Abstract
Aberrant hypermethylation of CpG islands (CGI) in human tumors occurs predominantly at repressed genes in the host tissue, but the preceding events driving this phenomenon are poorly understood. In this study, we temporally tracked epigenetic and transcriptomic perturbations that occur in a mouse model of liver carcinogenesis. Hypermethylated CGI events in the model were predicted by enrichment of the DNA modification 5-hydroxymethylcytosine (5hmC) and the histone H3 modification H3K27me3 at silenced promoters in the host tissue. During cancer progression, selected CGIs underwent hypo-hydroxymethylation prior to hypermethylation, while retaining H3K27me3. In livers from mice deficient in Tet1, a tumor suppressor involved in cytosine demethylation, we observed a similar loss of promoter core 5hmC, suggesting that reduced Tet1 activity at CGI may contribute to epigenetic dysregulation during hepatocarcinogenesis. Consistent with this possibility, mouse liver tumors exhibited reduced Tet1 protein levels. Similar to humans, DNA methylation changes at CGI in mice did not appear to be direct drivers of hepatocellular carcinoma progression, rather, dynamic changes in H3K27me3 promoter deposition correlated strongly with tumor-specific activation and repression of transcription. Overall, our results suggest that loss of promoter-associated 5hmC in liver tumors licenses reprograming of DNA methylation at silent CGI during progression. Cancer Res; 76(10); 3097–108. ©2016 AACR.
Introduction
Significant advances have been made in the study of epigenetic reprograming associated with specific cancer and tumor subtypes (1). Much of the focus has been on DNA methylation reprograming or alterations in the posttranslational modifications (PTM) of the DNA-bound histone proteins such as trimethylation at lysine 27 on Histone H3 (H3K27me3; refs. 2, 3). The identification of several new forms of modified cytosine in vertebrate DNA has added to this narrative by providing new patterns for interpretation, hypothesis building, and functional outcomes (4). In particular, 5-hydroxymethlycytosine (5hmC), generated from a 5–methylcytosine (5mC) precursor by the Ten-Eleven-Translocation (TET) enzymes, is hypothesized to be part of a predicted demethylation pathway and it differs significantly in its genomic distributions from that of 5mC (5–8). While 5mC tends to be found over heterochromatic and repetitive portions of the genome and has known roles in the maintenance of transcriptional silencing, 5hmC is largely restricted to the bodies of expressed genes, enhancer elements, and a cohort of promoter regions in many tissues (8–10). Beside a predicted role as a demethylation intermediate, the functional significance of the 5hmC modification is still largely unknown. However there is evidence that the genic levels of 5hmC correlate with the transcriptional activity of the associated genes in tissues while low-level direct enrichment of 5hmC over the core of promoters surrounding the transcription start site (TSS) appears to be related to transcriptional silencing events and maintenance of CpG hypomethylation (11–13).
Normal 5hmC patterns are dramatically altered in several human cancer types such as melanoma and hepatocellular carcinoma, as well as in cultured cancer cell lines and colon cancer (9, 11, 14–16). In previous work, we have identified hypermethylation prone promoters in various human cancer types that overlap with polycomb-regulated gene sets (17). We speculated that these genes are prone to methylation due to their inactivity in cancer host tissue and loss of a demethylase activity (2). A possibility is that 5hmC/Tet enzymes are involved in this process, as loss of Tet activity from such promoters may license the occurrence of aberrantly hypermethylatyed CpG islands (CGI) in cancer.
Previously, we carried out a number of studies investigating the epigenetic perturbations that occur in liver following exposure to the well-studied rodent nongenotoxic carcinogen (NGC), phenobarbital, to elucidate potential drug-induced epigenetic changes that anticipate cellular transformation (13, 18, 19). Phenobarbital is an antiepileptic drug that promotes hepatocarcinogenesis in rodents when administered subsequent to an initiating carcinogen such as N-nitrosodiethylamine (DEN; ref. 20). This results in selective clonal outgrowth of cells harboring activating mutations in the proto-oncogene Ctnnb1, encoding β-catenin, a key mediator in the canonical Wnt signaling pathway (21). In this study, we attempt to delineate the progression of hepatocarcinogenesis through combined epigenetic (5mC, 5hmC, and H3K27me3 patterns) and transcriptomic analysis. Our study tracks the initial epigenetic and transcriptomic perturbations throughout the early stages of toxicologic insult and follows them to an “end-state” that is characteristic of liver tumor formation in three models (21, 22). As a result, we link initial loss of 5hmC at CGI promoters with the occurrence of aberrant hypermethylated CGIs in tumors. This occurs at genes that are repressed in normal liver and remain silent in the resulting tumors, possibly due to the continuous presence of the inactivating mark H3K27me3.
Materials and Methods
Study material and animal treatment
For full description of mouse strains and treatment regimes, please see Supplementary Materials and Methods.
Methyl and hydroxymethyl DNA immunoprecipitation
Genomic DNA was extracted from frozen ground-up samples and fragmented to a range between 300 and 1,000 bp in size (Bioruptor, Diagenode) prior to immunoprecipitation with 5hmC (active motif #39769) or 5mC (Eurogentic # BI-MECY-1000) antibodies. For hydroxymethyl DNA immunoprecipitation (HmeDIP) and methyl DNA immunoprecipitation (MeDIP) protocols, see refs. 13 and 18. DNA was purified using DNA Clean & Concentrator (Zymo Research). 5hmC and 5mC patterns were then generated following dye labeling and hybridization to promoter specific microarrays, either on 2.1M Deluxe mouse promoter tiling microarrays (Roche Nimblegen) for the control, 12 week phenobarbital, Ctnnb1, and Ha-ras tumors or on 1M mouse promoter microarrays (Agilent) for control and NASH hepatocellular carcinoma (HCC) tumor samples.
