Skip to main content

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

Chromatin H3K27me3/H3K4me3 histone marks define gene sets in high-grade serous ovarian cancer that distinguish malignant, tumour-sustaining and chemo-resistant ovarian tumour cells

Abstract

In embryonic stem (ES) cells, bivalent chromatin domains containing H3K4me3 and H3K27me3 marks silence developmental genes, while keeping them poised for activation following differentiation. We have identified gene sets associated with H3K27me3 and H3K4me3 marks at transcription start sites in a high-grade ovarian serous tumour and examined their association with epigenetic silencing and malignant progression. This revealed novel silenced bivalent marked genes, not described previously for ES cells, which are significantly enriched for the PI3K (P<10−7) and TGF-β signalling pathways (P<10−5). We matched histone marked gene sets to gene expression sets of eight normal fallopian tubes and 499 high-grade serous malignant ovarian samples. This revealed a significant decrease in gene expression for the H3K27me3 and bivalent gene sets in malignant tissue. We then correlated H3K27me3 and bivalent gene sets to gene expression data of ovarian tumour ‘stem cell-like’ sustaining cells versus non-sustaining cells. This showed a significantly lower expression for the H3K27me3 and bivalent gene sets in the tumour-sustaining cells. Similarly, comparison of matched chemo-sensitive and chemo-resistant ovarian cell lines showed a significantly lower expression of H3K27me3/bivalent marked genes in the chemo-resistant compared with the chemo-sensitive cell line. Our analysis supports the hypothesis that bivalent marks are associated with epigenetic silencing in ovarian cancer. However it also suggests that additional tumour specific bivalent marks, to those known in ES cells, are present in tumours and may potentially influence the subsequent development of drug resistance and tumour progression.

Introduction

Ovarian cancer is associated with a poor prognosis and only around 40% of women diagnosed with ovarian cancer are alive after 5 years.1 This is partly because symptoms do not present until the malignancy is at an advanced stage. Conventional treatment usually involves cytoreductive surgery and chemotherapy with platinum-based compounds. Although, initially the majority of patients respond well and can continue to respond to multiple rounds of treatment, in most cases tumour eventually recurs in a chemo-resistant form.1, 2 It has been hypothesized that following chemotherapy of ovarian cancer, a specific subpopulation of cancer cells may remain having pluripotent embryonic stem (ES)-cell-like features.3

One of the hallmarks of ES cell chromatin is the presence of bivalent H3 lysine (K) 4 (H3K4me3) and K27 (H3K27me3) methylation, which frequently overlaps with developmental transcription factor genes, which are expressed at very low levels.4 This seems to depend largely on the presence of the repressive H3K27me3 mark. Furthermore it has been suggested that bivalent domains silence developmental genes, while keeping them poised for later activation, once differentiation is progressing or has been completed.4 Genes marked with bivalent marks in ES cells have been hypothesized to be more vulnerable to epigenetic silencing in tumours, including ovarian cancer.5, 6 H3K4 methylation is catalysed by members of the Trithorax, whereas H3K27 methylation is mediated via polycomb-group proteins.7 These proteins are responsible for the mitotic inheritance of lineage-specific gene expression programs4 and perform key developmental functions. Moreover, Polycomb-group proteins have an essential role in ES cell maintenance and upon differentiation show reduced expression. During differentiation, bivalent domains tend to resolve either into active H3K4 or silenced H3K27 methylation.4 Given the hypothesis that a subpopulation of tumour cells may have stem cell-like features, which are responsible for the sustained growth of ovarian cancer,3 we wanted to test whether the ES chromatin hallmark of bivalently marked genes, which is usually lost during differentiation, could be found in a primary ovarian tumour sample or whether different (possibly cancer related genes) may be involved.

For this study we made use of chromatin immunoprecipitation (ChIP) in combination with massively parallel DNA sequencing, ChIPseq,8 of a high-grade (G3) Stage III serous epithelial ovarian cancer sample and identified genes with H3K4me3 and H3K27me3 histone marks within 5 kb of the transcription start site. In order to test whether meaningful correlations with malignancy could be obtained, we matched the H3K4me3, H3K27me3 and bivalent gene sets obtained from our analysis to expression data sets from benign serous tumours and malignant ovarian high-grade serous carcinoma expression data sets. In addition, we investigated expression of these gene sets in subpopulations of ovarian tumour-sustaining cells.9 Such ovarian tumour-sustaining cells isolated from ovarian tumours and cell lines have stem cell-like properties, show more aggressive malignant characteristics such as increased anchorage-independent growth as spheroids and xenografts, and have a more invasive phenotype.10, 11, 12 These tumour-sustaining cells have also been implicated in acquisition of resistance to chemotherapy.3, 9 Ovarian tumour-sustaining cells from the IGROV1 ovarian cancer cell line over-express EZH2 and under-express PRC2 target genes.9 Moreover, knock-down of EZH2 expression reduced the growth of these ovarian tumour-sustaining cells as spheroids and as xenografts in NOD/SCID mice. The frequency of sustaining cells over-expressing EZH2 is increased in drug-resistant ovarian tumours from patients.9 We have therefore examined whether the expression of bivalently and H3K27me3-marked tumour-derived gene sets are enriched in ovarian tumour-sustaining or drug-resistant tumour cell populations.

