Mouse models for liver cancer

Abstract

Hepatocellular carcinoma (HCC), the most common form of primary liver cancer is the third leading cause of cancer‐related cell death in human and the fifth in women worldwide. The incidence of HCC is increasing despite progress in identifying risk factors, understanding disease etiology and developing anti‐viral strategies. Therapeutic options are limited and survival after diagnosis is poor. Therefore, better preventive, diagnostic and therapeutic tools are urgently needed, in particular given the increased contribution from systemic metabolic disease to HCC incidence worldwide. In the last three decades, technological advances have facilitated the generation of genetically engineered mouse models (GEMMs) to mimic the alterations frequently observed in human cancers or to conduct intervention studies and assess the relevance of candidate gene networks in tumor establishment, progression and maintenance. Because these studies allow molecular and cellular manipulations impossible to perform in patients, GEMMs have improved our understanding of this complex disease and represent a source of great potential for mechanism‐based therapy development. In this review, we provide an overview of the current state of HCC modeling in the mouse, highlighting successes, current challenges and future opportunities.

Highlights

• ► The treatment of HCC requires effective preventive, diagnostic and therapeutic tools.
• ► We review state‐of‐the‐art mouse models for HCC highlighting successes and challenges.
• ► We emphasize tumor‐host interactions and organ/systemic crosstalks.
• ► Cross‐species comparative studies will lead to knowledge‐based novel therapies.

1. Introduction

The World Health Organisation estimates that in 2008, liver cancer accounted for almost 700 000 deaths worldwide, half of these in China and more than 60 000 within Europe. It is the 6th most common cancer type and the third most common cause of death from cancer worldwide with, at the moment, the highest incidence in Asia and sub‐Saharan Africa. Primary liver cancer consists of several histologically different malignancies, such as cholangiocarcinoma, hepatoblastoma and hemangiosarcoma, although hepatocellular carcinoma (HCC) is by far the most common type, accounting for 70–80% of cases. Chronic hepatitis virus infections (HBV or HCV) and to a lesser extent, aflatoxin B‐contaminated dietary intake are the major risk factors in areas of high HCC incidence, while chronic alcohol consumption, obesity‐related fatty liver diseases, autoimmune hepatitis and genetically determined disorders, such as hemochromatosis, are important contributors, in particular in areas with lower incidence, such as Europe (Ferlay et al., 2010; Nordenstedt et al., 2012).

As chronic HBV or HCV infections currently account for more than 80% of HCC, the best strategy to prevent HCC is to eradicate viral infections. A vaccine against HBV has been developed in 1982 and is routinely used in many countries. Vaccination has reduced persistent HBV infections and led to marked reduction in hepatitis B‐related liver cancer (Romano et al., 2011). However, the millions of adults, who were infected before universal vaccination or that cannot afford it, are still at risk of developing HCC and the emergence of vaccine‐resistant hepatitis B surface antigen mutants is a serious concern. Hepatitis C‐associated cirrhosis is the leading cause of liver transplantation and despite encouraging results, no vaccine can currently protect against HCV infection (Feinstone et al., 2012). Fortunately, a treatment is available and effective in 50–80% of patients with persistent HCV infection, and better treatments are in sight (Dore, 2012). However, therapeutic options remain costly and patients with advanced liver pathologies will continue to require expensive disease management.

While the incidence of HCC in the areas endemic for viral infections has started to decline, its increase in developed countries is an important health concern. Different studies identify chronic alcohol ingestion, non‐alcoholic fatty liver disease (NAFLD), non‐alcoholic steatohepatitis (NASH), NASH‐related cryptogenic cirrhosis and even diabetes as relevant contributors for increased HCC incidence. NAFLD is the hepatic manifestation of the metabolic syndrome and the epidemiology of NAFLD mirrors the increased prevalence of obesity and diabetes. NAFLD has become the most common liver disorder in industrialized countries, affecting up to 30% of the adult population (Lazo and Clark, 2008). Given the increase in the prevalence of overweight, obesity and type 2 diabetes, the incidence of metabolic disease‐related HCC is expected to rise, further increasing the burden of liver diseases in years to come (Baffy et al., 2012). Understanding the cellular and molecular mechanisms leading to HCC, and most importantly those which are connected to systemic/metabolic influences has therefore become an urgent and imperative issue. Without this knowledge, developing efficient preventive, diagnostic and therapeutic counter‐measures is bound to fail.

Establishing animal models for HCC is essential for both basic and translational studies. Several rodent models have been used over the years to study HCC pathogenesis. The laboratory mouse is one of the best experimental systems, owing to the physiologic, molecular and genetic similarities to humans, its breeding capacity, short lifespan and the unlimited options offered by genetic engineering. More recently, an entirely sequenced genome, the possible use of defined genetic, but also environmental conditions, the establishment of non‐profit repositories and resource centers worldwide and the promising perspectives of national or transnational initiatives, for instance the International Knockout Mouse Consortium (IKMC) or the NCI's Mouse Models of Human Cancers Consortium (MMHC) have increased the range of potential benefits for using mouse models (GEMMs) in cancer research. In this review, we provide an overview of HCC modeling in the mouse, highlighting how simple and more sophisticated models have helped unraveling important aspects of disease development (Table 1). The utility of mouse models to establish and dissect HCC complexity and experimentally test clinically relevant hypotheses will be illustrated by selected examples and the challenges and future opportunities will be discussed. We also refer to excellent recent reviews, since by no means are we covering all aspects of HCC development.

Table 1.

Mouse models to study HCC development.

	Properties	Latency	Notes
Chemical
DEN/Phenobarbital DEN (single injection)	Genotoxic	5–10 m	When DEN is injected to adults promotion is needed, used in combination with genetic/dietary/environmental models
Aflatoxin	Genotoxic	>22 m	Often combined with genetic models
CDE diet, TAA, CCl4, peroxisome proliferators etc.	Associated with steatohepatitis, fibrosis, etc.	>12 m	Variability due to different experimental protocols, used for context‐specific modeling. Often combined with DEN or genetic models
GEMMs
HBV‐derived (sAg, HBx)	ER stress,focal necrosis, proliferation	12–24 m	Several lines/backgrounds with different penetrance/latency
HCV‐derived (Core)	Steatosis	12–24 m	Several lines/backgrounds with different penetrance/latency
Mdr2 knock‐out	Cholangitis	>12 m	Strong inflammatory component, strain dependent
Mosaic GEMMs
Somatic gene/molecules delivery: Virus‐based, RCAs/TVA, Hydrodynamic	Fast, cost‐effective	Few weeks	Can be combined with GEMMs, suitable for imaging
Implantation models: p53KO; mycTg hepatoblasts	Fast, cost‐effective	Few weeks	Suitable for imaging and large‐scale screens

More examples and details can be found in Newell et al., 2008; Li et al., 2011.

2. Conventional mouse models for liver cancer development

Inbred strains of mice differ greatly in their susceptibilities to spontaneous liver tumors (Drinkwater, 1988). This inter‐strain variation of wild‐type mice represents a valuable resource and is still used to identify liver tumor susceptibility genes (Peychal et al., 2009). Early on, efforts were made to develop better models, where tumors would develop more rapidly, with high penetrance and synchronous kinetics. The pathogenic alterations causing liver cancer were originally modeled through the use of chemicals or by classical transgenic approaches, where genes regarded as potentially relevant for liver cancer were introduced in the mouse genome. These early efforts were often assisted by serendipitous discoveries and unexpected findings.

