TNFα Signaling Exposes Latent Estrogen Receptor Binding Sites to Alter the Breast Cancer Cell Transcriptome

SUMMARY

The interplay between mitogenic and proinflammatory signaling pathways play key roles in determining the phenotypes and clinical outcomes of breast cancers. Using GRO-seq in MCF-7 cells, we defined the immediate transcriptional effects of crosstalk between estradiol (E2) and TNFα, identifying a large set of target genes whose expression is rapidly altered with combined E2+TNFα treatment, but not with either agent alone. The pleiotropic effects on gene transcription in response to E2+TNFα are orchestrated by extensive remodeling of the ERα enhancer landscape in an NF-κB- and FoxA1-dependent manner. In addition, expression of the de novo and synergistically regulated genes is strongly associated with clinical outcomes in breast cancers. Together, our genomic and molecular analyses indicate that TNFα signaling, acting in pathways culminating in the redistribution of NF-κB and FoxA1 binding sites across the genome, creates latent ERα binding sites that underlie altered patterns of gene expression and clinically relevant cellular responses.

INTRODUCTION

Estrogen signaling through estrogen receptor alpha (ERα), a nuclear transcription factor, has potent mitogenic effects in breast cancer cells. About two thirds of breast cancers are ERα- positive and exhibit estrogen-dependent growth at the time of diagnosis, making them good targets for anti-hormone therapies (Ignatiadis and Sotiriou, 2013). As such, ERα-status serves as an important diagnostic and prognostic indicator in breast cancer patients (Ignatiadis and Sotiriou, 2013). The estrogen signaling pathway exerts its growth-promoting effects by rapidly and transiently altering and expanding the breast cancer cell transcriptome, which leads to the altered expression of a wide variety of coding and non-coding RNAs (Hah et al., 2011; Hah et al., 2013). Estrogen signaling stimulates the binding of ERα to many thousands of sites across the genome that are pre-established and “marked” by pioneer transcription factors, such as FoxA1 or AP2γ (Hurtado et al., 2011; Tan et al., 2011), although access to new sites in the genome may be promoted by stimulation of mitogenic pathways (Ross-Innes et al., 2012). Once bound at these sites, ERα promotes the coordinated recruitment of coregulatory proteins and chromatin-modifying enzymes that establish an active enhancer, leading to chromatin looping and target gene transcription (Foulds et al., 2013; Fullwood et al., 2009; Hah et al., 2013; He et al., 2012; Shlyueva et al., 2014).

In contrast to the potent proliferative effects of estrogens in breast cancers, proinflammatory signaling in the tumor microenvironment can have either proliferative or antiproliferative effects, depending on the tumor context (Ben-Neriah and Karin, 2011; Coussens et al., 2013). Proinflammatory signaling in cancers derives from acute and chronic infections, tumor-infiltrating immune cells, and local adipose tissue, all of which are capable of supplying the tumor microenvironment with bioactive molecules that facilitate angiogenesis, invasion, and metastasis (Baumgarten and Frasor, 2012; Ben-Neriah and Karin, 2011). Consequently, inflammation is now considered a hallmark of cancer (Hanahan and Weinberg, 2011). One of the major integrators of proinflammatory signaling is the transcription factor NF-κB, which is found in almost all cell types and is activated by a variety of chemokines, cytokines, and growth factors, including tumor necrosis factor alpha (TNFα). In the canonical NF-κB pathway, liganded receptors in the cell membrane initiate signaling cascades that ultimately lead to the nuclear translocation of NF-κB subunits (e.g., p65 and p50), which bind to NF-κB enhancer elements that control target gene expression (Ben-Neriah and Karin, 2011; Natoli, 2012).

The interplay between mitogenic and proinflammatory signaling pathways plays key roles in determining the phenotypes and clinical outcomes of breast cancers. The distinct, but interrelated, responses to estrogens and TNFα (i.e., mitogenic versus proinflammatory) have the potential to create novel antagonistic or synergistic responses that can profoundly affect the biology of breast cancers (Baumgarten and Frasor, 2012). A number of recent studies have shown that proinflammatory signaling can lead to more aggressive hormone-dependent breast cancers. For example, constitutive activation of NF-κB is associated with more aggressive ERα-positive breast cancers and the development of resistance to endocrine therapy (Zhou et al., 2005). Furthermore, TNFα-activated NF-κB and estrogen-activated ERα can work together to potentiate the expression of a number of genes associated with proliferation, invasion, and metastasis in breast cancer cells (Frasor et al., 2009; Ross-Innes et al., 2012). As shown in gene-specific studies, crosstalk between the estrogen and TNFα signaling pathways is mediated, in part, by the ability of ERα and NF-κB to physically interact and bind together at enhancers (Kalaitzidis and Gilmore, 2005; Pradhan et al., 2012; Pradhan et al., 2010). However, the functional relationships and outcomes of the crosstalk between NF-κB/proinflammatory signaling and ERα/mitogenic signaling are complex and context-dependent, and they have the potential to promote clinical outcomes that are more favorable (Cunningham and Gilkeson, 2011; Frasor et al., 2009; Ross-Innes et al., 2012).

In differentiated cells, cell state-determining transcription factors, such as ERα, are thought to bind to a predetermined or pre-established enhancer repertoire, ensuring a constrained and biologically appropriate response to stimuli (Ghavi-Helm et al., 2014; Heinz et al., 2010; Hurtado et al., 2011; Shlyueva et al., 2014). How a pre-established enhancer repertoire may be altered by crosstalk with other signaling pathways is not well understood. In this regard, recent studies have identified “latent” enhancers, which are formed and activated de novo in response to cellular signals at genomic sites that are not pre-established (Kaikkonen et al., 2013; Nord et al., 2013; Ostuni et al., 2013). In the studies described herein, we used a variety of genomic, molecular, and biochemical approaches to (1) examine the nature and extent of crosstalk between the estrogen and TNFα signaling pathways in breast cancer cells, (2) determine how this crosstalk alters the ERα cistrome, creating new ERα binding sites and opportunities for gene regulation, and (3) and understand how the de novo transcriptome specifies biological outcomes.