Genome-wide ChIP sequencing of H3K27me3
Twenty-five micrograms of chromatin was taken for the control liver, a 12-week phenobarbital-treated liver, and a phenobarbital-exposed liver tumor and H3K27me3 chromatin immunoprecipitation (ChIP) carried out using 4 μg of antibody (Millipore, cat # 07-449) as described previously (23). Following this, Illumina libraries were prepared by Active Motif (Active Motif) and samples sequenced by HiSeq and mapped against the mouse reference genome (mm9 build).
Preparation of RNA for strand-specific RNA sequencing
Total RNA was extracted from the 2 control livers, 2 livers of mice treated with phenobarbital for 12 weeks, and 3 tumors that arose through sequential DEN/phenobarbital treatment. Strand-specific mRNA-seq libraries for the Illumina platform were generated through poly-A enrichment and sequenced at BaseClear BV (BaseClear). For details, see Supplementary Data.
Data access
All data for this study can be found at the Gene Expression Omnibus (GEO) in the super series GSE77731. Published mouse liver 5hmC datasets were analyzed from GEO series GSE40540 (13, 19). Published fetal liver RNAseq data were analyzed from GEO sample GSM850909. Mouse embryonic stem cell TET1 genome-wide datasets were analyzed from GEO sample GSM611192.
Results
5-methylcytosine and 5-hydroxymethylcytosine patterns are decoupled during hepatocarcinogenesis progression
As phenobarbital is thought to promote carcinogenesis through nongenotoxic mechanisms, epigenetic reprograming through long-term exposure is likely to be an important indicator of liver tumor formation. Our work has shown that mice exposed to tumor-promoting doses of phenobarbital exhibit changes to both their 5hmC and 5mC liver patterns at the promoter proximal regions of many genes (13, 18, 19). Here we expand upon these findings to investigate perturbations arising in hepatocellular tumors generated by a DEN/phenobarbital regimen (Supplementary Fig. S1A).
Immunohistologic staining of sections of mouse liver, following DEN/phenobarbital exposure, with antibodies against glutamine synthetase identify Ctnnb1-mutated hepatocellular tumor lesions, which are dependent on an activated Wnt/β-catenin signaling pathway (Fig. 1A; ref. 24). Staining with antibodies against 5mC or 5hmC reveals that these Ctnnb1-mutated tumors have reduced signals for both DNA modifications (Fig. 1A), an observation also reported in several types of human cancer including HCC (16). DNA modification patterns were directly investigated by immunoprecipitation with specific antibodies towards either 5hmC or 5mC prior to hybridization on high-density promoter microarrays with a set of control mouse liver DNAs (n = 2) and liver DNAs from mice that had been exposed to phenobarbital in their drinking water for 12 weeks (n = 2) or in the excised hepatic tumors from mice that had received phenobarbital for 35 weeks (n = 3). Analysis of overall Pearson correlation scores between the samples reveals that the 5hmC pattern is strongly perturbed following drug dosing (and in the tumors), while 5mC patterns are largely similar following drug dosing but differ greatly in the tumor samples (Fig. 1B). This result suggests that a change in 5hmC profiles might precede that of 5mC.
Levels and distributions of DNA modifications were altered following phenobarbital exposure and in the resulting liver tumors. A, immunostaining of liver sections with glutamine synthetase (GS) revealed Ctnnb1-mutated nodules. 5hmC and 5mC levels were reduced in the tumor relative to the surrounding regions. B, Pearson correlation clustering of datasets for 5hmC (i) and 5mC (ii) profiles. C, heatmap analysis of log2 5hmC (blue) and 5mC (red) signals over promoter regions (TSS ± 2 kb) ranked by tumor 5mC level (Tum 1–3, right). Cntl, normal control livers; PB, phenobarbital-exposed for 12 weeks; Tum, Ctnnb1-mutated liver tumors. CGIs are denoted by black bars in the far right plot. Average 5hmC and 5mC patterns are plotted below (solid black lines), against the mean signal from the control samples (dashed green line). Dashed box outlines promoters that lose 5hmC/gain 5mC in the tumors. D, changes in promoter core 5hmC versus 5mC are plotted for each gene promoter between replicate control livers (i), between average control: 12-week phenobarbital-treated livers (ii), and between average control and average tumor samples (iii). E, Venn diagram displaying the level of overlap in promoters that both lose 5hmC and gain 5mC in three Ctnnb1 tumor samples relative to the control livers. The total number of hypo-5hmC/hyper-5mC promoter cores in each tumor is shown between brackets. F, examples of two promoters that exhibit phenobarbital loss of 5hmc/tumor gain of 5mC. All data are plotted between log2 −1.5 and log2 +1.5.G. Heatmap analysis of log2 5hmC (blue) and 5mC (red) signals over the promoter regions (TSS ± 2 kb) present on the array ranked by Ha-ras tumor 5mC level for average normal control livers (n = 2), average Ctnnb1 tumors (n = 3), and a Ha-ras–mutated liver tumor. CGIs are denoted by black bars in the far right plot. Average 5hmC and 5mC patterns across all promoters are plotted below.