Results

Extracting histone modification profiles from ChIPseq data

Illumina GA2 sequencing gave collections of short reads for genomic DNA extracted from a primary high-grade serous ovarian tumour sample, selected via ChIP, based on the presence of H3K4me3 and H3K27me3 (see Materials and methods). Genes were associated to ChIP peaks, and subsequently assigned a specific histone modification status if the transcription start site lay within 5 kb (upstream or downstream) of that peak. Gene set 1 (n=580) comprises genes with the ‘bivalent’ signature of overlapping H3K4me3 and H3K27me3 marks, gene set 2 (n=913) contains genes marked exclusively with H3K27me3 (no proximal H3K4me3) and gene set 3 (n=11 066) comprises genes marked with H3K4me3 only (no proximal H3K27me3). Numbers of high-confidence peaks from each ChIP sample were determined as well as genes associated with a certain histone modification status/combination (Table 1). Full lists of these gene sets can be found in Supplementary Table S1. The tumour examined contained no normal adjacent tissue, but did consist of tumour, stromal and tumour-infiltrating cells. To address the possibility of generating ‘bivalent’ marks due to the presence of a mixed (for example, tumour/stromal) cell population, we examined the distribution of enrichment FDR-values evaluated by SICER over those H3K27me3 peaks that occur proximally to a gene (as defined in Materials and methods) with an overlapping H3K4me3 peak and those that occur proximally to genes alone. Comparing the two sets of log-enrichment values with a Student’s t-test revealed no significant difference between these value sets. Similarly, no significant difference was observed between gene-proximal H3K4me3 peak enrichment scores overlapping with H3K27me3 or without H3K27me3 proximal to the same gene. Finally, we determined the distribution of read counts per sequencing run corresponding to the H3K27me3 and H3K4me3 ChIP (see Materials and methods). The t-test showed no significant difference in the average normalized read-count scores for H3K27me3 peaks occurring with overlapping H3K4me3 peaks, when compared with the H3K27me3 peaks proximal to any gene without proximal H3K4me3 peaks. Although a significantly higher normalized read count for H3K4me3 peaks alone was found, this effect was also observed previously for ES bivalent H3K27me3 and H3K4me3 marks.4 Taken together, this argues that the identified bivalent marks are truly bivalent and most likely do not represent varying degrees of tumour heterogeneity.

Table 1 Total number of histone modification peaks and the number of genes associated with each histone modification/combination (see also Supplementary Table 1)

To validate the ChIPseq data, ChIP was done for H3K27me3, H3K4me3 and both marks on the same tumour as used for ChIPseq (Figure 1), followed by quantitative PCR (qPCR) for selected genes from each gene set: ALX1 and COCH representing Gene Set 1 (bivalent marks), EN1 and ZIC4 representing gene set 2 (H3K27me3) and FBX033 and IKBIP representing Gene Set 3 (H3K4me3). The data from ChIP qPCR confirms the ChIPseq data at the loci examined. Thus, ALX1 and COCH show approximately equivalent levels of H3K27me3 and H3K4me3 marks consistent with bivalent marks, while the loci selected from the other gene sets show increased levels of the appropriate mark. We also compared the ChIP qPCR data with that obtained from normal fallopian tissue, which high-grade serous epithelial ovarian cancers are believed to be derived from. Interestingly, while the permissive marked H3K4me3 genes appear also permissively marked in the normal fallopian tissue, the repressed marked loci in the tumour appear more bivalently marked in the normal tissue.

Figure 1
figure1

ChIP qPCR assay. Enrichment of histone marks H3K4me3, H3K27me3 and bivalent marks in selected candidate genes from tumour-specific gene sets in tumour used for ChIPseq (a) and normal fallopian tissue sample (b).

Effect of histone modification on gene expression

To determine whether the histone profiles obtained were in accordance with the gene expression profile for this tumour, we investigated RNA expression data (Data set 1) of the ovarian tumour sample for the ChIPseq gene sets (Bivalent, H3K4me3 and H3K27me3). Using mas5-expression calls (Affymetrix, High Wycombe, UK), probe-sets were categorized as either ‘present,’ ‘marginal’ or ‘absent’. The distribution of probe-sets in each expression category for each Gene Set was compared with the overall distribution across all genes represented on the array. Shown is the relative distribution of histone modification lists for each expression category (Figure 2). In line with previously published data,13 a significant overall shift from ‘present’ towards ‘absent’ is observed for genes with the silencing H3K27me3 mark present proximal to the transcription start site (χ2-test, P<2.2 × 10−16). Inversely, genes with the active H3K4me3 mark proximal to the transcription start site show a significant overall shift towards ‘present’ (χ2-test, P<2.2 × 10−16) when compared with the distribution for all genes. Taken together, the ChIPseq results obtained from a primary ovarian tumour sample correspond as expected to its gene expression data, confirming H3K27me3 as a repressive and H3K4me3 as an active mark. In line with the hypothesis that bivalent genes should be either very low expressed or silenced, the bivalent marked gene set of the tumour sample displayed a shift towards ‘absent’ (chi-squared test, P=8.1 × 10−8). However, this shift was observed to a lesser extent than for the H3K27me3 mark alone, possibly indicating a potential for active transcription of bivalently marked genes in a malignant tumour setting.

