Final Report Summary - MODHEP (Systems biology of liver cancer: an integrative genomic-epigenomic approach)
Executive Summary:
Executive summary
The founding concept of the MODHEP consortium was that the pathologic behavior of cancer cells is controlled by an abnormally reprogrammed genome, with aberrations at multiple levels of genome organization. These aberrations range from the genetic information and epigenetic structure of the genome to its higher-order folding and organization within the cell nucleus, and ultimately cause the changes in gene expression that drive tumor initiation and progression. A full understanding of the cancerous state will require that the molecular basis for genome reprogramming be described at all these levels, allowing a comprehensive model of genome reprogramming. MODHEP tackled this challenge in Hepatocellular Carcinoma (HCC) with an integrated, multi-layered analysis of genome organization in two mouse models of HCC, as well as in clinical HCC samples. A series of approaches based on advanced high-throughput sequencing technologies were used to produce genome-wide maps of alterations in DNA sequence, epigenetic marks, protein-DNA interactions, long range chromatin folding and gene expression. In parallel, imaging techniques were used to analyze changes in chromatin and nuclear organization at the single-cell level. The combination of these approaches to tractable tumor models allowed an unprecedented view on the dynamic changes in genome organization that are associated with tumor progression, as well as tumor regression. These studies led to the identification of a series of candidate genes, some of which were validated for their involvement in cellular transformation and cancer, paving the way for their therapeutic exploitation.
Summary description of project context and objectives
The concept behind MODHEP was that the pathologic behavior of cancer cells is controlled by an abnormally reprogrammed genome, with pathological changes occurring at the various levels of genome organization: genetic (DNA sequence), epigenetic (DNA-binding proteins and chromatin modifications), higher-order folding and organization of the genome within the nuclear space, and ultimately deregulated gene expression. Thus, one of the key objectives of MODHEP was to produce an integrated analysis of the nuclear changes occurring during tumor progression by generating coordinated, multi-faceted and quantitative series of datasets providing a complete description of the genetic, epigenetic and nuclear organizational aberrations associated with tumorigenesis in two complementary models of liver cancer.
MODHEP's scientific objectives, as initially defined by the scope of each WP, were the following:
• A preliminary WP (WP1) had a logistic objective, namely the set-up and management of the two experimental tumor models (tet-Myc and Mdr2-/- mice) and the coordination of sample preparation and distribution throughout the consortium, for the optimal achievement of all subsequent WPs.
• Defining global changes in the transcriptome (including coding and non-coding RNA species) during multiple stages of liver cancer development using next-generation sequencing (WP2).
• Defining the genomic alterations occurring at subsequent stages of liver cancer development using high-throughput sequencing (WP3).
• Profiling epigenetic features (and the transcriptional regulators responsible for their control) at multiple stages of liver cancer development, using chromatin immunoprecipitation coupled to high-throughput sequencing (WP4).
• Defining the changes in higher-order chromatin organization and nuclear architecture at high resolution using a combination of genome-wide and single gene approaches (WP5).
• Assessing the transferability of the conclusions to human liver cancer and pre-cancerous states using targeted as well as genome-wide approaches in a large number of patients (WP6).
• Providing a systems-level understanding of integrated transcriptional, genomic, epigenomic and nuclear architectural changes during liver cancer progression (WP7).
• Validating the impact of candidate gene products on tumour development in vivo using genetic approaches in the mouse (WP8).
The above scientific objectives were associated with a series of technical/technological objectives, originally defined as follows:
• Development and optimisation of techniques for transcriptional profiling of all RNA species (coding and non-coding) in multi-stage cancer models using next-generation sequencing technologies (RNAseq).
• Development and optimisation of techniques for high-throughput identification of emerging, low-frequency mutations during early stages of cancer development.
• Development of new experimental and computational tools for high-throughput 3-dimensional epigenome analysis.
• Devising a computational framework for reverse engineering of a gene regulatory network model from large scale transcriptional, epigenetics and genetic data.
• Devising a novel three-dimensional computational model to describe the changes in the epigenome during the carcinogenic process.
• Development of novel computational methodologies for modeling cancer gene networks in the context of the aberrant cancer epigenomes.
As described in the various intermediate reports as well as in this final report, the MODHEP work plan was adjusted during the funding period to face experimental contingencies and biological variations in the adopted tumor models, as well as to extend the analysis to tumor regression (and not simply progression), a key addition that was not included in our original proposal. The consortium thus went on to generate the diverse range of proposed datasets, gaining essential insight into the multiple layers of genome organization in liver tumorigenesis.
While part of the experiments, analyses and integration are still ongoing and will extend beyond the funding period, we can assert here that the consortium has significantly fulfilled its initial goals, and in some instances surpassed these with important additional achievements.
Project Context and Objectives:
WP1 Integration
Integrating the activity of individual groups has been one of the crucial aspects of this project, which required that mice, tissues and cells at various pre-tumoral and tumoral stages had to be promptly available to all partners to carry out specific experimental work.
The work package (WP1) was specifically dedicated to the integration of MODHEP data and resources, and was aimed at streamlining the interactions among partners by creating the essential enabling conditions for a productive exchange amongst the Consortium partners of the necessary resources and data generated. The principal aim of this WP was to define robust and reproducible experimental procedures for the delivery to all partners of tissue samples and cells obtained from the experimental mice. To this end, we first set up centralized colonies of HCC mice models (at IEO and IP, see below) and housed them throughout the duration of the project. Next, we provided participating groups with suitable material to perform experimental analyses throughout the duration of the project, setting up a central repository for data, accessible to all partners.
An important note of WP1 regards our choice of mouse models. These models had to (i.) provide insight into disease initiation and progression; (ii.) be clinically relevant; (iii.) allow access to sufficient amounts of homogeneous cell populations for chromatin/nuclear analysis and (iv.) produce datasets suitable for computational and systems biology modelling. No single mouse model can integrate all of these qualities, nor will it be completely comparable to the corresponding human disease. On the one hand, transgenic oncogene-driven tumors provided the advantage of relatively synchronous and homogeneous disease onset and progression. On the other, non-transgenic tumors driven by inflammation and/or spontaneous mutations more faithfully represented the succession of events occurring in human cancer. To take advantage of all features, we combined two different models. First, Tet-Myc mice (as extensively described below), which allowed the controlled activation of the c-myc oncogene in hepatocytes and highly synchronous and penetrant cancer development (Beer et al., 2004), provided access to pre-tumoral and tumoral stages. Second, Mdr2-/- mice, in which a chronic peri-portal inflammatory process leads to displasia, adenoma and adenocarcinoma, mimicked inflammation-driven liver cancer in humans (Mauad et al., 1994; Pikarsky and Ben-Neriah, 2006).
Task 1.1: Definition of standardized procedures for the production and delivery of samples to WPs 2-5 and 8
Start date: Month 1. End date: Month 9.
This task, which has been completed according to the original plan, consisted in the definition of the experimental procedures for the production and delivery of biological samples from the centralized mouse colonies that have been established at IEO and Institut Pasteur to all groups that performed experimental work on them (as described in WPs 2- 5 and 8).
Briefly, in addition to standard procedures as genomic DNA extraction and histological analysis, we have written a protocol (part of deliverable D1.1) for preparation of chromatin immunoprecipitation followed by deep sequencing (ChIP-seq) and for fixed samples that have been subjected to Chromosome conformation capture techniques. This protocol has been shared with other MODHEP partners to allow highly comparable analyses between the laboratories.
Task 1.2: Generation of samples for experimental analyses to be carried out by partners
Start date: Month 1. End date: Month 54.
Two centralized mouse facilities for the project have been established at IEO and Institut Pasteur. Each Institute has established and maintained the two mouse colonies - namely the Tet-Myc/LAPtTA and the Mdr2-/- strains - under identical veterinary and housing standards, and throughout the length of the project. The two facilities also produced and delivered to partner laboratories the experimental samples obtained at different pre-tumoral and tumoral stages. These samples have been used for the downstream analyses described in the other WPs. The dimension of normal, preneoplastic and neoplastic mouse livers was sufficient to divide each organ into multiple pieces to be divided among several labs for different analyses. Therefore, multiple datasets have been generated from each individual liver, allowing us to assess the coexistence of complex alterations within cells coming from the same organ. When the generation of an ultrapure hepatocyte population has been required for downstream analyses, we have obtained it by the portal vein perfusion technique, which allows organ dissociation in situ and short term culture of pure populations of non-proliferating hepatocytes.
Mdr2 mice (Natoli lab, IEO, Milan)
IEO has established Mdr2-/- and Mdr2+/+ mice colonies. We have collected Mdr2+/+ and Mdr2-/- liver samples from 10 animals per time point (8 weeks, 4 months, 6 months, 8 months and 15 months/HCC stage). Different type of samples (for transcriptome, genetic or epigenetic analyses) have been used or sent to partners. Significant results have been described in each WP section.
tet-Myc/LAPtTA (Amati lab, IIT, Milan)
General overview
We used the tet-Myc/LAPtTA mouse colony to produce samples of Myc-driven liver tumorigenesis, with a focus on both tumor progression and early regression. We have set up the tet-Myc/LAPtTA mouse colony in Milan (deliverable D1.2) and studied tumor development depending on the age of tet-Myc induction. tet-Myc induction in utero led to tumor development by the age of 5-6 weeks. During the course of the MODHEP funding period, initial results from transcriptome (WP2) and epigenome (WP5) analyses in the Amati lab (IIT) based on the overexpression of the wild-type Myc gene pointed to an essential role of Myc-dependent gene repression for Myc-induced liver tumorigenesis. Partner IIT therefore decided to include an additional mouse strain for further analyses. In addition to wild-type Myc, the mutant MycV394D (hereafter MycVD) was examined. MycVD does not bind to Myc’s partner protein Miz1 and is therefore compromised in target gene repression.
Task 1.3: Developing a data coordination center for the project.
Start date: Month 7. End date: Month 66.
A small database of distributed samples has been generated to allow tracing the origin of each sample and accessing information regarding the mice.
Ivica Letunic' s team at Biobyte set up and maintained a data coordination center for the use of MODHEP partners and for the dissemination of project results. The data center served as a repository for processed datasets and models, with primary data analysis performed by the respective groups and primary data archiving handled in the global data repositories (EBI, NCBI). A web based user interface to the database has been developed, which allowed partners to simply select relevant datasets based on various queries throughout the WPs.
Significant results, partner IIT (Amati)
To study the interplay between Myc activity, transcription and chromatin organization in liver tumors we used the tet-Myc/LAPtTA mouse model (tet-Myc WP1, Suppl. Fig. 1.1A) as originally described by others (van Riggelen J et al. Genes & Development. 2010 Jun 15;24(12):1281-94), but based on two different c-myc transgenes, tet-MycWT and tet-MycVD, the latter expressing the MycV394D mutant defective in Miz1 binding. Breeding of either of these tet-Myc strains with LAPtTA animals in the absence of doxycycline led to expression of the human c-myc mRNA in liver progenitor cells in utero in double-transgenic embryos (tet-Myc WP1, Suppl. Fig. 1.1B).
We first characterized tumor progression induced by wild-type Myc: at embryonic day E18.5 double-transgenic tet-MycWT/LAPtTA (dtg) embryos appeared phenotypically indistinguishable from control siblings (tet-Myc WP1, Suppl. Fig. 1.1C) retained a normal liver architecture at the histological level (data not shown), but showed slightly increased liver sizes (tet-Myc WP1, Suppl. Fig. 1.1D E). By 6 weeks of age, tet-MycWT/LAPtTA mice showed a gross increase in abdominal size and fully penetrant, gender-independent development of multi-nodal liver tumors that expressed high Myc levels (tet-Myc WP1, Fig. 1.1A-E Suppl. Fig. 1F). Closer histological assessment revealed both epithelial hepatoblastoma and HCC-like tumors. Tumors stained positive for E-cadherin (tet-Myc WP1, Fig. 1.1E) expressed mRNAs encoding markers of fetal progenitor cells, such as AFP, Dlk1, Sall4 (tet-Myc WP1, Suppl. Fig. 1.1H) and contained frequent mitotic and apoptotic cells (tet-Myc WP1, Suppl. Fig. 1.1G). As previously reported, Myc- driven liver tumors showed full oncogene addiction, as feeding tumor-bearing mice with doxycycline-containing food rapidly suppressed Myc expression (tet-Myc WP1, Fig. 1.1F) and induced tumor regression, with significant decreases in tumor mass within a week, and apparently normal livers and healthy conditions within 3-4 weeks (tet-Myc WP1, Fig. 1.1G). In summary, tet-Myc activation in utero caused a prenatal increase in liver size and the fully penetrant development of Myc- dependent hepatoblastomas by the age of 4-6 weeks.
Based on a distinct tet-Myc strain, but the same LAPtTA transgene and experimental scheme used here, others reported slower tumor development following Myc activation in utero. The reasons for these differences may be partly genetic (distinct myc transgenes), partly environmental, but are beyond the scope of this study.
We then addressed whether the V394D mutation affected Myc's oncogenic activity in the liver, as previously reported in lymphomas. Relative to tet-MycWT, tet-MycVD animals showed delayed tumorigenesis (tet-Myc WP1, Fig. 1.1B). Once formed, however, MycWT- and MycVD-induced tumors appeared equally aggressive, moribund animals showing comparable liver weights (tet-Myc WP1, Fig. 1.1H) and tumor load (tet-Myc WP1, Suppl. Fig. 1.3A). Albeit variable among individual tumors, wild-type and mutant Myc were expressed within similar ranges, either as mRNA (tet-Myc WP1, Fig. 1.1I) or protein (tet-Myc WP1, Suppl. Fig. 1.2B). As with MycWT, MycVD-induced tumors showed no significant gender differences (tet-Myc WP1, Suppl. Fig. 1.2C) and expressed the fetal hepatoblasts markers Afp, Dlk1, Sall4 (tet-Myc WP1, Suppl. Fig. 1.3E) and E-cadherin (tet-Myc WP1, Fig. 1.1E). By histological analysis, MycVD showed the same tumor types as MycWT, albeit with increased frequency of the HCC-like tumors compared to hepatoblastoma (tet-Myc WP1, Suppl. Fig. 1.2E). As a marker of apoptosis we monitored PARP cleavage, revealing variability among tumors, but no significant difference between the two myc genotypes (tet-Myc WP1, Suppl. Fig. 1.2B).
To address Myc function at a pre-tumoral stage, we purified LAPtTA/tet-MycWT and VD fetal hepatoblasts by magnetic cell sorting with an E-cadherin antibody. RT-PCR analysis revealed similar myc mRNA levels in both populations (tet-Myc WP1, Suppl. Fig. 1.1B) comparable activation of a Myc-induced gene (Smpdl3b) but defective repression of Cdkn1a by MycVD, as expected (tet-Myc WP1, Suppl. Fig. 1.2F). MycWT and MycVD led to similar increases in the number of fetal hepatoblasts recovered per embryo (tet-Myc WP1, Suppl. Fig. 1.2G) suggesting equivalent proliferative activities.
