimerization of HIF-2α, but not HIF-1α, with ARNT/HIF-1β and inhibits HIF-2α target gene expression (48). Similar to what has been reported in renal cell carcinoma cells (49), short-term treatment with PT2385 using the lower range of the dose tested in renal cell carcinoma models (48) significantly, although only minimally, reduced the growth of both MCF7L-TamR and T47D-TamR cell models, with no effect on P cell growth (SI Appendix, Fig. S12 A and B). In contrast, PT2385 led to robust reduction in clonogenicity of both MCF7L and T47D TamR cell models, while no effect was observed on P cells expressing lower HIF-2α (Fig. 7 A and B). In addition, pharmacologic HIF-2α inhibition also reduced clonogenicity of the MCF7L-FulR, T47D- estrogen deprivation-resistant (EDR), and T47D-FulR models, but not the MCF7L-EDR model (Fig. 7 C and D), suggesting that the antitumor activity of the HIF-2α inhibitor is not specific for TamR. Since our MCF7L-TamR cells are growth-dependent on Tam (33) and responsive to estrogen treatment (Fig. 7E), we further tested the efficacy of PT2385 under different endocrine regimens. Addition of PT2385 to estrogen, ED, or Ful, ±Tam, led to significant reduction in clonal formation of the MCF7L-TamR cells (Fig. 7E). Of note, in the endocrine-sensitive MCF7L-P cells, the addition of PT2385 further increased the antitumor effect in the presence of endocrine treatments (SI Appendix, Fig. S12C). Similarly, combination of PT2385 with Ful reduced clonogenicity of the MCF7L-EDR cells (Fig. 7F), in which the classic estrogen-dependent genes (e.g., PGR and BCL2) still operate (36), suggesting the necessity of cotargeting the HIF-2α and ER pathways in this model. Taken together, these data suggest the model-specific response to the selective HIF-2α inhibitor and signify the importance of combinatory therapy cotargeting the likely nonredundant HIF-2α and ER pathways in endocrine resistance.

Fig. 7.

HIF-2α inhibition suppresses aggressive phenotypes of endocrine-resistant BC cells. (A and B) Quantification of colony formation of MCF7L-P/TamR and T47D-P/TamR cells treated with the HIF-2α antagonist PT2385 at 1 or 5 µM for 14 d. (C and D) Quantification of colony formation of MCF7L-EDR/FulR and T47D-EDR/FulR cells treated with the HIF-2α antagonist PT2385 at 1 or 5 µM for 14 d. (E) Quantification of colony formation of MCF7L-TamR cells treated with estrogen (E2), estrogen-deprivation (ED), or fulvestrant (Ful), each ±Tam and ±PT2385 (5 µM). (F) Quantification of colony formation of MCF7L-EDR cells treated with ED or Ful, each ±PT2385 (5 µM). (G) Quantification of cell migration and invasion of MCF7L-TamR cells treated with PT2385. Data represent mean ± SEM. N.S., nonsignificant, *P < 0.05, **P < 0.01, ***P < 0.001, two-tailed t test. (H) Hierarchical clustering of RPPA data for the altered expression (two-way ANOVA, Tukey’s multiple comparisons, P < 0.05) of proteins (26 DN and 3 UP) in MCF7L-TamR cells treated with PT2385. (Right) Bar charts of altered expression (log₂ ratio) of the corresponding proteins in TamR cells upon FOXA1 KD. (I) Enriched KEGG pathways represented by the DN proteins upon PT2385 treatment. Significance in enrichment was calculated by a Fisher’s exact test (P = 0.05 is indicated by a dashed gray line). (J) A schematic diagram of H-FOXA1–induced enhancer and transcriptional reprogramming. FOXA1 augmentation leads to establishment of SEs and rearrangement of TFs to activate HIF-2α transcription, which in turn induces a prometastatic transcriptional program to mediate the impact of H-FOXA1 to promote ER⁺ BC progression from endocrine-sensitive to endocrine-resistant and metastatic disease. Targeting HIF-2α represents a strategy targeting deregulated transcriptional programs, resulting in blockade of H-FOXA1–induced enhancer (Enh.) reprogramming to circumvent endocrine-resistant and metastatic disease progression. ERE, estrogen response elements.