Levels and distributions of DNA modifications were altered following phenobarbital exposure and in the resulting liver tumors. A, immunostaining of liver sections with glutamine synthetase (GS) revealed Ctnnb1-mutated nodules. 5hmC and 5mC levels were reduced in the tumor relative to the surrounding regions. B, Pearson correlation clustering of datasets for 5hmC (i) and 5mC (ii) profiles. C, heatmap analysis of log2 5hmC (blue) and 5mC (red) signals over promoter regions (TSS ± 2 kb) ranked by tumor 5mC level (Tum 1–3, right). Cntl, normal control livers; PB, phenobarbital-exposed for 12 weeks; Tum, Ctnnb1-mutated liver tumors. CGIs are denoted by black bars in the far right plot. Average 5hmC and 5mC patterns are plotted below (solid black lines), against the mean signal from the control samples (dashed green line). Dashed box outlines promoters that lose 5hmC/gain 5mC in the tumors. D, changes in promoter core 5hmC versus 5mC are plotted for each gene promoter between replicate control livers (i), between average control: 12-week phenobarbital-treated livers (ii), and between average control and average tumor samples (iii). E, Venn diagram displaying the level of overlap in promoters that both lose 5hmC and gain 5mC in three Ctnnb1 tumor samples relative to the control livers. The total number of hypo-5hmC/hyper-5mC promoter cores in each tumor is shown between brackets. F, examples of two promoters that exhibit phenobarbital loss of 5hmc/tumor gain of 5mC. All data are plotted between log2 −1.5 and log2 +1.5.G. Heatmap analysis of log2 5hmC (blue) and 5mC (red) signals over the promoter regions (TSS ± 2 kb) present on the array ranked by Ha-ras tumor 5mC level for average normal control livers (n = 2), average Ctnnb1 tumors (n = 3), and a Ha-ras–mutated liver tumor. CGIs are denoted by black bars in the far right plot. Average 5hmC and 5mC patterns across all promoters are plotted below.
Next, we defined differentially hydroxymethylated regions (dHMR) or differentially methylated regions (dMR) between the control and phenobarbital-exposed livers or between the control and tumor samples (Supplementary Fig. S1B). Mapping of the resulting dHMRs and dMRs to one of five genomic compartments (intergenic, promoter distal, promoter proximal, promoter core, or intra-genic; Supplementary Fig. S1B) reveals that genomic regions of epigenetic perturbation are largely distinct for the two modifications. Following phenobarbital exposure both at 12 weeks and in the resulting liver tumors, 5hmC was strongly lost over the promoter proximal and core regions and both reduced and elevated in the bodies of genes (Supplementary Fig. S1B). In contrast, an increased number of 5mC peaks were observed in the tumor samples compared with those exposed to short-term phenobarbital. In addition, although 5mC was seen to be both acquired and lost at intergenic and intragenic loci following 12-week phenobarbital exposure, hypomethylation was typically observed at such sites in the resulting tumors.
Tumor-specific promoter hypermethylation occurs after phenobarbital-mediated loss of 5hmC
As the 5hmC-modified DNA is itself derived from TET-oxidized hydroxylation of a 5mC precursor, we set out to test the relative changes in both marks across all of the promoters. First, we identified a subset of promoter core regions that are marked by 5hmC in the normal livers in this study (Fig. 1C). Independent validation of select loci by glucosylation-mediated restriction enzyme–sensitive qPCR (gRES-qPCR; see Supplementary Experimental Procedures) reveals that these sites contain approximately 15% to 20% 5hmCpG, while a control region of low 5hmC and 5mC at Gapdh was only found to contain less than 5% 5hmCpG (Supplementary Figs. S2 and S3).
Analyses of the DNA modification patterns reveal that there is a dramatic loss in promoter core 5hmC levels in the livers of mice exposed to phenobarbital for 12 weeks as well as in the resulting tumors (Supplementary Fig. S4). In contrast, 5mC levels were unaffected following chronic phenobarbital dosing. However, we observed a strong acquisition of 5mC at a subset of CpG islands in the tumor samples, the majority of which were initially marked by 5hmC in the healthy liver (Fig. 1C and D and Supplementary Figs. S4 and S5; Supplementary Table S1). Extension of this analysis with published liver datasets from mice receiving different lengths of phenobarbital dosing reveals that loss of promoter core 5hmC does not occur following acute drug exposure (i.e., 1/7 days) but instead requires longer chronic exposure (i.e., 28-day dosing; Supplementary Fig. S6; ref. 19). A reciprocal 5hmC loss/5mC accumulation was also observed at a common set of promoter elements in the phenobarbital-exposed tumor samples (n = 2,037), indicating that such a “switch” may be a hallmark of hepatocarcinogenesis progression (Fig. 1E and F).
Promoter hypo-5hmC/hyper-5mC is a common feature of mouse liver tumor types with differing activating mutations
To test whether the promoter core loss of 5hmC and gain of 5mC is a feature of chemical exposure or is instead a more general hallmark of hepatocarcinogenesis, we also profiled 5hmC and 5mC patterns in two mouse liver tumors of differing pathology (Supplementary Fig. S7). One was a mouse liver tumor that had arisen following DEN induction only, resulting in a Ha-ras–mutated tumor (22). The second set of liver tumors resulted from an obesity-based mouse model in which neonatal male mice develop multiple HCCs following nonalcoholic steatohepatitis (NASH) onset (21). In both cases, we observed a gain of promoter core 5mC at loci normally marked by 5hmC in the healthy tissue (Fig. 1G and Supplementary Fig. S7). The reciprocal nature of the changes in 5hmC/5mC persist to some degree in all of the three tumor types with a general loss of 5hmC accompanied by a gain in 5mC (Supplementary Figs. S7 and S8). While there was a large degree of commonality in the promoter core spanning probes that exhibited a loss of 5hmC or gain of 5mC between the three tumor types, there were also a number of regions that were unique to the particular tumor type, which may reflect stratification of particular cancer subtypes (Supplementary Fig. S9).