WP2 - Transcriptome Analysis & WP6 Screening and validation in human samples
In the frame of the MODHEP consortium, we joined our efforts to produce and analyse a large amount of genetic and genomic data for human clinical HCC samples and mouse HCC models using unique high-throughput methodologies and instruments such as the CAGE technology and the retro-transposon capture sequencing, the latter having been developed and improved for many years by the MODHEP partners.
• Large scale transcriptome sequencing for human adult and mouse liver cancers
We sequenced CAGE libraries for 50 primary hepatocellular carcinoma (HCC) tumours and 50 matched non-tumours as well as 21 primary hepatoblastoma (HB) tumours and 21 matched non-tumours. In addition, we sequenced CAGE libraries for 37 liver samples including wild type, inflammation, adenoma, and HCC extracted from the Mdr-/- mouse model as well as 38 liver samples of wild type and HCC extracted from the Tet-Myc mouse model. The raw CAGE data have been submitted to the NCBI database of Genotypes and Phenotypes (dbGaP; http://www.ncbi.nlm.nih.gov/gap/) under accession number phs000885.v1.p1 for human HCC and the Gene Expression Omnibus (GEO; http://www.ncbi.nlm.nih.gov/geo/) under accession number GSE60982 for mouse HCC.
• Aberrant activation of LTR retroviral promoters in human adult HCC
We identified 64,366 distinct TSSs as the transcriptome of human HCC tumour tissues from the CAGE data. To distinguish potential non-coding RNAs from protein-coding genes, we classified the identified 64,366 TSSs into 3 groups using the Gencode12 annotation: (1) coding TSS, which overlaps with a protein coding TSS (+/- 100bp) in sense; (2) proximal TSS, which is located within 5kb of a coding TSS or overlaps with a coding exon; (3) distal TSS, which is neither coding nor proximal TSS. We compared expression levels between 50 tumours as a case group and 50 matched non-tumours as a control group using edgeR to explore mis-activated transcripts in tumours. We identified 11167 significantly up-regulated TSSs in tumours with the threshold FDR<0.05 of which 4942, 4779, and 4756 are coding, proximal, and distal TSSs, respectively. The most significantly up-regulated protein-coding gene is GPC3, a known candidate biomarker and therapeutic target for HCC. For distal TSSs, 34% of the top 100 most significantly up-regulated elements overlap with repetitive elements in their sense direction. To reveal the association between up-regulated TSSs and repetitive elements, we examined what fraction of up-regulated TSSs overlap with major repetitive elements (LINE, LTR, and SINE), which are annotated by RepeatMasker and RepBase. Interestingly, about 20% of up-regulated distal TSSs overlap with LTR elements in the sense direction (Fig 2.1). Furthermore, the fraction of LTRs increases from 20% to 30%, if limited to the most significantly up-regulated TSSs with FDR below 1.0E-08. These percentages are outstandingly higher than the percentages for (1) LTR elements in the anti-sense direction, (2) non-up-regulated distal TSSs, (3) LINE and SINE elements in sense and anti-sense (1~6%) and (4) randomized distal TSSs. This strongly indicates that the promoters derived from LTR elements are largely activated in HCC tumours. We further examined the expression levels of the LTR promoters in normal livers using 3 datasets: (1) normal liver tissues prepared in this study (n=5), (2) adult and fetus liver tissues produced by Fantom5 (n=2), (3) primary hepatocytes by Fantom5 (n=3). As expected, expression levels of LTRs in normal livers are much lower than those of non-tumours as well as tumours (Fig. 2). We thus conclude that the up-regulation of LTR promoters is a hallmark feature of HCCs.
• Molecular signature of HCC comprising LTR-derived transcripts
We selected human ncRNA candidates appropriate for determining a molecular signature of HCC using the following stringent criteria: (i) FDR threshold below 1.0E-10; (ii) fold change above 8.0; (iii) expression in at least 30 tumour samples. About one third (43/133) that passed the criteria coincided with LTRs (Fig. 2.3). According to HepG2 data produced in the ENCODE project, these LTR-derived transcripts seem to be nuclear ncRNAs (Fig. 4). The fact that the signature-LTRs are scarcely transcribed in normal liver tissues raises the question of the nature of the tissues in which they are intrinsically programmed for expression. According to FANTOM5, an expression atlas of various types of human primary cells and tissues (http://fantom.gsc.riken.jp/5/datafiles/latest/basic/) their expression is limited to 1 to 2% of the normal tissues and primary cells (Fig. 2.5). Intriguingly, reproductive tissues and primary cells occupy the top seven positions. Testis, which expresses about half of the 43 signature-LTRs, is in the first position followed by placenta, chorionic, and amniotic membranes (Fig. 2.6). Some of the highest-ranking signature-LTRs, such as LTR-003, LTR-004, and LTR-006, are exclusively expressed in reproduction related tissues (Fig. 2.7). In contrast, the top 6 signature-LTRs are widely detected in all hepatic and non-hepatic cancer cell lines, emphasizing the close link between these LTRs and carcinogenesis. We then submitted 15 of the 43 signature-LTRs to 3’ RACE validation. Deep sequencing of 3’ RACE products using MiSeq successfully determined the transcript structures of 14 of them, among which 13 had at least one splice site. The 25 transcripts detected, including splice variants, were all predicted to be ncRNAs with low coding potential by the Coding-Potential Assessment Tool (CPAT). The median exon size is of 154 bp, which is comparable to the average coding gene size (122bp), whereas the median intron size is of about 37 kb, i.e. much longer than the general protein-coding gene size (1 kb). Figure 2.8 shows an example of novel transcripts, LTR-002, determined by the 3’ RACE. The TSS is located at the middle of an LTR element, and the second and third exons are located far downstream of the first exon without encoding any viral proteins, suggesting that LTR-002 has become fixed as a promoter provider to the ncRNA. We validated the CAGE data by measuring the expression levels of selected four LTRs using quantitative real-time PCR in two different datasets. The expression of these LTRs was more than 10-fold higher in ~50-80% of the tumour samples compared to the matched NT in both datasets, which confirm that these LTRs are recurrently activated in HCC with different activation levels (Fig. 2.9). Finally, using clinical information, we show that the HCCs with high-level LTR is globally more severe than those with low-level LTR (Fig. 2.10). In particular, The LTR-high HCCs are significantly correlated with a high risk of recurrence, suggesting that the LTR expression level might be useful as a new characteristic in the definition of HCC molecular subclasses. This study was published in Genome Research [PMID:26510915].
• Somatic genomic alterations in paediatric HCC caused by bile salt export pump deficiency (BSEP-HCCs)
We extensively searched for potential cancer driver gene mutations in HCC tumors developed in humans and in mice using exome sequencing. We compared the somatic genomic alterations profiling in HCCs in the two species. In both HCC was induced by defects in hepatocyte biliary transporters, which expose hepatocytes to bile salts and inflammation. We mapped the cancer genomes of Mdr2 knock-out (KO) mice, which, similarly to human HCC associated with bile salt export pump (BSEP-HCC) deficiency [also known as progressive familial intrahepatic cholestasis (PFIC) type 2], develop HCC through the impairment of bile secretion. In BSEP deficiency, inherited mutations in the ABCB11 gene cause impairment of bile salt export from hepatocytes into bile, chronic inflammation and eventually liver cancer. Thus, a study of this type of liver cancer is an opportunity to map the acquired genomic modifications that trigger liver cancer in the absence of exogenous direct (viruses) or indirect (alcohol) mutagens.
We sequenced the whole exomes of six human BSEP-HCCs and nine murine HCCs as well as their corresponding non-tumor surrounding liver tissues. We also used genome-wide SNP arrays to investigate the occurrence of copy-number variations (CNVs) in seven BSEP-HCCs, including all tumors screened for point mutations and one additional lesion. In both human and mouse cancer genomes, we find a few somatic point mutations with no impairment of cancer genes (Fig. 11), and a pervasive occurrence of chromosomal rearrangements that led to massive gene amplification (Fig.12). In both cases, cancer genomes accumulated massive copy-number gain in contrast to very few somatic SNVs or small indels. Copy-number gains preferentially occur at late stages of cancer development and frequently target the MAPK signalling pathway, and in particular direct regulators of JNK. The pharmacological inhibition of JNK delays cancer progression in the mouse. The low levels of mutational instability in human and murine HCCs we found suggest that the acquisition of mutational instability is not the driving force for the development of this type of liver cancer. Such a genomic signature is remarkably different from that of the other HCCs previously sequenced, which acquire mutational instability and tend to accumulate gene deletions rather than amplifications. These findings confirm the genetic heterogeneity of liver cancers caused by different etiological agents and, at the same time, the remarkable analogy between human and mouse tumors with similar aetiopathogenesis. This study was published in Nature Communications [PMID:24819516].
• Somatic and germinal LINE1 retrotransposition in HCC
The same set of human HCCs, whose transcriptomes were analysed by CAGE and RNA Seq, was also explored for somatic rearrangements by RC-Seq. This enhanced retro-transposon capture sequencing (RC-seq) method was applied to the genomes of human HCCs in order to elucidate whether, and how, LINE1 retrotransposons contribute to insertional mutagenesis during liver carcinogenesis.
We studied T/NT HCC samples i.e. tissues originating from tumor (T) areas and non-tumor (NT) areas distant from the tumors from 19 patients with HCC. The underlying liver disease was Hepatitis B or C virus–induced cirrhosis. We demonstrated on this cohort the importance of endogenous L1-mediated retrotransposition in the germ line and somatic cells of patients with HCC. We constructed RC-seq libraries for the 19 T/NT HCC samples and then sequenced in multiplex on an Illumina HiSeq platform. We found two archetypal mechanisms revealing MCC (Mutated in Colorectal Cancers) and ST18 (Suppression of Tumorigenicity 18) as candidate genes for HCC. The first of these genes, MCC, which is expressed in the liver, is known to regulate the oncogenic β-catenin/Wnt signaling pathway frequently activated in HCC. 4 out of 19 (21.1%) donors presented germ line retrotransposition insertions in the tumor suppressor MCC pointing to an inhibition of the liver tumor suppressor function of MCC. The expression of MCC was ablated in each case, enabling oncogenic β-catenin/Wnt signaling in tumors and adjacent liver. The second gene, ST18, was activated by an intronic, tumor-specific L1 insertion. Experimental assays confirmed that the L1 insertion interrupted a negative feedback loop by blocking ST18 repression of its enhancer. ST18 was also demonstrated to be amplified and up-regulated in human (liver cancer cell lines) and mouse (Mdr2 KO) HCC model systems, consistent with ST18 functioning as a candidate liver oncogene. Our study substantiate LINE1-mediated retrotransposition as an important etiological factor in HCC. It showed that (i) Germ line L1 and Alu insertions in MCC activated β-catenin/Wnt signaling; (ii) L1 was broadly activated and transcribed in HCC coincident with L1 promoter hypomethylation ; (iii) L1 mobilization in tumor cells accelerated transformation of the HCC genome; (iii) A tumor-specific L1 insertion interrupted a negative feedback loop regulating ST18. This study was published in Cell [PMID:23540693]
WP3 Genome-wide Detection and Analysis of Somatic Mutations
We have conducted an extensive exome profiling of tumors deriving from Mdr2-/- mice (Mdr2-/- HCCs) at different stages of disease development, as well as in tumors from tet-Myc mice, to identify somatic mutations and copy number variations that are progressively acquired in the cancer genome. We have observed very few somatic point mutations in either model even at late cancer stages. In Mdr2-/- mice, we observed the progressive accumulation of gene amplification events affecting the MAPK signalling pathways, and in particular activators of the cJun-N terminal kinases (JNK).
Such a cancer mutational landscape characterised by very few somatic mutations not affecting any known cancer gene differs substantially from that of most human HCCs. To understand whether this is due to inter-species differences or, rather, to the tumor aetiology, we have sequenced the exomes of human HCCs associated with bile salt export pump (BSEP) deficiency, also known as progressive familial intrahepatic cholestasis type 2, which are etiologically similar to liver cancers associated with Mdr2 gene loss. BSEP deficiencies are a heterogeneous group of rare autosomal recessive disorders caused by inherited inactivating mutations in the hepatocyte membrane transporter genes. The disease usually appears in infancy or early childhood and manifests with hepatocellular damage and cholestasis due to defects in bile formation. In BSEP deficiency, inherited mutations in the ABCB11 gene cause impairment of bile salt export from hepatocytes into bile, leading to liver chronic inflammation and to the early onset of hepatocellular carcinoma. HCCs arisen in chronic liver disease in Mdr2-KO mice have an aetiopathogenesis similar to that of BSEP-HCCs, while it is clearly distinct from that of viral and metabolic disease-associated human HCCs. Consistent with this notion, we have observed BSEP-HCC genomes acquire massive gene amplification that affect components of signal transduction pathways, such as the ErbB, the PI3K/Akt and the mitogen-activated protein kinase (MAPK) signalling pathways. Interestingly, we have shown that the pharmacological inhibition of JNK in Mdr2-/- mice attenuates cancer progression and might offer a useful therapeutic intervention in BSEP deficiency patients waiting for liver transplantation. The study in Mdr2-/- mice has been published in Nature Communications (Iannelli et al. 2014; PMID:24819516).
WP4 Chromatin dynamics of transcriptional and epigenetic regulators during hepatocarcinogenesis
Genetic and expression profiling studies have identified several molecular pathways that are deregulated and/or mutated in HCC, and gene expression changes that drive tumor formation. This notwithstanding, the past decade of research has highlighted modifications of the epigenetic landscape as a hallmark of cancers. Indeed epigenetic mechanisms regulate normal development and homeostasis and are crucial for tissue-specific gene expression maintenance. The objectives of WP4 were to determine the chromatin dynamics of transcriptional and epigenetic regulators during hepatocarcinogenesis.
A first approach consisted in the generation of ChIP-seq databases of chromatin-associated features during liver cancer progression. Multiple histone modifications were chosen in order to obtain a comprehensive cartography of the epigenetic landscape relative to gene transcriptional activity (H3K4me1, H3K4me3, H3K27ac) in normal livers and to highlight modifications accompanying the alterations of transcriptional activity occurring during hepatocarcinogenesis, as identified in WP2. Along with this, chromatin binding profile of RNA polymerase II (RNAPII), and histones marks signing actively transcribed gene body, H3K36me3 and H3K79me2, were investigated as well.
A second approach was to analyse the binding of liver and cancer specific transcription factors. In the Tet-Myc model, first of all the founding role of Myc in Myc-induced hepatocarcinogenesis has been assayed on a genome-wide scale by systematic ChIP-seq. Tumors driven by the activation of the transcription factor Myc generally show oncogene addiction. To complete the second approach on the founding role of Myc in the tumor development and maintenance, further ChIP sequencing experiments, in tumors after short-term tet-Myc inactivation, have been performed for histone marks and regulatory factor, including Myc and RNA polymerase II. The work in tet-Myc mice was published in Cancer research (Kress et al. 1016 PMID: 27197165).
