Fig 1.
Identification of the ETV1 cistrome in oesophageal adenocarcinoma-derived OE33 cells.
(A) Pie charts showing the genomic distribution of the ETV1 binding peaks (left) and the expected ratio of genomic regions (right) (promoter defined as +/- 1 kb from TSS). (B) UCSC browser track showing ETV1 binding peaks (bottom tracks) at the DUSP6 and EGFR loci with enrichment over the input sample (top tracks). Intergenic (grey boxes) and promoter proximal (blue boxes) peaks are highlighted. (C) Top two motifs from de novo motif discovery at the ETV1 binding regions. (D) The top 10 gene ontology terms from the biological processes category of the genes associated with ETV1 binding regions. (E) Box plots showing the average gene expression levels from microarray analysis of the 398 direct ETV1 target genes in HET1A compared to OE33 cells. Horizontal line represents the median values; ***P-value = <0.001 (2 tailed T-test). (F) Heatmap showing gene expression changes of direct ETV1 target genes following siRNA-mediated ETV1 or ETV4 knockdown in OE33 cells. Only genes showing a fold change of >1.5 following depletion of ETV1 are shown. Data are normalised to OE33 treated with a non-targeting control siRNA (siNTC), and clustered. (G) Model showing a potential regulatory module acting at ETV1 binding elements consisting of AP1 and ETV1 transcription factors.
Fig 2.
ATAC-seq reveals open chromatin regulatory regions in OAC cell lines.
(A) Heatmap of normalised Tn5 cleavage events (log2) in 500 bp regions surrounding the summits of the significant differentially accessible regions (+/- linear fivefold change, p<0.05) between non-cancer and cancer cell lines. Replicate data are shown for three of the cell lines. Data are hierarchical clustered using 1-Pearson’s correlation and grouped according to either more “open” or “closed” in cancer. (B) Pie chart showing the genomic distribution of the differentially accessible regions (promoter = +/- 1 kb from TSS). (C) UCSC browser track showing open chromatin regions at the FAM110C locus in the indicated cell lines. Intergenic (grey boxes) and promoter proximal (blue boxes) peaks are highlighted. (D) The top 6 gene ontology terms from the disease category of the genes associated with the differentially accessible regions more open in cancer cells. (E) Box plots of expression (log2) from microarray data of the nearest gene to differentially accessible peaks in the indicated cell lines. Horizontal line indicates median. *** = P-value <0.001 (t test).
Fig 3.
Identification of AP1 as a potential regulator of differential chromatin accessibility in OAC-derived cells.
(A) Top four motifs from de novo motif discovery at the 987 differentially accessible chromatin regions (+/- 250 bp from the centre) which are more open in cancer. (B) Plot of normalised Tn5 cleavage events +/-75bp from the motif centre (blue arrow) around the AP-1 motifs found in differentially accessible chromatin regions that are more open in cancer cells. Data are plotted from OE33 (red) and HET1A (blue) cells. (C) ChIP-qPCR analysis of JUN binding to accessible chromatin regions associated with the indicated genes in OE33 cells. Non-specific IgG is used as a control and the FOXO4 locus is a control region not bound by AP1. Data are shown as means ±SD (n = 3); * = P-value <0.05 and ** = P-value <0.01 (t test). (D) Relative density of the indicated motifs around the centre of the differentially accessible chromatin regions which are more open in cancer cells. Data are plotted as numbers of motifs per base pair across the regions. (E) Scatterplot showing genes which change significantly (linear 1.5 fold change, q<0.01, FPKM>10) and are either up- or down-regulated following DN-FOS expression. (F) Box plots of expression (log2) of nearest gene to differentially accessible peaks found to be more open in OE33 cells compared to HET1A cell lines. Data are partitioned according to whether the associated differentially accessible peak contains an AP1 motif, and are calculated using genes showing significant changes in expression (>1.5 fold; q-value <0.01) in OE33 cells transduced with viral vectors encoding DN-FOS or empty vector (control). Horizontal line indicates median; *** = P-value <0.001 (t test). (G) The top 5 gene ontology terms from the molecular and cellular function category of the genes showing significantly changed expression in cells transduced with lentiviruses encoding DN-FOS. (H) Pie chart showing the effect of DN-FOS transduction on the expression of the 738 genes which are downregulated by treatment with siRNA against ETV1 (2 fold change). Up and downregulated genes are defined as >1.5 fold change (q-value <0.01). (I) Expression of putative AP1 target genes across OAC-derived tissue samples (microarray data, GSE13898; [4]). AP1 target genes are selected based on their association with nearby regions that are both more open in cancer cells and also contain an AP1 motif and in addition, exhibit reduced expression following DN-FOS expression (>1.3 fold). Data are row Z normalised and subjected to Eucladian clustering and three main subclusters are highlighted (SC1-3). SC3 delineates samples that show low level AP1 expression and low level target gene expression.
Fig 4.
ETV1 binding regions are associated with open chromatin.