2.1. Chemical carcinogenesis (DEN)

The liver is the primary target site for hundreds of chemicals including pesticides, food additives, pharmaceuticals and industrial intermediates. Identifying hepato‐carcinogenic compounds and understanding the cellular and molecular processes during carcinogenic transformation of hepatocytes is an ongoing challenge. The genotoxic drug diethylnitrosamine (DEN) is used since the 60s to induce HCCs in rodents (Rajewsky et al., 1966), and is the most widely used chemical to induce liver cancer in mice. DEN undergoes metabolic activation in hepatocytes by enzymes of the cytochrome P450 family and acts as a complete carcinogen, if injected into mice younger than 2 weeks, when hepatocytes are still actively proliferating. When administered later, tumor promotion is required and can be achieved by different ways, for instance by phenobarbital (PB), carbon tetrachloride, partial hepatectomy or, as recently demonstrated by high fat diet (HFD) feeding (Park et al., 2010). In addition to age and sex, the influence of the genetic background on DEN‐induced HCC has been documented early (Diwan et al., 1986) and experiments aiming at deciphering the underlying causes have led to important disease‐relevant discoveries (see also below). DEN is a DNA alkylating agent leading to the formation of mutagenic DNA adducts. In addition, DEN bioactivation by cytochrome P450 can generate reactive oxygen species (ROS) (Qi et al., 2008), which damage DNA, proteins and lipids and lead to hepatocyte death. Similar to what occurs in patients, HCC development in this model usually follows a slow multistep sequence, where cycles of necrosis and regeneration promote neoplastic transformation. The progression from early dysplastic lesions to fully malignant tumors is associated with an increased occurrence of genomic alterations (Farazi and DePinho, 2006; Thorgeirsson and Grisham, 2002). Mouse liver tumors induced by a single DEN injection frequently harbor activating Ha‐ras mutations (Chen et al., 1993), while tumors induced in a DEN/PB initiation–promotion protocol preferentially show activating β‐catenin mutations and genomic instability (Aleksic et al., 2011; Aydinlik et al., 2001). While mutation in the ras gene in human HCC are rather rare, activation of the ras pathway is more frequent and correlate with poor prognosis (Calvisi et al., 2006). On the other hand, approximately 30% of human HCCs display mutations activating the β‐catenin pathway (Taniguchi et al., 2002) and gene expression profiling indicated a good overlap between the DEN/PB‐induced mouse tumors and their human counterparts, further supporting the usefulness of this chemical carcinogenesis mouse model (Lee et al., 2004; Stahl et al., 2005).

2.2. Hepatitis virus transgenic mice: HBV/HCV

The technology for generating “transgenic” mouse lines carrying heritable manipulated DNA integrated into the mouse genome was established in the early 1980s and the first reports with transgenic mice carrying oncogenes few years later collectively substantiated the hypothesis that in the context of the whole organism, oncogene expression can lead to tumor development (Hanahan et al., 2007).

Mice are resistant to infection by human hepatitis B or C viruses. Transgenic mice were initially developed to model the chronic carrier state of HBV infection, where viral DNA sequences integrate in the host genome (Babinet et al., 1985; Chisari et al., 1985). Since these first reports in 1985, multiple transgenic mice have been generated carrying the full HBV genome and every HBV gene, e.g. encoding for surface envelope proteins (large, middle and small), X protein (HBx), core and pre‐core proteins, either under the control of the HBV promoter or liver‐specific host promoters, such as albumin. Likewise, the HCV polyprotein, and the core protein alone or in combination with envelope proteins have been expressed in transgenic mice (reviewed in Li et al., 2011; McGivern and Lemon, 2011). These studies have enabled the investigation of general pathogenetic mechanisms of liver injury and malignant transformation in vivo and the HBV or HCV transgenic models have provided definitive proof that viral genes can initiate and promote liver carcinogenesis.

Interestingly, only the large HBV envelope and the HBx protein were found carcinogenic and the HCV core protein is the major contributing factor to HCV‐related hepatocarcinogenesis (Chisari et al., 1989; Kim et al., 1991; Li et al., 2011; Moriya et al., 1998). Transgenic mice that overexpress the HBV large envelope polypeptide and accumulate HBV surface antigen (HBsAg) in hepatocytes display signs of chronic liver injury with variable degrees of necrosis, inflammation, endoplasmic reticulum stress, regenerative hyperplasia, transcriptional deregulation and aneuploidy, which inevitably leads to neoplasia (Dunsford et al., 1990). The incidence of HCC in this model correlates with the frequency, severity, and timing of liver cell injury, which itself corresponds to the intrahepatic concentration of HBsAg and is influenced by genetic background and sex. Thus, the inappropriate expression of a single structural viral gene is sufficient to cause malignant transformation. In contrast, preneoplastic cirrhosis and inflammation are not obvious in HBx transgenic mice (Kim et al., 1991). Most of the published data support the notion that HBx contributes to HBV‐carcinogenesis by promoting a pre‐neoplastic proliferative response that facilitates secondary transforming events. HBx can transactivate both viral and cellular genes in the nucleus and stimulates signal transduction pathways in the cytoplasm. HBx was shown to increase the activity of AP‐1, NFκ‐B, c‐myc, ras, Wnt/β‐catenin, to affect cell proliferation, survival, DNA stability and repair mechanisms and to sensitize hepatocytes to other HCC inducers, such as chemical carcinogens (Li et al., 2011). On the other hand hepatic steatosis, a characteristic feature of chronic HCV infection in human, and oxidative stress are proposed to play a pivotal role in HCC development in mouse models for chronic HCV infection. The HCV core protein promotes steatosis and oxidative stress by interfering with lipid and triglycerides turnover and affecting the activity of transcription factors controlling metabolic gene expression, such as SREBP‐1c, RXR and PPAR (Khan et al., 2010; McGivern and Lemon, 2011).

Finally, efforts have been dedicated to understand the mutagenic events linked among others, to HBV integration (Zhang, 2012) and more importantly, to establish mouse models permissive to HBV and HCV infection and replication. Progress in mouse genetic manipulation and the valuable information gathered from studies performed in vitro and in the clinic have facilitated the development of chimeric models where immune‐deficient mice are grafted with human hepatocytes (reviewed in Chayama et al., 2011; MacArthur et al., 2012). A more recent highlight, stemming from the progress in gene manipulation, is the generation of a humanized mouse model that can be infected by HCV (Dorner et al., 2011).

