Article Text

Download PDFPDF

Original article
Master transcription factors form interconnected circuitry and orchestrate transcriptional networks in oesophageal adenocarcinoma
  1. Li Chen1,
  2. Moli Huang2,
  3. Jasmine Plummer3,
  4. Jian Pan4,
  5. Yan Yi Jiang5,
  6. Qian Yang1,
  7. Tiago Chedraoui Silva3,
  8. Nicole Gull3,
  9. Stephanie Chen3,
  10. Ling Wen Ding5,
  11. Omer An5,
  12. Henry Yang5,
  13. Yulan Cheng6,
  14. Jonathan W Said7,
  15. Ngan Doan7,
  16. Winand NM Dinjens8,
  17. Kevin M Waters9,
  18. Richard Tuli10,
  19. Simon A Gayther3,
  20. Samuel J Klempner11,12,
  21. Benjamin P Berman3,
  22. Stephen J Meltzer6,
  23. De-Chen Lin1,
  24. H Phillip Koeffler1,5
  1. 1 Department of Medicine, Cedars-Sinai Medical Center, Los Angeles, California, USA
  2. 2 School of Biology and Basic Medical Sciences, Soochow University, Suzhou, China
  3. 3 Center for Bioinformatics and Functional Genomics, Cedars-Sinai Medical Center, Los Angeles, California, USA
  4. 4 Department of Hematology and Oncology, Children's Hospital of Soochow University, Suzhou, China
  5. 5 Cancer Science Institute of Singapore, National University of Singapore, Singapore, Singapore
  6. 6 Departments of Medicine and Oncology, Johns Hopkins University School of Medicine and Sidney Kimmel Comprehensive Cancer Center, Baltimore, Maryland, USA
  7. 7 Department of Pathology and Laboratory Medicine, David Geffen School of Medicine, University of California Los Angeles, Los Angeles, California, USA
  8. 8 Department of Pathology, Erasmus MC, University Medical Center, Rotterdam, The Netherlands
  9. 9 Department of Pathology and Laboratory Medicine, Cedars-Sinai Medical Center, Los Angeles, California, USA
  10. 10 Department of Radiation Oncology, Cedars-Sinai Medical Center, Los Angeles, California, USA
  11. 11 The Angeles Clinic and Research Institute, Cedars-Sinai Medical Center, Los Angeles, California, USA
  12. 12 Samuel Oschin Comprehensive Cancer Institute, Cedars-Sinai Medical Center, Los Angeles, California, USA
  1. Correspondence to Dr De-Chen Lin, Department of Medicine, Cedars-Sinai Medical Center, Los Angeles 90048, CA 90048-0750, USA; dchlin11{at}gmail.com; Dr Moli Huang, School of Biology and Basic Medical Sciences, Soochow University, Suzhou, 215123, China; huangml{at}suda.edu.cn

Abstract

Objective While oesophageal squamous cell carcinoma remains infrequent in Western populations, the incidence of oesophageal adenocarcinoma (EAC) has increased sixfold to eightfold over the past four decades. We aimed to characterise oesophageal cancer-specific and subtypes-specific gene regulation patterns and their upstream transcription factors (TFs). 

Design To identify regulatory elements, we profiled fresh-frozen oesophageal normal samples, tumours and cell lines with chromatin immunoprecipitation sequencing (ChIP-Seq). Mathematical modelling was performed to establish (super)-enhancers landscapes and interconnected transcriptional circuitry formed by master TFs. Coregulation and cooperation between master TFs were investigated by ChIP-Seq, circularised chromosome conformation capture sequencing and luciferase assay. Biological functions of candidate factors were evaluated both in vitro and in vivo.

Results We found widespread and pervasive alterations of the (super)-enhancer reservoir in both subtypes of oesophageal cancer, leading to transcriptional activation of a myriad of novel oncogenes and signalling pathways, some of which may be exploited pharmacologically (eg, leukemia inhibitory factor (LIF) pathway). Focusing on EAC, we bioinformatically reconstructed and functionally validated an interconnected circuitry formed by four master TFs—ELF3, KLF5, GATA6 and EHF—which promoted each other’s expression by interacting with each super-enhancer. Downstream, these master TFs occupied almost all EAC super-enhancers and cooperatively orchestrated EAC transcriptome. Each TF within the transcriptional circuitry was highly and specifically expressed in EAC and functionally promoted EAC cell proliferation and survival.

Conclusions By establishing cancer-specific and subtype-specific features of the EAC epigenome, our findings promise to transform understanding of the transcriptional dysregulation and addiction of EAC, while providing molecular clues to develop novel therapeutic modalities against this malignancy.

  • transcription factor
  • gene regulation
  • signal transduction
  • oesophageal cancer
View Full Text

Statistics from Altmetric.com

Significance of this study

What is already known on this subject?

  • Malignant transformation is associated with gains and losses in enhancer activity across the epigenome, resulting in widespread changes in transcriptional regulation.

  • A small number of master transcription factors (TFs) are instrumental for orchestrating gene expression programmes by regulating cell-specific (super)-enhancers.

  • The genomic landscapes of oesophageal squamous cell carcinoma (ESCC) and oesophageal adenocarcinoma (EAC) have been established and contrasted; however, their epigenomic features have not been comprehensively and systematically compared and analysed.

What are the new findings?

  • EAC and ESCC display distinct (super)-enhancer landscapes, which contribute to subtype-specific transcriptional dysregulation.