Computational analysis reported a strong increase in both binding intensity and the total number of binding sites at all accessible genomic regulatory elements in Myc-driven tumors, supporting the recently described phenomenon of “invasion” of Myc binding landscape upon its overexpression. Focusing on annotated promoter regions revealed a tight correlation between the presence of RNAPII, H3K4me3 and H3K27ac (in either C, T or Toff samples) and Myc binding in T, while promoters lacking those features remained unbound. Most importantly, active promoter marks pre-existed in C and were retained following Myc elimination in Toff. Thus, consistent with other studies, over-expressed Myc in T widely associated with already active promoters while inactive promoters remained unbound. In tumors, over-expressed Myc associated with active promoters in a widespread manner, but increases in binding neither discriminated Myc-dependent from -independent transcriptional responses, nor correlated with the extent of changes in mRNA levels. Furthermore, neither Myc- binding nor recruitment to enhancers (determined by nearest-neighbor analyses) in C or T discriminated primary from secondary Myc target genes. Taken together, no simple rule could be drawn that would link promoter/enhancer association by Myc with either primary or secondary gene regulation.
Myc has been suggested to regulate global RNA Pol II activity at the elongation step. In our system, up- or down-regulation of gene expression was paralleled by equivalent changes in RNA Pol II binding at promoters. During the C to T transition, either primary or secondary genes showed similar variations in Pol II binding, but only primary Myc-dependent genes were affected during the T to Toff transition. Thus indicating that Myc influences RNA Pol II loading on either activated or repressed genes.
Transcription can be regulated at all steps of the RNA Pol II life cycle, including loading, pausing and elongation. The relative contribution of these steps in our experimental system were calculated by RNA polII occupancy and stalling index (RNA Pol II occupancy on promoters relative to gene bodies). Consistent changes in Pol II abundance were identified not only on promoters - as mentioned above – but also on gene bodies. However, stalling indices were not consistently affected in any of the regulatory categories. Altogether, activated and repressed genes showed generally consistent changes in Pol II loading and elongation, whether regulated by Myc in a primary or secondary manner. Further studies will be requires to elucidate the contribution of each regulatory step to Myc activity.
To follow up the epigenetic changes in chromatin patterns along the tumorigenic process in Mdr2 -/- mice, we generated a panel of ChIP-seq data sets on liver samples prepared at different time points. In particular, we considered multiple biological replicates of adenoma, low-grade HCC (≤60% carcinoma content) and high-grade HCC (≥80% carcinoma content), all coming from 11-to-16 months old individuals. These samples were compared to age-matched WT livers, and to samples of 8 months old inflamed but non tumor-bearing livers.
Computational deconvolution of dynamic histone modifications was used in order to identify transcription factors (TFs) activated during the tumorigenic process, and therefore possibly involved in bringing about transcriptional changes associated with, or favoring, cancer. Deposition of H3K27Ac is due to TF binding to underlying regulatory elements and promoting local recruitment of histone acetyltransferases. Therefore, changes in H3K27Ac reflect changes in the activity or expression of specific TFs and may indicate TFs potentially relevant for the disease progression. Furthermore, to improve our classification of genomic regulatory regions, H3K4me3 (a mark for active TSS/promoters) and H3K4me1 (enriched at enhancers when not associated with H3K4me3) were profiled in WT, inflamed livers and HCC samples.
Principal component analysis (PCA) applied over ChIP-seq data for H3K4me1 and H3K4me3, highlighted that samples belonging to different disease steps were clustering separately, although in the case of H3K4me3 the most discriminative variable was PC2. In the case of H3K27Ac, for which data for two additional categories (“adenoma” and “high grade HCC”) were available, separation was still generally neat. Based on PCA, “outliers” samples failing to cluster close to the bulk of each category were excluded of the next analyses. We found that the highest number of induced or repressed peaks (with respect to WT) displayed by all samples at a given time point was found for H3K27Ac, and particularly for induced peaks, indicating that this mark is subjected to the highest variability specifically associated to tumorigenesis in Mdr2-KO mice. On the contrary, the number of common induced or repressed H3K4me1 and H3K4me3 peaks was very low both for inflamed and low-grade HCC samples. Gene ontology (GO) analysis for genes located in the vicinity of H3K27Ac peaks induced was performed at all stages. Results revealed a strong inter-class functional enrichment for biological processes associated with inflammation and immune response, suggesting that the inflammatory response started in the early disease stages and continued until tumors developed. We observed that, at all time points, the most enriched matrices corresponded to binding sites for the activator protein 1 (AP-1) TF family members (in particular Jun, JunB, JunD, JDP2, Fos, FosB, FosL1, FosL2 a BATF). AP-1 TFs are activated in response to a variety of stimuli, including cell stress and inflammation, exposure to cytokines or growth factors and infection, and control cell proliferation and apoptosis. Therefore, the finding that common H3K27Ac induced peaks were enriched for AP1 matrices is in agreement with the general activation of inflammatory pathways resulting from the GO analysis. The second class of transcriptions factors whose matrices were enriched in H3K27Ac common induced peaks was the erythroblast transformation-specific (ETS) TF family, and the Spi-like subfamily including Spi1 (PU.1).
Taken together, this evidence points to the establishment of an inflammatory/stress response program in hepatocytes starting from the pre-malignant, chronic inflammatory phase. This inflammatory signature is retained up to the more advanced HCC stage, and is accompanied by the activation of members of the AP-1 and ETS transcription factor family. PU.1 is a macrophage-specific TF, therefore we decided to better characterize the presence of macrophages and other inflammatory components in our samples and we are investigating how macrophages impact on the generated data.
Another aim of this WP was the identification of epigenomic changes during liver carcinogenesis in relation to silencing and cellular memory, through genome-wide profiling of one of the main factors involved in this process, the so-called polycomb group proteins (PcG). PcGs are epigenetic regulators fundamental for the maintenance of cell-specific gene expression pattern, cell identity, pluripotency and differentiation. Polycomb Repressive Complex 1 and 2 (PRC1, PRC2), are thought to act sequentially to stably maintain gene repression, whereby PRC1 is able to read the repressive H3K27me3 mark deposited by PRC2 and in turn catalyzes H2A mono-ubiquitination (H2Aub1), leading to a transcriptionally silent chromatin conformation.
ChIP-sequencing experiments were performed for PRC2 proteins, H3K27me3 and PRC1 proteins. Attempts to obtain reliable ChIP form the Mdr2-/- model failed, partly due to technical issues. Instead the contribution of Polycomb proteins to the Myc-induced tumor phenotype could be investigated.
Computational analysis of the ChIP-seq data demonstrated the existence multiple combinations of PCR2 and PRC1 complexes, and their differential association with repressive but also with active histone marks. We notably observed a strong correlation between binding of Suz12, Ezh2 and H3K27me3, as expected. Notwithstanding, our data also confirm the recent findings (Stojic et al, 2011, Mousavi et al, 2012, Xu et al, 2015) that PRC2 binding also clusters with active histone marks, a trait that we observed more particularly at promoters in adult liver cells. On the contrary we found that PRC2 characteristics are altered in the Myc-induced tumors. Indeed, we observed that Suz12 co-localization with active histone marks is strongly reduced in tumors relative to normal livers. This reduction is found mainly at promoters and to a lesser extent at potential enhancers. Recent studies have reported a physical interaction between Ezh2 and Myc family in cancers, and notably with c-Myc in HCC (Corvetta et al, 2013, Wang et al, 2014). Our results indicate that, in normal liver, Suz12 and Myc, naturally associate at a subset of PRC2 peaks and that this interaction is specifically observed at TSSs of transcriptionally active genes. Looking at the dynamics of Myc and PRC2 interplay in the Myc induced tumors, we observed that Myc-PRC2 association significantly increases in the tumor (from ˜15% to ˜42% of Suz12 peaks), along with an increased and preferential association (60%) with TSSs covered by non-active chromatin. Finally sites of Myc-PRC2 co-occupancy were found to be different in normal versus tumor livers. Another remarkable finding was a strong alteration of H3K27me3 genomic landscape between control and tumors, which appears to be regional. Further analyses of other PcG components and histone regulators are currently ongoing to obtain a more complete picture on Polycomb dynamics in liver tumorigenesis.
The MODHEP consortium also investigated the epigenomic status of a classic model of facultative heterochromatin, the inactive X chromosome, in the context of liver cancer. The X chromosome carries over 1000 genes, many of which are critical. The stable, epigenetic silencing of one of the two X chromosomes in females is essential for correct gene dosage. Prior studies have revealed that the inactive X shows aberrant organisation and expression in some tumors such as breast cancer (Chaligné et al, 2015) but no studies were carried out in the context of liver cancer. The challenge of assessing the inactive X chromosome lies in the ability to distinguish the active X in a population of cells where random X inactivation will lead to a mixture of cells with either of the two Xs inactivated. We derived mice carrying polymorphic X chromosomes by crossing the mdr2-/- or Tet-Myc mice (Fvb strain) to M. Castaneus mice, which provides F1 with approximately 1 SNP/100bp. Tumors are usually clonal and therefore have a defined Xi. For normal liver control samples, which are likely to be cell mosaics for parental Xi status, we generated mice carrying a knock-out of Xist on one X chromosome (B6 origin), thus ensuring that only one X (Cast origin) would be the inactive one and greatly facilitating allele-specific analyses in wild-type control livers. These samples have been processed for allele-specific Hi-C analysis, in parallel with the tumour samples isolated from both tumour models crossed with Cast mice. Adapting the ChIP-sequencing procedure for allele-specific analysis, the specific chromatin status of the inactive X chromosome in WT and tumour samples, is being profiled allele-specifically for histones marks and regulatory factors, as well as the chromosomal architectural protein CTCF.
These results will be instrumental for the 3D nuclear organisation WP5, and eventually for the modelling approaches in WP7.
WP5 Higher-order chromatin and nuclear architecture
Gene expression changes drive tumour formation. Gene expression, in turn, is influenced by the folding of the genome inside the cell nucleus. In work package 5 of the Modhep project we hypothesized that tumour formation is accompanied and possibly influenced by changes in the folding of the genome. In order to study this, several complementary approaches were pursued.
One approach involved the use of Hi-C, which at the start of the project was a relatively novel technique that can analyse contact frequencies between any pair of genomic segments across the entire genome in an unbiased manner. Hi-C therefore allows generating high-resolution 3D genome contact maps. Hi-C studies were focused on the Myc mouse model that develops liver tumours upon inducible overexpression of the Myc oncogene, one of the most pervasive cancer genes. The consortium had to optimize the Hi-C strategy and make it applicable to cells from the mouse liver before and after they developed liver tumours. We succeeded in optimizing the Hi-C protocol for mouse liver cells and generated high resolution 3D genome contact maps for healthy liver, Myc-induced liver tumors and Myc-depleted regressed liver tumors. Our studies are among the first to generate such high-resolution contact maps for primary tissue directly taken from an organism.
A second approach assumes that the functionally most important changes in 3D structure most likely take place at the genes that alter their expression during tumour formation. A dedicated capture-C method involving hybridization capture probes that allow targeted analysis of contacts specifically made by all gene promoters was successfully developed and applied to healthy liver and liver tumours.
Computational analysis of the data shows that tumour formation is accompanied by large scale changes in the shape of the genome. For example, we find that chromosomal regions change nuclear locations upon tumour formation. Many regions move between nuclear compartments that are known as the active (A) and the inactive (B) compartment. Interestingly, chromosomal segments that move to the active compartment often carry genes that increase their expression in the tumour whereas chromosomal regions that move to the inactive B compartment often contain genes that lower their expression in the tumour. Compartment switches surprisingly appeared irreversible, or at least were stable for 16 hours after depletion of Myc oncogene expression, a period during which already pronounced reversal of gene expression was observed. Although preliminary, this suggests that Myc-driven gene expression changes are not dependent and may precede changes in the overall topology of chromosomes.
Chromosomes are structurally subdivided into topologically associated domains, or TADs: these are the structural but also the functional units of chromosomes. At the level of TADs, chromosome structure was largely unaltered in the tumours, but TADs were found engaged in more inter-TAD contacts in the tumour. This gain in long-range contacts was lost again after tumour regression. Within TADs, chromatin loops between genes and regulatory sequences like enhancers are formed. Roughly 15% of these chromatin loops changed between the control (wild-type) liver and the tumours, with essentially similar numbers of loops that dissolved and appeared during oncogenesis.
A further intersection of the contact maps with the epigenetics and transcriptome datasets generated in other work packages is ongoing and expected to soon yield a unique, integrated picture of the dynamics between epigenetics, genome topology and gene expression that underlies tumorigenesis in a mouse model expressing an oncogene, Myc, leading to novel fundamental insight in the function of this pervasive oncogene, and fulfilling one of the essential goals of our consortium.
Another aim of the consortium was to develop, optimize and standardize techniques to explore the 3D organization of the active (Xa) and inactive (Xi) X chromosome in healthy livers and tumours isolated from female mice. For this, first allele-specific Hi-C analysis on F1 hybrid (129 x cast) ES cells and in clonal neural precursor stem cells (NPCs) was developed and applied. The studies revealed that the inactive X chromosome (Xi) adopts an unusual three-dimensional structure with a global absence of TADs, except at chromosomal regions that escape X inactivation. Xi was found to be partitioned into two “megadomains” that span several tens of megabases each, within which the chromatin fiber seems to be randomly folded. The boundary within the two megadomains lies at the DXZ4 macrosatellite locus, which also overlaps with a structural boundary on the human Xi. When this region is deleted, the Xi no longer folds into this bipartite structure. Oligo FISH probes spanning several megabases across the X chromosome enabled 3D FISH analysis to validate the HiC findings. This study thus provided the critical tools for the analysis of physical measurements of 3D chromatin structure at a chromosome wide scale as a means to investigate the Xi in liver tumours.
Two models of liver tumorigenesis were used in this project, the mdr2-/- and the Tet-Myc/Lap-tTA mouse models. The challenge of assessing the inactive X chromosome is that it must be distinguished from the active X in the same cell, and that in the case of populations of cells, random X inactivation will lead to a mixture of cells with either of the two Xs inactive. To avoid the issue of non-clonality we generated mice carrying a knock-out of Xist on one X chromosome (B6 origin), thus ensuring that only one X (Cast origin) would be the inactive one and greatly facilitating allele-specific analyses in wild-type livers. Tumors are often clonal and therefore have a defined Xi. These samples were processed for allele-specific Hi-C analysis, in parallel with the tumour samples isolated from both tumour model crossed with Cast mice. Tumour samples were found to have frequent partial or complete erosion of the megadomain partitioning, which does not appear to correspond to deletion of the DXZ4 boundary. The disruption could thus be epigenetic, and the chromatin status of this region in WT and tumour samples is under investigation.