2.3. Hepatic onco‐mice

Besides the efforts to emulate viral hepatitis‐associated events, several “oncomice” aiming at modeling hepatocarcinogenesis were generated by directing the expression of known oncogenes to the liver. The expression of SV40 T‐antigens, mutant H‐ras, or c‐myc under the control of albumin, alpha‐1‐antitrypsin or antithrombin III promoter fragments, led to pleiotropic effects in the liver including neoplasia (Dubois et al., 1991; Sandgren et al., 1989; Sepulveda et al., 1989). SV40 T transgenic mice were the first mouse model where hepatocarcinogenesis was observed after indirectly targeting tumor suppressors, as expression of SV40 large T simulates loss of p53 and Rb. This was also one of the earliest mouse models, where liver tumor metastasis to distant organs was observed (Dubois et al., 1991). Several other models were generated using similar strategies to achieve ectopic hepatic expression of oncogenes, cell cycle proteins, growth factors and in general genes predicted from human datasets or in vitro culture experiments to play important roles in liver cancer (Cadoret et al., 2001; Conner et al., 2000; Jhappan et al., 1990; Wang et al., 2001). Although conventional transgenic mouse models relying on random transgene integration suffer from several limitations, such as early onset and broad pattern of transgene expression, which may not recapitulate the pathogenesis of human HCC, they have been widely used to understand the disease and more generally liver physiology. For example, such models have proven useful to assess potential synergistic effects between oncogenes and growth factors through the generation of double transgenic mice (Murakami et al., 1993). Most importantly, bi‐transgenic models based on the tetracycline regulatory system have permitted a temporal control of transgene expression and allowed to assess, in vivo the requirement for the initial oncogenic event for tumor maintenance and the therapeutic relevance of targeting crucial oncogenes in HCC (Manickan et al., 2001; Shachaf et al., 2004; Tward et al., 2007; Wang et al., 2001). Finally, several elegant studies comparing the molecular profiles of human and mouse HCC revealed extensive overlaps at the gene expression level. Mouse tumor signatures could even be attributed to specific subclasses of human HCC and predicted patient clinical outcome (Coulouarn et al., 2008; Ivanovska et al., 2011; Lee et al., 2004). Such molecular profiling studies, which are being extended to proteomic and microRNA analyses highlight the potential value of investing in animal models of HCC (Liu et al., 2010; Pineau et al., 2010; Ritorto and Borlak, 2011).

3. Gene targeting technology opens new avenues and better GEMMs

In 1992 gene targeting technology converged with whole organism cancer research and led to the generation of more sophisticated mouse models for cancer research. Technological advances that allowed precise spatiotemporal control of gene expression and mutagenesis either in the germline or in a somatic context resulted in widespread promises toward more accurate modeling of human disease and functional exploration of carcinogenic events.

Efforts to systematically determine gene function and establish disease models culminated with transnational initiatives, such as the International Knockout Mouse Consortium (IKMC) and the International Mouse Phenotyping Consortium (IMPC). In 2007, the IKMC started the ambitious task to generate mutations in virtually every putative murine protein‐coding gene in a concerted worldwide action (Bradley et al., 2012). Within the IMPC, many centers worldwide will contribute to the exhaustive, large‐scale and standardized phenotyping of mouse mutants. This effort, ultimately generating and analyzing thousands mutants over the next decade will establish a functional encyclopedia of the mammalian genome and most importantly, will provide a centralized portal for free, unrestricted distribution of mouse resources and datasets to the scientific community (Mallon et al., 2012). Other local or transnational initiatives are progressing in parallel with further technological advancements, such as targeting non coding sequences, creating “driver” lines or in vivo RNA interference libraries and repositories and most importantly efforts toward the integration of industry in GEMM‐based preclinical research (Marks, 2009; Premsrirut et al., 2011; Prosser et al., 2011; Smedley et al., 2011).

The original gene targeting strategies entailed the disruption (knock‐out) or the modification (knock‐in) of an allele in embryonic stem (ES) cells. Such germline mutations are constitutive, are present in all cells of the mouse and are in principle more useful in studying familial cancer diseases, such as Li‐Fraumeni syndrome, exemplified by the p53‐deficient or missense germline mutants (Kenzelmann Broz and Attardi, 2010), than cancers arising from somatic mutations. Nevertheless, these early models represented an important method to functionally characterize individual genes and some of these mutants have unraveled unsuspected players in liver carcinogenesis.

Our aim here is not to review all mouse models for liver cancer that have been produced using technologies derived from gene targeting, but rather to provide examples illustrating how sophisticated mouse genetics can enable a more thorough understanding of interactions among genes, cells, tissues and environmental effectors that contribute to cancer susceptibility, disease progression and potential response to therapeutic interventions.

4. Metabolic function of the liver impacts on carcinogenesis: Mdr2 and Aox mouse models

Mdr2, encoded by abcb4, is a phospholipid flippase that promotes biliary secretion of phospholipids, which form mixed micelles with bile acids and cholesterol, thereby protecting the biliary epithelium from the damaging effects of bile acids. In humans, biallelic or monoallelic ABCB4 defects are associated with several diseases, such as type 3 progressive familiar intrahepatic cholestasis, intrahepatic cholestasis of pregnancy, drug‐induced liver injury, biliary fibrosis, or cirrhosis (Davit‐Spraul et al., 2010). Abcb4/Mdr2 knock‐outs, initially generated to unravel the function of Mdr2, display a complete absence of phospholipids from bile and a striking liver disease (Smit et al., 1993). Mdr2‐deficient mice develop cholestasis with features closely resembling human sclerosing cholangitis, hepatocellular injury, low grade hepatitis, periportal fibrosis and with a background‐dependent latency of several months, preneoplastic lesions slowly progressing to metastatic liver cancer (Mauad et al., 1994). Mdr2 knock‐out mice provide a valuable model to study inflammation‐associated hepatic carcinogenesis and were exploited in a pioneering study to genetically demonstrate the relevance of inflammation for liver carcinogenesis (Pikarsky et al., 2004). The cellular and molecular players involved and the relevance of this seminal discovery will be discussed in section 6.1.

Mdr2 knock‐out mice were extensively studied over the years as they provided a model for studying cholestasis, biliary fibrosis and carcinogenesis. Gene expression profiling of Mdr2‐deficient tumors indicated a good correlation with a subset of murine HCC models, as well as with human HCC datasets associated with better survival (Katzenellenbogen et al., 2007). More interestingly, molecular and metabolic profiling of diseased livers at the precancerous stage revealed a striking deregulation of pathways related to oxidative stress and proliferation, but also lipid metabolism (Katzenellenbogen et al., 2006; Moustafa et al., 2012). The age‐dependent progressive deregulation of genes controlling lipid metabolism likely correlates with the frequent steatotic dysplastic nodules in the livers of Mdr2 knock‐out mice. Furthermore, alterations in bile acids, lipid and triglyceride metabolism were causally linked to the cholestatic phenotype (Moustafa et al., 2012). Whether these deregulations also contribute to carcinogenesis in this model remain to be addressed.

Another model linking the hepatic metabolic function to tumor formation is the Aox‐deficient mouse model. Initially generated to model Zellweger syndrome, a peroxisomal genetic disorder, homozygous mutant mice lacking fatty acyl‐CoA oxidase (Aox), display severe steatosis, resulting in scattered cell death, steatohepatitis and hepatic adenomas and carcinomas (Fan et al., 1996). At the molecular level, Aox‐deficient livers accumulate unmetabolized very‐long chain fatty acyl‐CoAs and hydrogen peroxide, in particular during the early steatohepatitis phase, leading to sustained activation of peroxisome proliferator‐activated receptors (PPAR) and DNA damage. PPAR ligands or peroxisome proliferators are a broad class of compounds used in industry, pharmacy and agriculture that include ester plasticizers, herbicides and lipid‐lowering drugs, such as fibrates. Endogenous ligands for the PPARs include free fatty acids, eicosanoids and others. Prolonged exposure to peroxisome proliferators leads to HCC in rodents, while their carcinogenic role in human is not established. Nevertheless, the Aox‐deficient mouse provides a genetic model to study the potential contribution of peroxisome receptors in the pathogenesis of liver cancer, in particular when compared to animals treated with exogenous peroxisome proliferators or with other animal models of HCC. Gene and protein expression profiles in Aox‐deficient liver tumors were documented and revealed distinguishing features of this model. Comparative genomic profiling indicated that while gene expression patterns in tumors from Aox knock‐out or fibrate–treated mice defined a distinctive cluster with limited similarity to human HCCs (Lee et al., 2004), several genes involved in lipid, carbohydrate and amino acid metabolism, steroid hormone synthesis and stress response were deregulated (Chu et al., 2004; Meyer et al., 2003), indicating that hepato‐toxic byproducts of lipid metabolism can induce oxidative stress and DNA damage and contribute to hepatocarcinogenesis. Interestingly, hepatic steatosis is a prominent feature of HCV‐associated liver pathologies and PPARα was found essential to hepatic steatosis and HCC pathogenesis in an HCV transgenic mouse model (Tanaka et al., 2008). The relevance of this observation to human HCV‐related hepatocarcinogenesis remain to be established, as several reports, including using humanized PPARα mouse models have documented important inter‐species differences in PPARα biology (Cheung et al., 2004). However, studies using heterozygous animals for another peroxisome proliferator‐activated receptor, PPARγ indicate that PPARγ can function as a tumor‐suppressor in hepatocarcinogenesis (Yu et al., 2010), and correlation studies in patients suggest that PPARγ agonists, such as the anti‐diabetic thiazolidinediones might have a beneficial effect in HCC (reviewed in Wu et al., 2012).