  • An interconnected transcriptional circuitry in EAC formed by four master TFs (ELF3, KLF5, GATA6 and EHF) is identified and validated.

  • Master TFs occupy most of EAC super-enhancers and cooperatively orchestrated EAC transcriptome, thereby promoting the survival and proliferation of EAC cells.

  • Transcriptionally activated by master TFs through EAC-specific super-enhancers, LIF contributes to the malignant phenotypes of EAC cells.

Significance of this study

How might it impact on clinical practice in the foreseeable future?

  • This work uncovers many epigenomic features that may help develop novel therapeutic modalities for treating patients with EAC

  • Particularly, EAC-specific super-enhancers activate a number of oncogenes and signalling pathways, some of which may be exploited pharmacologically (eg, LIF pathway).

  • EAC cells are transcriptionally addicted to master TFs (ELF3, KLF5, GATA6 and EHF), and likely their associated transcriptional cofactors, which may offer a novel strategy to fight against this malignancy.

Introduction

Transcriptional dysregulation is a prominent hallmark of cancer. The cis-regulatory elements known as enhancers are key modulators of cell type-specific expression programmes. Recently, a unique group of enhancers, termed super-enhancers, have been identified.1 Super-enhancers recruit an exceptionally large number of transcription factors (TFs) and cofactors and can be identified by extensive active histone marks, such as histone 3 lysine 27 acetylation (H3K27Ac).1 2 Importantly, we3–5 and others6–9 have shown that malignant transformation is accompanied by locus-specific gains and losses in enhancer activity—particularly super-enhancers—across the epigenome, resulting in widespread changes in transcriptional output. Recent studies have suggested that a small number of TFs are critical for orchestrating specific gene expression programmes by regulating most cell-specific super-enhancers.1 10 These TFs, called ‘Master’ TFs, are often associated with super-enhancers themselves, and control their own transcription and that of other master TFs through an interconnected autoregulatory circuitry.11–13 This transcriptional paradigm is exemplified in embryonic stem cells, where master TFs (OCT4, SOX2 and NANOG) bind to their own and each other’s super-enhancers, forming an interconnected circuitry.11 In squamous cell carcinomas, our group has recently characterised δNp63 and SOX2 as master TFs that form an interconnected circuitry.14

As the eighth most common cancer and sixth leading cause of cancer-related mortality worldwide,15 oesophageal cancer is classified histologically as either adenocarcinoma (EAC) or squamous cell carcinoma (ESCC). While ESCC remains infrequent in Western populations, the incidence of EAC has strikingly increased sixfold to eightfold in Western countries over the past four decades.16 The prognosis of patients with EAC remains very poor, with a 5-year survival rate of 17% in the USA.15 Many EAC genomic drivers have been identified, including mutations in TP53, KRAS, CDKN2A, ARID1A and SMAD4,5 17–21 offering new insights into the pathogenesis of this malignancy. However, in stark contrast to genomic alterations, our understanding of the EAC epigenome is largely confined to DNA methylation changes at selected genome loci.22 23 Very recently, Britton et al 24 performed Assay for Transposase-Accessible Chromatin using sequencing (ATAC-Seq) and identified open chromatin regions in three EAC cell lines and six EAC tumour samples, revealing important upstream TFs (activator protein 1 (AP1) and E26 transformation-specific (ETS) factors). Nevertheless, EAC-associated cistrome aberrations and their biological significance remain poorly characterised. The current study addressed these crucial questions by comprehensively and integratively analysing the molecular features of the epigenome of EAC.

Materials and methods are described in online supplementary data.

Results

Characterisation of enhancer landscapes in oesophageal cancer

To establish the landscapes of active regulatory elements in oesophageal cancer, we profiled 11 fresh-frozen EAC tumour specimens and 9 well-annotated EAC cell lines using chromatin immunoprecipitation sequencing (ChIP-Seq) of H3K27Ac. To identify subtype-specific features of chromatin modification, we reprocessed H3K27Ac ChIP-Seq data from six ESCC cell lines that we had generated previously4 5 25 (see online supplementary table 1). After peak identification, we separately explored elements that were either transcriptional start site (TSS)-proximal (within 2 kb of TSS, putative active promoters) or TSS-distal (beyond 2 kb of any TSS, putative active enhancers), given their different roles in the regulation of gene expression programmes. Notably, enhancer peaks exhibited much greater variability between groups of samples than did promoter peaks. For example, with FDR <0.001 and fold change of peak intensity >4, there were 19 617 differential enhancer peaks when comparing EAC and ESCC samples, 17 times greater than differential promoter peaks (n=1151). The increased variability in enhancer region was not simply due to the larger number of enhancer elements (n=47 162/sample) compared with promoter elements (n=10 286/sample; see online supplementary table 1). These results suggest pervasive genome-wide alterations in enhancer (but not promoter) activities between EAC and ESCC samples; therefore, we focused on enhancer element for subsequent analyses.