RESULTS

The E2-Regulated Transcriptome and Cellular Mitogenic Responses are Dramatically Altered by TNFα in Breast Cancer Cells

To determine the interplay between mitogenic and proinflammatory signaling in breast cancer cells at the transcriptional level, we monitored the direct transcriptional output in ERα-positive MCF-7 cells treated with 17-β-estradiol (E2), tumor necrosis factor alpha (TNFα), or both E2 + TNFα (Figure S1A) using global run-on coupled with deep sequencing (GRO-seq). By limiting the treatment times to 40 min., when both E2 and TNFα transcriptional responses are maximal for most genes (Danko et al., 2013; Hah et al., 2011; Luo et al., 2014), our analyses capture primary transcriptional outcomes. Across the genome, TNFα-induced proinflammatory signaling and E2-induced mitogenic signaling converged on many genes, including mRNA genes and long non-coding RNA (lncRNA) genes, leading to numerous instances of antagonism or enhancement of gene expression (Figures 1A, 1B, and S1B). Notably, the immediate and extensive E2-regulated gene expression program was dramatically altered by the addition of TNFα, leading to the up- or down-regulation of a new set of genes (de novo or “gained” genes, including 1,660 mRNA genes and 773 lncRNA genes) that were un-affected by either treatment alone (Figure 1B, left; Figure S1C, left). In addition, we observed a group of genes (synergistically-regulated genes, including 2,104 mRNA genes and 649 lncRNA genes) whose E2-mediated up- or down-regulation was synergistically enhanced by TNFα (Figure 1B, right; Figure S1C, right). These responses led to an E2 + TNFα-regulated transcriptome that is much larger than those observed with E2 or TNFα alone. The genes within this unique transcriptome are involved in molecular and cellular processes ranging from signaling and transcription to metabolism, proliferation, and apoptosis (Table S1).

Figure 1. Crosstalk Between Mitogenic and Proinflammatory Signaling in Breast Cancer Cells: Effects on mRNA Expression, Cell Growth, and Breast Cancer Patient Outcomes.

MCF-7 cells were treated with vehicle (Veh or V), estradiol (E or E2), TNFα (T or TNF), or both E2 + TNFα (E+T) and subjected to GRO-seq or cell proliferation assays. The GRO-seq data were analyzed for patterns of gene expression and related to breast cancer patient outcomes.

(A) Heatmap (left) and Venn diagram (right) showing the regulation of all protein coding genes under each treatment condition as revealed by GRO-seq. The changes in expression shown in the heatmap were centered relative to the control (vehicle) condition and scaled relative to the maximum absolute value of the expression change for each gene. Thus, the values range between +1 and -1, where +1 or -1 denote the maximum or minimum value, respectively.

(B) Box plots showing changes in GRO-seq signals for genes that are newly (de novo or gained) or synergistically regulated by E2 + TNFα. Bars marked with different letters are significantly different from each other (Wilcoxon rank sum test, p < 1 × 10^-6). For example, bars labeled with ‘a’ are different from those labeled with ‘b’ or ‘c’, but not from other bars labeled with ‘a’.

(C) MCF-7 cell proliferation assay. Each data point represents the mean ± SEM, n = 3.

(D and E) Kaplan-Meier survival analyses for different breast cancer types using three distinct gene sets as inputs: (1) a proinflammatory gene set defined by gene ontologies (panel D), (2) the E2 + TNFα de novo and synergistically up-regulated gene set (panel E, left and middle), and (3) the E2 + TNFα de novo and synergistically down-regulated gene set (panel E, right). The breast cancer outcome-linked gene expression data were accessed using the Gene Expression-Based Outcome for Breast Cancer Online (GOBO) tool. In some cases, high expression levels of the gene sets are predictive of better outcomes in a particular breast cancer type (e.g., panel D, left; panel E, left), while in others it is predictive of a poor outcome (e.g., panel D, right; panel E, middle).

See also Figure S1 and Table S1.

The altered patterns of E2-regulated gene expression in response to TNFα were associated with corresponding changes in the growth of MCF-7 cells. Specifically, TNFα completely blocked the potent mitogenic response to E2 in a dose-dependent manner (Figures 1C and S1D). The inhibitory effect of TNFα was also observed in two other ERα-positive breast cancer cells lines (i.e., ZR-75-1 and MDA-MB-231; Figure S1E). To determine if the unique E2 + TNFα-regulated transcriptome has a broader clinical significance, we mined breast cancer outcome-linked gene expression data using the Gene Expression-Based Outcome for Breast Cancer Online (GOBO) tool, expressing the outcomes in Kaplan-Meier survival plots. As a reference point, we determined if high expression levels of a proinflammatory gene set (defined by gene ontologies) are associated with good or poor outcomes in breast cancers. As expected given the complex nature of inflammation in breast cancers, we found that high expression levels of this gene set is associated with good outcomes in some breast cancers (Figure 1D, left; Figure S1F) and poor outcomes in others (Figure 1D, right; Figure S1F). Likewise, the expression levels of E2 + TNFα de novo and synergistically regulated gene sets are strong predictors of clinical outcomes, but good versus poor outcomes vary based on the particular breast cancer type (Figures 1E and S1G). For example, high level expression of the E2 + TNFα up-regulated gene set predicts good outcomes in ERα-positive, lymph node-positive breast cancers (Figure 1E, left) and poor outcomes in untreated breast cancers (Figure 1E, middle). Interestingly, the E2 + TNFα de novo and synergistically regulated gene sets are largely non-overlapping with the proinflammatory gene set defined by gene ontologies (Figure S1H). Collectively, these results indicate that the gene expression program induced by co-stimulation with E2 and TNFα is tied to robust biological outcomes that control the growth of breast cancer cells and predict distinct clinical outcomes across a range of breast cancer types. The studies described below are aimed at understanding the molecular mechanisms and the functional interplay between the estrogen and TNFα signaling pathways.