In summary, the Mdr2 and Aox knock‐out mouse models, which constitute “atypical” HCC models in the sense that the original gene modification is not strictly HCC‐related, provide valuable insights into several aspects of liver cancer pathogenesis and demonstrate how perturbations in the liver metabolic function can trigger or promote cancer development. Given the emerging role of fatty acids, bile acids and their receptors in hepatocyte survival, proliferation and carcinogenesis, modulating bile/fatty acid toxicity or intervening on critical aspects of liver metabolism could provide therapeutic means to reduce the incidence of liver cancer in high‐risk populations.

5. Modeling a proper tumor microenvironment using mouse models

Despite the valuable information derived from the use of the previously described “early” HCC models, these models still harbor significant differences with human HCC pathogenesis. One important distinction is that the genetic modifications used to mimic the genetic context of the tumor cell or to assess the interaction of genes with tumor‐inducing agents within the so called “tumor initiating cell”, were very often already present in the mouse germline or at least in a large proportion of liver cells. This represents a non‐physiological situation, in which the pre‐neoplastic cell does not evolve in the context of a normal or at least genetically different environment. In addition, some of these “whole‐body mutants” displayed non‐hepatic pathologies that precluded their utilization for liver research. Several efforts have attempted to circumvent this shortcoming and better recapitulate the tumor cell interactions with the surrounding environment and some will be illustrated below. However, it should be noted that this limitation has paradoxically facilitated the identification of some initiation‐related molecular events, pathways and/or molecular signatures by providing abundant homogeneous primary material at times where starting experimental material was the limiting factor.

5.1. Tissue‐specific GEMMs

Early attempts to limit the gene targeted population has focused on targeting specific liver cell types within the parenchymal or non‐parenchymal fractions using strategies largely derived from cre/lox conditional targeting. Sophisticated multi‐allelic mouse strains were developed, where transgene(s) expression can be achieved in a space and time‐restricted fashion. Such “conditional” approaches allowed circumventing the early lethality of conventional knock‐out mice that precluded the assessment of gene function in adult liver. For instance, c‐Jun, a component of transcription factor AP‐1 and proto‐oncogene, was suspected to be a critical mediator of liver homeostasis, as c‐Jun‐deficient mice die in utero with massive fetal liver cell death (Hilberg et al., 1993; Johnson et al., 1993). Mice specifically lacking c‐jun in the liver were viable and subsequent studies demonstrated that c‐Jun is a critical regulator of hepatocyte proliferation and survival during liver regeneration, inflammation and cancer (Behrens et al., 2002; Eferl et al., 2003; Hasselblatt et al., 2007; Machida et al., 2011; Min et al., 2012). Furthermore, the essential role of c‐Jun in liver cancer development appeared to be restricted to the tumor initiation phase, as genetic inactivation of c‐jun during tumor promotion did not affect liver cancer development (Eferl et al., 2003; Min et al., 2012).

Further improvement in modeling HCC in mice included models where the genetic modification was directed to only a small fraction of liver cells, modeling the stochastic nature of tumorigenesis. One interesting example of how such strategies have been particularly useful to understand liver cancer development relates to the numerous efforts to genetically interfere with the wnt/β‐catenin pathway in the whole liver or in subsets of the hepatocyte population. Constitutive loss of Apc, Axin1 or gain of β‐catenin function results in early lethality, while β‐catenin mutation or overexpression does not give rise to liver tumors, unless combined with other oncogenic events (Nejak‐Bowen and Monga, 2011; Thompson and Monga, 2007). In contrast, conditional postnatal liver‐specific deletion of Apc results in a β‐catenin‐dependent increase in hepatocyte proliferation, hepatomegaly and mortality within days, possibly due to perturbed protein and ammonia metabolism (Colnot et al., 2004; Reed et al., 2008). Consistent with genetic profiling data associating β‐CATENIN and AXIN1 mutations to different human HCC subsets, deletion of Axin1 in the liver of adult mice resulted in late onset HCC, but showed limited phenotypic and molecular overlap with Apc deletion (Feng et al., 2012). Strikingly, mosaic Apc deletion in few hepatocytes using limited Adenovirus‐mediated somatic Cre recombinase delivery was sufficient to cause HCC, demonstrating that activating mutations in the wnt/β‐catenin pathway can initiate hepatocarcinogenesis, when limited to a small number of cells (Colnot et al., 2004).

5.2. Mosaic GEMMs

Several approaches have been explored to either generate mosaic models targeting a small subset of cells or to somatically deliver coding sequences and/or interfering elements to the liver. Somatic delivery was also explored as a complementary method to genetic manipulation, to rapidly generate mouse mutants for cancer gene discovery or to functionally assess candidate pathways within a pre‐existing model. The mouse liver is well‐suited for these strategies, since it is easily targeted with viruses and small‐molecules using vascular delivery systems (Pathak et al., 2008). One elegant technique makes use of the avian replication‐competent avian sarcoma‐leukosis virus long terminal repeat with splice acceptor/tumor virus A (RCAS/TVA) system. Transgenic animals expressing the chicken receptor in a tissue‐specific manner can be infected with a RCAS virus bearing genes of interest. This method has been successfully applied to HCC modeling (Lewis et al., 2005) and improvements based on combinations with the Cre/loxP or the tetracycline systems have been described to allow a more versatile and a better control of somatic delivery (von Werder et al., 2012).

5.3. Non‐GEMMs

In parallel, complementing methods not relying on genetic modification of the host have been developed. Several employ viral vectors, for instance lentiviruses, Adenoviruses, and Adeno‐associated viruses (AAV) that have been further engineered to allow efficient transduction, larger sequences inclusion and time and cell‐specific expression. Recent advances in AAV vector technology allow 90–100% transduction of hepatocytes and long‐term gene expression without toxicity following a single systemic administration of recombinant virus. A particularly elegant study has documented how this system can be exploited for systemic delivery of therapeutic miRNAs using a Myc‐based HCC mouse model (Kota et al., 2009). We have recently used an Adeno‐virus‐based transient delivery method to address the potential therapeutic value of targeting the histone deacetylase Sirt6 or the anti‐apototic regulator Survivin, specifically during the initiation phase of DEN‐induced HCC. We have shown that transiently interfering with Sirt6 or Survivin affected tumor burden several weeks later. As we observed a correlation between the expression of c‐Jun, c‐Fos, Sirt6 and Survivin in a fraction of human pre‐neoplastic HCC lesions, our findings indicate that Sirt6 or Survivin could be therapeutically relevant in the early phases of the disease (Min et al., 2012).