Hierarchical clustering using active enhancers with the most variable intensities (ie, the top 10 000) clearly separated ESCC and EAC samples (figure 1A). We readily reproduced this clustering pattern with a different statistical cut-off when selecting the most variable enhancer signals (using FDR <0.001; see online supplementary figure 1), demonstrating the robustness of the clustering approach. Moreover, EAC tumour specimens and cell lines clustered together, suggesting a convergent enhancer state markedly distinct from ESCC. This convergence of epigenomic state between tumour samples and cell lines has been observed in other types of cancers, such as rhabdomyosarcoma26 and neuroblastoma.27 The similarity of enhancer landscapes between EAC tumour specimens and cell lines suggests that the difference between EAC and ESCC samples is not simply due to cell culture, and allowed us to use cell line models to investigate some of the most important EAC-specific enhancer features, as discussed below.

Figure 1

Enhancer and super-enhancer landscapes of oesophageal cancers. (A) Hierarchical clustering using enhancers with the most variable intensities (top 10 000). (B) Left: pie chart showing the number of gained enhancers in each group; right: pathway enrichment of gained enhancers. Dot size denotes the number of genes enriched. (C) Box plot of mRNA levels of genes associated with changed enhancers in EAC and ESCC samples from TCGA. (D) Box plot of DNA methylation levels of changed enhancer loci in EAC and ESCC samples from TCGA. P value was determined by Wilcoxon test. (E) Venn diagram of the number of super-enhancers annotated in each group. (F) Inflection plot ranking enhancer intensities, and only group-specific super-enhancers are displayed as examples. (G) IGV plots of the H3K27Ac ChIP-Seq profiles of group-specific super-enhancers. Each line represents one sample; values of normalised ChIP-Seq signal intensities are shown on the upper left corner; genomic structure of the genes associated with super-enhancer is shown at the bottom. ChIP-Seq, chromatin immunoprecipitation sequencing; EAC, oesophageal adenocarcinoma; ESCC, oesophageal squamous cell carcinoma; Fpkm, fragments per kilobase of transcript per million mapped reads; H3K27Ac, histone 3 lysine 27 acetylation; IGV, integrative genomics viewer; mRNA, messenger RNA; TCGA, The Cancer Genome Atlas.

By focusing on protein-coding genes, differential analysis identified 6537 gained enhancers (assigned to 3189 genes) and 6876 lost enhancers (assigned to 3587 genes) in EAC versus ESCC samples (fold change >2, FDR <0.001; see online supplementary table 2). Although this differential usage of enhancers is partially contributed by the different cell identity, alterations in enhancer landscapes also clearly reflect subtype-specific cancer biology. Particularly, these subtype-specific enhancers were associated with genes enriched in signalling pathways displaying subtype-specific features (figure 1B). For example, EAC-high enhancer genes were over-represented in the HIF-1α, cytokine,28 FOXA1,22 TNF, PDGFR22 and ARF622 pathways, while ESCC-high enhancers were more significantly enriched in Hippo,4 22 δNp63,5 25 29 Rac1, focal adhesion and NRF25 22 29–31 signalling. These data were highly consistent with previous reports characterising differential activities of signalling pathways between EAC and ESCC (eg, FOXA1 and ARF6 signalling were stronger in EAC, while Hippo and NRF2 activities were more elevated in ESCC22).

We next investigated whether differences in enhancer landscapes between groups reflected changes in transcriptomic output by reanalysing RNA-Seq data of EAC (n=88) and ESCC (n=90) samples from The Cancer Genome Atlas (TCGA).22 Importantly, genes associated with gained enhancers were expressed at significantly higher levels in the corresponding group (figure 1C). We and others have previously shown that active enhancers generally exhibit lower DNA methylation levels compared with inactive enhancers and other silenced chromatin elements,32–35 implying a dynamic competition between TF binding and DNA methylation. Indeed, gained enhancer elements generally exhibited lower DNA methylation levels in the corresponding group (figure 1D).

Considering that the different enhancer profiles between EAC and ESCC were partially attributable to their different cell types, we next performed H3K27Ac ChIP-Seq analyses on five frozen samples from non-malignant gastro-oesophageal junction (NGEJ) which have the same columnar cell type with EAC. Importantly, a total of 1703 NGEJ-high and 485 NGEJ-low enhancers were identified when compared with EAC (fold change >2, false discovery rate (FDR) <0.1; see online supplementary table 3), and hierarchical clustering successfully separated these two types of samples (see online supplementary figure 1B). Pathway enrichment analysis revealed that EAC-high enhancer-associated genes were over-represented in several oncogenic pathways, including the signalling of epidermal growth factor receptor (EGFR), wingless-INT (Wnt) and epithelial mesenchymal transition (EMT). In contrast, processes specific to GI mucin-secreting cells, such as O-linked glycosylation of mucin, were specifically enriched in the NGEJ-high enhancer set. Moreover, bone morphogenic protein signalling, which is important for both normal oesophagus development36 and intestinal metaplasia of gastro-oesophageal junction cells,37 was only enriched in NGEJ-high genes (see online supplementary figure 1C). These results highlight dysregulated enhancer landscape in EAC, which is associated with both cancer-specific and subtype-specific transcriptional networks apparently contributing to tumour biology.