Since FOXA1 KD reduced the TamR cell migration (10), we asked whether targeting HIF-2α could achieve a similar effect. Indeed, PT2385 showed marked efficacy in reducing cell migration of both MCF7L and T47D TamR models in a dose-dependent manner (Fig. 7G and SI Appendix, Fig. S13A), but had no effect on their less migratory P cell counterparts (SI Appendix, Fig. S13 B and C). The suppression of cell migration in these two TamR cell models treated with PT2385 was recapitulated by HIF-2α KD using two siRNA sequences (SI Appendix, Fig. S13 D and E), supporting the on-target efficacy of the HIF-2α antagonist. Furthermore, the invasiveness of MCF7L-TamR cells was drastically repressed upon either PT2385 treatment or HIF-2α KD (Fig. 7G and SI Appendix, Fig. S13F).

To further explore the alteration in cellular signaling pathways underlying these phenotypic changes, we performed RPPA on MCF7L-TamR cells treated with two doses of PT2385. Pairwise and clustering analysis revealed a dose-dependent effect of PT2385 on proteins that were DN (n = 26) in treated vs. untreated TamR cells (Fig. 7H and Dataset S5). The majority (>50%) of these DN proteins were also DN upon FOXA1 KD in TamR cells (10), including those belonging to GFR-related signaling, such as EGFR and downstream phosphorylated AKT and MAPK (Fig. 7 H, Right). Western blots further confirmed the DN of these proteins in TamR cells upon HIF-2α KD (SI Appendix, Fig. S14A), in line with UP in P cells upon HIF-2α OE (SI Appendix, Fig. S14B), supporting the on-target effect of PT2385. Finally, KEGG pathway analysis showed the significant enrichment of VEGF, ERBB2, and focal adhesion signaling in the proteins DN upon PT2385 treatment (Fig. 7I). Importantly, these same pathways were also enriched in the proteins UP upon HIF-2α or FOXA1 OE in P cells (Fig. 6B) and in the gene set DN upon HIF-2α KD in TamR cells (Fig. 6E), supporting the notion that high HIF-2α and FOXA1 at least partly converge on the same signaling that can be impeded by the specific HIF-2α antagonist.

Discussion

In this study, we characterized the H-FOXA1–induced epigenomic and transcriptional alterations in endocrine resistance using BC cell models with endogenous or exogenous FOXA1 OE. We showed the coordination of H-FOXA1–induced enhancer and transcriptional reprogramming, resulting in gene-expression profiles associated with embryonic development and prometastatic processes. We identified HIF-2α as the top SE-targeted TF induced by H-FOXA1, its role in mediating H-FOXA1 signaling, and its therapeutic potential in treating endocrine-resistant and metastatic BC (Fig. 7J). The effectiveness of a specific HIF-2α antagonist in treating our H-FOXA1–expressing endocrine-resistant models, and our findings of high HIF-2α signaling in endocrine-resistant metastatic BC, highlight the importance of developing HIF-2α–targeted therapy to improve patient outcome.

Recent clinical and preclinical studies by us and others suggest that FOXA1 augmentation, via gene amplification and OE, is a driver for endocrine-resistant and metastatic ER⁺ BC (10, 28). Substantial enrichment of activating FOXA1 genomic aberrations has been recently reported in advanced hormone-resistant prostate cancer using large primary and metastatic cohorts (50, 51). FOXA1 assists and orchestrates lineage-specific TF binding and action in both embryonic development and tumorigenesis (17, 52), the two biological events possibly sharing traits induced by H-FOXA1 signaling. Interestingly, our study shows that H-FOXA1 induces genome-wide enhancer reprogramming in endocrine-resistant BC to promote transcriptional programs associated with early developmental processes. Our findings are in line with what has been shown in metastatic pancreatic cancer, where H-FOXA1 drives a retrograde developmental transition for disease progression and metastasis (27). Both mammary gland development and embryonic endoderm differentiation involve the branching morphogenesis of rapidly proliferating epithelial cells that invade into mesenchymal tissues (53). Such cellular characteristics share commonality with the traits of metastatic cancer cells, possibly explaining the biologic consequences of H-FOXA1–induced enhancer reprogramming by hacking into developmental programs to promote ER⁺ disease progression and metastasis.