In mice, rapid high volume‐infusions of naked plasmid DNA, termed “hydrodynamic delivery” is a very efficient method for in vivo gene transfer (Zhang et al., 1999). While the mechanism of DNA uptake is poorly understood, the liver accounts up to 95% of the expressed transgenic DNA (Suda and Liu, 2007). Further improvement includes the addition of liver‐specific and/or regulatable promoter sequences, selection markers, imaging cassettes or the use of transposons to enables long‐term gene expression (Bell et al., 2007; Wangensteen et al., 2008). One recent application of this system was used to demonstrate that, while somatic Akt over‐expression led to HCC development in mice, additional activation of the Notch pathway converted normal hepatocytes into rapidly progressing, lethal intrahepatic cholangiocarcinomas, a primary liver tumor with increased incidence and poor prognosis (Fan et al., 2012). The oncogenic cooperation between the Ras and Akt pathways was mechanistically dissected using a similar strategy and highlighted the functional significance of mammalian target of rapamycin complex (mTORC1) activation in this context (Ho et al., 2012). Such somatic–delivery approaches are particularly useful, when applied to parallel series of GEMMs to interrogate the therapeutic value of a gene or cell‐type manipulation. For example, Kang et al. delivered oncogenic Ras‐encoding transposons to the livers of GEMMs lacking different immune cell lineages and identified a key role for the adaptive immune response in premalignant hepatocytes clearance (Kang et al., 2012).

A recent improvement in addressing the interaction of tumor cells with their environment and, more importantly, to rapidly identify tumor‐associated events using screen‐like approaches are combinations of mouse to mouse transplantation with other genetic tools, like gene targeting and RNAi interference. One elegant and powerful system has been developed over the years in S. Lowe's laboratory and was successfully used to identify novel HCC tumor suppressors using in vivo forward genetic screens (Zender et al., 2008). The system is based on multiple ex vivo genetic manipulation of mouse embryonic liver progenitor cells, followed by seeding into the livers of recipient mice. Orthotopic implantation of progenitor cells harboring defined genotypes, such as combinations of p53 inactivation with Myc, activated Akt or oncogenic Ras expression led to rapid tumor formation in recipient mice that can be monitored by non invasive imaging (Zender et al., 2006). The advantage of this system is that it allows the rapid generation of tumor models with complex genotypes without the cost and effort associated to intercrossing genetically modified strains. This is particularly attractive when considering preclinical drug screening. In addition, further studies of tumor/host interactions are possible, since neoplastic cells develop in a recipient liver environment that can be either normal or independently manipulated, to reflect aspects of human tumor development.

6. Tumor‐host interactions and organ crosstalks

Like most carcinomas, HCCs are composed of transformed hepatocytes and a variety of other cell types and extracellular matrix components that form the tumor microenvironment. Besides identifying critical cell types and molecules in the tumor microenvironment, it is important to understand how these cells and molecules contribute to the different stages of HCC development to design efficient diagnostic and therapeutic tools. Because they are amenable to complex experimental manipulations, GEMMs have increased our understanding how transformed hepatocytes arise in the context of the whole liver. More recently, GEMMs demonstrated that tumor–host interactions extend well beyond the liver microenvironment in HCC. These now include other organs, such as the gonads, the adipose tissue and the gut, as well as living microorganisms, such as the bacterial flora in the intestine.

6.1. Liver inflammation: lessons from GEMMs and cell‐specific crosstalk

A role for inflammation in tumor development is now generally accepted and an inflammatory environment appears to be an essential component of all tumors, including those, where a direct causal relationship with inflammation is not yet proven (DiDonato et al., 2012; Mantovani et al., 2008). The role of inflammation in HCCs is well documented, although the molecular basis and the contribution of distinct cell types within the liver is still being investigated, in particular with the help of GEMMs (Farazi and DePinho, 2006; He and Karin, 2011; Luedde and Schwabe, 2011). At first glance several GEMMs have revealed somewhat counter‐intuitive results, since it was found that genes, known to have tumor‐promoting activities, turned out to be tumor‐suppressive (Ben‐Neriah and Karin, 2011; Feng, 2011).

6.1.1. NF‐κB and HCC development

One of the best known pathways playing an essential role in inflammation, the NF‐κB pathway can have a dual role in HCC development, depending on the liver cell‐type, but also on the pathological setting underlying HCC development (Ben‐Neriah and Karin, 2011; DiDonato et al., 2012). Early studies employing GEMMs showed that NF‐κB has a tumor‐promoting role, such as in the aforementioned Mdr2 knock‐out model (Pikarsky et al., 2004), as well as when HCC is triggered by sustained lymphotoxin signaling (Haybaeck et al., 2009), or when IKKβ was deleted from all liver cells using the Mx‐cre transgene in the DEN chemical carcinogenesis model (Maeda et al., 2005). However, when IKKβ inactivation was restricted to hepatocytes, a dramatic increase in HCC development was reported in the DEN model (Maeda et al., 2005), while early lethality and aggravated liver disease was observed in an Mdr2‐deficient background (Ehlken et al., 2011). Furthermore, hepatocyte‐specific deletion of another NFκB activator IKKγ, led to spontaneous liver damage and HCCs (Luedde et al., 2007). These findings indicate that NF‐κB can act as a tumor‐suppressor in hepatocytes and it is proposed that NF‐κB suppresses liver cancer development by promoting hepatocyte survival during the injury/tumor‐initiation phase. When NF‐κB activation is impaired in hepatocytes, cell death and subsequent compensatory proliferation increase, leading to enhanced carcinogenesis. A critical pro‐survival mechanism controlled by NF‐κB in hepatocytes involves the oxidative stress response, as demonstrated by experiments utilizing anti‐oxidants in the aforementioned mouse models (Luedde et al., 2007; Maeda et al., 2005). Additionally, defective NF‐κB signaling in hepatocytes leads to sustained JNK activation, which is proposed to increase hepatocyte proliferation by acting on the cell cycle inhibitor p21 and other targets (Hui et al., 2008; Sakurai et al., 2006; Seki et al., 2012). Paradoxically, in models where liver cancer develops in a context of chronic low grade inflammation and limited injury, such as the Mdr2 knock‐out or the lymphotoxin transgenic mice, NF‐κB in hepatocytes appears to be rather a tumor‐promoter and essential for sustaining the production of inflammatory cytokines, which induce hepatocyte proliferation, in particular during the tumor progression phase (Haybaeck et al., 2009; He and Karin, 2011; Pikarsky et al., 2004).

Liver injury and hepatocyte cell death is an important source for inflammatory signals, such as IL‐1α, which signals to the surrounding microenvironment. The liver's non‐parenchymal cells, in particular resident macrophages or Kupffer cells, respond to these signals by producing cytokines, chemokines and growth factors that favor hepatocyte proliferation and tissue repair. GEMMs have demonstrated that NF‐κB is essential in Kupffer cells to control the expression of IL‐6 and TNFα, and NF‐κB is therefore a paracrine tumor promotor in Kupffer cells, by enhancing compensatory proliferation in hepatocytes, possibly through the activation of STAT3 (He and Karin, 2011; Maeda et al., 2005; Sakurai et al., 2008). Of note, liver injury also results in activation of fibrogenic myofibroblasts, which play a central role in hepatic fibrosis and produce hepato‐mitogenic factors such as HGF (Brenner, 2009). As NF‐κB has been shown to modulate the survival of hepatic myofibroblasts in vitro (Watson et al., 2008), it is likely that studies using GEMMs to dissect the role of NF‐κB in myofibroblasts during HCC will unravel further unexpected results.