Proinflammatory Signaling Alters the Repertoire of ERα Binding Sites and Unveils New Sites of Molecular Crosstalk

To explore the molecular crosstalk between the estrogen and TNFα signaling pathways at the genomic level, we used ChIP-seq for ERα (a transcription factor mediator of the estrogen signaling pathway) and the p65 subunit of NF-κB (a transcription factor mediator of the TNFα signaling pathway). Although the majority of the E2-induced ERα binding events was unchanged (“maintained”) by cotreatment with TNFα, the cotreatment caused a gain of 1,664 new ERα binding sites and a loss of 1,938 ERα binding sites (Figures 2A and 2B). We confirmed the presence of the gained sites, which were of particular interest given the transcription results in Figure 1A, in another ERα-positive breast cancer cell line, ZR-75-1, by ChIP-qPCR (Figure S2A).

Figure 2. Co-Stimulation of the Estrogen and TNFα Signaling Pathways Alters the ERα Cistrome in MCF-7 Cells.

(A) (Left) Heatmaps of ERα ChIP-seq reads from MCF-7 cells treated with E2, TNFα, or both. (Right) Heatmap of p65 ChIP-seq reads at the genomic loci where ERα is bound.

(B) Box plot of the normalized read counts for three distinct sets of ERα binding sites (gained, maintained, lost). Bars marked with different letters are significantly different (Wilcoxon rank sum test, p < 1 × 10^-6).

(C) De novo motif analyses of the three sets of ERα binding sites noted in (B) were performed using MEME. The predicted motifs were matched to known motifs using TOMTOM.

See also Figure S2.

About half of the gained ERα binding sites were also occupied by NF-κB upon cotreatment with E2 and TNFα, while the lost ERα binding sites generally not occupied by NF-κB (Figures 2A and 2B). Interestingly, TNFα-mediated NF-κB binding was enhanced in the presence of E2 at the gained (and to a lesser extent, maintained) ERα binding sites (Figure 2B, top and middle). These results suggest functional interactions between two signal-regulated transcription factors (i.e., ERα and NF-κB) across the genome. Such observations fit with studies of individual ERα binding sites showing that ERα and NF-κB can physically interact on chromatin, allowing for changes in gene expression (Kalaitzidis and Gilmore, 2005; Pradhan et al., 2012; Pradhan et al., 2010). DNA sequence analyses of the gained, maintained, and lost ERα binding sites showed an expected enrichment of ERα (ESR1), NF-κB (RELA), and FoxA1 (FOXA1) motifs, with the most prevalent motifs at the gained ERα binding sites being RELA and ESR1 (Figure 2C). Similar to what we observed with the ERα cistrome upon treatment with TNFα, we also observed alterations in the TNFα-induced NF-κB cistrome by cotreatment with E2, leading to the gain or loss of NF-κB binding sites across the genome (Figure S2, B and C). Collectively, these results illustrate how distinct signaling pathways converge on nuclear transcription factors to alter their patterns of binding, creating new cistromes and, ultimately, target transcriptomes.

E2-Dependent ERα Binding Sites Gained Upon Cotreatment with TNFα Produce Enhancer RNAs and Are Associated with Target Gene Expression

Signal-regulated enhancers can modulate target gene expression from distal positions in the genome through looping events that bring distal enhancers into contact with proximal promoters through the formation of chromatin loops (de Laat and Duboule, 2013). Recent studies from our lab and others have shown that enhancers with histone modifications indicative of activity, such as H3K4me1 and H3K27ac, are more likely to produce enhancer transcripts (eRNAs) and loop to target gene promoters (Hah et al., 2013; Kieffer-Kwon et al., 2013; Kim et al., 2010). Thus, eRNA production is a good marker of active enhancers that regulate target genes (Hah et al., 2013). To assess whether the E2-dependent ERα binding sites gained upon cotreatment with TNFα are active and functional enhancers, we leveraged the power of GRO-seq to assay enhancer transcription and target gene transcription upon ERα binding (Figure 3A).

Figure 3. ERα Binding Sites Gained by Co-treatment with E2 and TNFα are Positively Correlated with Transcription of the Enhancers and Nearby Target Genes.

(A) Overview of the pipeline for integrating ChIP-seq data with GRO-seq data to link ERα binding with enhancer and target gene transcription in MCF-7 cells.

(B) Box plots showing GRO-seq RPKM values for enhancer transcription and nearest neighbor up-regulated gene transcription at maintained and gained ERα binding sites. Bars marked with different letters are significantly different (Wilcoxon rank sum test, p < 0.001).

(C) Three dimensional (3D) box plots showing the positive relationships among ERα binding (called by ChIP-seq; from Figure 2B), enhancer transcription (called by GRO-seq at the ERα binding sites; from panel B), and nearest gene transcription (called by GRO-seq; from panel B) for the top 50% of ERα binding sites (see Supplemental Experimental Procedures). Solid lines represent the interquartile range; dotted lines represent Q1 - 1.5 × IQR or Q3 + 1.5 × IQR.

(D) Relative displacement (i.e., fold change) in E2 versus E2 + TNFα (left) with associated p-values (right) in each dimension for the 3D box plots shown in panel C. See Figure S3F for the actual median log2 fold change values, as well as the difference (Δ) in median log2 fold change values for E2 versus E2 + TNFα.

See also Figure S3.