The role of H-FOXA1 in enhancer reprogramming, especially via SEs in endocrine resistance, further supports the link between embryonic development and tumorigenesis. Originally identified as large docking sites for lineage-specific TFs in pluripotent ESCs, SEs constitute a cluster of stitched enhancers controlling gene expression to define mammalian cell identity (34). It has been shown that estrogen induces SE formation in MCF7-P cells around the core of the ERE (54). Activation of SEs has also been shown in ER⁺ BC cell models made resistant to AIs (55). Our study using the FOXA1-amplifed TamR cell model reveals a reconfiguration of SEs, highly associated with FOXA1 binding and altered transcriptome, as one of the mechanisms to promote endocrine-resistant and metastatic phenotypes. The engagement of SEs in such epigenetic and transcriptomic reprogramming involves multiple key TFs. Specifically, we found the highly enriched binding motifs of AP-1 and other developmental TFs (e.g., STAT5 and SOX9) at newly established enhancers and SEs in TamR cells. We have recently shown that transcription of the embryonic factor SOX9 is stimulated by the ER–RUNX2 complex in endocrine resistance (33). Interestingly, several master TFs in ESCs are often driven by and also bound to SEs, thus forming a self-regulatory circuit to amplify transcriptional programs (34). This raises the possibility that H-FOXA1 adopts a similar mechanism, as shown in embryonic development, to establish SEs engaged by multiple self-reinforcing TFs and to activate transcriptional reprogramming to promote endocrine resistance and metastasis. Future studies are needed to further advance our understanding of the involvement of 3D chromatin architecture (56) in enhancer reprogramming upon endocrine resistance driven by H-FOXA1 signaling, including by endogenous or exogenous FOXA1 OE and amplification.

FOXA1 up-regulation has been documented in immature/progenitor luminal BC stem cells (57) and in murine transgenic luminal tumors enriched for a luminal progenitor gene signature (58). In metastatic pancreatic cancer, the role of FOXA1 up-regulation in promoting enhancer reprogramming is associated with the progenitor subtype of pancreatic cancer, whereas FOXA1 expression is low in the squamous subtype (27). Increased luminal progenitor/stem cell activity has been documented in endocrine-resistant BC (59). Regardless of the ER binding, the AP-1 binding motif was mostly enriched at the gained FOXA1 enhancers/SEs in TamR cells, in line with our previous findings for the strong dependency of endocrine resistance on AP-1 (31) and the fact that the AP-1 motif was also enriched at the gained enhancers in metastatic pancreatic tumors expressing H-FOXA1 (27). In addition, other TFs, such as SIX1, with an active role in mammary luminal progenitor cells and metastatic tumors (60), were identified as potential components residing at the gained FOXA1 enhancers without ER binding (Fig. 3G), suggesting ER-independent events of H-FOXA1–induced enhancer reprogramming in endocrine resistance. It has recently been reported that up-regulated GATA5 cooperates with H-FOXA1 to activate enhancer reprogramming in metastatic models of pancreatic cancer (27). Interestingly, we have previously reported that the expression of GATA3, a lineage-specific TF in luminal-subtype BC, is substantially decreased in ER⁺ P cells upon FOXA1 OE and in the FOXA1-amplified MCF7-TamR cells (10, 61). The decreasing GATA3, via genetic and epigenetic mechanisms, may also explain our observed loss of a subset of FOXA1 binding sites, as shown in the H-FOXA1–induced FOXA1 cistrome, where GATA3 and FOXA1 conjointly dictate the luminal-lineage gene expression (62, 63). Whether other GATA family members or other lineage-specific TFs are engaged in H-FOXA1–induced enhancer reprogramming is yet to be determined. The interplay between H-FOXA1, AP-1, and other key TFs—in both an ER-dependent and -independent manner, as we have shown at the enhancer level, to drive a luminal progenitor transcriptional program and thereby contribute to endocrine resistance—remains to be further investigated.