Distinct super-enhancer landscapes in EAC and ESCC

We recently showed that super-enhancers play prominent roles in regulating the expression of a key array of oncogenes important for the malignant phenotype of cancer cells.3 4 25 Thus, we next annotated super-enhancers in EAC and ESCC samples using the ROSE method1 38 (figure 1E–G; see online supplementary tables 4a–b) and revealed that subtype-specific super-enhancers accounted for 55.8% (871/1561) and 53.8% (803/1493) of all super-enhancers in EAC and ESCC samples (figure 1E), respectively. Notably, unique sets of key oncogenes were associated with subtype-specific super-enhancers, many of which reflected subtype-specific cancer biology. For example, ERBB2, MET, GATA6, ETV6 and HNF1B were associated with EAC-specific super-enhancers; CTTN, FGFR2, TP73 and WNT5A were assigned to ESCC-specific ones (figure 1F,G). We similarly found that super-enhancer reservoirs were substantially altered between EAC and NGEJ samples, with only 30.4% (554/1819) being shared (see online supplementary figures 1D–F and tables 4c).

Identification of interconnected transcriptional circuitry formed by master TFs in EAC

After establishing the landscape of enhancers in EAC, we next sought to determine which upstream TFs control the activity of these regulatory elements. Considerable evidence demonstrates that cell-type specific gene expression programmes are dominated by a small number of master TFs in each respective cell type.2 11 13 Master TFs are often associated with super-enhancers themselves and form interconnected autoregulatory loops (also known as core regulatory circuitry) by binding to each other’s super-enhancers.11 13 33 39 Master TFs are also highly expressed in their corresponding cell types. Taking into account these known biological phenomena, we modified a previously established mathematical method40 and performed integrative circuitry analysis of EAC samples (see online supplementary methods), thereby identifying a small set (n=10) of candidate master TFs (figure 2, online supplementary figure 2A), including several TFs with known oncogenic functions, such as GATA6, KLF5, FOXA1 and HES1. Compared with other TFs, these candidate master TFs had higher predicted transcriptional connectivity (defined by the magnitude of enrichment of the binding motif of candidate TFs in super-enhancer regions along the genome; see online supplementary figure 2B).

Figure 2

Master TFs form interconnected transcriptional circuitry in EAC. (A) Integrative methods for identification of candidate master TFs. (B) Heatmap of Pearson correlation coefficient between candidate master TFs in TCGA EAC (n=88), stomach adenocarcinoma (STAD, n=415) or breast cancer samples (n=1100). (C) Heatmap of fold changes of mRNA levels of master TFs and c-Myc and EVX1 (non-master TFs, negative control) following siRNA knockdown of each master TF (left) or three different shRNAs against ELF3 (right). (D) Western blot validating the coregulation among master TFs in Eso26 cells. The numbers denote the densitometric quantitation of band intensity, normalised by actin levels. (E) IGV plot of ChIP-Seq showing co-occupancy (shaded) of ELF3, KLF5 and GATA6 at the super-enhancers of their own gene and the other three master TFs. Antibodies against endogenous KLF5 and GATA6 were used. A flag antibody for exogenous ELF3-Flag was used because of the poor quality of ELF3 antibody for ChIP-Seq. (F) Schematic graph of the model of interconnected circuitry, with rectangles and ovals representing enhancer elements and proteins, respectively. BE, Barrett’s oesophagus; ChIP-Seq, chromatin immunoprecipitation sequencing; EAC, oesophageal adenocarcinoma; ESCC, oesophageal squamous cell carcinoma; IGV, integrative genomics viewer; mRNA, messenger RNA; NGEJ, non-malignant gastro-oesophageal junction; shRNA, short hairpin RNA; siRNA, small interfering RNA; TCGA, The Cancer Genome Atlas; TFs, transcription factors.

We reasoned that in a fully interconnected circuitry, the RNA expression of each member should show strong positive correlation in relevant cell/tissue types. We thus interrogated the TCGA EAC RNA-Seq data set and noted that the expression of four candidates (ELF3, KLF5, GATA6 and EHF) displayed prominently positive correlations with each other (figure 2, online supplementary figure 2B; all Pearson correlation coefficients >0.2). These correlations were expectedly observed in stomach adenocarcinoma (STAD) given the known molecular similarity between EAC and STAD,22 41 but were absent in other cancer types, such as breast cancer (figure 2B). Moreover, in a pan-cancer RNA-Seq analysis in both TCGA and Cancer Cell Line Encyclopedia samples, these four candidates were in general expressed highly in EAC samples relative to most other tumour types (see online supplementary figures 3 and 6C). We therefore next focused on characterising these four high-confidence master TFs.

To validate direct transcriptional regulation among these candidates, we first performed ChIP-Seq to map the genome-wide occupancy of these four master TFs in Eso26, an EAC cell line. We could not generate high-quality ChIP-Seq data of EHF because of lack of ChIP-grade antibody. Strikingly, ELF3, KLF5 and GATA6 co-occupied the super-enhancers of all four master TFs including themselves (figure 2E), forming an interconnected circuitry, as we had predicted. Moreover, the super-enhancer regions of all of these four TFs were highly specific to both EAC tumour samples and cell lines, as they were substantially weaker in either ESCC cells or NGEJ samples. Serving as additional controls for EAC cells, we further generated H3K27Ac ChIP-Seq data in two Barrett’s oesophagus (BE) cell lines (ChTRT and GihTRT), which again exhibited negligible signals when compared with EAC samples (figure 2E).

To directly confirm their interconnected transcriptional regulation, each TF was silenced using small interfering RNA (siRNA). Knockdown of any single TF decreased the expression of the other three members (figure 2C). However, c-MYC, a well-known oncogenic TF in EAC but not predicted within the transcriptional circuitry, was not affected (figure 2C, left panel). Similarly, EVX1, a novel oncogenic TF in EAC cells (manuscript in preparation) but a non-master TF, was not consistently regulated by these four factors. This interconnected circuitry was further confirmed at the protein level using additional individual siRNAs (figure 2D). These results were also verified by shRNA-mediated knockdown (figure 2C,D). Together, these data identified and validated an interconnected transcriptional circuitry consisting of four master TFs in EAC (figure 2F).