Promoter epigenetic dysregulation events are related to only a handful of transcriptional perturbations
To test the relationship between promoter core epigenetic dysregulation and transcriptional perturbation during hepatocarcinogenesis progression, we carried out RNA-sequencing (RNA-seq) on matched control livers (n = 2), phenobarbital-treated livers (n = 2), and DEN/phenobarbital-induced Ctnnb1-mutated liver tumor samples (n = 3). Pearson correlation analysis and principal component analysis (PCA) reveals that the global transcriptomic patterns of the Ctnnb1-mutated tumors were distinct from both control and phenobarbital-treated liver samples (Supplementary Fig. S10). No clear relationship was evident between changes in the epigenetic state at promoter core elements with expression alterations of the associated genes (Fig. 2A). This was most notable at the promoters of genes exhibiting no significant change in gene expression following phenobarbital exposure or in the resulting tumors (Fig. 2A, gray plots). However there is a significant retention of 5hmC levels over the promoter cores of the genes that are repressed in the tumor samples (Fisher P, 1.99E−0.7) and a reduction of 5mC levels over tumor-induced genes (Fisher P, 2.52E−0.15); implying that there may be a functional relationship between epigenetic and transcriptomic perturbations in the progression of hepatocarcinogenesis (Fig. 2B). A handful of genes exhibit a strong epigenetic remodeling associated with transcriptional activation, in particular, the cytochrome P450 genes, Cyp2b10 and Cyp2c55, which in addition to being two of the most strongly induced genes following phenobarbital exposure, undergo a strong loss of both promoter core 5hmC and 5mC following phenobarbital exposure (Fig. 2C). In agreement with recent research, the genes that were identified as containing promoter core elements that lost 5hmC and gained 5mC upon tumorigenesis correspond to low expressed/transcriptionally silent genes in the healthy control liver (Fig. 2D), highlighting that aberrant methylation at these loci is not directly linked to a change in their transcriptional status in the tumor (17).
Epigenetic perturbations weakly associate with changes in transcriptional state during hepatocarcinogenesis. A, scatter plot analysis of change in promoter core 5hmC (i) and 5mC (ii) signal versus log2 fold change in expression following phenobarbital (PB) exposure or resulting tumors. Green, induced (>log2 2-fold elevated); red, repressed (>log2 2-fold repressed); gray, unchanged expression. Select genes are highlighted. B, boxplot of change in promoter core 5hmC (i) and 5mC (ii) over the total gene set (gray), induced (green), and repressed (pink) genes between control and tumor samples. Although typically following the global trend (i.e., loss of 5hmC/gain in 5mC in all three gene sets), there is significantly more 5hmC over the promoter cores of repressed genes and less 5mC over induced gene sets. Fisher P values are shown between plots. C, example of epigenetic perturbation and transcriptional changes at the Cyp2b10 gene. 5mC and 5hmC microarray data plotted from log2 −1.5 to log2 +1.5. RNA-seq data plotted from RPKM of 0 to 2,000. D, average RPKM value across total gene set for control (C1, C2, greens), phenobarbital (PB1, PB2, blues), and tumor (T1, T2, and T3, reds).
Epigenetic perturbations weakly associate with changes in transcriptional state during hepatocarcinogenesis. A, scatter plot analysis of change in promoter core 5hmC (i) and 5mC (ii) signal versus log2 fold change in expression following phenobarbital (PB) exposure or resulting tumors. Green, induced (>log2 2-fold elevated); red, repressed (>log2 2-fold repressed); gray, unchanged expression. Select genes are highlighted. B, boxplot of change in promoter core 5hmC (i) and 5mC (ii) over the total gene set (gray), induced (green), and repressed (pink) genes between control and tumor samples. Although typically following the global trend (i.e., loss of 5hmC/gain in 5mC in all three gene sets), there is significantly more 5hmC over the promoter cores of repressed genes and less 5mC over induced gene sets. Fisher P values are shown between plots. C, example of epigenetic perturbation and transcriptional changes at the Cyp2b10 gene. 5mC and 5hmC microarray data plotted from log2 −1.5 to log2 +1.5. RNA-seq data plotted from RPKM of 0 to 2,000. D, average RPKM value across total gene set for control (C1, C2, greens), phenobarbital (PB1, PB2, blues), and tumor (T1, T2, and T3, reds).
Key signaling pathways are perturbed during hepatocarcinogenesis progression
We next identified differentially expressed genes that arise following phenobarbital exposure (control to phenobarbital) or upon tumor formation (control to tumor; >log2 2-fold change with associated unadjusted P values <0.05). Genes exhibiting a change in their gene expression patterns were grouped into one of six classes (“i”–“vi”, Fig. 3A and B; Supplementary Table S2) based on their profile in control, phenobarbital-treated samples, and the tumors. Functional analysis of the genes in each class reveals that distinct pathways are deregulated following phenobarbital exposure as well as in the phenobarbital-exposed liver tumors (Fig. 3B). For example, genes deregulated exclusively in the Ctnnb1-mutated tumors are linked to pathways involved in production of cell adhesion molecules, diabetes, PPAR signaling, and TGFβ signaling pathway, as well as genes frequently mutated or transcriptionally perturbed in the progression of many cancers (such as the Wnt signaling protein axin1 and the cell-cycle regulator ccnd1; refs. 25, 26). Close inspection of the data reveals strong changes in the expression of genes related to control of proliferation (Cdkn1a), apoptosis, and DNA damage response (Trp53 and Gadd45b), as well as genes associated with known cancer-related signaling pathways (the Tfgα gene, the Tgfβ pathway–related tumor suppressor, Chd1, and the Wnt signaling component, Ctnnb1). In addition, we observe strong phenobarbital-specific responses at select genes previously reported to exhibit transcriptional changes following toxicologic challange (Gtl2, Prom1, and Cytochrome P450 genes; Fig. 3C). Together, these data highlight progressive transcriptomic misregulation that is indicative of physiologic changes and a detoxification response.