Further dissection of H-FOXA1–engaged SEs revealed HIF-2α as the top TF of SE targets to mediate H-FOXA1 signaling in endocrine resistance. Recent studies have illustrated SEs as hot spots of cis-regulatory elements driving the transcription of key oncogenes, such as MYC (64), in tumorigenesis. A link between HIF-2α expression and SE activation was unexpected, as the HIF family member HIF-1α is mainly induced by hypoxia at the protein level via the oxygen-dependent hydroxylase pathway (65). However, we show that HIF-2α up-regulation is concordant with H-FOXA1 across our multiple endocrine-resistant cell models grown under nonhypoxic conditions. Noticeably, high GFR-related signaling, such as EGFR, has been shown to induce HIF-2α expression in endocrine-resistant BC cells (40). Therefore, both H-FOXA1–induced SEs and high GFR-related signaling (10), which also activates FOXA1, may converge to activate HIF-2α transcription and signaling in endocrine-resistant BC. In addition, we found that perturbation of either HIF-2α or FOXA1 in our endocrine-resistant cell models led to a substantial overlapping of concordantly altered GFR-related signaling, such as ERBB2, and prometastatic VEGF and focal adhesion pathways. This suggests that HIF-2α is a key mediator of H-FOXA1 signaling, potentially forming a vicious cycle involving high GFR-related signaling to further accelerate H-FOXA1–induced enhancer reprogramming. Interestingly, in the endocrine-sensitive MCF7L-P cells, PT2385 further enhanced the efficacy of antiestrogen treatments (SI Appendix, Fig. S12C), possibly by counteracting the early partial induction of HIF-2α upon short-term endocrine therapy (SI Appendix, Fig. S8D), as an early adaptive mechanism of resistance. These data are also supported by previous findings showing HIF-2α reduction in ER⁺ BC cells upon estrogen stimulation (66). Nevertheless, at the time of acquired resistance the induction in HIF-2α levels is much greater, partly due to the H-FOXA1–induced SE activation.

We found the HIF-2α–dependent GS to be highly associated with prometastatic processes, with poor outcome of ER⁺ tumors treated with endocrine therapy, and with endocrine-resistant metastatic ER⁺ tumors harboring wild-type but not mutant ESR1. This raises the hypothesis that the ESR1 mutations, mainly found in metastatic post-AI treatment tumors, and the activation of H-FOXA1/HIF-2α signaling represent two different mechanisms in driving ER⁺ disease progression. Interestingly, we found that the HIF-2α antagonist PT2385 inhibited clonogenicity of MCF7L-TamR and FulR cells, but not the EDR cells, which have been shown to be hypersensitive to residual low levels of estrogen (67, 68). In combination with Ful, the efficacy of PT2385 in reducing clonogenicity was enhanced in MCF7L-TamR cells, and was also observed in MCF7L-EDR cells (Fig. 7 E and F). These findings suggest that cotargeting the ER pathway in ER⁺-resistant tumors can enhance the efficacy of anti–HIF-2α or at least sensitize tumors to such treatment. Continuous accrual of clinical cohorts of advanced ER⁺ tumors with sequencing profiles and therapeutic records is therefore essential to test our hypothesis and to guide future HIF-2α–targeted therapy, especially in combination with Ful or new selective ER degraders (36), to prevent or treat endocrine-resistant and metastatic disease. Preferentially targeting SEs or SE-targeted genes, instead of targeting one gene at a time, has been proven to be a powerful way to treat aggressive tumors driven by aberrant transcriptional programs (64). Although FOXA1 is currently undruggable, we have identified HIF-2α as a top H-FOXA1 downstream signaling component, and showed that a specific HIF-2α antagonist, currently in clinical trial for advanced kidney cancer, suppressed the endocrine-resistant metastatic cell propensities, including clonogenicity (especially in the presence of Ful), as well as cell migration and invasion. By characterizing H-FOXA1–induced reprogramming and its top downstream target, our findings support a highly effective and less toxic therapeutic strategy of targeting HIF-2α and its prometastatic transcriptional program, to circumvent endocrine-resistant metastatic BC with H-FOXA1 signaling.