Finally, constitutive heterochromatin organization inside the cell nucleus is appreciable under the microscope by DAPI-staining, with these chromosomal regions carrying inactive chromatin forming densely stained foci in the nucleus. Nuclear reorganization of constitutive heterochromatin is a hallmark of many tumours. An objective of the consortium was to gain more insight into this reorganization. To investigate this we focused on the Myc-driven mouse tumour model. In addition to DAPI staining and immunofluorescence assays we set-up fluorescence in situ hybridization (FISH) to visualize the genomic loci corresponding to the constitutive heterochromatin pericentric domains and the neighboring centric domains. We found that organization of pericentric domains displays major changes with a loss of clustering between pericentric domains and a loss of nuclear periphery localization. To independently analyse the spatial organization of pericentromeric heterochromatin we developed a 4C variant that exclusively analyses contacts made by the satellite repeat sequences that flank the centromeres on each chromosome (Sat-4C). Using Sat-4C we confirmed a genomic reorganization of genomic domains associated with pericentromeric heterochromatin. Taken together these results indicate that in myc tumours a major reorganization and nuclear localization of constitutive heterochromatin occurs and accompanies redistribution or reshuffling of pericentric-associated genomic loci. Applying the same approaches to the Myc-off regressed tumors showed that the organization of constitutive heterochromatin is plastic and can dynamically return to a ‘normal’ rather than ‘tumoral’ organization.
WP7 Computational and Systems-level analyses
With the advent of multiple techniques for comprehensive characterization of genetics, gene expression and epigenetic programs in tumors, a major challenge become the integration of multiple data sources toward models of tumorigenesis. In work package 7 of the Modhep project we developed techniques for computational analysis of multi-layered data and applied them to the analysis of the two mouse HCC models, as well as of human HCC samples.
We developed several techniques for analysing chromosome conformation data as assessed by 3C, 4C and Hi-C experiments. Such data provide opportunities for inferring physical contact structures that ties up together complex regulatory regions involving multiple genes and control elements. Modhep research provided us with new techniques for analysing Hi-C data in order to infer the contact structures in such regions. We were specifically interested in the contacts linking gene promoter with gene regulatory elements (or enhancers). Such contacts can be assessed effectively by targeted Hi-C analysis, but in order to understand their potential impact, we embed these targeted experiments within global analysis of HiC matrices, models for topological domains, and integration of ChIP experiments. Modhep research thus provided one of the first comprehensive attempts to understand gene de-regulation during carcinogenesis by combining most of the available strategies for interrogating gene regulation and epigenomic landscapes.
When integrating transcriptional maps of MYC induced liver carcinogenesis with such comprehensive epigenomic modelling, we observed how this key oncogene can drive gene regulation at two levels. First, we observed classical targets of MYC being induced through increase in MYC occupancy and associated promoter histone remodelling, but not de-novo recruitment of enhancers or enhancer-promoter contacts. Such genes were quickly repressed following reduction in MYC levels. On the other hand, a second class of genes recruited MYC and de-novo enhancer contacts, as part of a possible more profound epigenetic reprogramming process. These genes exhibited a more stable transcriptional aberration, which persisted, at least partly, even following reduction in MYC level. Additional analysis of these two classes of genes is on going, but this model provide possible explanation to the effect of targeted therapies on oncogenes, which may lead to direct correction of the gene expression program of direct oncogenes targets, but not to proper regulation of genes that may carry epigenomic memory that is not directly dependent on stable oncogene expression.
Another important research direction in WP7 targeted the role repetitive elements - or 'mobile DNA' - play in hepatocellular carcinoma (HCC). Targeted DNA sequencing of tumour samples obtained from human HCC patients, and from a mouse model of HCC was performed as part of an attempt to link the activity of mobile DNA and the carcinogenic process. Crucially, this analysis revealed new copies of a mobile DNA element called LINE-1 in the tumours that were not found in normal blood samples taken from the same individuals. These new LINE-1 mutations were found to occur in genes involved in preventing tumour formation and growth, including a novel cancer gene called ST18, were a LINE-1 mutation turned the gene on and contributed to tumour growth.
Further integrative analysis aim at linking transcription factor (TF) activation/deactivation patters with inferred post-translation modifications (PTMs). Capturing this kind of regulatory interactions using only transcriptional data, such as gene expression profiles (GEPs), is considered challenging since GEPs are further downstream of the PTM event and only indirectly linked to it. We developed new approaches based on Multi-Information that directly quantifies the co-regulation among N genes with a single positive value (0 indicating no co-regulation) by assessing the extent of statistical dependency among them all. Application of this approach to infer a hierarchical network of interactions between key transcriptional regulators and signalling molecules altered in hepatocellular carcinoma (HCC) identified 29 TFs with an abnormal activity in HCC. The methodology then identified for each one of the 29 TFs which were the kinases that are more likely to be modulating them and thus active in HCC. The inferred hierarchical network of interactions between the identified aberrant transcriptional regulators and signalling pathways (kinases) was shown to be altered in hepatocellular carcinoma. This is providing experimentalists with new hypotheses for further exploration of the regulatory dynamics leading to HCC.
WP8: Preclinical validation in the mouse tumour models
Multidrug resistance P-glycoproteins, Mdr2 in mouse and MDR3 in human, are the choline-containing phospholipid phosphatidylcholine (PC) transporters that belong to the subfamily B4 of ATP-binding cassette transporters (ABC). In the liver, Mdr2/MDR3 are involved in the translocation of PC from the inner leaflet to the outer leaflet of the canalicular membrane of hepatocytes for direct extraction by bile acids (BA). BA and PC are essential components in the bile that form bile acid micelles to reduce the toxic detergent activity of BA for hepatocytes and cholangiocytes and to establish physiological bile flow. Mutations in the ABCB4 gene in human result in deficiency of PC in the bile, causing diseases including progressive familial intrahepatic cholestasis type 3 (PFIC3), intrahepatic cholestasis of pregnancy and low-phospholipid-associated cholelithiasis (LPAC). A constitutive Mdr2 knockout mouse has been generated, which successively develops liver inflammation, hepatic fibrosis and hepatocellular carcinoma (HCC) at late ages. The pathogenesis in Mdr2-/- mice is commonly considered to be caused by toxic BA in the biliary canaliculus, which damages hepatocytes and cholangiocytes and provokes liver inflammation. However, the contribution of ABCB4 deficiency to the development of the pathologies has been largely unknown. Previous observations suggest that Mdr2/MDR3 are not only the PC transporters for bile formation, but also play a role in the maintenance of lipid homeostasis. Indeed, a previous report showed association of alterations in lipid metabolism and liver pathogenesis in Mdr2-/- mice. Although hepatocytes are the major cell type in which Mdr2/MDR3 are expressed, Mdr2/MDR3 mRNAs have been detected in many tissues such as adrenal glands, muscle, tonsil and spleen, and various cell types including T and B lymphocytes, epithelial cells and fibroblasts. In this study, we used mouse embryonic fibroblasts (MEF) isolated from Mdr2-/- embryos to investigate if Mdr2 deficiency impacts on other cell functions than bile formation. We found accumulated reactive oxygen species (ROS), increased lipid peroxidation and DNA damages in Mdr2-/- MEFs. These cells display enhanced proliferation, resistance to doxorubicin-induced apoptosis and spontaneously undergo transformation at late passages. The phenotypes in MEFs are correlated with the findings in Mdr2-/- mice in which were observed enhanced levels of lipid peroxidation and DNA damage in the liver and increased susceptibility of carcinogen-induced tumor development in the intestine. Our results imply that Mdr2 deficiency can contribute to liver diseases through perturbation of ROS homeostasis and underscore possible extrahepatic functions of Mdr2, as well as of MDR3 in humans (Tebbi et al. 2016 Carcinogenesis, PMID: 26542370).
Project Results:
WP1 Integration
This WP had a logistic purpose, essential to support the operations of the MODHEP consortium and for the development of other WPs: by itself, WP1 produced no specific impact or exploitation.
WP2 Transcriptome Analysis & WP6 Screening and validation in human samples
In the frame of the MODHEP consortium, we produced a large amount of genetic and genomic data for human clinical HCC samples and mouse HCC models using various high-throughput methodologies and instruments such as the CAGE technology and the retro-transposon capture sequencing, which have been developed and improved for many years by the MODHEP partners.
The data provide precise positions of transcription start sites (TSSs), and revealed a high complexity of the non-coding transcriptome in HCC. In particular, identified LTR-derived ncRNAs can be biomarkers associated with clinical conditions such as recurrence and differentiation states. The data were released to public databases as a unique resource for the functional characterization of the ncRNAs.
WP3 Genome-wide Detection and Analysis of Somatic Mutations
The data obtained in this WP showed no significant contribution of somatic mutations to tumorigenesis in the mouse models of HCC. We foresee no specific use of this information in the clinic since human HCCs are associated with a spectrum of well-defined genetic lesions. On the other hand, our finding that genomic instability targeting the MAPK signalling pathway is associated with tumorigenesis in Mdr2-/- mice (Iannelli et al. 2014, Nat. Comm.) provides new insight into this pathology, and more specifically in the subset of human HCC associated with defects in hepatocyte biliary transporters, pointing to possible therapeutic exploitation of MAPK signalling inhibitors in these particular pediatric tumors.
WP4 Chromatin dynamics of transcriptional and epigenetic regulators during hepatocarcinogenesis
Sustained and continuous analyses of the data are still in progress. The ongoing investigation will permit to complement our understanding of the incidence of the epigenetic modifications on liver tumor outcome. Some of the results obtained in this WP, together with results generated within other WPs, have been already used for dissemination under the form of poster presentations but also as a first publication in a high impact journal in the field of cancer research (Kress et al, Cancer Research 2016). Other results obtained are in completion and are being related to other WPs data in collaboration with the partners of the consortium. We expect to understand better how these different layers of genome organization influence each other, and impact on cellular physiology and tumor progression, a cornestone of our consortium. In addition having profiled similar histone modifications in different liver models could also be instrumental to highlight differences in the behavior of the epigenetic machineries depending of the events raising the tumor (sustained inflammation, oncogene upregulation), and help identify differential targets for future therapeutic development. Notably, unlike genetic mutations, epigenetics aberrations are reversible in nature. This characteristic has led to the emergence of investigations on possible epigenetic therapy.
Finally, we expect our study on the Chromatin dynamics of transcriptional and epigenetic regulators during hepatocarcinogenesis to be the basis of several new publications available to the scientific community.
WP5 Higher-order chromatin and nuclear architecture
The 3D genome is increasingly appreciated as an important epigenetic contributor to genome functioning. Our collaborative efforts to study the 3D genome in the mouse liver, and changes therein upon induction of liver tumours, mark one of the first studies, if not the first, that aim to understand the dynamics and role of genome organization in cancer. The Modhep work sparkled the development of novel technologies that enable to study important aspects of genome structure in greater depth. Among these novel technologies is the promoter capture-C method, which is dedicated to the detailed analysis of DNA contacts made by all promoters of all genes across the genome. Sat-4C is another method developed in this context, which allows uncovering the genomic regions that associate with pericentromeric heterochromatin, a compartment in the cell nucleus where inactive chromatin is stored. Novel Hi-C computational analyses tools were developed, including Hi-C-Pro, an optimized and flexible pipeline for Hi-C data processing. Collectively, these novel tools are highly instrumental to the ever-increasing scientific community that studies nuclear organization.
Dedicated bioinformatics efforts are still ongoing to fully exploit and integrate the unique comprehensive dataset generated in the tet-Myc tumor model. Following up from our initial description of 1D-chromatin and transcriptional profiles in (Kress et al. 2016, Cancer research), these 3D genome analyses are soon expected to create a unique, integrated picture of the dynamics between epigenetics, genome topology and gene expression that underlies Myc-induced tumorigenesis, leading to novel fundamental insight in the function of this pervasive oncogene, and fulfilling one of the essential goals of our consortium. Conceptually, this insight will be relevant for our understanding of the molecular processes that underlie and accompany tumour formation in general, not only in Myc-driven tumours but also in other types of cancers.
WP7 Computational and Systems-level analyses
Progress in genomics and epigenomics is promising to lead to a new generation of cancer therapies, and to comprehensive and detailed mapping of tumor cells functionality and response to treatment. The technology that is emerging is capable of generating data on the genetic and molecular phenotypes of tumors, but the interpretation of such data remains difficult. Over the last few years, research on tumor genetics and epigenomics, including research by MODHEP, indicated that in order to utilize modern techniques for the understanding of cancer, it will be essential to integrate multiple layers of tumor profiling and modelling. For example, it is clear that tumor genetics alone, while extremely important and informative to personalized treatment, cannot fully guide clinical outcome without further characterization of the cellular and epigenomic state of the tumor cells. Likewise, epigenomic profiling must be performed comprehensively such that both local signatures at oncogenes and tumor supressors, and global characteristics of the chromosomal loci containing them, will be achieved. The interdisciplinary challenge of combining genetics, epigenomics data with novel computational models, toward effective development of cancer treatment is driven forward by the research performed in MODHEP, and the joint understanding between biologists, clinicians and mathematicians, as formed by the consortium, is going to continue and serve as a basis for further progress in this field.
WP8 Preclinical validation in the mouse tumour models
The identification of broader functions of the PC transporters Mdr2/MDR3 may have impact in the appreciation of human diseases associated with MDR3 mutations. Although extrahepatic phenotypes have never been reported in patients harboring homozygous mutations in the MDR3 gene, analysis of potential effects of MDR3 defects in other tissues than the liver may provide full insight into disease pathogenesis. In particular, the question of whether the functions of immune cells are affected by MDR3 mutations may be worth to be addressed, as these cells express the ABCB4 gene and are key factors in mediating inflammatory response in the diseased liver caused by MDR3 mutations.
PFIC3 is one of the three types of Progressive Familial Intrahepatic Cholestasis. With estimated incidence between 1/50,000 and 1/100,000 births, PFICs are rare but severe diseases of childhood, which leads to death from liver failure at ages usually ranging from infancy to adolescence. There are currently no effective cures available. PFIC is closely associated with hepatic inflammation, which is an essential element in inducing liver fibrosis and HCC. Our results from Mdr2-/- mice show that Mdr2-/- liver outnumbers WT liver in CD8 T cell population and that both effectors and memory cells of CD8 T cell population are increased in Mdr2-/- livers. There are more interferon gamma-secreting CD8 T cells derived from the liver and the spleen of Mdr2-/- mice than those from WT mice, indicative of cytotoxic T cell functions. This knowledge might be applied for appropriate control of inflammation in mice, which would open the horizon for clinical translation of these results to improve the severity of phenotypes for the benefit of PFIC patients.