Figure 2
figure2

Distribution of mas5 gene-expression calls for probe sets on the HGU133plus2 microarray across each gene set derived from the histone modification profile obtained for the primary tumour sample. The distribution of present, marginal and absent calls across the entire array (all genes) is shown to represent the ‘baseline distribution’ against which each gene set is then compared. Statistically significant differences between the distributions for each gene set are described in the text.

Expression of marked genes in malignant ovarian tumour compared with benign ovarian lesions

Gene expression data taken from eight benign serous ovarian lesions as well as the profiled malignant tumour sample were used to establish whether the gene sets defined via histone marks would reveal any differential expression between the tumour and the benign samples (Data set 2). To compare gene expression data from different samples, we used RMA (Robust Multichip Average) normalization.14 For each of the histone modification gene sets, each gene’s difference in expression level between the profiled malignant tumour and the benign serous ovarian tumours was calculated. Table 2 shows the averages of the per-probe set differences for each histone modification gene set. The presented P-values result from a t-test comparing the values of gene expression changes for each gene set versus the distribution of expression changes of all genes. Gene expression differences are displayed for each histone modification signature as the average of the constituent genes’ differences between the level in the profiled tumour sample and the median level across the benign samples. The bivalent as well as the H3K27me3 marked genes show significantly lower expression in the malignant tumour than the benign samples. For the genes defined by the active H3K4me3 mark a slight, although not significant, shift towards increased expression is detected.

Table 2 Comparison of the profiled tumour gene expression with combined expression data of eight benign ovarian cysts

Expression of histone-marked gene sets in high-grade serous ovarian cancer

To assess whether the association between expression and histone marks would validate across a larger tumour cohort, we interrogated the Cancer Genome Atlas (TCGA) ovarian cancer gene expression data set, consisting of 499 high-grade serous ovarian malignant tumours and eight normal fallopian tube samples from unaffected individuals.15 Raw expression data for all samples were normalized together using RMA. Using the bivalent, as well as the H3K27me3 gene sets defined from a single malignant tumour sample, we confirm increased silencing of these genes to be a general feature of high-grade serous tumours when compared with non-cancerous ovarian tissue (Table 3).

Table 3 ROAST differential expression statistics (P-values shown) across the Cancer Genome Atlas ovarian-cancer gene-expression data set for gene sets defined by different histone marks

In ES cells, bivalent chromatin domains, containing H3K4me3 and H3K27me3 marks, silence developmental genes, while keeping them poised for later activation following differentiation.4 Therefore, we addressed whether the primary malignant tumour defined bivalent gene set was indeed typical for human ES (hES) cell genes or whether high-grade serous ovarian cancer samples had a unique set of bivalently marked genes. For this we compared our gene sets with published lists of genes with bivalent marks in hES cells,16 and genes with H3K27me3 in hES cells:13 see Supplementary Table S1. Of the 580 bivalent genes we identified in the primary tumour sample, 215 were reported as being bivalent in hES cells,16 however, the remaining 365 bivalent genes are observed in the ChIPseq-profiled tumour, but not in hES cells (tumour bivalent). Functional enrichment analysis using PANTHER17 showed that the tumour bivalent group of genes is significantly enriched for the PI3K (P<10−7) and the TGF-β signalling pathway (P<10−5).

H3K27me3 gene set silencing in ovarian cancers is associated with PRC2 complex expression levels

We wanted to address whether expression of the PRC2-complex is associated with the observed silencing of the described gene sets, given its key role in maintaining the H3K27me3 mark.18 For this we correlated the gene expression levels of PRC2 members (EZH1, EZH2, SUZ12, EED, RBBP7) with the expression status of potential PRC2 target genes defined via the presence of H3K27me3 in our investigated ovarian tumour sample. Here, PRC2, as well as H3K27me3 gene expression signature z-scores were calculated using the method of19 for two independent publicly available ovarian cancer gene expression data sets, ‘TCGA’15 and ‘Tothill’.20 Signature scores were calculated as described in Materials and methods. Both the data sets (TCGA n=499, Tothill n=211), showed across the serous epithelial ovarian carcinomas, a negative correlation of the PRC2 signature z-scores with the H3K27me3 signature z-scores, with Pearson correlation coefficients of −0.32 (P=1.1 × 10−9) and −0.38 (P=2.4 × 10−11) for the TCGA and Tothill data sets, respectively. Here, expression levels of PRC2 complex members (at least one probe-set for each of EZH1, RBBP7, EZH2, EED and SUZ12) were significantly negatively correlated with the H3K27me3 expression signature in at least one of the data sets, with two probe sets present on both arrays significant in both data sets (one for EZH2 and one for SUZ12). This analysis supports the H3K27me3 mark, regulated by PRC2, having a key role in mediating gene silencing of this particular set of genes in ovarian cancers.