First, we assayed enhancer transcription at the gained, maintained, and lost ERα binding sites. We observed that the amount of enhancer transcription in response to vehicle (V), E2 (E), TNFα (T), or E2 + TNFα (E+T) was positively correlated with the extent of ERα (or p65) binding under the same conditions; in general, more enhancer transcription was observed with more ERα binding, as determined by the correlation analyses shown in Figure S3A (see also Figure 3B, middle, compare to the extent of ERα binding in Figure 2B). For example, with the gained ERα binding sites, the levels of enhancer transcription were greatest with the TNFα or E2 + TNFα treatments (Figure 3B, middle bottom), reflecting the occupancy of those sites by p65 (for TNFα alone; Figure 2B, top right) or ERα and p65 (for E2 + TNFα; Figure 2B, top).

Next, we used GRO-seq to determine the levels of transcription at the nearest neighboring gene (upstream or downstream) from each ERα binding site. For the genes located nearest to the gained and maintained ERα binding sites, we observed that the amount of target gene transcription in response to the four treatment conditions was proportional to the extent of enhancer transcription under the same conditions (i.e., in general, more target gene transcription was observed with more enhancer transcription) (Figure 3B). Furthermore, we found that gained ERα binding sites are significantly enriched for neighboring genes whose expression is stimulated by E2 + TNFα treatment (Figure S3B). Lost ERα binding sites, which showed persistent transcription of the nearest neighboring genes with E2 + TNFα treatment in spite of reduced enhancer transcription, were an exception (Figure S3C), suggesting a dissociation of ERα binding from gene expression in the presence of TNFα for this particular set of ERα binding sites. Gene-specific analyses confirmed that the E2 + TNFα-dependent gene regulatory events require ERα (Figure S3, D and E).

Together, these analyses link ERα binding (called by ChIP-seq), enhancer transcription (called by GRO-seq at the ERα binding sites), and nearest gene transcription (called by GRO-seq), which is illustrated further using three-dimensional box plots of the data (Figures 3C, 3D, and S3F). This representation also shows the impact that cotreatment with TNFα has on gained ERα binding sites and their target genes, producing new ERα binding events, as well as stimulating enhancer and target gene transcription (Figure 3, C and D; note the significant displacement in all three dimensions with E2 + TNFα for the gained enhancers; see also Figure S3F). Collectively, these results indicate that the gained ERα binding sites are bona fide enhancers, representing a unique inducible ERα cistrome that controls the expression of a distinct and biologically relevant gene set.

The Classical Estrogen-induced ERα Cistrome is Demarcated Prior to Stimulus, While the De Novo ERα Cistrome Requires TNFα Stimulation for Chromatin Accessibility

Previous models of enhancer function suggested that signal-activated transcription factors, such as ERα, bind within a predetermined repertoire of accessible genomic loci that are established upon cellular differentiation (Ghavi-Helm et al., 2014; Heinz et al., 2010; Hurtado et al., 2011; Shlyueva et al., 2014). In addition, recent studies have identified “latent” enhancers, which are formed and activated de novo in response to cellular signals at genomic sites that are not pre-established (Kaikkonen et al., 2013; Nord et al., 2013; Ostuni et al., 2013). Accessibility at pre-established enhancers is maintained by lineage-specific transcription factors and pioneer transcription factors that function to maintain an open chromatin architecture (Heinz et al., 2010; Hurtado et al., 2011; Shlyueva et al., 2014). In this regard, the Forkhead protein FoxA1, a well established pioneer factor for ERα, is constitutively bound at ERα enhancers prior to estrogen stimulation (Hurtado et al., 2011). Thus, we sought to determine the chromatin state in the basal (i.e., unstimulated) condition at the maintained (i.e., classical) and gained ERα enhancers prior to receptor binding. To do so, we aligned publicly available deep sequencing data sets for several enhancer features (FoxA1, DNase1 accessibility, chromatin marks, and CBP) to the maintained and gained ERα cistrome (Figures 4A and 4B). In untreated MCF-7 cells, the maintained ERα enhancers have an open chromatin architecture (i.e., enhanced DNase1 accessibility), and are enriched in FoxA1 binding and features of active enhancers (e.g., H3K27ac and CBP), prior to stimulation with E2 or TNFα (Figures 4A and 4B). These features suggest that the maintained ERα enhancers are constitutively accessible in these cells under basal conditions, prior to signal-induced ERα binding (Figure 4A and 4B). We observed similar results for the lost ER enhancers (data not shown). In contrast, the gained ERα enhancers have a less accessible (“closed”) chromatin architecture, and are not significantly enriched in FoxA1 binding or features of active enhancers (e.g., H3K27ac and CBP), prior to stimulation with E2 or TNFα (Figure 4A and 4B). Interestingly, treatment with TNFα or TNFα + E2, but not E2 alone, drove NF-κB and FoxA1 binding to the gained ERα enhancers (Figure 4C; see Figure S4, A - D for specific examples). In addition, as discussed in more detail below, we observed some differences in the FoxA1 responses to the treatments depending on the level of NF-κB enrichment at these sites (low versus high; Figure S4, A - C). These results suggest that TNFα or TNFα + E2 is required for the mobilization of NF-κB and FoxA1 to new genomic loci, where they help to create new ERα binding sites that are not part of the classical estrogen-induced ERα cistrome. These results fit well with the gene expression analyses described above. The de novo sites of ERα binding in response to E2 + TNFα in MCF-7 cells have the features of recently described latent enhancers (Heinz et al., 2010; Kaikkonen et al., 2013; Ostuni et al., 2013).

Figure 4. Latent ERα Binding Sites Exposed by Cotreatment with E2 and TNFα Have a Restrictive Chromatin State Prior to Treatment.

(A) Heatmap showing gained and maintained ERα binding sites aligned with genomic data (ChIP-seq and DNase1-seq) for known enhancer features from untreated MCF-7 cells (basal state).