Potential Impact:
http://www.modhep.eu
Main contacts:
Prof. Bruno Amati - bruno.amati@iit.it
modhep@novamen.eu
List of Websites:
Link to project videos: http://www.modhep.eu/videos.php
Executive summary
The founding concept of the MODHEP consortium was that the pathologic behavior of cancer cells is controlled by an abnormally reprogrammed genome, with aberrations at multiple levels of genome organization. These aberrations range from the genetic information and epigenetic structure of the genome to its higher-order folding and organization within the cell nucleus, and ultimately cause the changes in gene expression that drive tumor initiation and progression. A full understanding of the cancerous state will require that the molecular basis for genome reprogramming be described at all these levels, allowing a comprehensive model of genome reprogramming. MODHEP tackled this challenge in Hepatocellular Carcinoma (HCC) with an integrated, multi-layered analysis of genome organization in two mouse models of HCC, as well as in clinical HCC samples. A series of approaches based on advanced high-throughput sequencing technologies were used to produce genome-wide maps of alterations in DNA sequence, epigenetic marks, protein-DNA interactions, long range chromatin folding and gene expression. In parallel, imaging techniques were used to analyze changes in chromatin and nuclear organization at the single-cell level. The combination of these approaches to tractable tumor models allowed an unprecedented view on the dynamic changes in genome organization that are associated with tumor progression, as well as tumor regression. These studies led to the identification of a series of candidate genes, some of which were validated for their involvement in cellular transformation and cancer, paving the way for their therapeutic exploitation.
Summary description of project context and objectives
The concept behind MODHEP was that the pathologic behavior of cancer cells is controlled by an abnormally reprogrammed genome, with pathological changes occurring at the various levels of genome organization: genetic (DNA sequence), epigenetic (DNA-binding proteins and chromatin modifications), higher-order folding and organization of the genome within the nuclear space, and ultimately deregulated gene expression. Thus, one of the key objectives of MODHEP was to produce an integrated analysis of the nuclear changes occurring during tumor progression by generating coordinated, multi-faceted and quantitative series of datasets providing a complete description of the genetic, epigenetic and nuclear organizational aberrations associated with tumorigenesis in two complementary models of liver cancer.
MODHEP's scientific objectives, as initially defined by the scope of each WP, were the following:
• A preliminary WP (WP1) had a logistic objective, namely the set-up and management of the two experimental tumor models (tet-Myc and Mdr2-/- mice) and the coordination of sample preparation and distribution throughout the consortium, for the optimal achievement of all subsequent WPs.
• Defining global changes in the transcriptome (including coding and non-coding RNA species) during multiple stages of liver cancer development using next-generation sequencing (WP2).
• Defining the genomic alterations occurring at subsequent stages of liver cancer development using high-throughput sequencing (WP3).
• Profiling epigenetic features (and the transcriptional regulators responsible for their control) at multiple stages of liver cancer development, using chromatin immunoprecipitation coupled to high-throughput sequencing (WP4).
• Defining the changes in higher-order chromatin organization and nuclear architecture at high resolution using a combination of genome-wide and single gene approaches (WP5).
• Assessing the transferability of the conclusions to human liver cancer and pre-cancerous states using targeted as well as genome-wide approaches in a large number of patients (WP6).
• Providing a systems-level understanding of integrated transcriptional, genomic, epigenomic and nuclear architectural changes during liver cancer progression (WP7).
• Validating the impact of candidate gene products on tumour development in vivo using genetic approaches in the mouse (WP8).
The above scientific objectives were associated with a series of technical/technological objectives, originally defined as follows:
• Development and optimisation of techniques for transcriptional profiling of all RNA species (coding and non-coding) in multi-stage cancer models using next-generation sequencing technologies (RNAseq).
• Development and optimisation of techniques for high-throughput identification of emerging, low-frequency mutations during early stages of cancer development.
• Development of new experimental and computational tools for high-throughput 3-dimensional epigenome analysis.
• Devising a computational framework for reverse engineering of a gene regulatory network model from large scale transcriptional, epigenetics and genetic data.
• Devising a novel three-dimensional computational model to describe the changes in the epigenome during the carcinogenic process.
• Development of novel computational methodologies for modeling cancer gene networks in the context of the aberrant cancer epigenomes.
As described in the various intermediate reports as well as in this final report, the MODHEP work plan was adjusted during the funding period to face experimental contingencies and biological variations in the adopted tumor models, as well as to extend the analysis to tumor regression (and not simply progression), a key addition that was not included in our original proposal. The consortium thus went on to generate the diverse range of proposed datasets, gaining essential insight into the multiple layers of genome organization in liver tumorigenesis.
While part of the experiments, analyses and integration are still ongoing and will extend beyond the funding period, we can assert here that the consortium has significantly fulfilled its initial goals, and in some instances surpassed these with important additional achievements.
Project Context and Objectives:
WP1 Integration
Integrating the activity of individual groups has been one of the crucial aspects of this project, which required that mice, tissues and cells at various pre-tumoral and tumoral stages had to be promptly available to all partners to carry out specific experimental work.
The work package (WP1) was specifically dedicated to the integration of MODHEP data and resources, and was aimed at streamlining the interactions among partners by creating the essential enabling conditions for a productive exchange amongst the Consortium partners of the necessary resources and data generated. The principal aim of this WP was to define robust and reproducible experimental procedures for the delivery to all partners of tissue samples and cells obtained from the experimental mice. To this end, we first set up centralized colonies of HCC mice models (at IEO and IP, see below) and housed them throughout the duration of the project. Next, we provided participating groups with suitable material to perform experimental analyses throughout the duration of the project, setting up a central repository for data, accessible to all partners.
An important note of WP1 regards our choice of mouse models. These models had to (i.) provide insight into disease initiation and progression; (ii.) be clinically relevant; (iii.) allow access to sufficient amounts of homogeneous cell populations for chromatin/nuclear analysis and (iv.) produce datasets suitable for computational and systems biology modelling. No single mouse model can integrate all of these qualities, nor will it be completely comparable to the corresponding human disease. On the one hand, transgenic oncogene-driven tumors provided the advantage of relatively synchronous and homogeneous disease onset and progression. On the other, non-transgenic tumors driven by inflammation and/or spontaneous mutations more faithfully represented the succession of events occurring in human cancer. To take advantage of all features, we combined two different models. First, Tet-Myc mice (as extensively described below), which allowed the controlled activation of the c-myc oncogene in hepatocytes and highly synchronous and penetrant cancer development (Beer et al., 2004), provided access to pre-tumoral and tumoral stages. Second, Mdr2-/- mice, in which a chronic peri-portal inflammatory process leads to displasia, adenoma and adenocarcinoma, mimicked inflammation-driven liver cancer in humans (Mauad et al., 1994; Pikarsky and Ben-Neriah, 2006).
Task 1.1: Definition of standardized procedures for the production and delivery of samples to WPs 2-5 and 8
Start date: Month 1. End date: Month 9.
This task, which has been completed according to the original plan, consisted in the definition of the experimental procedures for the production and delivery of biological samples from the centralized mouse colonies that have been established at IEO and Institut Pasteur to all groups that performed experimental work on them (as described in WPs 2- 5 and 8).
Briefly, in addition to standard procedures as genomic DNA extraction and histological analysis, we have written a protocol (part of deliverable D1.1) for preparation of chromatin immunoprecipitation followed by deep sequencing (ChIP-seq) and for fixed samples that have been subjected to Chromosome conformation capture techniques. This protocol has been shared with other MODHEP partners to allow highly comparable analyses between the laboratories.
Task 1.2: Generation of samples for experimental analyses to be carried out by partners
Start date: Month 1. End date: Month 54.
Two centralized mouse facilities for the project have been established at IEO and Institut Pasteur. Each Institute has established and maintained the two mouse colonies - namely the Tet-Myc/LAPtTA and the Mdr2-/- strains - under identical veterinary and housing standards, and throughout the length of the project. The two facilities also produced and delivered to partner laboratories the experimental samples obtained at different pre-tumoral and tumoral stages. These samples have been used for the downstream analyses described in the other WPs. The dimension of normal, preneoplastic and neoplastic mouse livers was sufficient to divide each organ into multiple pieces to be divided among several labs for different analyses. Therefore, multiple datasets have been generated from each individual liver, allowing us to assess the coexistence of complex alterations within cells coming from the same organ. When the generation of an ultrapure hepatocyte population has been required for downstream analyses, we have obtained it by the portal vein perfusion technique, which allows organ dissociation in situ and short term culture of pure populations of non-proliferating hepatocytes.
Mdr2 mice (Natoli lab, IEO, Milan)
IEO has established Mdr2-/- and Mdr2+/+ mice colonies. We have collected Mdr2+/+ and Mdr2-/- liver samples from 10 animals per time point (8 weeks, 4 months, 6 months, 8 months and 15 months/HCC stage). Different type of samples (for transcriptome, genetic or epigenetic analyses) have been used or sent to partners. Significant results have been described in each WP section.
tet-Myc/LAPtTA (Amati lab, IIT, Milan)
General overview
We used the tet-Myc/LAPtTA mouse colony to produce samples of Myc-driven liver tumorigenesis, with a focus on both tumor progression and early regression. We have set up the tet-Myc/LAPtTA mouse colony in Milan (deliverable D1.2) and studied tumor development depending on the age of tet-Myc induction. tet-Myc induction in utero led to tumor development by the age of 5-6 weeks. During the course of the MODHEP funding period, initial results from transcriptome (WP2) and epigenome (WP5) analyses in the Amati lab (IIT) based on the overexpression of the wild-type Myc gene pointed to an essential role of Myc-dependent gene repression for Myc-induced liver tumorigenesis. Partner IIT therefore decided to include an additional mouse strain for further analyses. In addition to wild-type Myc, the mutant MycV394D (hereafter MycVD) was examined. MycVD does not bind to Myc’s partner protein Miz1 and is therefore compromised in target gene repression.
Task 1.3: Developing a data coordination center for the project.
Start date: Month 7. End date: Month 66.
A small database of distributed samples has been generated to allow tracing the origin of each sample and accessing information regarding the mice.
Ivica Letunic' s team at Biobyte set up and maintained a data coordination center for the use of MODHEP partners and for the dissemination of project results. The data center served as a repository for processed datasets and models, with primary data analysis performed by the respective groups and primary data archiving handled in the global data repositories (EBI, NCBI). A web based user interface to the database has been developed, which allowed partners to simply select relevant datasets based on various queries throughout the WPs.
Significant results, partner IIT (Amati)
To study the interplay between Myc activity, transcription and chromatin organization in liver tumors we used the tet-Myc/LAPtTA mouse model (tet-Myc WP1, Suppl. Fig. 1.1A) as originally described by others (van Riggelen J et al. Genes & Development. 2010 Jun 15;24(12):1281-94), but based on two different c-myc transgenes, tet-MycWT and tet-MycVD, the latter expressing the MycV394D mutant defective in Miz1 binding. Breeding of either of these tet-Myc strains with LAPtTA animals in the absence of doxycycline led to expression of the human c-myc mRNA in liver progenitor cells in utero in double-transgenic embryos (tet-Myc WP1, Suppl. Fig. 1.1B).
We first characterized tumor progression induced by wild-type Myc: at embryonic day E18.5 double-transgenic tet-MycWT/LAPtTA (dtg) embryos appeared phenotypically indistinguishable from control siblings (tet-Myc WP1, Suppl. Fig. 1.1C) retained a normal liver architecture at the histological level (data not shown), but showed slightly increased liver sizes (tet-Myc WP1, Suppl. Fig. 1.1D E). By 6 weeks of age, tet-MycWT/LAPtTA mice showed a gross increase in abdominal size and fully penetrant, gender-independent development of multi-nodal liver tumors that expressed high Myc levels (tet-Myc WP1, Fig. 1.1A-E Suppl. Fig. 1F). Closer histological assessment revealed both epithelial hepatoblastoma and HCC-like tumors. Tumors stained positive for E-cadherin (tet-Myc WP1, Fig. 1.1E) expressed mRNAs encoding markers of fetal progenitor cells, such as AFP, Dlk1, Sall4 (tet-Myc WP1, Suppl. Fig. 1.1H) and contained frequent mitotic and apoptotic cells (tet-Myc WP1, Suppl. Fig. 1.1G). As previously reported, Myc- driven liver tumors showed full oncogene addiction, as feeding tumor-bearing mice with doxycycline-containing food rapidly suppressed Myc expression (tet-Myc WP1, Fig. 1.1F) and induced tumor regression, with significant decreases in tumor mass within a week, and apparently normal livers and healthy conditions within 3-4 weeks (tet-Myc WP1, Fig. 1.1G). In summary, tet-Myc activation in utero caused a prenatal increase in liver size and the fully penetrant development of Myc- dependent hepatoblastomas by the age of 4-6 weeks.
Based on a distinct tet-Myc strain, but the same LAPtTA transgene and experimental scheme used here, others reported slower tumor development following Myc activation in utero. The reasons for these differences may be partly genetic (distinct myc transgenes), partly environmental, but are beyond the scope of this study.
We then addressed whether the V394D mutation affected Myc's oncogenic activity in the liver, as previously reported in lymphomas. Relative to tet-MycWT, tet-MycVD animals showed delayed tumorigenesis (tet-Myc WP1, Fig. 1.1B). Once formed, however, MycWT- and MycVD-induced tumors appeared equally aggressive, moribund animals showing comparable liver weights (tet-Myc WP1, Fig. 1.1H) and tumor load (tet-Myc WP1, Suppl. Fig. 1.3A). Albeit variable among individual tumors, wild-type and mutant Myc were expressed within similar ranges, either as mRNA (tet-Myc WP1, Fig. 1.1I) or protein (tet-Myc WP1, Suppl. Fig. 1.2B). As with MycWT, MycVD-induced tumors showed no significant gender differences (tet-Myc WP1, Suppl. Fig. 1.2C) and expressed the fetal hepatoblasts markers Afp, Dlk1, Sall4 (tet-Myc WP1, Suppl. Fig. 1.3E) and E-cadherin (tet-Myc WP1, Fig. 1.1E). By histological analysis, MycVD showed the same tumor types as MycWT, albeit with increased frequency of the HCC-like tumors compared to hepatoblastoma (tet-Myc WP1, Suppl. Fig. 1.2E). As a marker of apoptosis we monitored PARP cleavage, revealing variability among tumors, but no significant difference between the two myc genotypes (tet-Myc WP1, Suppl. Fig. 1.2B).
To address Myc function at a pre-tumoral stage, we purified LAPtTA/tet-MycWT and VD fetal hepatoblasts by magnetic cell sorting with an E-cadherin antibody. RT-PCR analysis revealed similar myc mRNA levels in both populations (tet-Myc WP1, Suppl. Fig. 1.1B) comparable activation of a Myc-induced gene (Smpdl3b) but defective repression of Cdkn1a by MycVD, as expected (tet-Myc WP1, Suppl. Fig. 1.2F). MycWT and MycVD led to similar increases in the number of fetal hepatoblasts recovered per embryo (tet-Myc WP1, Suppl. Fig. 1.2G) suggesting equivalent proliferative activities.