6.1.2. MAP kinases and HCC development

Mitogen‐activated kinases respond to paracrine signals and modulate the expression of essential survival and proliferation genes, as well as cytokines, chemokines and growth factors. Members of the MAPKs; JNK, p38 and ERK families and their targets, for instance AP‐1 (Jun/Fos), are likely playing complex, inter‐connected cell‐ and stage‐dependent roles in HCC development. For instance, increased HCC development was observed in liver‐ and hepatocyte‐specific p38α knock‐outs, likely due to enhanced JNK activation (Hui et al., 2007; Sakurai et al., 2008). On the other hand, studies using complete JNK1 and JNK2 knock‐out mice or compound JNK1/2 point to a potentially more complex role of JNK in HCC development (Seki et al., 2012). DEN‐induced carcinogenesis is comparable between wild‐type and JNK2−/− mice while JNK1‐deficient and mice with liver‐specific deletion of JNK1 in a JNK2−/− background develop less HCCs. These experiments indicate a tumor‐promoting function for JNK1, but not JNK2, which has been confirmed using human HCC cell lines and xenografts (Das et al., 2011; Hui et al., 2008; Sakurai et al., 2006). Interestingly, hepatocyte‐specific deletion of JNK1 in a JNK2−/− background increased tumor burden without affecting incidence and multiplicity, suggesting that JNK can function in transformed hepatocytes to limit tumorigenesis (Das et al., 2011). This study also indicates that JNK activity in non‐parenchymal liver cells is likely an important modulator of hepatic inflammation that certainly merits further investigation, including in paradigms of liver injury where JNK activity has been observed, for instance acute hepatitis (Seki et al., 2012). Studies where gene manipulation is achieved in a cell‐ or tissue‐specific manner will likely provide further examples and shed additional light on the importance of crosstalks between the different liver cell types at different stages of liver cancer development. Preliminary observations indicate that the AP‐1 member c‐Fos might be added to the list of molecules playing opposing, context‐ and cell type‐specific roles in liver disease (Wagner lab unpublished). In conclusion, these few examples highlight the essential role of GEMMs to mechanistically and molecularly dissect the function of genes in different liver cells with the goal to provide the basis for novel combined strategies to combat this detrimental disease.

6.2. Understanding the gender bias in HCC

One intriguing universal epidemiologic characteristic of HCC is the prominent male dominance, with a male to female ratio between 2:1 and over 4:1 (Nordenstedt et al., 2012). This difference applies regardless of the etiologic factor and cannot be simply explained by gender‐specific differences in exposure to risk factors, e.g. viral hepatitis or alcohol abuse. Furthermore, the gender bias is not limited to HCC incidence, but extends to prognosis and survival and the advantages for female patients tend to be attenuated after menopause, supporting an important contribution of sex hormones to HCC pathogenesis.

The liver is characterized by sexually dimorphic gene expression translating into sex‐specific differences in lipid, drug, steroid hormone, and xenobiotic metabolism, potentially resulting in distinct responses of males and females to environmental challenges. This characteristic is conserved in rodents and most mouse models for HCC harbor a clear male predominance. Furthermore, early studies using chemical carcinogens indicated that gonadectomy can attenuate gender‐related differences in HCC development. Therefore, mouse models provided a valuable tool to investigate the liver‐intrinsic and ‐extrinsic factors causing gender disparity in hepatocarcinogenesis, with the aim of unraveling potentially targetable mechanisms that could be used in patients.

The overall conclusion that can be drawn from the experimental and epidemiological studies is that both androgen and estrogen sex steroids and their respective receptors/downstream pathways contribute to the gender disparity of HCC, with specific effects in each gender. For example, castration, administration of estrogen or anti‐androgen agents as well as genetic ablation of androgen receptor (AR) in hepatocytes, limits HCC development in male rodents (Kalra et al., 2008; Ma et al., 2008; Yeh and Chen, 2010). Conversely, ovariectomy, testosterone supplementation or genetic inactivation of estrogen receptor alpha (ERα) increase HCC development in female mice (Kalra et al., 2008; Naugler et al., 2007; Poole and Drinkwater, 1996; Yeh and Chen, 2010). Several studies using sophisticated GEMMs have attempted to decipher how and in which cell types sex hormones modulate liver cancer development. In hepatocytes, ERα and AR appear to co‐regulate multiple pathways involved in carcinogenesis, including xenobiotic metabolism, carcinogen detoxification, oxidative stress, DNA damage and repair, cell death and proliferation (Li et al., 2012; Ma et al., 2008). Recently, Li et al. reported a central role for the winged helix transcription factors Foxa1/Foxa2 in recruiting AR and ERα to multiple target genes in transformed hepatocytes (Li et al., 2012). Strikingly, while Foxa1/2 hepatocyte‐deficient mice maintained an almost normal dimorphic gene expression profile in the absence of carcinogenic treatment, dimorphic HCC development and gene expression profile was completely reversed upon chemical carcinogenesis. This comprehensive genome‐wide transcriptional binding and profiling analysis was supported by analysis of human samples, where certain single nucleotide polymorphisms at FOXA2 binding sites correlated with increased HCC incidence in woman (Li et al., 2012).

Animal and clinical studies also indicate that the situation is more complex. While mice lacking hepatic AR, displayed less HCCs, male mutants still developed more HCCs than their female counterparts and the tumors that developed were more undifferentiated and exhibited increased lung metastasis potential, resulting in increased mortality (Ma et al., 2012). In addition, tumor burden in Foxa1/2‐deficient females exceeded that of control male mice, indicating that estrogens can also exert a tumor promoting effect, in line with association‐based population studies indicating that both estrogens (oral contraceptives) and androgens (anabolic steroids) can induce liver adenomas and ultimately HCC in otherwise healthy individuals (Kalra et al., 2008; Li et al., 2012).

An important aspect that should not be overlooked is the impact of sexual dimorphism beyond the hepatocyte and/or the liver. Males and females show differences in the prevalence of many diseases that have important inflammatory components and female mammals have different immune reactions than males (Chrousos, 2010). The decline in ovarian function with menopause is associated with increased production of ciculating pro‐inflammatory cytokines, including TNFa, IL‐6 and IL‐1β (Pfeilschifter et al., 2002). One study using animal models elegantly demonstrated that estrogen can also protect hepatocytes from malignant transformation by down‐regulating IL6 secretion by Kupffer cells (Naugler et al., 2007). Because these models allow such analyses of complex, multi‐organs interactions in vivo, we anticipate that GEMMs will unravel additional cellular and molecular players within and most importantly, beyond the liver and will help understanding how sexual dimorphism contributes to HCC development.

Several clinical trials have tested the efficacy of hormonal treatment in HCC with rather conflicting results, partly due to the small size and heterogeneous nature of the patient population and the limited information regarding sex hormone receptor expression in these studies (Kalra et al., 2008). While data from patients and mouse models point more toward a context‐dependent involvement of gender‐associated events in HCC, a better understanding of the contribution of sexual dimorphism to HCC biology at a local and systemic level is needed to design appropriate clinical trials.