Expression of epigenetically regulated genes in tumour-sustaining cells

We have isolated tumour-sustaining (side-population) cells with stem cell-like properties from the ovarian tumour cell line, IGROV1.9 These tumour-sustaining cells are more resistant to the drugs carboplatin and paclitaxel and have enhanced ability to grow in an anchorage-independent manner, as well as xenografts in NOD/SCID mice. In addition, the sustaining cell population over-express stem cell markers (NOS signature) and members of PRC2, including EZH2.9 Over-expression of EZH2 is also observed in side population isolated directly from malignant ovarian tumour cells present in patient ascites and is increased in resistant tumours at relapse. Therefore, we examined expression of the tumour-derived gene sets bivalently and H3K27me3 marked in the IGROV1 tumour-sustaining cell populations. Gene-expression data from the Affymetrix HGU133plus2 microarray were previously published for three independent ‘side-population’ samples, as well as for three corresponding ‘non-side-populations’.9 We used the raw gene expression data from this study, normalizing the data from all arrays using RMA. Differences between the median expression level across the side-population samples and the median expression level across the non side-population samples were calculated for each probe set. These differences for probe sets mapping to any gene in a given gene set were compared with the differences for all probe sets using a Student’s t-test. Genes marked with H3K27me3 in the profiled tumour were generally expressed at a significantly lower level in the tumour-sustaining side-population than the non-side-population (P=2 × 10−5), being in line with the EZH2 over-expression in the side-population.9 These effects are illustrated in Figure 3, using the average signature z-scores (calculated as described in Materials and methods) to reflect relative expression in the different cell population samples. Similarly, genes associated with the bivalent mark in the profiled tumour also had a tendency to be expressed at a lower level in the side-population versus the non-side-population (P<0.0001), while genes marked with H3K4me3 in the profiled tumour were generally expressed at a higher level in the side-population than the non-side-population (P=6 × 10−18). Separating the tumour-identified and the hES bivalent marks, it is apparent that the tumour-sustaining side-population shows an increase in silencing of the known hES cell bivalent gene set when compared with the novel tumour identified bivalent marked gene set.

Figure 3
figure3

Average expression signature z-scores in IGROV1 side-population and non-side-population for genes associated with different histone modification states in profiled tumour. The Bivalent Gene Set has been further separated into marks identified in the ChIPseq tumour sample and already known hES bivalently marked genes.

Association of histone modification profiles with resistance to chemotherapy

It has been hypothesized that in ovarian cancer, tumour stem cell-like subpopulations may be responsible for chemo-resistance.3, 9 To detect expression features associated with acquired drug resistance and related to our defined ChIPseq gene sets, we interrogated the PEO1/PEO4 chemo-resistance cell line model. This cell line pair offers the opportunity to study the molecular basis of the acquisition of resistance to platinum-based chemotherapy in ovarian cancer, as both cell lines were derived from the same patient: PEO1 was derived, while the patient still responded to platinum-based chemotherapy, whereas PEO4 was derived when the tumour had acquired resistance to chemotherapy.21

Gene expression profiles were obtained for 4 samples from each of the PEO1 and PEO4 cell lines using the Affymetrix HT-HGU133a GeneChip (Data set 3).22 Raw expression data were normalized using RMA. Signature z-scores were calculated (as described in Materials and methods) for these samples for each of our histone modification gene sets. Here, genes identified by bivalent or repressive (H3K27me3) chromatin marks were generally expressed at a lower level in the resistant PEO4 cell line than the sensitive PEO1 cell line, and genes categorized as H3K4me3 marked were generally expressed at a higher level in PEO4 than PEO1 (P=0.0001, P=0.002 and P<1 × 10−8, for the bivalent, H3K27me3 and H3K4me3 gene sets, respectively). Differential expression statistics were calculated using LIMMA23 for all genes represented on the microarray categorized as H3K27me3 group. 48 of the probe sets representing these genes showed significant differential expression between the PEO1 and PEO4 samples (P<0.05, after Benjamini–Hochberg multiple testing correction). Of these, 33 were expressed at a higher level in PEO1 than PEO4 and only 15 were expressed at a higher level in PEO4 than PEO1. There was a significant over-representation of PEO4 silencing for H3K27me3 marked genes, when compared with the overall distribution of genes that were significantly differentially expressed between the PEO1 and PEO4 samples (χ2-test, P=0.0057, expected numbers 23 and 25). As shown in Figure 4, genes marked by bivalent or H3K27me3 chromatin marks were generally expressed at a lower level in the resistant PEO4 cell line compared with the sensitive PEO1 cell line, and genes categorized as H3K4me3 group were generally expressed at a higher level in PEO4 than PEO1. Unlike the tumour-sustaining side-population (Figure 2), genes with tumour identified, but not hES bivalent marks are not significantly less silenced in the chemo-resistant cell line PEO4 (when compared with the chemo-sensitive line PEO1) than those genes with tumour-identified and hES bivalent marks. The chemo-sensitive PEO1 cell line shows increased expression of hES bivalent as well as H3K27me3 marks, which is consistent with the pattern in the non-side population discussed earlier.

Figure 4
figure4

Average expression signature z-scores for PEO1 and PEO4 samples profiled using Affymetrix HT-HGU133a GeneChips. The bivalent gene set has been separated into marks identified in the ChIPseq tumour sample and already known hES bivalently marked genes.