The normal liver transcriptome is perturbed following phenobarbital (PB) exposure and in resulting tumor samples. A, sets of genes, which change their transcriptional activity following either phenobarbital treatment, in the phenobarbital tumors, or in both sets of samples, were stratified into one of 6 groups (i–vi) plotted as change in expression relative to the control liver group. Examples of select induced genes (blue) and repressed genes (red) are shown under each group in a dashed box. B, plots of P values associated with significantly enriched pathway-based analysis of groups of genes induced (i–iii) or repressed (iv–vi) following phenobarbital treatment or in the tumor samples relative to control livers. No pathways were found enriched in the gene set from group vi (down phenobarbital only). C, plots of RPKM values (reads per kilobase per million) over candidate genes of interest that have potential roles in cell signaling, cancer progression, and xenobiotic response. Roman numeral above each gene is related to the groups identified in Fig. 3A. D, plots of change in the levels of established fetal and adult liver–specific transcripts between control livers to those exposed to phenobarbital for 12 weeks (blue bars), to phenobarbital-exposed liver tumors (red bars), or to fetal livers (yellow bar). Change in expression is plotted as log2 fold change versus average signal from the control livers. E, fold change in expression signals for fetal and adult liver transcripts taken from published microarrays relative to control livers for phenobarbital-exposed Ctnnb1-mutated liver tumors (red), non-phenobarbital–driven Ha-Ras–mutated tumors (pink), and long-term phenobarbital-exposed livers without tumors (gray).
The normal liver transcriptome is perturbed following phenobarbital (PB) exposure and in resulting tumor samples. A, sets of genes, which change their transcriptional activity following either phenobarbital treatment, in the phenobarbital tumors, or in both sets of samples, were stratified into one of 6 groups (i–vi) plotted as change in expression relative to the control liver group. Examples of select induced genes (blue) and repressed genes (red) are shown under each group in a dashed box. B, plots of P values associated with significantly enriched pathway-based analysis of groups of genes induced (i–iii) or repressed (iv–vi) following phenobarbital treatment or in the tumor samples relative to control livers. No pathways were found enriched in the gene set from group vi (down phenobarbital only). C, plots of RPKM values (reads per kilobase per million) over candidate genes of interest that have potential roles in cell signaling, cancer progression, and xenobiotic response. Roman numeral above each gene is related to the groups identified in Fig. 3A. D, plots of change in the levels of established fetal and adult liver–specific transcripts between control livers to those exposed to phenobarbital for 12 weeks (blue bars), to phenobarbital-exposed liver tumors (red bars), or to fetal livers (yellow bar). Change in expression is plotted as log2 fold change versus average signal from the control livers. E, fold change in expression signals for fetal and adult liver transcripts taken from published microarrays relative to control livers for phenobarbital-exposed Ctnnb1-mutated liver tumors (red), non-phenobarbital–driven Ha-Ras–mutated tumors (pink), and long-term phenobarbital-exposed livers without tumors (gray).
We also observed specific expression of a series of genes in the tumor samples normally associated with fetal liver (Fig. 3D). For example, α-fetoprotein 1 (Afp1) and Glypican-3 (Gpc3), as well as the hepatocyte nuclear factors 1a, 1b, and 4a (Hnf1a, 1b, and 4a), are highly expressed in the developing fetal liver compared with the mature adult tissue, while there is a reduced expression of albumin (Alb) and CCAAT/enhancer-binding protein alpha (C/EBPa), which are normally found in the adult liver (Fig. 3D). This may represent either a dedifferentiation of mature hepatocytes into a more fetal-like state or positive selection of liver cancer stem cells during tumorigenesis, which would have a potential growth advantage due to their elevated proliferative abilities (27). Similar changes were also observed in published expression array datasets for both phenobarbital-exposed (Ctnnb1 mutated) and non-phenobarbital–exposed (Ha-ras mutated) liver tumors (Fig. 3E; ref. 28).
A unique polycomb-mediated repression signature at CGIs is established in phenobarbital-exposed mouse liver tumors
It has been widely reported that the global levels and genome wide patterns of the repressive histone modification, H3K27me3), as well as the enzymes responsible for its deposition, are altered in many cancer types including HCC (29–31). We carried out genome-wide ChIP sequencing for the H3K27me3 modification to determine whether these polycomb-mediated silencing mechanisms are also perturbed in the phenobarbital-derived liver tumors. Peak finding analysis indicates a general elevation in the levels of H3K27me3 signals in the phenobarbital liver tumors compared with control liver (94,341 (200 bp) window regions gain H3K27me3, 31,995 (200 bp) windows lose H3K27me3; Fig. 4A and Supplementary Fig. S11).
Polycomb signatures are perturbed in the phenobarbital (PB)-exposed and Ctnnb1-mutated liver tumors. A, distribution of hyper- and hypo-H3K27me3 peak windows normalized to the number of each feature. Hyper-H3K27me3 peaks, blue bars; hypo-H3K27me3 peaks, yellow bars. Promoter distal, TSS −2 kb to −1 kb, promoter proximal; TSS −1 kb to −250 bp, promoter core TSS ± 250 bp; intragenic, TSS + 250 bp to TES; intergenic, all remaining features. B, box plot of promoter core H3K27me3 signals for control (gray), 12-week phenobarbital (yellow), and tumor (maroon) samples. C, density scatter plot of change in promoter-core DNA modification state versus change in H3K27me3 signal for average 5hmC and 5mC upon phenobarbital exposure (i and iii) or in the tumor samples (ii and iv). D, plots of average control and tumor H3K27me3 patterns across the total promoter cohort, promoters identified as hypo-5hmC/hyper-5mC in the tumors, promoters of tumor-induced genes, and promoters of tumor-repressed genes. Control liver patterns, black line; tumor patterns, brown line; control liver H3K27me3 pattern from total promoter set, dashed green line. Plots cover promoter regions TSS ± 2 kb. E, examples of H3K27me3 ChIP-seq patterns, 5hmC, and 5mC microarray patterns and RNA-seq expression states at loci identified in D. F, average patterns of several histone modifications in the control livers at the tumor hypo-5hmC/hyper-5mC promoter regions. Black line, patterns across the total promoter set. Plots shown for TSS ± 2 kb regions. G, heatmap analysis of the bivalency modifications H3K27me3 and H3K4me3, DNA modifications, and expression status over the tumor hypo-5hmC/hyper-5mC promoter core regions. A random selection of 1,000 promoters was tested as a comparison. H3K27me3, green; H3K4me3, purple; 5hmC and 5mC, blue (high) and red (low), respectively; RNAseq RPKM, gray.