(B) Normalized ChIP-seq or DNase1-seq read counts for known enhancer features at gained versus maintained ERα binding sites in untreated MCF-7 cells. Bars marked with different letters are significantly different (Wilcoxon rank sum test, p < 0.001).

(C) Heatmap (top) and box plots (below) of NF-κB and FoxA1 occupancy at gained ERα binding sites, as determined by ChIP-seq. Bars marked with different letters are significantly different (Wilcoxon rank sum test, p < 0.001).

See also Figure S4.

Establishment of TNFα-Induced Latent ERα Enhancers Requires NF-κB and FoxA1

The TNFα-dependent binding of NF-κB and FoxA1 at latent ERα enhancers suggested that these factors may act to increase chromatin accessibility and promote ERα binding. To determine the roles of NF-κB and FoxA1 at gained/latent ERα enhancers in MCF-7 cells, we separated the enhancers into those with relatively high NF-κB co-occupancy after stimulation and those with relatively low NF-κB co-occupancy (Figure 5A). We then determined the role of NF-κB or FoxA1 at these enhancers using siRNA-mediated knockdown of NF-κB p65 or FoxA1, coupled with ChIP-qPCR (to assay the binding of p65, FoxA1, and ERα) and RT-qPCR (to assay gene expression outcomes for neighboring genes). ERα binding in the presence of E2 + TNFα at gained/latent enhancers with high NF-κB occupancy (e.g., those near POU3F1, EFNA1, and B4GALT1) was dramatically reduced upon p65 knockdown (Figure 5C; Figures 5B and S5A, middle panels), but was unaffected at maintained enhancers (e.g., those near FMN1, P2RY2, and GREB1; Figures 5B and S5A, top panels) or gained/latent enhancers with low NF-κB occupancy (e.g., those near PTGER3, GALNTL4, and PPP1R3C; Figures 5B and S5A, bottom panels). We also observed a corresponding effect of the expression of the nearest neighboring gene in the presence of E2 + TNFα, namely that reduced ERα binding correlated with reduced expression upon p65 knockdown (Figures 5D and S5B, middle panels; compare to the largely unaffected genes in the top and bottom panels). These results indicate that NF-κB is required for ERα binding at a subset of latent/gained enhancers, as well as corresponding target gene expression in response to E2 + TNFα.

Figure 5. NF-κB Helps to Unveil Some Latent ERα Enhancers in MCF-7 Cells After Treatment with TNFα.

(A) Genome browser views of ERα and NF-κB (p65) ChIP-seq data for different types of ERα enhancers.

(B) NF-κB facilitates ERα binding at TNFα-responsive enhancers. ChIP-qPCR for ERα and NF- κB in MCF-7 cells with or without RNAi-mediated knockdown (KD) of p65, followed by treatment with E2 + TNFα. The enhancers are designated by the nearest neighboring gene. Each bar represents the mean + SEM, n = 3. The asterisks indicate significant differences from the corresponding control (Student’s t-test, p-value < 0.05).

(C) Western blot showing knockdown of p65 using siRNAs.

(D) NF-κB is required for the efficient expression of genes neighboring latent E2 + TNFα-responsive ERα enhancers. Target gene expression with or without RNAi-mediated knockdown (KD) of p65, with E2 + TNFα treatment. Each bar represent the mean + SEM, n = 3. The asterisks indicate significant differences from the corresponding control (Student’s t-test, p-value < 0.05).

See also Figure S5.

As noted above, gained/latent enhancers with low NF-κB occupancy showed little dependence on NF-κB for ERα binding. We hypothesized that at these enhancers FoxA1, rather than NF-κB, might play a critical role in establishing a chromatin environment permissive for ERα binding. We tested this in MCF-7 cells using a set of experiments similar to those described above for NF-κB. At maintained ERα enhancers, FoxA1 was bound constitutively (Figure 4A) and as expected, siRNA-mediated knockdown of FoxA1 dramatically reduced ERα binding, as well as the expression of the nearest neighboring gene (Figure 6B; Figures 6A and 6C, top panels; Figures S6A and S6B, top panels). Likewise, at gained/latent enhancers with low NF-κB occupancy, knockdown of FoxA1 dramatically reduced ERα binding, as well as the expression of the nearest neighboring gene (Figures 6A and 6C, bottom panels; Figures S6A and S6B, bottom panels). In contrast, at gained/latent enhancers with high NF-κB occupancy, knockdown of FoxA1 had little effect on ERα binding or target gene expression (Figures 6A and 6C, middle panels; Figures S6A and S6B, middle panels). A similar pattern of FoxA1 dependence was observed for gained ERα enhancers in ZR-75-1 cells (Figures S6, C through F). These results indicate that FoxA1 is required for ERα binding at a subset of latent/gained enhancers with low NF-κB occupancy, as well as corresponding target gene expression in response to E2 + TNFα.

Figure 6. Cytokine-Induced FoxA1 Binding at Latent Enhancers is Required for ERα Binding.

(A) FoxA1 facilitates ERα binding at TNFα-responsive enhancers. ChIP-qPCR for ERα and FoxA1 in MCF-7 cells with or without RNAi-mediated knockdown (KD) of FoxA1, followed by treatment with E2 + TNFα. The enhancers are designated by the nearest neighboring gene. Each bar represent the mean + SEM, n = 3. The asterisks indicate significant differences from the corresponding control (Student’s t-test, p-value < 0.05).

(B) Western blot showing knockdown of FoxA1 using siRNAs.

(C) FoxA1 is required for the efficient expression of genes neighboring latent E2 + TNFα-responsive ERα enhancers. Target gene expression with or without RNAi-mediated knockdown (KD) of FoxA1, with E2 + TNFα treatment. Each bar represent the mean + SEM, n = 3. The asterisks indicate significant differences from the corresponding control (Student’s t-test p-value < 0.05).