Discussion

Our analysis has shown the power of interrogating a single primary ovarian tumour sample by ChIPseq against two histone modifications (H3K4me3 and H3K27me3) in combination with publicly available ovarian cancer gene expression data sets. We have identified gene sets based on histone marks, which not only correspond well with the predicted gene expression pattern in the tumour, but importantly also in a large independent ovarian cancer gene expression set. We have identified novel cancer specific bivalent chromatin marks in a malignant high-grade serous epithelial ovarian cancer, which have not been described in hES cells before, and which are enriched for pathways relevant to ovarian cancer: PI3K and TGF-β signalling.24, 25, 26 We have identified cancer-specific H3K27me3-marked genes, which are likely to be targeted by PRC2, as the overall expression level of this gene set is inversely proportional to the expression level of PRC2 members in high-grade serous ovarian tumours. Neither the ChIPseq nor expression data sets are generated from purified tumour cells, but rather from macrodissected samples, which will also contain stroma and infiltrating normal cells. Future studies should explore potential to select out specific subpopulations for analysis or quantifying enrichment of the modifications during tumour progression. Interestingly, increased EZH2 expression in either tumour cells or in tumour vasculature is predictive of poor clinical outcome for ovarian cancer patients.27

Eighty-two (of 1121) known hES H3K27me3 target genes were significantly silenced across the 499 TCGA tumour samples when compared with normal tissue. This represents a significant enrichment over what would be expected by chance (P=0.00039, Supplementary Table S2), implying that hES H3K27me3 target genes are consistently affected across a wide range of tumour samples. Of 1570 genes with the H3K27me3 mark in the profiled primary tumour sample, 60 genes were found to be significantly silenced in the TCGA tumour samples. However, only 20 of these were hES marked (19 of which were bivalent), whereas 40 were tumour specific (20 of which appear to be bivalent). Taking these observations together this favors the idea of an aberrant and possibly random establishment of H3K27me3-mediated silencing through EZH2 over-expression, eventually leading to a subpopulation of cells acquiring hES cell-like features, thereby potentially driving tumourgenesis.

Our data substantiates the observation of tumour-sustaining side-populations having stem cell-like features which require EZH2 expression for their maintenance.9 Here, we found that the tumour-based H3K27me3 gene set showed significantly lower expression in the tumour-sustaining cell population than the non-side population. This tendency was also seen for the bivalent gene set, and the known hES bivalent targets were generally significantly more silenced than the genes with novel tumour-specific bivalent marks.

Finally, the comparison of the chemo-resistant (PEO4) versus the chemo-sensitive (PEO1) cell line showed similar significant differences in silencing of the H3K27me3 and the bivalent Gene Set, as seen in the comparison of tumour-sustaining side-populations versus non-side populations. However, silencing of the novel tumour defined bivalent marks is increased in the resistant cell line over that seen in the tumour-sustaining side-population, and the preferential silencing of hES cell bivalent marks compared with tumour identified bivalent marks is diminished.

Although speculative, the fact that the tumour-sustaining side-population displays a significant increase in silencing of hES bivalent targets, which is accompanied by only very few tumour specific bivalent genes, may point towards a subpopulation of tumour-sustaining cells, which indeed originates from a stem-cell. However, the cancer-stem cell has acquired further bivalently marked genes additional to those observed in hES cells and similar to those found in the primary tumour. Therefore, chemo-resistance in ovarian cancer may originate from a small stem cell population, which has acquired certain types of tumourigenic features, selected by and/or induced by chemotherapy.

Gene silencing by DNA methylation has been implicated as an epigenetic drivers of drug resistance.28 Genes that are bivalent marked in tumour cells at presentation may make these genes more vulnerable to epigenetic fixing of silencing by DNA methylation, as has been suggested for bivalent marks in ES cells and tumour DNA methylation.5, 6 Thus, we suggest that gene sets defined via the H3K27me3 and the bivalent mark may have a role in the malignant progression of ovarian cancer as well as the acquisition and/or maintenance of stem cell-like features, which may make ovarian cancer particularly susceptible to acquired chemo-resistance.

Materials and methods

Tumour sample

The high-grade serous epithelial ovarian tumour from an untreated patient at primary presentation used for ChIPseq was obtained from the Hammersmith Hospital Ovarian Cancer Tumour Bank. Tumour collection for this study was approved by local Ethics Committee and informed written consent was obtained from patient.