Polycomb signatures are perturbed in the phenobarbital (PB)-exposed and Ctnnb1-mutated liver tumors. A, distribution of hyper- and hypo-H3K27me3 peak windows normalized to the number of each feature. Hyper-H3K27me3 peaks, blue bars; hypo-H3K27me3 peaks, yellow bars. Promoter distal, TSS −2 kb to −1 kb, promoter proximal; TSS −1 kb to −250 bp, promoter core TSS ± 250 bp; intragenic, TSS + 250 bp to TES; intergenic, all remaining features. B, box plot of promoter core H3K27me3 signals for control (gray), 12-week phenobarbital (yellow), and tumor (maroon) samples. C, density scatter plot of change in promoter-core DNA modification state versus change in H3K27me3 signal for average 5hmC and 5mC upon phenobarbital exposure (i and iii) or in the tumor samples (ii and iv). D, plots of average control and tumor H3K27me3 patterns across the total promoter cohort, promoters identified as hypo-5hmC/hyper-5mC in the tumors, promoters of tumor-induced genes, and promoters of tumor-repressed genes. Control liver patterns, black line; tumor patterns, brown line; control liver H3K27me3 pattern from total promoter set, dashed green line. Plots cover promoter regions TSS ± 2 kb. E, examples of H3K27me3 ChIP-seq patterns, 5hmC, and 5mC microarray patterns and RNA-seq expression states at loci identified in D. F, average patterns of several histone modifications in the control livers at the tumor hypo-5hmC/hyper-5mC promoter regions. Black line, patterns across the total promoter set. Plots shown for TSS ± 2 kb regions. G, heatmap analysis of the bivalency modifications H3K27me3 and H3K4me3, DNA modifications, and expression status over the tumor hypo-5hmC/hyper-5mC promoter core regions. A random selection of 1,000 promoters was tested as a comparison. H3K27me3, green; H3K4me3, purple; 5hmC and 5mC, blue (high) and red (low), respectively; RNAseq RPKM, gray.
Focusing on promoter core regions, we find that these are relatively unchanged in response to phenobarbital but elevated in the resulting tumor (Fig. 4B). Studies suggest that promoter regions marked by H3K27me3 in mouse embryonic stem cells become hypermethylated upon mammary cancer formation (32). To test this possibility in our mouse liver tumor, we directly compared our H3K27me3 and 5hmC/5mC datasets. Plots of promoter core 5hmC and 5mC change against changes in the H3K27me3 signal following phenobarbital exposure reveal little change in the histone modification, even though 5hmC levels are lost at a number of loci (Fig. 4C). However, compared with the normal genomic profile, there is a strong increase in the number of hyper-H3K27me3 peaks at promoter spanning loci upon tumor progression. Interestingly a significant proportion of promoters acquire 5mC without a change in H3K27me3 occupancy (Fig. 4C). Intriguingly, we observed only a very modest elevation of the H3K27me3 mark at promoters following phenobarbital exposure but a strong elevation in signal is observed at many promoters in the resulting tumors (Fig. 4D and Supplementary Fig. S12). Promoters that become hypermethylated and hypohydroxymethylated in the tumors do not exhibit this elevation as they are already strongly enriched for H3K27m3 in normal healthy livers (Fig. 4D and E). In agreement with the notion that the H3K27me3 mark is inhibitory towards transcription, analysis of the promoter regions of genes induced in the tumor indicates full loss of H3K27me3 is coincident with gene activation in the resulting tumors. Conversely the genes that become repressed in the tumor have dramatic gains of H3K27me3 over their promoters in the resulting tumors (Fig. 4D and E and Supplementary Fig. S12).
To further test the chromatin landscape of this promoter set in normal control livers, we analyzed the profiles of several other commonly studied histone tail modifications using publicly available mouse liver epigenomics datasets produced by the ENCODE project (http://www.genome.gov/encode/; Fig. 4F). Although there was no observable difference, compared with the total promoter set, in the levels of Histone H3 lysine 4 mono-methylation (H3K4me1), H3 lysine 27 acetylation (H3K27ac), or H3 lysine 36 tri-methylation (H3K36me3), there was an enrichment at these promoters for the typically euchromatic mark H3 lysine 4 tri-methylation (H3K4me3). This suggests that a proportion of these 5hmC-marked promoters are bivalent (simultaneously enriched for H3K4me3 and H3K27me3).Combined analysis of the promoters that lose 5hmC and gain 5mC in the tumors (n = 2,035) reveals that a strong epigenetic transition occurs during hepatocarcinogenesis at promoter core elements originally bivalent/5hmC marked and transcriptionally silent genes in the normal healthy liver, without leading to expression changes (Fig. 4G and Supplementary Figs. S13 and S14).