Master TFs cooperatively orchestrate the transcriptional network of EAC

To understand the significance of interconnected circuitry in the regulation of the EAC transcriptome, we next explored in depth the cistromes of three TFs within the circuitry, namely ELF3, KLF5 and GATA6. As expected, motif analysis found highly significant enrichment of their own recognition sequences within the corresponding ChIP-Seq peaks (figure 3A). Notably, the binding motif of each single factor was also strongly enriched in the peaks from the other master TFs (figure 3A), suggesting that occupancies of these master TFs lie in close proximity to each other. In contrast, either minimum or no enrichment was observed in non-master TFs in EAC (eg, E2F1 and TP63). Indeed, along the genome, ELF3, KLF5 and GATA6 exhibited a prominent co-occupancy pattern (figure 3C). Furthermore, metagene analysis showed that the distribution of these binding peaks strongly aligned (figure 3B), suggesting their functional interplay in EAC cells. To investigate the transcriptional implications of the occupancy of ELF3, KLF5 and GATA6, we correlated their binding profiles with H3K27Ac ChIP-Seq data generated from the same Eso26 cell line and observed prominently enriched H3K27Ac signals adjacent to the regions occupied by these master TFs (figure 3B,C). Specifically, the majority of ELF3 (87.1%, 5702/6543; p<2.2e-16), KLF5 (68.7%, 15 215/22 141; p<2.2e-16) and GATA6 peaks (76.6%, 3375/4404; p<2.2e-16; all χ2 test) were associated with H3K27Ac signals, suggesting that transcriptional activation was associated with the binding of these three TFs. Indeed, transcripts assigned to the binding of any single TF were expressed at significantly higher levels than those assigned to none (figure 3D).

Figure 3

Master TFs orchestrate cooperatively EAC transcriptional network. (A) Position weight matrix and heatmap showing the p values of enriched motifs in either ELF3, KLF5, GATA6 or co-occupied genomic regions in Eso26 cells. The enrichment of TP63 and E2F1 motifs is shown as negative controls. (B) Line plots showing the distribution of indicated ChIP-Seq signals at GATA6 peak regions (centred at the summit of GATA6 peaks). (C) Heatmap showing ChIP-Seq signals at GATA6 peak regions (±3 kb of peak centre), rank ordered by intensity of GATA6 peaks based on reads per million mapped reads (RPM). Lines, peaks; colour scale of peak intensity is shown at the bottom. (D) Box plot of mRNA levels of genes associated with each group of peaks in Eso26 cells. (E) Fold ratio of the percentage of super-enhancers (SE) over typical enhancers (TE) bound by individual master TFs either alone or together. ChIP-Seq, chromatin immunoprecipitation sequencing; EAC, oesophageal adenocarcinoma; Fpkm, fragments per kilobase or transcript per million mapped reads; mRNA, messenger RNA; TFs, transcription factors.

We next explored the transcriptional impact of the co-occupancy of the master TFs. Importantly, transcripts bound by all three TFs were expressed at the highest levels (figure 3D). Given the prominent co-occupying pattern of ELF3, KLF5 and GATA6 in active regulatory regions, we assigned their binding peaks to either super-enhancer or typical enhancer elements to gain additional insights into their transcriptional cooperation. Akin to a few master TFs identified in other cell types (eg, PHOX2B, HAND2 and GATA3 in neuroblastoma27), EAC master TFs each occupied a significant proportion of super-enhancers but interacted with only a small fraction of typical enhancers (see online supplementary figure 4A). In fact, all of the annotated super-enhancers were occupied by at least one of these three TFs. Notably, this preference of interacting with super-enhancers was even more profound when considering co-occupied regions. Specifically, regulatory elements occupied by more TFs had a higher likelihood of lying within super-enhancers than did typical enhancers (figure 3E). These data demonstrate that master TFs interconnect via coregulation within the circuitry, and cooperatively regulate gene expression programmes by preferentially activating super-enhancers along the genome.