WP2 - Transcriptome Analysis & WP6 Screening and validation in human samples
In the frame of the MODHEP consortium, we joined our efforts to produce and analyse a large amount of genetic and genomic data for human clinical HCC samples and mouse HCC models using unique high-throughput methodologies and instruments such as the CAGE technology and the retro-transposon capture sequencing, the latter having been developed and improved for many years by the MODHEP partners.
• Large scale transcriptome sequencing for human adult and mouse liver cancers
We sequenced CAGE libraries for 50 primary hepatocellular carcinoma (HCC) tumours and 50 matched non-tumours as well as 21 primary hepatoblastoma (HB) tumours and 21 matched non-tumours. In addition, we sequenced CAGE libraries for 37 liver samples including wild type, inflammation, adenoma, and HCC extracted from the Mdr-/- mouse model as well as 38 liver samples of wild type and HCC extracted from the Tet-Myc mouse model. The raw CAGE data have been submitted to the NCBI database of Genotypes and Phenotypes (dbGaP; http://www.ncbi.nlm.nih.gov/gap/) under accession number phs000885.v1.p1 for human HCC and the Gene Expression Omnibus (GEO; http://www.ncbi.nlm.nih.gov/geo/) under accession number GSE60982 for mouse HCC.
• Aberrant activation of LTR retroviral promoters in human adult HCC
We identified 64,366 distinct TSSs as the transcriptome of human HCC tumour tissues from the CAGE data. To distinguish potential non-coding RNAs from protein-coding genes, we classified the identified 64,366 TSSs into 3 groups using the Gencode12 annotation: (1) coding TSS, which overlaps with a protein coding TSS (+/- 100bp) in sense; (2) proximal TSS, which is located within 5kb of a coding TSS or overlaps with a coding exon; (3) distal TSS, which is neither coding nor proximal TSS. We compared expression levels between 50 tumours as a case group and 50 matched non-tumours as a control group using edgeR to explore mis-activated transcripts in tumours. We identified 11167 significantly up-regulated TSSs in tumours with the threshold FDR<0.05 of which 4942, 4779, and 4756 are coding, proximal, and distal TSSs, respectively. The most significantly up-regulated protein-coding gene is GPC3, a known candidate biomarker and therapeutic target for HCC. For distal TSSs, 34% of the top 100 most significantly up-regulated elements overlap with repetitive elements in their sense direction. To reveal the association between up-regulated TSSs and repetitive elements, we examined what fraction of up-regulated TSSs overlap with major repetitive elements (LINE, LTR, and SINE), which are annotated by RepeatMasker and RepBase. Interestingly, about 20% of up-regulated distal TSSs overlap with LTR elements in the sense direction (Fig 2.1). Furthermore, the fraction of LTRs increases from 20% to 30%, if limited to the most significantly up-regulated TSSs with FDR below 1.0E-08. These percentages are outstandingly higher than the percentages for (1) LTR elements in the anti-sense direction, (2) non-up-regulated distal TSSs, (3) LINE and SINE elements in sense and anti-sense (1~6%) and (4) randomized distal TSSs. This strongly indicates that the promoters derived from LTR elements are largely activated in HCC tumours. We further examined the expression levels of the LTR promoters in normal livers using 3 datasets: (1) normal liver tissues prepared in this study (n=5), (2) adult and fetus liver tissues produced by Fantom5 (n=2), (3) primary hepatocytes by Fantom5 (n=3). As expected, expression levels of LTRs in normal livers are much lower than those of non-tumours as well as tumours (Fig. 2). We thus conclude that the up-regulation of LTR promoters is a hallmark feature of HCCs.
• Molecular signature of HCC comprising LTR-derived transcripts
We selected human ncRNA candidates appropriate for determining a molecular signature of HCC using the following stringent criteria: (i) FDR threshold below 1.0E-10; (ii) fold change above 8.0; (iii) expression in at least 30 tumour samples. About one third (43/133) that passed the criteria coincided with LTRs (Fig. 2.3). According to HepG2 data produced in the ENCODE project, these LTR-derived transcripts seem to be nuclear ncRNAs (Fig. 4). The fact that the signature-LTRs are scarcely transcribed in normal liver tissues raises the question of the nature of the tissues in which they are intrinsically programmed for expression. According to FANTOM5, an expression atlas of various types of human primary cells and tissues (http://fantom.gsc.riken.jp/5/datafiles/latest/basic/) their expression is limited to 1 to 2% of the normal tissues and primary cells (Fig. 2.5). Intriguingly, reproductive tissues and primary cells occupy the top seven positions. Testis, which expresses about half of the 43 signature-LTRs, is in the first position followed by placenta, chorionic, and amniotic membranes (Fig. 2.6). Some of the highest-ranking signature-LTRs, such as LTR-003, LTR-004, and LTR-006, are exclusively expressed in reproduction related tissues (Fig. 2.7). In contrast, the top 6 signature-LTRs are widely detected in all hepatic and non-hepatic cancer cell lines, emphasizing the close link between these LTRs and carcinogenesis. We then submitted 15 of the 43 signature-LTRs to 3’ RACE validation. Deep sequencing of 3’ RACE products using MiSeq successfully determined the transcript structures of 14 of them, among which 13 had at least one splice site. The 25 transcripts detected, including splice variants, were all predicted to be ncRNAs with low coding potential by the Coding-Potential Assessment Tool (CPAT). The median exon size is of 154 bp, which is comparable to the average coding gene size (122bp), whereas the median intron size is of about 37 kb, i.e. much longer than the general protein-coding gene size (1 kb). Figure 2.8 shows an example of novel transcripts, LTR-002, determined by the 3’ RACE. The TSS is located at the middle of an LTR element, and the second and third exons are located far downstream of the first exon without encoding any viral proteins, suggesting that LTR-002 has become fixed as a promoter provider to the ncRNA. We validated the CAGE data by measuring the expression levels of selected four LTRs using quantitative real-time PCR in two different datasets. The expression of these LTRs was more than 10-fold higher in ~50-80% of the tumour samples compared to the matched NT in both datasets, which confirm that these LTRs are recurrently activated in HCC with different activation levels (Fig. 2.9). Finally, using clinical information, we show that the HCCs with high-level LTR is globally more severe than those with low-level LTR (Fig. 2.10). In particular, The LTR-high HCCs are significantly correlated with a high risk of recurrence, suggesting that the LTR expression level might be useful as a new characteristic in the definition of HCC molecular subclasses. This study was published in Genome Research [PMID:26510915].
• Somatic genomic alterations in paediatric HCC caused by bile salt export pump deficiency (BSEP-HCCs)
We extensively searched for potential cancer driver gene mutations in HCC tumors developed in humans and in mice using exome sequencing. We compared the somatic genomic alterations profiling in HCCs in the two species. In both HCC was induced by defects in hepatocyte biliary transporters, which expose hepatocytes to bile salts and inflammation. We mapped the cancer genomes of Mdr2 knock-out (KO) mice, which, similarly to human HCC associated with bile salt export pump (BSEP-HCC) deficiency [also known as progressive familial intrahepatic cholestasis (PFIC) type 2], develop HCC through the impairment of bile secretion. In BSEP deficiency, inherited mutations in the ABCB11 gene cause impairment of bile salt export from hepatocytes into bile, chronic inflammation and eventually liver cancer. Thus, a study of this type of liver cancer is an opportunity to map the acquired genomic modifications that trigger liver cancer in the absence of exogenous direct (viruses) or indirect (alcohol) mutagens.
We sequenced the whole exomes of six human BSEP-HCCs and nine murine HCCs as well as their corresponding non-tumor surrounding liver tissues. We also used genome-wide SNP arrays to investigate the occurrence of copy-number variations (CNVs) in seven BSEP-HCCs, including all tumors screened for point mutations and one additional lesion. In both human and mouse cancer genomes, we find a few somatic point mutations with no impairment of cancer genes (Fig. 11), and a pervasive occurrence of chromosomal rearrangements that led to massive gene amplification (Fig.12). In both cases, cancer genomes accumulated massive copy-number gain in contrast to very few somatic SNVs or small indels. Copy-number gains preferentially occur at late stages of cancer development and frequently target the MAPK signalling pathway, and in particular direct regulators of JNK. The pharmacological inhibition of JNK delays cancer progression in the mouse. The low levels of mutational instability in human and murine HCCs we found suggest that the acquisition of mutational instability is not the driving force for the development of this type of liver cancer. Such a genomic signature is remarkably different from that of the other HCCs previously sequenced, which acquire mutational instability and tend to accumulate gene deletions rather than amplifications. These findings confirm the genetic heterogeneity of liver cancers caused by different etiological agents and, at the same time, the remarkable analogy between human and mouse tumors with similar aetiopathogenesis. This study was published in Nature Communications [PMID:24819516].
• Somatic and germinal LINE1 retrotransposition in HCC
The same set of human HCCs, whose transcriptomes were analysed by CAGE and RNA Seq, was also explored for somatic rearrangements by RC-Seq. This enhanced retro-transposon capture sequencing (RC-seq) method was applied to the genomes of human HCCs in order to elucidate whether, and how, LINE1 retrotransposons contribute to insertional mutagenesis during liver carcinogenesis.
We studied T/NT HCC samples i.e. tissues originating from tumor (T) areas and non-tumor (NT) areas distant from the tumors from 19 patients with HCC. The underlying liver disease was Hepatitis B or C virus–induced cirrhosis. We demonstrated on this cohort the importance of endogenous L1-mediated retrotransposition in the germ line and somatic cells of patients with HCC. We constructed RC-seq libraries for the 19 T/NT HCC samples and then sequenced in multiplex on an Illumina HiSeq platform. We found two archetypal mechanisms revealing MCC (Mutated in Colorectal Cancers) and ST18 (Suppression of Tumorigenicity 18) as candidate genes for HCC. The first of these genes, MCC, which is expressed in the liver, is known to regulate the oncogenic β-catenin/Wnt signaling pathway frequently activated in HCC. 4 out of 19 (21.1%) donors presented germ line retrotransposition insertions in the tumor suppressor MCC pointing to an inhibition of the liver tumor suppressor function of MCC. The expression of MCC was ablated in each case, enabling oncogenic β-catenin/Wnt signaling in tumors and adjacent liver. The second gene, ST18, was activated by an intronic, tumor-specific L1 insertion. Experimental assays confirmed that the L1 insertion interrupted a negative feedback loop by blocking ST18 repression of its enhancer. ST18 was also demonstrated to be amplified and up-regulated in human (liver cancer cell lines) and mouse (Mdr2 KO) HCC model systems, consistent with ST18 functioning as a candidate liver oncogene. Our study substantiate LINE1-mediated retrotransposition as an important etiological factor in HCC. It showed that (i) Germ line L1 and Alu insertions in MCC activated β-catenin/Wnt signaling; (ii) L1 was broadly activated and transcribed in HCC coincident with L1 promoter hypomethylation ; (iii) L1 mobilization in tumor cells accelerated transformation of the HCC genome; (iii) A tumor-specific L1 insertion interrupted a negative feedback loop regulating ST18. This study was published in Cell [PMID:23540693]
WP3 Genome-wide Detection and Analysis of Somatic Mutations
We have conducted an extensive exome profiling of tumors deriving from Mdr2-/- mice (Mdr2-/- HCCs) at different stages of disease development, as well as in tumors from tet-Myc mice, to identify somatic mutations and copy number variations that are progressively acquired in the cancer genome. We have observed very few somatic point mutations in either model even at late cancer stages. In Mdr2-/- mice, we observed the progressive accumulation of gene amplification events affecting the MAPK signalling pathways, and in particular activators of the cJun-N terminal kinases (JNK).
Such a cancer mutational landscape characterised by very few somatic mutations not affecting any known cancer gene differs substantially from that of most human HCCs. To understand whether this is due to inter-species differences or, rather, to the tumor aetiology, we have sequenced the exomes of human HCCs associated with bile salt export pump (BSEP) deficiency, also known as progressive familial intrahepatic cholestasis type 2, which are etiologically similar to liver cancers associated with Mdr2 gene loss. BSEP deficiencies are a heterogeneous group of rare autosomal recessive disorders caused by inherited inactivating mutations in the hepatocyte membrane transporter genes. The disease usually appears in infancy or early childhood and manifests with hepatocellular damage and cholestasis due to defects in bile formation. In BSEP deficiency, inherited mutations in the ABCB11 gene cause impairment of bile salt export from hepatocytes into bile, leading to liver chronic inflammation and to the early onset of hepatocellular carcinoma. HCCs arisen in chronic liver disease in Mdr2-KO mice have an aetiopathogenesis similar to that of BSEP-HCCs, while it is clearly distinct from that of viral and metabolic disease-associated human HCCs. Consistent with this notion, we have observed BSEP-HCC genomes acquire massive gene amplification that affect components of signal transduction pathways, such as the ErbB, the PI3K/Akt and the mitogen-activated protein kinase (MAPK) signalling pathways. Interestingly, we have shown that the pharmacological inhibition of JNK in Mdr2-/- mice attenuates cancer progression and might offer a useful therapeutic intervention in BSEP deficiency patients waiting for liver transplantation. The study in Mdr2-/- mice has been published in Nature Communications (Iannelli et al. 2014; PMID:24819516).
WP4 Chromatin dynamics of transcriptional and epigenetic regulators during hepatocarcinogenesis
Genetic and expression profiling studies have identified several molecular pathways that are deregulated and/or mutated in HCC, and gene expression changes that drive tumor formation. This notwithstanding, the past decade of research has highlighted modifications of the epigenetic landscape as a hallmark of cancers. Indeed epigenetic mechanisms regulate normal development and homeostasis and are crucial for tissue-specific gene expression maintenance. The objectives of WP4 were to determine the chromatin dynamics of transcriptional and epigenetic regulators during hepatocarcinogenesis.
A first approach consisted in the generation of ChIP-seq databases of chromatin-associated features during liver cancer progression. Multiple histone modifications were chosen in order to obtain a comprehensive cartography of the epigenetic landscape relative to gene transcriptional activity (H3K4me1, H3K4me3, H3K27ac) in normal livers and to highlight modifications accompanying the alterations of transcriptional activity occurring during hepatocarcinogenesis, as identified in WP2. Along with this, chromatin binding profile of RNA polymerase II (RNAPII), and histones marks signing actively transcribed gene body, H3K36me3 and H3K79me2, were investigated as well.
A second approach was to analyse the binding of liver and cancer specific transcription factors. In the Tet-Myc model, first of all the founding role of Myc in Myc-induced hepatocarcinogenesis has been assayed on a genome-wide scale by systematic ChIP-seq. Tumors driven by the activation of the transcription factor Myc generally show oncogene addiction. To complete the second approach on the founding role of Myc in the tumor development and maintenance, further ChIP sequencing experiments, in tumors after short-term tet-Myc inactivation, have been performed for histone marks and regulatory factor, including Myc and RNA polymerase II. The work in tet-Myc mice was published in Cancer research (Kress et al. 1016 PMID: 27197165).