Tet1 protein levels are reduced in mouse HCC
To better understand the epigenetic dysregulation occurring during hepatocarcinogenesis, we investigated the transcriptomic changes in a selection of key epigenetic-modifying enzymes (Fig. 5A). In general, the majority did not exhibit a strong change in gene expression (>2-fold change vs. control tissue). We did, however, observe a strong induction of the histone H3 lysine 9 demethylase Jmjd1c and the histone deacetylase Hdac11 alongside a low to moderate elevation (1.5–2 fold) of several HDACs, methyl-binding proteins (Mbd1), Trithorax (Mll3), and polycomb group proteins (Suz12). The levels of the methyltransferases Dnmt-1 (1.64 FC), -3a (1.51 FC), and -3b (0.98 FC) were not significantly altered. We did not observe changes in the levels of expression of the Tet genes (Tet1, 2, and 3) responsible for the conversion of 5hmC from 5mC. This result was validated both by microarray and qRT-PCR; thus, loss of 5hmC in mouse HCC cannot be explained by the misexpression of the Tet genes (Supplementary Fig. S15; refs. 14, 19, 33). As levels of the Tet1 protein were previously shown to be reduced in human HCC, we carried out IHC for Tet1 in mouse liver tumors (Fig. 5B; ref. 16). In agreement with the results of the human study, we did observe a strong loss of Tet1 staining in MAPK-positive tumor cells. Although levels of Tet1 expression in the mouse liver are low (supplementary Fig. S15), this result indicates that protein levels are readily detectable in the normal healthy liver tissue. Published in vitro studies have shown that loss of Tet1 is coincident with the loss of 5hmC from promoter regions (9, 34, 35). Moreover, in human embryonic carcinoma NCCIT cells, Tet1 loss was seen to correlate with hypermethylation of the CGIs and CpG island shores (9). Interestingly, analysis of Tet1-binding sites in published mESC data reveals an enrichment over the promoters that we identified as becoming aberrantly hypermethylated in mouse liver tumors (Supplementary Fig. S16). To test for the consequences resulting from loss of the Tet1 protein observed in the liver tumors, we profiled 5hmC levels across the promoter regions in control and Tet1 KO mouse livers by hmeDIP-seq (36). This revealed that loss of Tet1 in the mouse liver results in a dramatic reduction of 5hmC over a large number of promoters (Fig. 5C). This was also observed over the promoter regions identified as becoming aberrantly methylated following a loss of 5hmC during tumorigenesis (Supplementary Fig. S17), suggesting that direct reduction in Tet1 protein levels or inhibition of TET activity during the progression of tumorigenesis could account for a decrease of promoter-associated 5hmC and enable subsequent gain of 5mC.
The role of Tet1 in the maintenance of epigenetic state during hepatocarcinogenesis. A, heatmap of expression change (fold change in RPKM) for key epigenetic proteins following phenobarbital (PB) exposure and in tumor samples. Green, induced; red, repressed. B, immunostaining of control healthy liver and two liver tumors for glutamine synthase (GS; hepatocytes at central vein), MAPK (tumor-positive cells), and Tet1. C, heatmap analysis of 5hmC landscape at promoters (TSS ± 2 kb) in control healthy liver versus those in a Tet1-deficient mouse (Tet1−/−). 5hmC levels ranked by control promoter core level. Blue, high; red, low. D, model illustrating how Tet1 may regulate the CGI epigenetic landscape in liver. Examples of epigenetic landscapes are shown for the promoter regions of genes transcriptionally induced or repressed during hepatocarcinogenesis or at the promoters of genes, which exhibit aberrant promoter hypohydroxymethylation/hypermethylation in the liver tumors. H3K27me3, orange triangle; H3K4me3, green oval; 5hmCpG, blue lollypop; 5mCpG, black lollypop; unmodified CpG, white circle. Possible mechanisms to explain the role of Tet1 at CGIs are outlined. Transcription factor (TF)-binding events are also shown. For reference, the global changes in the three epigenetic marks are shown alongside with a gain in a mark represented by a green arrow and a loss in the mark by a red arrow.
The role of Tet1 in the maintenance of epigenetic state during hepatocarcinogenesis. A, heatmap of expression change (fold change in RPKM) for key epigenetic proteins following phenobarbital (PB) exposure and in tumor samples. Green, induced; red, repressed. B, immunostaining of control healthy liver and two liver tumors for glutamine synthase (GS; hepatocytes at central vein), MAPK (tumor-positive cells), and Tet1. C, heatmap analysis of 5hmC landscape at promoters (TSS ± 2 kb) in control healthy liver versus those in a Tet1-deficient mouse (Tet1−/−). 5hmC levels ranked by control promoter core level. Blue, high; red, low. D, model illustrating how Tet1 may regulate the CGI epigenetic landscape in liver. Examples of epigenetic landscapes are shown for the promoter regions of genes transcriptionally induced or repressed during hepatocarcinogenesis or at the promoters of genes, which exhibit aberrant promoter hypohydroxymethylation/hypermethylation in the liver tumors. H3K27me3, orange triangle; H3K4me3, green oval; 5hmCpG, blue lollypop; 5mCpG, black lollypop; unmodified CpG, white circle. Possible mechanisms to explain the role of Tet1 at CGIs are outlined. Transcription factor (TF)-binding events are also shown. For reference, the global changes in the three epigenetic marks are shown alongside with a gain in a mark represented by a green arrow and a loss in the mark by a red arrow.
Discussion
DNA methylation reprograming has been previously noted for a host of tumor types including HCC (2, 14, 16, 37). In addition, changes to the normal 5hmC patterns have also been reported in both mouse and rat livers following exposure to either nongenotoxic or genotoxic agents (13, 19, 38). In this current study, we build on these early observations to show for the first time in mouse models of liver cancer that early perturbations in the epigenetic landscape over transcriptionally silent CGIs that are marked by low levels of 5hmC and H3K27me3 predicate hypermethylation that occurs in tumors. We utilized several rodent liver tumor models (phenobarbital-exposed Ctnnb1-mutant tumors, DEN-treated Ha-Ras–mutant tumors, and NAFLD-driven tumors) to demonstrate the same observation for each. The reciprocal nature of these changes fits with the notion that these two modifications exist within a programed DNA methylation/demethylation pathway that is mediated by Tet enzyme activity (39). A reduction of Tet1 protein levels in mouse HCC may account for the observed epigenetic disruption; however, this occurs without obvious changes in Tet transcript levels (Figs. 1 and 5). We propose that reduced Tet1 binding and/or activity at target CGIs may be responsible for aberrant epigenetic events in many cancers (Fig. 5D). Direct loss of Tet1 function in liver results in a similar dramatic reduction in promoter 5hmC levels (Fig. 5C). In phenobarbital-exposed mice, reduction of Tet1 and 5hmC levels may also be linked to an elevated rate of proliferation in tumor cells and altered cellular intermediary metabolism (28, 40). Indeed, it was recently shown that both Tet1 expression and global 5hmC levels are reduced in proliferating cells in culture as well as in rapidly proliferating hepatocytes following partial hepatectomy (41, 42). Finally, the function of the Tet enzymes may also be affected by altered metabolic programs in these tumors, such as changes in the levels and utilization of the Tet enzyme cofactor α-ketoglutarate (28).