ChIPseq sample preparation

The −80 °C-stored tumour sample was fragmented, tumour pieces transferred into phosphate-buffered saline/formaldehyde solution at room temperature and incubated for 20 min. To stop the reaction glycine (final 125 mM) was added and following 5 min incubation at room temperature, spun down at 2000, g for 5 min. The sample was washed twice with ice-cold phosphate-buffered saline and then lysed in lysis buffer (25 mM Hepes pH 7.9, 1.5 mM MgCl2, 10 mM KCl, 0.1% NP-40, freshly added protease inhibitor from Sigma, St Louis, MO, USA), homogenized on ice and spun at 3000 g for 5 min at 4 °C. The pellet was re-suspended in sonication buffer (50 mM Hepes, 140 mM NaCl, 1 mM EDTA, 1% Triton X-100, 0.1% Na-deoxycholate, 0.1% SDS, freshly added protease inhibitors) and sonicated for at least 20 min. The sample was then spun down for 10 min at 4 °C. The chromatin containing supernatant was stored at −80 °C until immunoprecipitation. For immunoprecipitation the29 Nature protocol was followed, with magnetic beads instead of agarose. Rabbit polyclonal antibodies used were H3K4me3 (Millipore, Cat No. 07-473) and H3K27me3 (Millipore, Cat No. 07-449). End-repair of DNA, adding ‘A’ bases to the DNA, ligation of sequencing adaptors to DNA fragments and amplification of adaptor modified DNA by PCR was done according to the recommended Illumina protocol. For gel purification of the ‘SolexaPreGel for ChIPSeq’ a 2% agarose Tris-acetate-EDTA gel was loaded with the entire sample using loading buffer containing Xylene cyanol (Sigma, X4126). Band fragments of 200–300 bp were excised using a Dark Reader and DNA was purified with the Qiagen MinElute Gel Extraction Kit (Qiagen, West Sussex, UK). DNA was eluted with 15 μl EB preheated to 50 °C. To check for optimal concentration and that no unwanted adapters are present the library was subject to a Bioanalyzer DNA 1000 assay (Agilent Technologies, Cheshire, UK).

ChIPseq run

The Illumina (San Diego, CA, USA) Genome Analyser II (Solexa) high throughput DNA sequencing system, CSC Genomics Core Laboratory (GCL) (Imperial College London, London, UK) was used for sequencing.

Peak finding

Individual 38-base reads were aligned to the human genome GRCh37 (hg19) using Bowtie.30 Locations of significant peaks of enriched DNA in each of the ChIP samples were identified using SICER (spatial clustering approach for the identification of ChIP-enriched regions),31 utilizing comparison against a collection of short reads corresponding to unselected input DNA and filtering for significance at FDR <0.001.

ChIP qPCR

ChIP was performed as described in ‘ChIPseq Sample Preparation’. For real-time measurements the 2 × iQ SYBR Green Supermix (Bio-Rad Laboratories, Hemel Hempstead, Hertfordshire, UK), 150 nM or 300 nM primers and 0.4 μl of DNA per 20 μl reaction was used. Low-white 96-well plates (Bio-Rad) on a CFX96 Real-time System/C1000 Thermal Cycler (Bio-Rad) were used: 95 °C for 3 min; 95 °C for 10 min, 56 °C for 10 min, 72 °C for 30 min for 42 cycles followed by a melting curve from 72–95 °C. Each measurement was done in triplicate and the primers used can be found in Supplementary Table 3. Results were calculated as a fold increase of the input control with error bars indicating s.d. in the H3K4me3 or H3K27me3 DNA fraction, respectively.

Data set 1

The transcriptome of the tumour profiled with ChIPseq was profiled using the Affymetrix HGU133plus2 GeneChip. Sample labelling, microarray hybridization and imaging were carried out by the Genome Institute of Singapore, following manufacturer’s instructions.

Data set 2

A set of samples from benign ovarian lesions were collected from patients in Hammersmith Hospital and profiled using the Affymetrix HGU133plus2 GeneChip, as Data set 1.

Data set 3

Four samples were taken from cultures of each of the PEO1 and PEO4 ovarian cancer cell lines, and profiled using the Affymetrix HT-HGU133a GeneChip. About 3 × 106 cells from each line were seeded in 10-ml RPMI1640 (10% fetal Calf Serum, penicillin-streptomycin, L-glutamine). Samples were sent to Lawrence Berkeley National Laboratory microarray facility for RNA extraction, labelling, microarray hybridization and imaging.

Read-count normalization

Due to representation biases across the genome, to assess whether or not bivalent marks appeared to arise from heterogeneous mixtures of cells bearing mutually-exclusive H3K4me3 or H3K27me3 marks, a normalized read-count was calculated as shown in Eq. (1). Here, the max depthChIP represents the greatest number of overlapped reads (either H3K4me3 or H3K27me3) that map to any single base within the peak region. Max depthcontrol represents the equivalent for the input DNA for the same region, total #readsChIP represents the number of aligned reads from the relevant ChIP sequencing run and total #readscontrol represents the number of aligned reads from the input DNA sequencing run.

Gene set test statistics

Statistical tests of differential expression of the gene sets between tumour and normal samples of the TCGA data set were performed using the ROAST method32 as implemented in the ‘limma’ package of Bioconductor. The expression data set was fully quantile normalized to remove any systematic bias between the tumour samples and the normal samples.

Calculation of signature scores

Gene expression z-scores were calculated following the improved method of Gene Set Enrichment Analysis described in Irizarry et al.19 Negative z-scores imply an overall down-regulation of genes in the signature and positive z-scores imply an overall up-regulation of genes in the respective signature. The greater the absolute value of the z-score, the more extreme the observed over/under-expression of the signature.

Publication of data

Raw data from sequencing runs available on NCBI Sequence Read Archive, accession SRP016075. All gene expression data available on Gene Expression Omnibus, accession GSE41500.