(D) ~30% of gained ERα enhancers show significant reductions in ERα binding upon FoxA1 knockdown (“affected”) (FDR < 0.05 using SICER). The results are from ChIP-seq of ERα in MCF-7 cells with or without siRNA-mediated knockdown of FoxA1 and treated with or without E2 + TNFα.

(E) Average fold changes in ERα binding at gained ERα binding sites affected versus unaffected upon FoxA1 knockdown. Bars marked with different letters are significantly different (Wilcoxon rank sum test, p < 0.001).

(F) Box plots showing p65 occupancy at gained ERα binding sites affected versus unaffected upon FoxA1 knockdown. Bars marked with different letters are significantly different (Wilcoxon rank sum test, p < 0.001).

See also Figure S6.

In ChIP-seq assays in MCF-7 cells, we found that ERα binding at ~30% of the gained enhancers was significantly reduced upon siRNA-mediated knockdown of FoxA1 (FDR < 0.05; Figure 6, D and E; affected versus unaffected by FoxA1 knockdown). These results separated the gained/latent ERα enhancers into two groups: (1) FoxA1-dependent (i.e., affected by FoxA1 knockdown) and (2) FoxA1-independent (i.e., unaffected by FoxA1 knockdown). Both groups have similar levels of ERα binding with E2 + TNFα (Figure S6G) and similar levels of nearest neighboring gene transcription (Figure S6H). With these genomic data, we were able to test the hypothesis that gained enhancers with high NF-κB occupancy do not require FoxA1. We found that FoxA1-dependent latent ERα enhancers have lower occupancy of p65 than the FoxA1-independent enhancers, as expected (Figure 6F), and are sensitive to FoxA1 knockdown.

Our results showing loading of FoxA1 at new sites in the genome in response to TNFα treatment point to key links between cellular signaling pathways and the activity of pioneer factors. To explore the connections between TNFα signaling and FoxA1 loading on chromatin in more detail, we biochemically fractionated E2- and TNFα-treated MCF-7 cells into distinct cellular compartments (i.e., cytoplasm, nucleoplasm, and chromatin). As expected, FoxA1 did not localize to the cytoplasm and the treatments (i.e., Veh, E2, TNFα, or E2+ TNFα) did not alter the levels of total FoxA1 protein, as determined by Western blotting for FoxA1 (Figure 7A; see “Whole Cell Extract” and “Cytoplasm”). Also as expected, FoxA1 was bound to chromatin under basal/unstimulated conditions and showed enhanced binding or stabilization upon E2 treatment (Figure 7A; see “Chromatin”). Interestingly, treatment with TNFα, alone or in the presence on E2, resulted in a dramatic increase in chromatin-bound FoxA1 and a corresponding decrease of soluble FoxA1 in the nucleoplasm (Figure 7A; see “Chromatin” versus “Nucleoplasm”). These results indicate that TNFα-treatment can drive more FoxA1 to chromatin, providing new sites of accessible chromatin for the binding of ERα.

Figure 7. Stimulation with TNFα Drives FoxA1 to Chromatin in MCF-7 Cells.

(A) Western blots of different subcellular protein fractions showing a decrease in nucleoplasmic FoxA1 and a corresponding increase in chromatin bound FoxA1 in cells treated with TNFα in the presence or absence of E2. Fraction-specific markers and loading controls include β-tubulin (β-Tub), histone H3 (H3), and SNRP70.

(B) A model for cytokine induced latent ERα enhancers in breast cancer cells, as described in the text.

Taken together, our data suggest a model where the majority of the E2-regulated target genes are controlled by poised ERα enhancers that are located in open regions of chromatin and are maintained by constitutively bound FoxA1 and other lineage-specific transcription factors (Figure 7B, top left). A subset of latent ERα enhancers, however, are located in less accessible regions of the genome and require TNFα signaling to promote NF-κB and FoxA1 binding, leading to enhanced chromatin accessibility and subsequent ERα binding (Figure 7B, top right and bottom).

DISCUSSION

In the studies described herein, we have defined the molecular mechanisms that underlie crosstalk between estrogen-mediated mitogenic signaling and TNFα-mediated proinflammatory signaling in breast cancers, as well as the functional consequences and clinical relevance of this interplay. We have defined the immediate transcriptional effects of crosstalk between E2 and TNFα, identifying a large set of target genes whose expression is rapidly and dramatically altered (up-regulated or down-regulated) with combined E2 + TNFα treatment, but not with either agent alone. The target genes include classical mitogenic and proinflammatory genes, as well as long non-coding RNA genes (Figure 1A and S1B). The pleiotropic effects on gene transcription in response to E2 + TNFα were orchestrated by the extensive remodeling of the ERα enhancer landscape (Figure 2, A and B) in an NF-κB- and FoxA1-dependent manner (Figures 5, 6, S5, and S6). The altered patterns of gene regulation in the presence of TNFα inhibit the estrogen-dependent growth of MCF-7 cells. In addition, high-level expression of the de novo and synergistically regulated genes is associated with good outcomes in some types of breast cancer and poor outcomes in other types of breast cancer. Together, our genomic and molecular analyses indicate that TNFα signaling, acting in pathways culminating in the redistribution of NF-κB and FoxA1 binding sites across the genome, creates new E2-dependent ERα binding sites that underlie altered patterns of gene expression.

TNFα Signaling Exposes Latent ERα Enhancers in Breast Cancer Cells: Requirement for NF-κB and FoxA1

In the “classical” pathway for estrogen-induced gene activation, the estrogen-induced ERα cistrome is demarcated by FoxA1 binding prior to hormone exposure. In these cases, FoxA1 acts as a pioneer factor that facilitates chromatin opening, providing access to the genome for the liganded receptor (Hurtado et al., 2011). Once bound to estrogen response elements, liganded ERα promotes the formation of active enhancers, which includes recruitment of coregulators (e.g., SRCs and p300/CBP, Mediator) (Foulds et al., 2013), histone modifications and additional chromatin opening (He et al., 2012), recruitment of RNA polymerase II and enhancer transcription (Hah et al., 2013; Wang et al., 2011), and looping to target gene promoters (Fullwood et al., 2009).