Emerging data suggest that the deregulation of normal Polycomb group proteins (PcG), critical mediators of the “silencing” chromatin modification H3K27me3, plays causative roles in oncogenesis (31, 43). EZH2 is often highly expressed in HCC and can have important roles in tumor progression and fetal liver development (44). H3K27me3 profiling during mouse phenobarbital exposure implies that an “epigenetic switch” takes place at the promoters of genes, which are linked to changes in their expression levels during hepatocarcinogenesis; gains in H3K27me3 are associated with repression of genes in the tumors, while loss is accompanied with their activation (Fig. 4D and E). In contrast to the promoters that exhibit aberrant hypermethylation/hypohydroxymethylation in the liver tumors, we observed a strong enrichment of H3K27me3 levels in the healthy tissue, which is not significantly altered in the tumors (Fig. 4C, D, and G). Fittingly, these promoters tend to be associated with transcriptionally silent genes, a result that matches with observations in human cancers (17). These findings are also in line with recent reports in which H3K27me3-modified histone tails were found to be associated with 5mC-marked DNA in cultured cancer cells (45, 46). Our data build on previous work that indicated that promoters that are marked with H3K27me3 in embryonic stem cells are more likely to gain DNA methylation during differentiation and carcinogenesis than those lacking H3K27me3 (47, 48). We suggest that Tet1 binding along with PRC2 components mark promoters that are destined to become hypermethylated in cancer; this can be inferred by the presence of 5hmC at these regions in control liver samples. Similar findings have been published in human cancer cell lines where promoters marked by H3K27me3 and H3K4me3 were initially enriched for 5hmC but lose this modification upon depletion of Tet1 by siRNA (9). The finding that 5hmC- and H3K27me3-marked promoters mark sites that later become hypermethylated in the tumors supports a model in which general aberrant methylation at CGIs is not linked with silencing of tumor suppressor genes, but instead indicates decoupling of TET function from PRC2 complexes at these genes. These relationships between DNA and H3K27 methylation clearly warrant more careful molecular examination in vivo, especially as culturing of somatic cells in vitro leads to a reduction in Tet levels and subsequent de novo methylation at H3K27me3-marked promoters (42).
A recent study in human colon cancer reports that promoter proximal regions approximately 1 kb upstream of the TSS marked by 5hmC in the normal colon appear to correlate to sites that are resistant to hypermethylation in colon cancer progression. These results contrast with both our findings and recent observations in mouse mammary tumors and human embryonic carcinoma cells (9, 32), in which polycomb-marked promoters in the normal state may become hypermethylated in the resulting cancer. This indicates that epigenetic disruption during cancer progression is linked to the type of cancer in question and is likely dependent on a combination of TET protein misregulation, tissue background, and genetic mutation. Further combined epigenetic and transcriptomic studies, particularly in cancers arising in different tissues, will be essential to better understand the precise contribution of epigenetic pathways to hepatocarcinogenesis in particular and to carcinogenesis in general.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Disclaimer
All IMI-MARCAR consortium partners had a role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Authors' Contributions
Conception and design: J.P. Thomson, H. Lempiäinen, J.G. Moggs, R.R. Meehan
Development of methodology: J.P. Thomson, R. Ottaviano, E. Unterberger, R. Terranova
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.P. Thomson, E. Unterberger, H. Lempiäinen, R. Terranova, M.J. Lyall, A.J. Drake, C.R. Wolf, J.G. Moggs, M. Schwarz
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.P. Thomson, R. Illingworth
Writing, review, and/or revision of the manuscript: J.P. Thomson, R. Ottaviano, H. Lempiäinen, A.J. Drake, C.R. Wolf, J.G. Moggs, M. Schwarz, R.R. Meehan
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.P. Thomson, R. Ottaviano, A. Muller, A.R. Kerr, M.J. Lyall
Study supervision: J.P. Thomson, R.R. Meehan
Other (performed the experiments): R. Ottaviano
Other (developed tools for data analysis): S. Webb
Acknowledgments
The authors thank Nick Hastie for comments, Florian Halbritter for help with development of the Geneprof RNAseq analysis software and Angie Fawkes, Richard Clark, and Lee Murphy for assistance with Ion Proton sequencing.
Grant Support
This work is partly funded by the Innovative Medicine Initiative Joint Undertaking (IMI JU) under grant agreement number 115001 (MARCAR project; URL: http://www.imi-marcar.eu/). Novartis and the MRC are full participants in the IMI consortium and Novartis provided financial contribution to the scientific program. This work was also partly funded by the MRC. J.P. Thomson was a recipient of IMI-MARCAR–funded career development fellowships and is now funded by a grant from CEFIC. R.R. Meehan is supported by the Medical Research Council. Work in R.R. Meehan's laboratory is supported by IMI-MARCAR, the BBSRC, CEFIC, and the MRC. H. Lempiäinen is the recipient of a NIBR Postdoctoral Fellowship.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.