Materials and Methods

Cell Lines.

Endocrine-resistant cell models (MCF7L, T47D, and 600MPE) were established as described previously (10, 36, 37). Cells were maintained in phenol red-free medium supplemented with 1% penicillin/streptomycin/glutamine (Gibco), 10% charcoal-stripped FBS (for EDR), and 100 nM 4-OH-tamoxifen (Sigma) or 100 nM fulvestrant (Sigma) for TamR and FulR cells, respectively. The isogenic MCF7L lines expressing Dox-inducible HIF-2α or FOXA1 were constructed as described previously (10). All of the cells passed authentication before the experiments.

ChIP-Seq.

Cells were grown to 85% confluence in 15-cm plates and cross-linked with 1% formaldehyde (10 min) followed by quenching (125 mM glycine). Cells were washed in cold PBS and harvested by a cell scraper in cold PBS with protease inhibitors (Roche). Cell pellets were resuspended in cytosolic and then nuclear lysis buffer, and sheared at 4 °C using a Bioruptor (Diagenode) for 20 min at high output (30-s on, 30-s off). Sonicated lysates were cleared by centrifuging at 20,000 × g for 10 min and diluted 4 times before preincubation with protein-A/G beads (Santa Cruz) for 30 min at 4 °C. ChIP was performed by overnight incubation at 4 °C with antibody against human FOXA1 (Abcam, ab23738), H3K27ac (Active Motif, #39134), or H3K4me1 (Abcam, ab176877), followed by an additional 1-h incubation with protein-A/G beads. For FOXA1 ChIP-seq, we added spike-in Drosophila melanogaster chromatin along with the antibody against histone variant H2Av (Active Motif, # 61752). Beads were washed consecutively with low- and high-salt wash buffer, once with LiCl wash buffer (20 mM Tris pH 8.0, 1 mM EDTA, 250 mM LiCl, 1% Nonidet P-40, 1% sodium deoxycholate), and once with TE buffer. DNA was eluted in elution buffer (50 mM NaHCO₃ and 1% SDS) and then supplemented with 300 mM NaCl. Cross-links were reversed overnight at 67 °C. RNA was digested at 37 °C with 0.1 mg/mL RNase A for 30 min. DNA was purified with a PCR purification kit (Qiagen).

Indexed libraries were prepared from ChIP DNA using the KAPA Hyper Library Preparation Kit (Kapa Biosystems). Libraries were amplified by 12 cycles of PCR, and then assessed for size distribution using the 4200 TapeStation High Sensitivity D1000 ScreenTape (Agilent Technologies) and quantified using the Qubit dsDNA HS Assay Kit (ThermoFisher). The indexed libraries were multiplexed, 10 libraries per pool. The pool was quantified by qPCR using the KAPA Library Quantification Kit (KAPA Biosystems) and then sequenced on the Illumina NextSeq500 using the high-output 75-bp single-read configuration.

Additional information on ChIP-seq data analyses, ChIP-qPCR, RNA-seq and analyses, GSEA, CRISPR/Cas9-mediated enhancer disruption, qRT-PCR, immunohistochemistry, RNA interference, Kaplan–Meier curves, clonogenic assay, cell migration and invasion assays, RPPA assay, Western blotting, and statistical analyses is provided in SI Appendix, Materials and Methods.