Herein, we show that the TNFα signaling pathway can redirect the pioneer factor FoxA1 to new sites in the genome that were not established during the ontology of the cell (Figure 7B, right). The interface of the signaling pathway with the transcription factor is evident at the genomic level, with a redistribution of FoxA1 to new loci, creating new opportunities for ERα binding. These results are consistent with a latent enhancer model, in which activation of cellular signaling pathways promotes the sequential binding of stimulus-activated and lineage-determining transcription factors (Heinz et al., 2010; Ostuni et al., 2013). Importantly, de novo or gained ERα binding at these latent enhancers is ligand-dependent; the sites newly demarcated by FoxA1 in response to TNFα cannot be accessed by ERα in the absence of ligand. Thus, in this case, FoxA1 retains its pioneer functions, but they are not determined as a consequence of development or differentiation, but rather as an endpoint of proinflammatory signaling. The TNFα-induced redistribution of FoxA1 may occur in response to posttranslational modification of FoxA1 as an endpoint of TNFα signaling pathway, although further analyses are needed to test this hypothesis.

We also observed that liganded ERα can also access an additional set of latent enhancers demarcated by NF-κB upon TNFα signaling (Figure 7B, bottom left). Recent studies have suggested that proinflammatory signaling, primarily through NF-κB, plays a role in promoting an open chromatin state allowing for cell plasticity in response to environmental cues, however the role of NF-κB as a pioneer factor has been debated (Lee et al., 2012; Molinero et al., 2012; Natoli, 2012; Ndlovu et al., 2009; Rao et al., 2011). Nonetheless, TNFα-mediated NF-κB binding, alone or perhaps in conjunction with other transcription factors, is required for ERα binding at these latent enhancers, contributing to extensive crosstalk between the estrogen and TNFα signaling pathways observed in the cistrome and the transcriptome.

The salient features of the latent ERα enhancers are (1) limited chromatin accessibility (i.e., low FoxA1 occupancy and DNase1 sensitivity) and limited enhancer activity (i.e., low H3K27ac and CBP) prior to stimulation (Figure 4, A and B) and (2) and rapid and robust enhancer activation (as illustrated increased enhancer transcription) and ERα binding after stimulation (Figure 3B). The signal-dependent unveiling of these latent ERα enhancers results in the formation of bona fide active enhancers with dramatic effects on target gene transcription (Figure 3C). The transition from a quiescent to activated chromatin landscape at the latent ERα enhancers is largely mediated by TNFα-induced recruitment of NF-κB or FoxA1 to these sites (Figures 5 and 6).

Distinct Breast Cancer Cell Transcriptomes Are Controlled by Poised Versus Latent ERα Enhancers

Our results show that the majority of E2-regulated target genes in MCF-7 cells are controlled by poised (“maintained”) ERα enhancers that are located in open regions of chromatin and are constitutively bound by FoxA1 (or other lineage-specific transcription factors). The estrogen-dependent expression of the genes targeted by these enhancers is largely unaffected by TNFα signaling (Figure 3C, left). In contract, a smaller set of estrogen-dependent latent ERα enhancers, which are created in response to TNFα signaling and are dependent on the actions of NF-κB and FoxA1, specify a distinct transcriptome that is uniquely responsive to the combined actions of E2 and TNFα (Figure 3C, right). This unique transcriptome, which is created upon the convergence of mitogenic and proinflammatory signaling, contains mRNA and lncRNA genes that are newly or synergistically regulated (either up or down) only upon the combined actions of E2 and TNFα.

The genes within this unique transcriptome, which are involved in molecular and cellular processes ranging from signaling and transcription to metabolism, proliferation, and apoptosis (Table S1), are likely to control the unique cellular outcomes upon combined treatment with E2 and TNFα (e.g., inhibition of the mitogenic growth of MCF-7 cells; Figure 1C). Although it is formally possible that TNFα signaling acts dominantly to E2 signaling to repress cell proliferation through the expression of a TNFα-regulated target gene set, rather than through the unique E2 + TNFα-regulated transcriptome described above, we think this is unlikely for the following reasons. First, the unique E2 + TNFα-regulated gene set is a strong predictor of clinical outcomes in breast cancers, independent of the classical TNFα-regulated proinflammatory gene set (Figures 1E and S1F). Second, the unique E2 + TNFα-regulated gene set is enriched for gene ontologies related to cell growth control (Table S1). Third, the FoxA1 dependence of the effects we observed, coupled with the key role that FoxA1 plays in ERα-dependent breast cancer cell growth (Hurtado et al., 2011; Ross-Innes et al., 2012), supports a role for positive functional interplay between E2/ERα, TNFα, and FoxA1 in controlling a transcriptome that dictates cell growth responses.