Computational analysis reported a strong increase in both binding intensity and the total number of binding sites at all accessible genomic regulatory elements in Myc-driven tumors, supporting the recently described phenomenon of “invasion” of Myc binding landscape upon its overexpression. Focusing on annotated promoter regions revealed a tight correlation between the presence of RNAPII, H3K4me3 and H3K27ac (in either C, T or Toff samples) and Myc binding in T, while promoters lacking those features remained unbound. Most importantly, active promoter marks pre-existed in C and were retained following Myc elimination in Toff. Thus, consistent with other studies, over-expressed Myc in T widely associated with already active promoters while inactive promoters remained unbound. In tumors, over-expressed Myc associated with active promoters in a widespread manner, but increases in binding neither discriminated Myc-dependent from -independent transcriptional responses, nor correlated with the extent of changes in mRNA levels. Furthermore, neither Myc- binding nor recruitment to enhancers (determined by nearest-neighbor analyses) in C or T discriminated primary from secondary Myc target genes. Taken together, no simple rule could be drawn that would link promoter/enhancer association by Myc with either primary or secondary gene regulation.
Myc has been suggested to regulate global RNA Pol II activity at the elongation step. In our system, up- or down-regulation of gene expression was paralleled by equivalent changes in RNA Pol II binding at promoters. During the C to T transition, either primary or secondary genes showed similar variations in Pol II binding, but only primary Myc-dependent genes were affected during the T to Toff transition. Thus indicating that Myc influences RNA Pol II loading on either activated or repressed genes.
Transcription can be regulated at all steps of the RNA Pol II life cycle, including loading, pausing and elongation. The relative contribution of these steps in our experimental system were calculated by RNA polII occupancy and stalling index (RNA Pol II occupancy on promoters relative to gene bodies). Consistent changes in Pol II abundance were identified not only on promoters - as mentioned above – but also on gene bodies. However, stalling indices were not consistently affected in any of the regulatory categories. Altogether, activated and repressed genes showed generally consistent changes in Pol II loading and elongation, whether regulated by Myc in a primary or secondary manner. Further studies will be requires to elucidate the contribution of each regulatory step to Myc activity.
To follow up the epigenetic changes in chromatin patterns along the tumorigenic process in Mdr2 -/- mice, we generated a panel of ChIP-seq data sets on liver samples prepared at different time points. In particular, we considered multiple biological replicates of adenoma, low-grade HCC (≤60% carcinoma content) and high-grade HCC (≥80% carcinoma content), all coming from 11-to-16 months old individuals. These samples were compared to age-matched WT livers, and to samples of 8 months old inflamed but non tumor-bearing livers.
Computational deconvolution of dynamic histone modifications was used in order to identify transcription factors (TFs) activated during the tumorigenic process, and therefore possibly involved in bringing about transcriptional changes associated with, or favoring, cancer. Deposition of H3K27Ac is due to TF binding to underlying regulatory elements and promoting local recruitment of histone acetyltransferases. Therefore, changes in H3K27Ac reflect changes in the activity or expression of specific TFs and may indicate TFs potentially relevant for the disease progression. Furthermore, to improve our classification of genomic regulatory regions, H3K4me3 (a mark for active TSS/promoters) and H3K4me1 (enriched at enhancers when not associated with H3K4me3) were profiled in WT, inflamed livers and HCC samples.
Principal component analysis (PCA) applied over ChIP-seq data for H3K4me1 and H3K4me3, highlighted that samples belonging to different disease steps were clustering separately, although in the case of H3K4me3 the most discriminative variable was PC2. In the case of H3K27Ac, for which data for two additional categories (“adenoma” and “high grade HCC”) were available, separation was still generally neat. Based on PCA, “outliers” samples failing to cluster close to the bulk of each category were excluded of the next analyses. We found that the highest number of induced or repressed peaks (with respect to WT) displayed by all samples at a given time point was found for H3K27Ac, and particularly for induced peaks, indicating that this mark is subjected to the highest variability specifically associated to tumorigenesis in Mdr2-KO mice. On the contrary, the number of common induced or repressed H3K4me1 and H3K4me3 peaks was very low both for inflamed and low-grade HCC samples. Gene ontology (GO) analysis for genes located in the vicinity of H3K27Ac peaks induced was performed at all stages. Results revealed a strong inter-class functional enrichment for biological processes associated with inflammation and immune response, suggesting that the inflammatory response started in the early disease stages and continued until tumors developed. We observed that, at all time points, the most enriched matrices corresponded to binding sites for the activator protein 1 (AP-1) TF family members (in particular Jun, JunB, JunD, JDP2, Fos, FosB, FosL1, FosL2 a BATF). AP-1 TFs are activated in response to a variety of stimuli, including cell stress and inflammation, exposure to cytokines or growth factors and infection, and control cell proliferation and apoptosis. Therefore, the finding that common H3K27Ac induced peaks were enriched for AP1 matrices is in agreement with the general activation of inflammatory pathways resulting from the GO analysis. The second class of transcriptions factors whose matrices were enriched in H3K27Ac common induced peaks was the erythroblast transformation-specific (ETS) TF family, and the Spi-like subfamily including Spi1 (PU.1).
Taken together, this evidence points to the establishment of an inflammatory/stress response program in hepatocytes starting from the pre-malignant, chronic inflammatory phase. This inflammatory signature is retained up to the more advanced HCC stage, and is accompanied by the activation of members of the AP-1 and ETS transcription factor family. PU.1 is a macrophage-specific TF, therefore we decided to better characterize the presence of macrophages and other inflammatory components in our samples and we are investigating how macrophages impact on the generated data.
Another aim of this WP was the identification of epigenomic changes during liver carcinogenesis in relation to silencing and cellular memory, through genome-wide profiling of one of the main factors involved in this process, the so-called polycomb group proteins (PcG). PcGs are epigenetic regulators fundamental for the maintenance of cell-specific gene expression pattern, cell identity, pluripotency and differentiation. Polycomb Repressive Complex 1 and 2 (PRC1, PRC2), are thought to act sequentially to stably maintain gene repression, whereby PRC1 is able to read the repressive H3K27me3 mark deposited by PRC2 and in turn catalyzes H2A mono-ubiquitination (H2Aub1), leading to a transcriptionally silent chromatin conformation.
ChIP-sequencing experiments were performed for PRC2 proteins, H3K27me3 and PRC1 proteins. Attempts to obtain reliable ChIP form the Mdr2-/- model failed, partly due to technical issues. Instead the contribution of Polycomb proteins to the Myc-induced tumor phenotype could be investigated.
Computational analysis of the ChIP-seq data demonstrated the existence multiple combinations of PCR2 and PRC1 complexes, and their differential association with repressive but also with active histone marks. We notably observed a strong correlation between binding of Suz12, Ezh2 and H3K27me3, as expected. Notwithstanding, our data also confirm the recent findings (Stojic et al, 2011, Mousavi et al, 2012, Xu et al, 2015) that PRC2 binding also clusters with active histone marks, a trait that we observed more particularly at promoters in adult liver cells. On the contrary we found that PRC2 characteristics are altered in the Myc-induced tumors. Indeed, we observed that Suz12 co-localization with active histone marks is strongly reduced in tumors relative to normal livers. This reduction is found mainly at promoters and to a lesser extent at potential enhancers. Recent studies have reported a physical interaction between Ezh2 and Myc family in cancers, and notably with c-Myc in HCC (Corvetta et al, 2013, Wang et al, 2014). Our results indicate that, in normal liver, Suz12 and Myc, naturally associate at a subset of PRC2 peaks and that this interaction is specifically observed at TSSs of transcriptionally active genes. Looking at the dynamics of Myc and PRC2 interplay in the Myc induced tumors, we observed that Myc-PRC2 association significantly increases in the tumor (from ˜15% to ˜42% of Suz12 peaks), along with an increased and preferential association (60%) with TSSs covered by non-active chromatin. Finally sites of Myc-PRC2 co-occupancy were found to be different in normal versus tumor livers. Another remarkable finding was a strong alteration of H3K27me3 genomic landscape between control and tumors, which appears to be regional. Further analyses of other PcG components and histone regulators are currently ongoing to obtain a more complete picture on Polycomb dynamics in liver tumorigenesis.
The MODHEP consortium also investigated the epigenomic status of a classic model of facultative heterochromatin, the inactive X chromosome, in the context of liver cancer. The X chromosome carries over 1000 genes, many of which are critical. The stable, epigenetic silencing of one of the two X chromosomes in females is essential for correct gene dosage. Prior studies have revealed that the inactive X shows aberrant organisation and expression in some tumors such as breast cancer (Chaligné et al, 2015) but no studies were carried out in the context of liver cancer. The challenge of assessing the inactive X chromosome lies in the ability to distinguish the active X in a population of cells where random X inactivation will lead to a mixture of cells with either of the two Xs inactivated. We derived mice carrying polymorphic X chromosomes by crossing the mdr2-/- or Tet-Myc mice (Fvb strain) to M. Castaneus mice, which provides F1 with approximately 1 SNP/100bp. Tumors are usually clonal and therefore have a defined Xi. For normal liver control samples, which are likely to be cell mosaics for parental Xi status, we generated mice carrying a knock-out of Xist on one X chromosome (B6 origin), thus ensuring that only one X (Cast origin) would be the inactive one and greatly facilitating allele-specific analyses in wild-type control livers. These samples have been processed for allele-specific Hi-C analysis, in parallel with the tumour samples isolated from both tumour models crossed with Cast mice. Adapting the ChIP-sequencing procedure for allele-specific analysis, the specific chromatin status of the inactive X chromosome in WT and tumour samples, is being profiled allele-specifically for histones marks and regulatory factors, as well as the chromosomal architectural protein CTCF.
These results will be instrumental for the 3D nuclear organisation WP5, and eventually for the modelling approaches in WP7.
WP5 Higher-order chromatin and nuclear architecture
Gene expression changes drive tumour formation. Gene expression, in turn, is influenced by the folding of the genome inside the cell nucleus. In work package 5 of the Modhep project we hypothesized that tumour formation is accompanied and possibly influenced by changes in the folding of the genome. In order to study this, several complementary approaches were pursued.
One approach involved the use of Hi-C, which at the start of the project was a relatively novel technique that can analyse contact frequencies between any pair of genomic segments across the entire genome in an unbiased manner. Hi-C therefore allows generating high-resolution 3D genome contact maps. Hi-C studies were focused on the Myc mouse model that develops liver tumours upon inducible overexpression of the Myc oncogene, one of the most pervasive cancer genes. The consortium had to optimize the Hi-C strategy and make it applicable to cells from the mouse liver before and after they developed liver tumours. We succeeded in optimizing the Hi-C protocol for mouse liver cells and generated high resolution 3D genome contact maps for healthy liver, Myc-induced liver tumors and Myc-depleted regressed liver tumors. Our studies are among the first to generate such high-resolution contact maps for primary tissue directly taken from an organism.
A second approach assumes that the functionally most important changes in 3D structure most likely take place at the genes that alter their expression during tumour formation. A dedicated capture-C method involving hybridization capture probes that allow targeted analysis of contacts specifically made by all gene promoters was successfully developed and applied to healthy liver and liver tumours.
Computational analysis of the data shows that tumour formation is accompanied by large scale changes in the shape of the genome. For example, we find that chromosomal regions change nuclear locations upon tumour formation. Many regions move between nuclear compartments that are known as the active (A) and the inactive (B) compartment. Interestingly, chromosomal segments that move to the active compartment often carry genes that increase their expression in the tumour whereas chromosomal regions that move to the inactive B compartment often contain genes that lower their expression in the tumour. Compartment switches surprisingly appeared irreversible, or at least were stable for 16 hours after depletion of Myc oncogene expression, a period during which already pronounced reversal of gene expression was observed. Although preliminary, this suggests that Myc-driven gene expression changes are not dependent and may precede changes in the overall topology of chromosomes.
Chromosomes are structurally subdivided into topologically associated domains, or TADs: these are the structural but also the functional units of chromosomes. At the level of TADs, chromosome structure was largely unaltered in the tumours, but TADs were found engaged in more inter-TAD contacts in the tumour. This gain in long-range contacts was lost again after tumour regression. Within TADs, chromatin loops between genes and regulatory sequences like enhancers are formed. Roughly 15% of these chromatin loops changed between the control (wild-type) liver and the tumours, with essentially similar numbers of loops that dissolved and appeared during oncogenesis.
A further intersection of the contact maps with the epigenetics and transcriptome datasets generated in other work packages is ongoing and expected to soon yield a unique, integrated picture of the dynamics between epigenetics, genome topology and gene expression that underlies tumorigenesis in a mouse model expressing an oncogene, Myc, leading to novel fundamental insight in the function of this pervasive oncogene, and fulfilling one of the essential goals of our consortium.
Another aim of the consortium was to develop, optimize and standardize techniques to explore the 3D organization of the active (Xa) and inactive (Xi) X chromosome in healthy livers and tumours isolated from female mice. For this, first allele-specific Hi-C analysis on F1 hybrid (129 x cast) ES cells and in clonal neural precursor stem cells (NPCs) was developed and applied. The studies revealed that the inactive X chromosome (Xi) adopts an unusual three-dimensional structure with a global absence of TADs, except at chromosomal regions that escape X inactivation. Xi was found to be partitioned into two “megadomains” that span several tens of megabases each, within which the chromatin fiber seems to be randomly folded. The boundary within the two megadomains lies at the DXZ4 macrosatellite locus, which also overlaps with a structural boundary on the human Xi. When this region is deleted, the Xi no longer folds into this bipartite structure. Oligo FISH probes spanning several megabases across the X chromosome enabled 3D FISH analysis to validate the HiC findings. This study thus provided the critical tools for the analysis of physical measurements of 3D chromatin structure at a chromosome wide scale as a means to investigate the Xi in liver tumours.
Two models of liver tumorigenesis were used in this project, the mdr2-/- and the Tet-Myc/Lap-tTA mouse models. The challenge of assessing the inactive X chromosome is that it must be distinguished from the active X in the same cell, and that in the case of populations of cells, random X inactivation will lead to a mixture of cells with either of the two Xs inactive. To avoid the issue of non-clonality we generated mice carrying a knock-out of Xist on one X chromosome (B6 origin), thus ensuring that only one X (Cast origin) would be the inactive one and greatly facilitating allele-specific analyses in wild-type livers. Tumors are often clonal and therefore have a defined Xi. These samples were processed for allele-specific Hi-C analysis, in parallel with the tumour samples isolated from both tumour model crossed with Cast mice. Tumour samples were found to have frequent partial or complete erosion of the megadomain partitioning, which does not appear to correspond to deletion of the DXZ4 boundary. The disruption could thus be epigenetic, and the chromatin status of this region in WT and tumour samples is under investigation.