Accession codes

Accessions

Gene Expression Omnibus

Sequence Read Archive

References

  1. 1

    Aletti GD, Gallenberg MM, Cliby WA, Jatoi A, Hartmann LC . Current management strategies for ovarian cancer. Mayo Clin Proc 2007; 82: 751–770.

    Article  Google Scholar 

  2. 2

    Berkenblit A, Cannistra SA . Advances in the management of epithelial ovarian cancer. J Reprod Med 2005; 50: 426–438.

    PubMed  Google Scholar 

  3. 3

    Agarwal R, Kaye SB . Ovarian cancer: strategies for overcoming resistance to chemotherapy. Nat Rev Cancer 2003; 3: 502–516.

    CAS  Article  Google Scholar 

  4. 4

    Bernstein BE, Mikkelsen TS, Xie X, Kamal M, Huebert DJ, Cuff J et al. A bivalent chromatin structure marks key developmental genes in embryonic stem cells. Cell 2006; 125: 315–326.

    CAS  Google Scholar 

  5. 5

    Ohm JE, McGarvey KM, Yu X, Cheng L, Schuebel KE, Cope L et al. A stem cell-like chromatin pattern may predispose tumor suppressor genes to DNA hypermethylation and heritable silencing. Nat Genet 2007; 39: 237–242.

    CAS  Article  Google Scholar 

  6. 6

    Widschwendter M, Fiegl H, Egle D, Mueller-Holzner E, Spizzo G, Marth C et al. Epigenetic stem cell signature in cancer. Nat Genet 2007; 39: 157–158.

    CAS  Article  Google Scholar 

  7. 7

    Ringrose L, Paro R . Epigenetic regulation of cellular memory by the Polycomb and Trithorax group proteins. Annu Rev Genet 2004; 38: 413–443.

    CAS  Article  Google Scholar 

  8. 8

    Ho JW, Bishop E, Karchenko PV, Negre N, White KP, Park PJ . ChIP-chip versus ChIP-seq: lessons for experimental design and data analysis. BMC Genomics 2011; 12: 134.

    CAS  Article  Google Scholar 

  9. 9

    Rizzo S, Hersey JM, Mellor P, Dai W, Santos-Silva A, Liber D et al. Ovarian cancer stem cell-like side populations are enriched following chemotherapy and overexpress EZH2. Mol Cancer Ther 2011; 10: 325–335.

    CAS  Article  Google Scholar 

  10. 10

    Hu L, McArthur C, Jaffe RB . Ovarian cancer stem-like side-population cells are tumourigenic and chemoresistant. Br J Cancer 2010; 102: 1276–1283.

    CAS  Article  Google Scholar 

  11. 11

    Szotek PP, Pieretti-Vanmarcke R, Masiakos PT, Dinulescu DM, Connolly D, Foster R et al. Ovarian cancer side population defines cells with stem cell-like characteristics and Mullerian Inhibiting Substance responsiveness. PNAS 2006; 103: 11154–11159.

    CAS  Article  Google Scholar 

  12. 12

    Zhang S, Balch C, Chan MW, Lai HC, Matei D, Schilder JM et al. Identification and characterization of ovarian cancer-initiating cells from primary human tumors. Cancer Res 2008; 68: 4311–4320.

    CAS  Article  Google Scholar 

  13. 13

    Lee TI, Jenner RG, Boyer LA, Guenther MG, Levine SS, Kumar RM et al. Control of developmental regulators by Polycomb in human embryonic stem cells. Cell 2006; 125: 301–313.

    CAS  Article  Google Scholar 

  14. 14

    Irizarry RA, Hobbs B, Collin F, Beazer-Barclay YD, Antonellis KJ, Scherf U et al. Exploration, normalization, and summaries of high density oligonucleotide array probe level data. Biostatistics 2003; 4: 249–264.

    Article  Google Scholar 

  15. 15

    Cancer Genome Atlas Research Network, Integrated genomic analyses of ovarian carcinoma. Nature 2011; 474: 609–615.

    Article  Google Scholar 

  16. 16

    Zhao XD, Han X, Chew JL, Liu J, Chiu KP, Choo A et al. Whole-genome mapping of histone H3 Lys4 and 27 trimethylations reveals distinct genomic compartments in human embryonic stem cells. Cell Stem Cell 2007; 1: 286–298.

    CAS  Article  Google Scholar 

  17. 17

    Thomas PD, Kejariwal A, Guo N, Mi H, Campbell MJ, Muruganujan A et al. Applications for protein sequence-function evolution data: mRNA/protein expression analysis and coding SNP scoring tools. Nucleic Acids Res 2006; 34: W645–W650.

    Article  Google Scholar 

  18. 18

    Schwartz YB, Pirrotta V . Polycomb silencing mechanisms and the management of genomic programmes. Nat Rev Genet 2007; 8: 9–22.

    CAS  Article  Google Scholar 

  19. 19

    Irizarry RA, Wang C, Zhou Y, Speed TP . Gene set enrichment analysis made simple. Stat Methods Med Res 2009; 18: 565–575.