Proinflammatory Signaling in Breast Cancers: Unexpected Clinical Outcomes

Recent studies have identified important effects of proinflammatory signaling in cancers, which supplies the tumor microenvironment with bioactive molecules that facilitate angiogenesis, invasion, and metastasis (Baumgarten and Frasor, 2012; Ben-Neriah and Karin, 2011). In this regard, a number of recent studies have shown that proinflammatory signaling can lead to more aggressive hormone-dependent breast cancers (Coussens et al., 2013; Frasor et al., 2009; Kalaitzidis and Gilmore, 2005; Pradhan et al., 2012; Ross-Innes et al., 2012; Zhou et al., 2005). However, as our data show, the functional relationships and outcomes of the crosstalk between NF-κB/proinflammatory signaling and ERα/mitogenic signaling in breast cancers are complex and context-dependent. This is observed at the level of the proinflammatory transcriptome, even in the absence of estrogen signaling, where high level expression of a proinflammatory gene set can be associated with good or poor clinical outcomes depending on the breast cancer type (Figures 1D and S1F). We have identified a unique transcriptome, which is created upon the convergence of mitogenic and proinflammatory signaling, that is a robust predictor of clinical outcomes. Similar to the proinflammatory transcriptome, this gene set can be associated with good or poor clinical outcomes depending on the breast cancer type (Figures 1D and S1F). This variability may be controlled, in part, by differences in the length of exposure to the proinflammatory mediators (e.g., chronic, versus acute as in our studies herein), the type of cytokine (e.g., TNFα versus IL4), and the presence of other activated signaling pathways (e.g., with or without ligand-activated ERα). Taken together, our results highlight the potential varied responses and clinical outcomes of breast cancers upon proinflammatory signaling.

EXPERIMENTAL PROCEDURES

Additional details on the experimental procedures can be found in the Supplemental Materials

Cell Culture, Treatments, and RNAi-Mediated Knockdown

MCF-7 breast cancer cells were obtained from the ATCC and maintained in Minimum Essential Medium Eagle supplemented with 5% calf serum. Prior to all treatments, the cells were grown for 3 days in phenol red-free MEM Eagle medium supplemented with 5% charcoal-dextran-treated calf serum. All treatments were performed for 40 min with 100 nM 17β-estradiol, 25 ng/mL TNFα, or both. For siRNA transfections, commercially available siRNA oligonucleotides (Sigma) were transfected at a final concentration of 10 nM using Lipofectamine RNAiMAX reagent (Invitrogen).

Antibodies

The antibodies used for both Western blotting and ChIP were as follows: ERα (rabbit polyclonal generated in the Kraus Lab), p65 (Abcam ab7970), FoxA1 (Abcam ab23738), SNRP70 (Abcam ab83306), β-tubulin (Abcam ab6046), and histone H3 (Abcam ab1791).

Cell Proliferation Assays

Treatments and cell collections were performed at two day intervals. Cells were fixed with 10% formaldehyde and stained with a 0.1% crystal violet. Incorporated crystal violet was extracted using 10% glacial acetic acid and the absorbance was read at 595 nm.

Kaplan-Meier Analyses

Kaplan-Meier estimators (Dinse and Lagakos, 1982) were generated using the Gene Expression-Based Outcome for Breast Cancer Online (GOBO) tool (http://co.bmc.lu.se/gobo/) (Ringner et al., 2011). Gene sets determined by GRO-seq were provided as inputs to assess patient outcomes in ER-positive breast cancer subtypes.

Global Nuclear Run-on Sequencing (GRO-seq)

Nuclei isolation, nuclear run-on, and library preparation were performed as previously described (Danko et al., 2013; Hah et al., 2011), with modifications (Danko et al., 2013; Luo et al., 2014). The libraries were generated from two biological replicates using a circularized ligation based protocol for adaptor ligation used to improve the efficiency of library preparation, reduce sequence bias, and allow for barcoding (Luo et al., 2014; Wang et al., 2011). The GRO-seq data were analyzed using the groHMM software package described previously (Danko et al., 2013; Hah et al., 2011; Luo et al., 2014). The effects of E2, TNFα, and E2 + TNFα on the expression of coding and non-coding genes were analyzed using edgeR (Robinson et al., 2010). Further GRO-seq data analyses, including data visualization, are described in detail in the Supplemental Materials.

Chromatin Immunoprecipitation (ChIP)

Chromatin Immunoprecipitation was performed as previously described (Kininis et al., 2007) with a few modifications. For ChIP involving siRNA knockdown of p65 or FoxA1, cells were transfected with siRNA complexes 48 hrs prior to E2 and/or TNFα treatment. The enriched genomic DNA was analyzed by qPCR using the enhancer-specific primers listed in the Supplemental Materials.

ChIP-Seq

ChIP-seq libraries were generated from two biological replicates for each condition. Fifty ng of ChIPed DNA was used to generate libraries for sequencing, as previously described (Quail et al., 2008), with some modifications. Sequencing adaptors were attached using the Illumina TruSeq DNA Sample Prep Kit and sequenced using an Illumina HiSeq 2000. The ChIP-seq libraries were aligned to hg19 genome using default parameters of BOWTIE (Langmead et al., 2009). Uniquely mappable reads were converted into bigWig files using BEDTools for visualization in the UCSC genome browser (Quinlan and Hall, 2010). ERα, p65, and FoxA1 ChIP-seq samples from the untreated condition were used as controls for peak calling using MACS software (Zhang et al., 2008). Further ChIP-seq data analysis is described in detail in the Supplemental Materials.

mRNA Expression Analyses by Real-Time Quantitative PCR (RT-qPCR)

Total RNA was isolated using Trizol Reagent (Invitrogen), reverse transcribed, and subjected to real-time quantitative PCR (RT-qPCR) using the gene-specific primers listed in the Supplemental Materials. Target gene transcripts were normalized to the β-actin transcript.

Subcellular Fractionation and Analyses

Whole cell extracts were prepared in 50 mM Tris•HCl, pH 8, 150 mM NaCl, 1% NP-40, 0.5% sodium deoxycholate, and 0.1% SDS containing protease inhibitors. Cytoplasmic and nuclear extracts were made using the CelLytic NuCLEAR™ Extraction Kit from Sigma. The remaining chromatin pellet was solubilized with benzonase nuclease (Sigma) and boiled in 1X SDS sample buffer. The different subcellular fractions were analyzed for FoxA1 by Western blotting using a set of fraction-specific markers and loading controls (β-tubulin, histone H3, and SNRP70).