6.3. Systemic metabolism: obesity and diabetes

Obesity characterized by an abnormal increase in body mass index causes a number of pathological disorders, including metabolic syndrome, type 2 diabetes, NAFLD and NASH. Obesity was recognized as a major risk factor for several cancers of which HCC had the highest relative risk increase. Similar epidemiological data show an increased risk of HCC among patients with diabetes, particularly type 2 (Caldwell et al., 2004; Nordenstedt et al., 2012).

GEMMs provide a powerful tool to mechanistically analyze the connections between liver and/or systemic metabolism and HCC development. Similar to obese diabetic patients, genetically obese leptin‐deficient ob/ob mice have an increased incidence of HCC and long term exposure of wild‐type mice to high fat diet (HFD) was found to increase the rate of spontaneous HCC in a strain‐ and gender‐dependent fashion (Hill‐Baskin et al., 2009). Furthermore, chronic HFD dramatically increased the incidence of chemically‐induced HCC in rodents, regardless whether it was administered before or after the carcinogen (Herranz et al., 2010; Park et al., 2010). Moreover, the Mdr2 and Aox GEMMs demonstrated that changes in the metabolic function of the liver, in particular lipid handling were likely causally involved in HCC development.

HCC associated with steatosis is thought to result from a series of steps that start with lipid accumulation in the liver, either from dietary lipids or from increased lipolysis. Subsequently, toxic byproducts of lipid metabolism induce oxidative stress, DNA damage and hepatocytes death, triggering compensatory proliferation and local inflammation. Diets that are rich in saturated fats or natural genetic variants in lipid metabolism genes further exacerbate liver injury and this continuous process ultimately leads to HCC. Interestingly, one major lipid peroxidation product 4‐HNE, was shown to form DNA adducts at codon 249 of p53, a unique mutational hotspot in HCC (Hu et al., 2002). In addition, increased lipids in hepatocytes may interfere with cellular signaling and transcriptional pathways, such as those controlled by the PPAR family or provide an energetic advantage to tumor cells that extends beyond the initiation stage. Strikingly, increased expression of genes implicated in lipogenesis correlates with poor prognosis in HCC (Baffy et al., 2012; Yamashita et al., 2009).

Obesity is characterized by a low‐grade, chronic inflammatory response that has been generally implicated in increased cancer risk. The pathogenic role of obesity‐associated inflammation in liver cancer development has been mainly established using GEMMs and has led to the identification of IKK and JNK signaling pathways as pivotal mediators of obesity‐mediated inflammation acting in hepatocytes, adipocytes, but also liver and adipose tissue macrophages (Sun and Karin, 2012). The role of adipose‐derived cytokines TNF and IL‐6 in the development of HCC has recently been demonstrated in an experimental mouse model, where either dietary or genetic obesity promoted malignant liver tumor growth induced by DEN (Park et al., 2010). These cytokines also lie at the core of the association between obesity and insulin resistance, a key contributor in the development of obesity‐related HCC. Loss of TNF‐α function improves insulin sensitivity in obese mice, possibly through interaction with insulin receptor signaling (Hotamisligil et al., 1994; Uysal et al., 1997). Activation of the IGF‐1/IGF‐1 Receptor and a PI3K/Akt/mTOR signaling are also critically associated with obesity, insulin resistance and HCC while metformin, an AMPK/mTOR modulator has been shown to decrease the risk and improve HCC prognosis in diabetic patients (Baffy et al., 2012; Shimizu et al., 2013; Tovar et al., 2010). However, the exact mechanisms underlying the activation of these different diabetes‐associated pathways, their stage‐specificity and their connection with metabolic and inflammatory signals need further characterization.

Increased free fatty acid release from adipocytes and deregulated adipokine balance are also hallmarks of obesity. The involvement of adipose tissue‐derived hormones, such as leptin, adiponectin, resistin, chemerin and sfrp5m is increasingly recognized in the pathophysiology of HCC (Sun and Karin, 2012). For example, adiponectin levels are inversely associated with poor HCC prognosis in obese patients and adiponectin‐deficient animals developed early‐stage NASH with increased fibrosis on HFD, and accelerated tumorigenesis on a carcinogenic diet (Asano et al., 2009; Kamada et al., 2007; Saxena et al., 2010). Adiponectin is anti‐inflammatory and anti‐diabetic, and modulates several pathways relevant to hepatocarcinogenesis, such as JNK, IKK and AMPK/mTOR (Haugen and Drevon, 2007; Saxena et al., 2010). Improved GEMMs where the expression of specific adipokines/adipokine receptors can be modulated in a time and cell‐specific manner will certainly provide valuable information on the exact molecular and cellular contribution of adipose‐derived signaling molecules to liver cancer pathogenesis that could be used in a therapeutic setting.

6.4. Dissecting signals from the gut and gut‐resident micro‐organisms

Bile acids are produced in the liver and are unique to vertebrates, providing the host with the ability to digest and utilize a variety of dietary substrates. Bile acid levels are tightly regulated by several mechanisms to ensure the coordination of their production and release with dietary inputs. One such mechanism is the bile acid‐induced secretion by cells of the small intestine of FGF19 (or its mouse ortholog Fgf15), a hormone‐like molecule that acts on the hepatocytes to decrease bile acids synthesis (Inagaki et al., 2005). Several reports indicate that FGF15/19 can additionally regulate energy homeostasis, thus physiologically connecting the gut with the liver, adipose tissues and possibly brain and muscle. Therefore, FGF19 mimetics are being considered for diabetes treatment (Potthoff et al., 2012). However, transgenic mice expressing FGF19 in skeletal muscle developed liver dysplasia and HCCs and FGF19 administration stimulated hepatocyte proliferation in wild‐type mice (Nicholes et al., 2002). Furthermore, while FGF15/19 is not expressed in hepatocytes (Inagaki et al., 2005), FGF19 is amplified and over‐expressed in a significant fraction of human HCCs and both mouse forward‐genetics and xenograft experiments confirmed the relevance and therapeutic value of this finding (Sawey et al., 2011). It appears that FGF19 promotes liver cancer through a cell autonomous, possibly autocrine mechanism within transformed hepatocytes. In addition, data from GEMMs support the possibility of a paracrine tumor modulating function for FGF15/19 exerted from the gut in response to, for example inappropriate dietary inputs. Interestingly, bile acids are also mitogenic for hepatocytes (Huang et al., 2006) and are considered as tumor promoters and several mouse models with increased bile acid synthesis spontaneously develop HCCs (Liu et al., 2012). Therefore, FGF15/19 may also exert some beneficial effects in liver disease through the repression of bile acid production or even by affecting energy homeostasis. Further experiments using GEMMs, and controlled dietary conditions will certainly help dissect further the roles of bile acids and FGF15/19 in HCC development.

Besides its function in nutrient absorption and hormone production, the mammalian gut is also unique in that its lower part is a natural reservoir for diverse microbes playing a fundamental role in the host's well‐being. The constituents of the microbiota; bacteria, viruses and eukaryotes, interact with each other and with the host immune system and metabolic functions in both health and disease states (Clemente et al., 2012). Strikingly, the intestinal flora plays a crucial role in shaping the immune system, modifying bile acids, producing short‐chain fatty acids and vitamins, digesting complex nutrients and metabolizing orally administered drugs, all of which are important for liver homeostasis (Nicholson et al., 2012). In turn, dietary changes have been shown to have significant effect on the microbiota. For example, mice shifting to a high‐fat, high‐sugar “Western style” diet change their microbiota within a day (Turnbaugh et al., 2009). A recent study indicates that diet‐induced changes in microbiota composition can be partly attributed to the liver's bile acid synthesis activity (Devkota et al., 2012).