ELF3, KLF5 and GATA6 cooperatively activate the super-enhancer of ELF3

Given the above finding that master TF circuitry preferentially activates super-enhancer elements relative to typical enhancers, we next focused on characterising ELF3 super-enhancer loci, because ELF3 itself is a top-ranked master TF and has a massive super-enhancer in EAC samples (figure 4B). In contrast, the H3K27Ac modification of these enhancer loci was markedly weaker in ESCC samples (figure 4B), suggesting that this is an EAC-specific active chromatin state. Next, we employed circularised chromosome conformation capture (4C) assays to explore the interaction landscape of this ELF3 super-enhancer in Eso26 cells, using its promoter as the 4C bait (Viewpoint). Importantly, by cross-referencing H3K27Ac ChIP-Seq data generated in the same cell line, we successfully identified five enhancer constituents (E1–E5) interacting with the ELF3 promoter (figure 4C). Moreover, these five regions were always co-occupied by both ELF3 and KLF5 (figure 4C). GATA6 also had strong interaction with two of these five enhancer constituents (E1 and E4), suggesting that the activities of these enhancer elements were under control of the three master TFs. These co-occupancy patterns were validated by ChIP-qPCR in multiple additional EAC cell lines (see online supplementary figure 4B). Strikingly, these strong and extensive enhancer–promoter interactions were strictly confined within this super-enhancer window (figure 4A), indicating that this cluster of enhancer elements was dedicated to activating the transcription of ELF3 in EAC cells. We subsequently cloned individual constituent enhancer elements into the luciferase reporter vector and observed robust activities of E1 and E4 in different EAC cells (figure 4D). Consistently, these reporter activities were not detected in ESCC cells (figure 4D). To test direct transcriptional regulation of this super-enhancer on ELF3, we used CRISPR interference system where single guide RNAs (sgRNAs) guide dCas9/KRAB complex to suppress targeted cis-regulatory elements.42 We designed sgRNAs against E1 and E4 because (1) these two enhancer elements exhibited the highest reporter activities and (2) they were the only two regions bound by all three TFs. Importantly, targeting either E1 or E4 significantly reduced the expression of ELF3 (figure 4D). The expression levels of the other three master TFs were also decreased, again supporting the interconnected coregulation between these factors. These results support strong and complex regulation of ELF3 super-enhancer region, which is EAC-specific and controlled by EAC master TFs. In parallel, we performed another 4C assay to characterise enhancer–promoter interactions flanking KLF5 super-enhancer region. Again, we validated that multiple enhancer constituents, co-occupied by master TFs, interacted with KLF5 promoter in EAC cells (see online supplementary figure 5).

Figure 4

Master TFs cooperatively activate the super-enhancer of ELF3. (A) 4C assay showing the long-range interactions anchored on ELF3 promoter in Eso26 cells. Deeper red colour indicates higher interaction frequency. (B) ChIP-Seq profiles for H3K27Ac (in different groups of samples) and master TFs at ELF3 super-enhancer loci. (C) Zoom-in view of ChIP-Seq signals in Eso26 cells. Connecting lines showing the interactions detected by 4C. Five constituent enhancers (E1–E5) and one negative control (Ctrl) region were separately cloned into luciferase reporter vector. (D) Enhancer activity measured by luciferase reporter assays in indicated EAC cells and KYSE510 cells. Mean±SD are shown, n=2. *P<0.05, **p<0.01, ***p<0.001. (E) Eso26 cells expressing dCas9/KRAB vector with sgRNAs targeting E1 and E4 or control vector were subject to qRT-PCR to quantify the mRNA expression of master TFs. 4C assay, circularised chromosome conformation capture assay; ChIP-Seq, chromatin immunoprecipitation sequencing; EAC, oesophageal adenocarcinoma; ESCC, oesophageal squamous cell carcinoma; H3K27Ac, histone 3 lysine 27 acetylation; qRT-PCR, quantitative real-time PCR; sgRNA, single guide RNA; TFs, transcription factors.

Master TFs have strong progrowth functions in EAC cells

Considering the prominent roles of the master TFs in controlling EAC transcriptional network, particularly their preference in the regulation of super-enhancers, we hypothesised that these factors are required for the viability and proliferation of EAC cells. To test this, we first focused on the investigation of ELF3, whose functional significance in EAC remains unknown. Importantly, depletion of endogenous ELF3 expression by independent siRNAs markedly reduced cell proliferation (figure 5A) and colony growth (figure 5B) in different EAC cells, and the results were verified by doxycycline-inducible expression of multiple independent short hairpin RNAs (shRNAs) (figure 5E,F). Fluorescence-activated cell sorting analysis showed that silencing of ELF3 increased significantly EAC cell apoptosis (figure 5C) and cell cycle arrest at S-phase (figure 5D). In xenograft assays, induction of the expression of shRNA against ELF3 potently inhibited EAC xenograft growth in mice (figure 5G–I). These data characterise that ELF3, a master TF highly expressed in EAC, has strong protumour functions in this cancer. Prompted by the notable prosurvival and proproliferation capacities of ELF3, we next tested the functionality of the other three master TFs (KLF5, EHF and GATA6). Importantly, silencing of any of the three factors inhibited strongly the proliferation and colony formation of multiple different EAC cell lines but not BE cell lines (figure 5J–L, online supplementary figure 6A,B).

Figure 5

Master TFs have strong pro-EAC functions. (A) Knockdown of ELF3 by individual siRNAs decreased cell proliferation and colony growth (B), and increased cell apoptosis (C) and S-phase arrest (D) in different EAC cell lines. (E) Silencing of ELF3 by inducible shRNAs in Eso26 cells decreased cell proliferation, colony growth (F) as well as xenograft growth in vivo (G–I). (G) Weights, (H) images and (I) growth curves of resected tumours from both groups. (J–L) Knockdown of other three master TFs by individual siRNAs decreased cell proliferation and colony growth. Mean±SD are shown, n=3. *P<0.05, **p<0.01, ***p<0.001. EAC, oesophageal adenocarcinoma; shRNA, short hairpin RNA; TFs, transcription factors.