Finally, constitutive heterochromatin organization inside the cell nucleus is appreciable under the microscope by DAPI-staining, with these chromosomal regions carrying inactive chromatin forming densely stained foci in the nucleus. Nuclear reorganization of constitutive heterochromatin is a hallmark of many tumours. An objective of the consortium was to gain more insight into this reorganization. To investigate this we focused on the Myc-driven mouse tumour model. In addition to DAPI staining and immunofluorescence assays we set-up fluorescence in situ hybridization (FISH) to visualize the genomic loci corresponding to the constitutive heterochromatin pericentric domains and the neighboring centric domains. We found that organization of pericentric domains displays major changes with a loss of clustering between pericentric domains and a loss of nuclear periphery localization. To independently analyse the spatial organization of pericentromeric heterochromatin we developed a 4C variant that exclusively analyses contacts made by the satellite repeat sequences that flank the centromeres on each chromosome (Sat-4C). Using Sat-4C we confirmed a genomic reorganization of genomic domains associated with pericentromeric heterochromatin. Taken together these results indicate that in myc tumours a major reorganization and nuclear localization of constitutive heterochromatin occurs and accompanies redistribution or reshuffling of pericentric-associated genomic loci. Applying the same approaches to the Myc-off regressed tumors showed that the organization of constitutive heterochromatin is plastic and can dynamically return to a ‘normal’ rather than ‘tumoral’ organization.
WP7 Computational and Systems-level analyses
With the advent of multiple techniques for comprehensive characterization of genetics, gene expression and epigenetic programs in tumors, a major challenge become the integration of multiple data sources toward models of tumorigenesis. In work package 7 of the Modhep project we developed techniques for computational analysis of multi-layered data and applied them to the analysis of the two mouse HCC models, as well as of human HCC samples.
We developed several techniques for analysing chromosome conformation data as assessed by 3C, 4C and Hi-C experiments. Such data provide opportunities for inferring physical contact structures that ties up together complex regulatory regions involving multiple genes and control elements. Modhep research provided us with new techniques for analysing Hi-C data in order to infer the contact structures in such regions. We were specifically interested in the contacts linking gene promoter with gene regulatory elements (or enhancers). Such contacts can be assessed effectively by targeted Hi-C analysis, but in order to understand their potential impact, we embed these targeted experiments within global analysis of HiC matrices, models for topological domains, and integration of ChIP experiments. Modhep research thus provided one of the first comprehensive attempts to understand gene de-regulation during carcinogenesis by combining most of the available strategies for interrogating gene regulation and epigenomic landscapes.
When integrating transcriptional maps of MYC induced liver carcinogenesis with such comprehensive epigenomic modelling, we observed how this key oncogene can drive gene regulation at two levels. First, we observed classical targets of MYC being induced through increase in MYC occupancy and associated promoter histone remodelling, but not de-novo recruitment of enhancers or enhancer-promoter contacts. Such genes were quickly repressed following reduction in MYC levels. On the other hand, a second class of genes recruited MYC and de-novo enhancer contacts, as part of a possible more profound epigenetic reprogramming process. These genes exhibited a more stable transcriptional aberration, which persisted, at least partly, even following reduction in MYC level. Additional analysis of these two classes of genes is on going, but this model provide possible explanation to the effect of targeted therapies on oncogenes, which may lead to direct correction of the gene expression program of direct oncogenes targets, but not to proper regulation of genes that may carry epigenomic memory that is not directly dependent on stable oncogene expression.
Another important research direction in WP7 targeted the role repetitive elements - or 'mobile DNA' - play in hepatocellular carcinoma (HCC). Targeted DNA sequencing of tumour samples obtained from human HCC patients, and from a mouse model of HCC was performed as part of an attempt to link the activity of mobile DNA and the carcinogenic process. Crucially, this analysis revealed new copies of a mobile DNA element called LINE-1 in the tumours that were not found in normal blood samples taken from the same individuals. These new LINE-1 mutations were found to occur in genes involved in preventing tumour formation and growth, including a novel cancer gene called ST18, were a LINE-1 mutation turned the gene on and contributed to tumour growth.
Further integrative analysis aim at linking transcription factor (TF) activation/deactivation patters with inferred post-translation modifications (PTMs). Capturing this kind of regulatory interactions using only transcriptional data, such as gene expression profiles (GEPs), is considered challenging since GEPs are further downstream of the PTM event and only indirectly linked to it. We developed new approaches based on Multi-Information that directly quantifies the co-regulation among N genes with a single positive value (0 indicating no co-regulation) by assessing the extent of statistical dependency among them all. Application of this approach to infer a hierarchical network of interactions between key transcriptional regulators and signalling molecules altered in hepatocellular carcinoma (HCC) identified 29 TFs with an abnormal activity in HCC. The methodology then identified for each one of the 29 TFs which were the kinases that are more likely to be modulating them and thus active in HCC. The inferred hierarchical network of interactions between the identified aberrant transcriptional regulators and signalling pathways (kinases) was shown to be altered in hepatocellular carcinoma. This is providing experimentalists with new hypotheses for further exploration of the regulatory dynamics leading to HCC.
WP8: Preclinical validation in the mouse tumour models
Multidrug resistance P-glycoproteins, Mdr2 in mouse and MDR3 in human, are the choline-containing phospholipid phosphatidylcholine (PC) transporters that belong to the subfamily B4 of ATP-binding cassette transporters (ABC). In the liver, Mdr2/MDR3 are involved in the translocation of PC from the inner leaflet to the outer leaflet of the canalicular membrane of hepatocytes for direct extraction by bile acids (BA). BA and PC are essential components in the bile that form bile acid micelles to reduce the toxic detergent activity of BA for hepatocytes and cholangiocytes and to establish physiological bile flow. Mutations in the ABCB4 gene in human result in deficiency of PC in the bile, causing diseases including progressive familial intrahepatic cholestasis type 3 (PFIC3), intrahepatic cholestasis of pregnancy and low-phospholipid-associated cholelithiasis (LPAC). A constitutive Mdr2 knockout mouse has been generated, which successively develops liver inflammation, hepatic fibrosis and hepatocellular carcinoma (HCC) at late ages. The pathogenesis in Mdr2-/- mice is commonly considered to be caused by toxic BA in the biliary canaliculus, which damages hepatocytes and cholangiocytes and provokes liver inflammation. However, the contribution of ABCB4 deficiency to the development of the pathologies has been largely unknown. Previous observations suggest that Mdr2/MDR3 are not only the PC transporters for bile formation, but also play a role in the maintenance of lipid homeostasis. Indeed, a previous report showed association of alterations in lipid metabolism and liver pathogenesis in Mdr2-/- mice. Although hepatocytes are the major cell type in which Mdr2/MDR3 are expressed, Mdr2/MDR3 mRNAs have been detected in many tissues such as adrenal glands, muscle, tonsil and spleen, and various cell types including T and B lymphocytes, epithelial cells and fibroblasts. In this study, we used mouse embryonic fibroblasts (MEF) isolated from Mdr2-/- embryos to investigate if Mdr2 deficiency impacts on other cell functions than bile formation. We found accumulated reactive oxygen species (ROS), increased lipid peroxidation and DNA damages in Mdr2-/- MEFs. These cells display enhanced proliferation, resistance to doxorubicin-induced apoptosis and spontaneously undergo transformation at late passages. The phenotypes in MEFs are correlated with the findings in Mdr2-/- mice in which were observed enhanced levels of lipid peroxidation and DNA damage in the liver and increased susceptibility of carcinogen-induced tumor development in the intestine. Our results imply that Mdr2 deficiency can contribute to liver diseases through perturbation of ROS homeostasis and underscore possible extrahepatic functions of Mdr2, as well as of MDR3 in humans (Tebbi et al. 2016 Carcinogenesis, PMID: 26542370).
Project Results:
WP1 Integration
This WP had a logistic purpose, essential to support the operations of the MODHEP consortium and for the development of other WPs: by itself, WP1 produced no specific impact or exploitation.
WP2 Transcriptome Analysis & WP6 Screening and validation in human samples
In the frame of the MODHEP consortium, we produced a large amount of genetic and genomic data for human clinical HCC samples and mouse HCC models using various high-throughput methodologies and instruments such as the CAGE technology and the retro-transposon capture sequencing, which have been developed and improved for many years by the MODHEP partners.
The data provide precise positions of transcription start sites (TSSs), and revealed a high complexity of the non-coding transcriptome in HCC. In particular, identified LTR-derived ncRNAs can be biomarkers associated with clinical conditions such as recurrence and differentiation states. The data were released to public databases as a unique resource for the functional characterization of the ncRNAs.
WP3 Genome-wide Detection and Analysis of Somatic Mutations
The data obtained in this WP showed no significant contribution of somatic mutations to tumorigenesis in the mouse models of HCC. We foresee no specific use of this information in the clinic since human HCCs are associated with a spectrum of well-defined genetic lesions. On the other hand, our finding that genomic instability targeting the MAPK signalling pathway is associated with tumorigenesis in Mdr2-/- mice (Iannelli et al. 2014, Nat. Comm.) provides new insight into this pathology, and more specifically in the subset of human HCC associated with defects in hepatocyte biliary transporters, pointing to possible therapeutic exploitation of MAPK signalling inhibitors in these particular pediatric tumors.
WP4 Chromatin dynamics of transcriptional and epigenetic regulators during hepatocarcinogenesis
Sustained and continuous analyses of the data are still in progress. The ongoing investigation will permit to complement our understanding of the incidence of the epigenetic modifications on liver tumor outcome. Some of the results obtained in this WP, together with results generated within other WPs, have been already used for dissemination under the form of poster presentations but also as a first publication in a high impact journal in the field of cancer research (Kress et al, Cancer Research 2016). Other results obtained are in completion and are being related to other WPs data in collaboration with the partners of the consortium. We expect to understand better how these different layers of genome organization influence each other, and impact on cellular physiology and tumor progression, a cornestone of our consortium. In addition having profiled similar histone modifications in different liver models could also be instrumental to highlight differences in the behavior of the epigenetic machineries depending of the events raising the tumor (sustained inflammation, oncogene upregulation), and help identify differential targets for future therapeutic development. Notably, unlike genetic mutations, epigenetics aberrations are reversible in nature. This characteristic has led to the emergence of investigations on possible epigenetic therapy.
Finally, we expect our study on the Chromatin dynamics of transcriptional and epigenetic regulators during hepatocarcinogenesis to be the basis of several new publications available to the scientific community.
WP5 Higher-order chromatin and nuclear architecture
The 3D genome is increasingly appreciated as an important epigenetic contributor to genome functioning. Our collaborative efforts to study the 3D genome in the mouse liver, and changes therein upon induction of liver tumours, mark one of the first studies, if not the first, that aim to understand the dynamics and role of genome organization in cancer. The Modhep work sparkled the development of novel technologies that enable to study important aspects of genome structure in greater depth. Among these novel technologies is the promoter capture-C method, which is dedicated to the detailed analysis of DNA contacts made by all promoters of all genes across the genome. Sat-4C is another method developed in this context, which allows uncovering the genomic regions that associate with pericentromeric heterochromatin, a compartment in the cell nucleus where inactive chromatin is stored. Novel Hi-C computational analyses tools were developed, including Hi-C-Pro, an optimized and flexible pipeline for Hi-C data processing. Collectively, these novel tools are highly instrumental to the ever-increasing scientific community that studies nuclear organization.
Dedicated bioinformatics efforts are still ongoing to fully exploit and integrate the unique comprehensive dataset generated in the tet-Myc tumor model. Following up from our initial description of 1D-chromatin and transcriptional profiles in (Kress et al. 2016, Cancer research), these 3D genome analyses are soon expected to create a unique, integrated picture of the dynamics between epigenetics, genome topology and gene expression that underlies Myc-induced tumorigenesis, leading to novel fundamental insight in the function of this pervasive oncogene, and fulfilling one of the essential goals of our consortium. Conceptually, this insight will be relevant for our understanding of the molecular processes that underlie and accompany tumour formation in general, not only in Myc-driven tumours but also in other types of cancers.
WP7 Computational and Systems-level analyses
Progress in genomics and epigenomics is promising to lead to a new generation of cancer therapies, and to comprehensive and detailed mapping of tumor cells functionality and response to treatment. The technology that is emerging is capable of generating data on the genetic and molecular phenotypes of tumors, but the interpretation of such data remains difficult. Over the last few years, research on tumor genetics and epigenomics, including research by MODHEP, indicated that in order to utilize modern techniques for the understanding of cancer, it will be essential to integrate multiple layers of tumor profiling and modelling. For example, it is clear that tumor genetics alone, while extremely important and informative to personalized treatment, cannot fully guide clinical outcome without further characterization of the cellular and epigenomic state of the tumor cells. Likewise, epigenomic profiling must be performed comprehensively such that both local signatures at oncogenes and tumor supressors, and global characteristics of the chromosomal loci containing them, will be achieved. The interdisciplinary challenge of combining genetics, epigenomics data with novel computational models, toward effective development of cancer treatment is driven forward by the research performed in MODHEP, and the joint understanding between biologists, clinicians and mathematicians, as formed by the consortium, is going to continue and serve as a basis for further progress in this field.
WP8 Preclinical validation in the mouse tumour models
The identification of broader functions of the PC transporters Mdr2/MDR3 may have impact in the appreciation of human diseases associated with MDR3 mutations. Although extrahepatic phenotypes have never been reported in patients harboring homozygous mutations in the MDR3 gene, analysis of potential effects of MDR3 defects in other tissues than the liver may provide full insight into disease pathogenesis. In particular, the question of whether the functions of immune cells are affected by MDR3 mutations may be worth to be addressed, as these cells express the ABCB4 gene and are key factors in mediating inflammatory response in the diseased liver caused by MDR3 mutations.
PFIC3 is one of the three types of Progressive Familial Intrahepatic Cholestasis. With estimated incidence between 1/50,000 and 1/100,000 births, PFICs are rare but severe diseases of childhood, which leads to death from liver failure at ages usually ranging from infancy to adolescence. There are currently no effective cures available. PFIC is closely associated with hepatic inflammation, which is an essential element in inducing liver fibrosis and HCC. Our results from Mdr2-/- mice show that Mdr2-/- liver outnumbers WT liver in CD8 T cell population and that both effectors and memory cells of CD8 T cell population are increased in Mdr2-/- livers. There are more interferon gamma-secreting CD8 T cells derived from the liver and the spleen of Mdr2-/- mice than those from WT mice, indicative of cytotoxic T cell functions. This knowledge might be applied for appropriate control of inflammation in mice, which would open the horizon for clinical translation of these results to improve the severity of phenotypes for the benefit of PFIC patients.
Potential Impact:
http://www.modhep.eu
Main contacts:
Prof. Bruno Amati - bruno.amati@iit.it
modhep@novamen.eu
List of Websites:
Link to project videos: http://www.modhep.eu/videos.php