    Article  Google Scholar 

  20. 20

    Tothill RW, Tinker AV, George J, Brown R, Fox SB, Lade S et al. Novel molecular subtypes of serous and endometrioid ovarian cancer linked to clinical outcome. Clin Cancer Res 2008; 14: 5198–5208.

    CAS  Article  Google Scholar 

  21. 21

    Langdon SP, Lawrie SS, Hay FG, Hawkes MM, McDonald A, Hayward IP et al. Characterization and properties of nine human ovarian adenocarcinoma cell lines. Cancer Res 1988; 48: 6166–6172.

    CAS  Google Scholar 

  22. 22

    Stronach EA, Alfraidi A, Rama N, Datler C, Studd JB, Agarwal R et al. HDAC4-Regulated STAT1 activation mediates platinum resistance in ovarian cancer. Cancer Res 2011; 71: 4412–4422.

    CAS  Article  Google Scholar 

  23. 23

    Smyth GK . Linear models and empirical bayes methods for assessing differential expression in microarray experiments. Stat Appl Genet Molec Biol 2004; 3: 3.

    Article  Google Scholar 

  24. 24

    Matsumura N, Huang Z, Mori S, Baba T, Fujii S, Konishi I et al. Epigenetic suppression of the TGF-beta pathway revealed by transcriptome profiling in ovarian cancer. Genome Res 2011; 21: 74–82.

    CAS  Article  Google Scholar 

  25. 25

    Meng Q, Xia C, Fang J, Rojanasakul Y, Jiang BH . Role of PI3K and AKT specific isoforms in ovarian cancer cell migration, invasion and proliferation through the p70S6K1 pathway. Cell Signal 2006; 18: 2262–2271.

    CAS  Article  Google Scholar 

  26. 26

    Zhang HY, Zhang PN, Sun H . Aberration of the PI3K/AKT/mTOR signaling in epithelial ovarian cancer and its implication in cisplatin-based chemotherapy. Eur J Obstet Gynaecol Reprod Biol 2009; 146: 81–86.

    CAS  Article  Google Scholar 

  27. 27

    Lu C, Han HD, Mangala LS, Ali-Fehmi R, Newton CS, Ozbun L et al. Regulation of tumor angiogenesis by EZH2. Cancer Cell 2010; 18: 185–197.

    CAS  Article  Google Scholar 

  28. 28

    Zeller C, Dai W, Steele NL, Siddiq A, Walley AJ, Wilhelm-Benartzi CS et al. Candidate DNA methylation drivers of acquired cisplatin resistance in ovarian cancer identified by methylome and expression profiling. Oncogene 2012; 31: 4567–4576.

    CAS  Article  Google Scholar 

  29. 29

    Nelson JD, Denisenko O, Bomsztyk K . Protocol for the fast chromatin immunoprecipitation (ChIP) method. Nat Protoc 2006; 1: 179–185.

    CAS  Article  Google Scholar 

  30. 30

    Langmead B, Trapnell C, Pop M, Salzberg SL . Ultrafast and memory-efficient alignment of short DNA sequences to the human genome. Genome Biol 2009; 10: R25.

    Article  Google Scholar 

  31. 31

    Zang C, Schones DE, Zeng C, Cui K, Zhao K, Peng W . A clustering approach for identification of enriched domains from histone modification ChIP-Seq data. Bioinformatics 2009; 25: 1952–1958.

    CAS  Article  Google Scholar 

  32. 32

    Wu D, Lim E, Vaillant F, Asselin-Labat M-L, Visvader JE, Smyth GK . ROAST: rotation gene set tests for complex microarray experiments. Bioinformatics 2010; 26: 2176–2182.

    CAS  Article  Google Scholar 

Download references

Acknowledgements

This work was supported by funding from Ovarian Cancer Action; and Cancer Research UK; as well as the Imperial Experimental Cancer Research and Biomedical Research Centres. We are grateful for help from the Gynaecological Oncology team at Hammersmith Hospital and members of the Ovarian Cancer Action Research Centre, in particular Elham Shamsaei, Nona Rama, and Kay Dawson from Dept Pathology. We thank Prof Edison T Liu, Genome Institute of Singapore, and the Agency for Science Technology and Research of Singapore for supporting the microarray expression data on the ovarian tumour and benign samples. Gene expression data generated by The Cancer Genome Atlas (TCGA) Pilot Project established by the NCI and NHGRI were used in the study.

Author information

Affiliations

Authors

Corresponding author

Correspondence to R Brown.

Ethics declarations

Competing interests

The authors declare no conflict of interest.

Additional information

Supplementary Information accompanies the paper on the Oncogene website

Supplementary information

Rights and permissions

Reprints and Permissions

About this article

Cite this article

Chapman-Rothe, N., Curry, E., Zeller, C. et al. Chromatin H3K27me3/H3K4me3 histone marks define gene sets in high-grade serous ovarian cancer that distinguish malignant, tumour-sustaining and chemo-resistant ovarian tumour cells. Oncogene 32, 4586–4592 (2013). https://doi.org/10.1038/onc.2012.477

Download citation

Keywords

  • bivalent genes
  • ChIPseq
  • tumor sustaining cells
  • chemo-resistant cells
  • ovarian cancer

Further reading

Search

Quick links