Upregulated by master TFs via super-enhancers, LIF promotes EAC growth and migration

Following the identification and characterisation of the upstream master TF circuitry, we next focused on investigating the downstream signalling pathways activated by EAC-specific enhancers, inspired by previous work,3 4 7 43 demonstrating that tumour-specific enhancers converge on activating cancer hallmarks and associated signalling pathways. Among the pathways enriched by EAC-specific enhancers, we were particularly interested in the cytokine-mediated signalling since it was top-ranked in the EAC group (figure 1D). Careful examination of the overlapping pathway components (n=113; online supplementary table 5) identified many established protumour factors, including LIF, LYN, SYK, JAK2, IL1B and so on. Among these 113 components, LIF was the highest-ranked super-enhancer-assigned cytokine specific to EAC samples (online supplementary table 4). Moreover, LIF super-enhancer contained cobinding peaks of ELF3 and KLF5 (figure 6A). In contrast, these enhancer elements were much weaker in either ESCC, NGEJ or BE samples (figure 6A). Consistently, LIF expression was significantly upregulated in EAC samples (online supplementary figure 7A). Immunohistochemistry staining observed that LIF protein was strongly expressed in EAC tumours but not in either NGEJ or normal oesophageal squamous samples (figure 6G). Importantly, silencing of any of the four master TFs markedly inhibited the expression of LIF at both messenger RNA (mRNA) (figure 2C) and protein levels (figure 2D), strongly suggesting that these master TFs regulate the transcription of this top-ranked super-enhancer gene.

Figure 6

Upregulated by master TFs via super-enhancers, LIF promotes EAC growth and migration. (A) IGV plots of ChIP-Seq showing EAC-specific LIF super-enhancer which was co-occupied by master TFs. (B) Cell viability assay testing EC330, a LIF inhibitor, in EAC and ESCC cell lines. IC50 values are shown in the right panel. (C) Silencing of LIF with siRNA decreased different EAC cell proliferation and (D) colony growth. (E) LIF stimulated EAC cell migration, which was neutralised by an anti-LIF antibody. (F) Western blotting showing that LIF stimulated STAT3 and AKT phosphorylation in EAC cell lines. Mean±SD are shown, n=3. *P<0.05, **p<0.01, ***p<0.001. (G) IHC staining of LIF in EAC (n=35), non-malignant oesophagus squamous mucosa (NESQ, n=10) and NGEJ samples (n=7). BE, Barrett’s oesophagus; ChIP-Seq, chromatin immunoprecipitation sequencing; EAC, oesophageal adenocarcinoma; ESCC, oesophageal squamous cell carcinoma; H3K27Ac, histone 3 lysine 27 acetylation; IHC, immunohistochemistry; NGEJ, non-malignant gastro-oesophageal junction; TFs, transcription factors.

LIF is a well-established pleiotropic cytokine which regulates the differentiation of haematopoietic and neuronal cells. Interestingly, during the preparation of the present manuscript, a report was published associating higher LIF level in the serum of patients with EAC with worse response to neoadjuvant therapy.44 However, its biological functions and associated molecular pathways have not been investigated in EAC. To address this, we first depleted LIF transcript by siRNA in multiple EAC cell lines, and its knockdown drastically impaired EAC cell proliferation and colony growth (figure 6C,D). In contrast, silencing LIF produced much weaker effect on ESCC cell proliferation, suggesting its EAC-specific role (see online supplementary figure 7B). Recently, a steroidal LIF-specific small-molecule inhibitor (EC330) was developed45 46 (figure 6B). Importantly, this LIF inhibitor displayed potent antineoplastic activity in EAC cells in vitro, with IC50 ranging from 28 to 565 nM (figure 6B). Again validating the functional specificity of LIF in EAC cells, EC330 barely showed cytotoxicity against ESCC cells (figure 6B). Furthermore, exogenous LIF stimulation prominently enhanced cell migration, which was neutralised by coexposure to an anti-LIF antibody (figure 6, online supplementary figure 7C), confirming the specificity of the results. Lastly, exogenous LIF potently stimulated the phosphorylation of STAT3 and AKT pathways (figure 6F). Given that both STAT347 48 and AKT signalling are well-established progrowth cascades for EAC cells, these data characterise LIF as a key super-enhancer-driven factor which is upregulated by EAC master TFs and promotes EAC proliferation and migration.

Discussion

Despite numerous new insights gained from genomic analyses of patients with EAC,5 17–21 preventive or therapeutic strategies have not substantially improved outcomes. EAC exhibits high intertumour and intratumour genomic heterogeneity,19 49 50 increasing the barriers to exploiting targetable genomic lesions. Clearly, alternative molecular approaches in addition to genomic profiling are required to further decipher EAC pathophysiology for the development of more innovative and effective regimens.

To this end, we performed comprehensive epigenome profiling of EAC tumour samples and cell lines, and contrasted them against ESCC and NGEJ samples. We found widespread and pervasive alterations in EAC enhancer and super-enhancer landscapes, which were strongly associated with cancer-specific and subtype-specific biological states. We identified a myriad of novel EAC-promoting genes as well as oncogenic signalling pathways which may be exploited pharmacologically. Among these, cytokine signalling represents a particularly important pathway containing many components associated with either EAC-specific enhancers (such as LYN, JAK2 and IL1B) or super-enhancers (such as LIF and LIF receptor subunit alpha (LIFR)). Importantly, IL1B-associated signalling was shown to directly promote EAC development and progression in a transgenic mouse model, where additional cytokines (eg, IL6 and IL8) were also significantly upregulated.28 Here, we identified and validated LIF as a top-ranked super-enhancer-driven cytokine that is uniquely upregulated by master TFs in EAC samples. LIF has strong protumour functions specifically in EAC but not ESCC cells, which can be suppressed by a specific small-molecule inhibitor. Notably, in addition to LIF, the cytokine signalling pathway has a number of components which may be ‘druggable’ (eg, JAK2, YES1, SYK), highlighting the power of our integrative approach to discover novel actionable targets, some of which are under early clinical investigation (eg, NCT02693535).