(A) ChIP-seq (left) and ATAC-seq (right) tag densities compared in a 2 kb window around summit (indicated by arrow) of ETV1 binding regions. (B) Plot of normalised Tn5 cleavage events +/-1 kb from the peak summits of the ETV1 binding regions. Data are plotted from OE33 (red) and HET1A (blue) cells. (C) Heat map showing the relative accessibility (normalised ATAC-seq reads) found in 500 bp windows surrounding the summit of ETV1 binding regions in the HET1A (normal derived) cell line and the three cancer derived cell lines OE33, OE19 and MFD-1. Replicates were merged for OE33 and HET1A. Regions are clustered using hierarchical clustering. (D) UCSC browser track showing ETV1 binding peaks compared to input sample in OE33 cells (top two tracks) at the DUSP6 and ADAP1 loci and ATAC-seq signal in the same regions in the OE19, OE33, FLO1 and HET1A cells. Intragenic (grey boxes) and promoter proximal (blue boxes) peaks are highlighted. (E) Plot of normalised Tn5 cleavage events +/-75bp from the motif centre around the AP-1 (left) and ETS (right) motifs found at ETV1 binding regions. Data are plotted from OE33 (red) and HET1A (blue) cells. (F) Heatmap of relative gene expression data for the indicated ETV1 target genes and PEA3 and AP1 family transcription factors from 73 OAC biopsy samples. Target genes were selected based on being associated with an ETV1 binding region which is open in OE33 cancer cells and contains either an AP1 or ETS motif (or both) (indicated by yellow boxes). Data were generated by RT-qPCR on the BioMark HD System (Fluidigm) and are plotted as row Z scores of–ΔCT (GAPDH normalised) values. Data are clustered according to Pearson’s correlations and prominent subclusters indicated (SC1-3).
Fig 5.
AP1 and ETS transcription factors are upregulated in human OAC samples.
(A) Heatmap of relative gene expression data for the indicated PEA3 and AP1 family transcription factors from 73 OAC tumour and 17 normal oesophageal biopsy samples. Data were generated by RT-qPCR on the BioMark HD System (Fluidigm) and are plotted as row Z scores of–ΔCT (GAPDH normalised) values. (B) Heatmap of relative expression of the indicated PEA3 and AP1 family transcription factors from microarray gene expression data from normal, Barrett’s and OAC-derived tissue samples (GSE 13898; [4]). In both heatmaps, the rows are clustered by hierarchical clustering and the tumour samples by k-means clustering. (C) Box plots of expression (log2) of ETV4 and FOSL1 expression from microarray data [4] from normal, Barrett’s or OAC-derived samples. Horizontal line indicates median; ** = P-value <0.05 (t test). (D) To determine putative regulatory TFs, the enrichment of transcription factor motifs (-log10 binomial P-value)(x-axis) is plotted against fold difference (log2) in the mean gene expression (from microarray data; [4]) of the transcription factors which potentially bind to the motif, in cancer cells compared to normal cells (y-axis). Those highlighted in red show are deemed significant (linear 1.5 fold change in expression in the same direction as change in accessibility of the associated motif).
Fig 6.
ATAC-seq reveals in patient samples identifies AP1 and ETS proteins as regulators of OAC.
(A) PCA analysis of the 9 tissue samples across the top 50,000 accessibility regions derived from merging reads from all samples and recalling peaks using MACS2. Tumour samples are shown in red and normal samples in blue. (B) Pie chart showing the genomic distribution of the top 50,000 accessible regions across all tissue samples (bottom) or the 1015 differentially more accessible regions (top) in OAC cancer samples (promoter = +/- 1 kb from TSS). (C) Heatmap of normalised Tn5 cleavage events (log2) in 500 bp windows at the regions showing significant differential accessibility between normal and tumour tissue samples (+/- linear fivefold change, P<0.05). Data are subjected to hierarchical clustering using 1-Pearson’s correlation. (D) UCSC browser track showing open chromatin regions at the IHH, ZNF471 and ZFP28 loci in the indicated patient-derived normal and tumour tissue samples. Intergenic and intragenic (grey boxes) and promoter proximal (blue boxes) peaks are highlighted. Aggregated data from either two normal (-N) or three OAC (-T) cell lines is also shown. (E) Heatmap (left) and average tag profile (right) of normalised H3K27ac ChIP-seq tag density (GSM1013127) in normal oesophageal tissue (shown in a +/- 2 kb region around summit of each differentially accessible peak). Average profiles are shown for regions that are either more open in cancer (red line) or more open in normal (blue line) samples. (F) AP-1 and ETS motifs identified via de novo motif discovery, at regions that are more open in cancer (n = 962) against CpG matched background. (G) Heatmaps (left) and average ATAC-seq cleavage events (right) of ATAC-seq tag density in normal (blue line) and tumour (red line) samples from patient 006. Data are shown in a +/- 1 kb region relative to the summit of the ETV1 binding peaks defined by ChIP-seq in OE33 cells. Regions in the heatmaps are ranked according to ETV1 ChIP-seq signal (shown on the left).