We established and functionally validated an interconnected transcriptional circuitry formed by master TFs (ELF3, KLF5, GATA6 and EHF), which orchestrates the dysregulation of EAC transcriptome in a cooperative manner. These master TFs promote the expression of each other by interacting with their super-enhancers. Indeed, their mRNA levels significantly correlate with each other and are generally high in EAC tumours compared with other forms of human cancers. These master TFs operate in concert and often co-occupy enhancer elements in a cooperative fashion. Notably, these factors favour the regulation of super-enhancers over typical enhancers, such that virtually all of the super-enhancers annotated in Eso26 cells were bound by at least one of these master TFs. This biased pattern of interacting with super-enhancers over typical enhancers by master TFs was more conspicuous when considering the co-occupied regions. Transcripts bound by all three TFs were expressed at the highest levels, further supporting cooperation (figure 3).

Because of this pivotal role of master TFs in the regulation of EAC transcriptomic network (particularly through controlling super-enhancers), all of them are, not surprisingly, required for the survival and proliferation of EAC cells (figure 5). Although GATA6 has been established as a strong oncogene in EAC,51 the other three factors (ELF3, EHF and KLF5) remain hitherto unexplored in this cancer. Notably, GATA6 and KLF5 have been shown to interact with each other and promote both activities in gastric cancer,52 which shares a certain degree of genomic similarity with EAC.22 Both ELF3 and EHF belong to the ETS TF family. Intriguingly, both of them have seemingly opposing roles in cancer biology in different tumour types. For example, ELF3 suppresses the activity of androgen receptor and inhibits the proliferation of prostate cancer cells.53 In ampullary carcinoma, genomic sequencing suggests ELF3 as a tumour suppressor,54 while in hepatocellular cancer ELF3 has oncogenic activities and promotes cellular malignant phenotypes, suggesting a tissue context-specific role.55 Similarly, EHF also has been observed to have opposite functions depending on different tumour types.56 57 We reason that these disparities may be because ELF3 and EHF have different transcriptional cofactors/partners in distinct cell types, which results in their different cistromes and downstream genes. In contrast, KLF5 appears to have a consensus oncogenic role in different cancer types. KLF5 has also been recently identified as a master regulator driven by a super-enhancer in low-grade pancreatic ductal adenocarcinoma.58 Notably, in addition to being activated epigenetically, KLF5 exons harbour hotspot oncogenic mutations. Furthermore, the super-enhancer of KLF5 was found to be genetically amplified.59 Interestingly, we also observed that the KLF5 locus is significantly amplified in TCGA EAC samples (data not shown), supporting the notion that prominent driver genes can be altered in cancer cells through multiple different mechanisms.

In summary, by comprehensively establishing the epigenomic state of EAC and its upstream master regulators and downstream signalling pathways, this work promises to transform our understanding of the transcriptional dysregulation and addiction of EAC, while providing potential future therapeutic strategies against this deadly malignancy.

Acknowledgments

We thank Huynh Carissa for her coordination of sample collection.

References

View Abstract

Footnotes

  • LC, MH, JP, JP and YYJ contributed equally.

  • Contributors D-CL conceived and devised the study. D-CL, LC and MH designed the experiments and analysis. LC, JaP, JiP, YYJ, SC, WDL, JWS and ND performed the experiments. MH, TCS, OA, QY, HY and BB performed the bioinformatics and statistical analysis. YC, WNMD, KW, RT, SG, SJK and SJM contributed reagents and materials. LC, MH, JaP, D-CL and HPK analysed the data. D-CL and HPK supervised the research and wrote the manuscript.

  • Funding This research is supported by the National Research Foundation Singapore under its Singapore Translational Research (STaR) Investigator Award (NMRC/STaR/0021/2014) and administered by the Singapore Ministry of Health’s National Medical Research Council (NMRC), the NMRC Centre Grant awarded to National University Cancer Institute of Singapore, the National Research Foundation Singapore and the Singapore Ministry of Education under its Research Centres of Excellence initiatives (to HPK). This research is additionally supported by the RNA Biology Center at the Cancer Science Institute of Singapore, NUS, as part of funding under the Singapore Ministry of Education’s Tier 3 grants (grant number MOE2014-T3-1-006). SJM is supported by the Emerson Research Foundation and NIH grants DK118250, CA190040 and CA211457; he is also the Harry and Betty Myerberg Professor and American Cancer Society Clinical Research Professor. D-CL is supported by the DeGregorio Family Foundation, the Price Family Foundation as well as Samuel Oschin Comprehensive Cancer Institute (SOCCI) at Cedars-Sinai Medical Center through the Translational Pipeline Discovery Fund; he is member of UCLA Jonsson Comprehensive Cancer Center, UCLA Molecular Biology Institute as well as UCLA Cure: Digestive Diseases Research Center. SJK is supported by the Howard H Hall fund for oesophageal cancer research.

  • Competing interests None declared.

  • Patient consent for publication Obtained.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Data availability statement Data are available in a public, open access repository.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.