The implication of gut microbes in immune modulation and in systemic diseases, such as obesity and diabetes provided a reasonable rationale to hypothesize that these might also play an important role in modulating liver cancer development, at least indirectly. GEMMs provided a unique tool for assessing the effects of the gut microbiota on host physiology, as they can be engineered to lack or express molecules important for the host's interactions with bacteria. Furthermore, the improvement of animal husbandry and the generalization of extensive health monitoring resulted in a better understanding of the microbiological flora. GEMMs can be exploited in conventional, specific pathogen‐free or germ‐free settings. Moreover, germ‐free mice can be colonized either with selected microbial species or whole communities from healthy or diseased mice or humans. Using such approaches, it was demonstrated that gut‐derived bacterial products mediate hepatic fibrosis and inflammation, acting through TLR4, a member of a receptor family recognizing pathogen‐associated molecular patterns (Seki et al., 2007). Interestingly, a polymorphism in TLR4 correlates with reduced risk for fibrosis progression in patients with chronic hepatitis C virus infection (Huang et al., 2007). As hepatic injury has been associated with a defective intestinal barrier, bacterial translocation from the gut and increased hepatic exposure to bacterial products, targeting the bacterial flora might represent a valuable strategy to prevent chronic liver disease progression to HCC. In support of this idea, two recent studies experimentally demonstrated that gut microbes affect liver cancer development in mice exposed to hepatocarcinogens (Dapito et al., 2012; Fox et al., 2010). In the first study, intestinal colonization by Helicobacter hepaticus accelerated Aflatoxin‐ and HCV transgene‐induced HCC in mouse models (Fox et al., 2010). H. hepaticus is a well known mouse pathogen that translocates from the gut to the liver and induces hepatitis and HCC after a long latency in susceptible mouse strains (Ward et al., 1994). Interestingly, neither bacterial translocation to the liver nor induction of hepatitis was necessary to promote tumors initiated by aflatoxin or HCV‐transgene. Rather, it appeared that the bacteria were able to affect from their enteric niche, innate and adaptive immunity and possibly the liver's metabolic functions (Fox et al., 2010; Rogers, 2011). Conversely, Dapito et al. demonstrated that gut sterilization or TLR4 inactivation resulted in a significant reduction in HCC development in a DEN/CCl₄ tumor initiation/promotion model. Strikingly, tumor reduction was also observed when intestinal bacteria were eliminated using antibiotics during the tumor progression stage, suggesting that microbiome targeting could be used for HCC treatment in patients (Dapito et al., 2012). It remains to be established, if specific bacterial species are associated with liver cancer in patients and the precise mechanisms how the microbiome modulates distant tumor development need to be better elucidated. Nevertheless, the balance between gut flora, intestinal permeability, immune response and hepatocyte function appears decisive for the maintenance of liver homeostasis and clearly impacts on liver pathology.

7. The promises of onco‐genomics and other “omics”

Technological breakthroughs that allow simultaneous examination of thousands of genes, transcripts or proteins with high‐throughput techniques and computer‐based analytical tools have been the major scientific highlights of recent modern biology, in the wake of the human genome sequencing efforts. As a result, the traditional scientific approach to solve problems by relying on studying smaller, simpler units of a complex problem is being challenged by the emergence of two alternative/complementary scientific paradigms: high‐dimensional biology and systems biology (Strange, 2005). High‐dimensional biology (HDB) refers to the simultaneous omic‐based study of genetic and epigenetic variants, transcripts, proteins and metabolites of an organ, tissue, or an organism in health and disease, while systems biology can be viewed as the integration of experimental data derived from HDB with theoretical models aiming at predicting a biological outcome.

A number of high‐throughput omics technologies have been developed and applied to liver cancer research, aiming at the discovery of candidate biomarkers for cancer staging, prognosis, and therapeutic targeting. Large amounts of data on genetic and epigenetic abnormalities, gene and microRNA expression profiles, proteomics and more recently metabolomics have been collected. Similar sophisticated high‐throughput techniques have been applied to tumors and liver samples from GEMMs, allowing the simultaneous measurement of thousands of genes, transcripts and proteins. However, these approaches come with new challenges, for instance the heterogeneity of human samples with respect to etiology, pathogenesis stage and accompanying pathological conditions, and the requirement for advanced computational and statistical analysis tools as well as robust high‐throughput validation strategies to accommodate the enormous amount of data generated. An important concept that has emerged from these large‐scale studies is that cross‐species comparative and integrative strategies are tremendously helpful to identify molecular signatures relevant to HCC. In this respect, mouse models have provided a good source of comparative datasets, where an abundant, homogeneous, genetically and pathologically defined material can be used to “filter out” the “noise” from heterogeneous patient samples. In addition, the same strategies can be used to identify the most appropriate animal model to study a particular subtype of human HCC or as we foresee, even a particular stage of disease development, all of which are crucial, when attempting to evaluate therapeutic intervention strategies in a preclinical setting (Figure 1). Most of the studies published so far using this type of integrative analysis have utilized cross species datasets derived from transcriptional and/or genomic profiling. We have already mentioned some relevant, but certainly not exhaustive examples on how these approaches have led to important progress in understanding HCC pathogenesis (reviewed in Lee et al., 2005; Zender and Lowe, 2008). We anticipate that such HDB/systems biology strategies will become less costly and more routinely applied and will be further expanded to include additional species, such as rat models (Colak et al., 2010) or datasets, such as epigenetic, microRNAs, proteomics and metabolomics.

Figure 1.

Integration of data obtained from mouse models and human patient analyses toward better liver cancer prevention, diagnostic and treatment. Cross‐species comparative studies, either large‐scale, unbiased (omic‐type) or knowledge‐based (candidate approaches) should be carried out at every step along the process of improving mouse modeling and patient analyses. Improved mouse models ‐ GEMMs, mosaic, xenografts ‐ can be designed to recapitulate specific features of human HCC, while improved human disease characterization will allow better patient and disease stage stratification. Improved disease characterization and animal modeling will allow better, ultimately personalized, design of studies that aim at chemoprevention, biomarker discovery and drug testing.

Another useful application of high dimensional studies is the generation of publically available databases that can be used by the scientific community. For example a comprehensive catalog of common aberrations from both human and rodents (mouse and rat) is freely accessible in the OncoDB.HCC database (Su et al., 2007). This database provides a useful, validated, and graphical integration of published data derived from loss of heterozygosity analyses, array‐ comparative genomic hybridization, gene expression microarrays and proteomics. However, due to molecular diversity of the alterations underlying HCC development together with a multiplicity of samples, preparation and analysis tools, a major obstacle remains to optimize, control and harmonize data acquisition, interpretation and exchange, in order to provide a better basis for an efficient functional evaluation of the results. Nevertheless, the integration of multiple high dimensional independent datasets across species will certainly provide novel insights into molecular mechanisms in liver cancer and will lead to the identification of biomarkers and therapeutic targets that are likely to change the clinical management of the disease.

8. Challenges and implications with clinical perspectives

Liver cancer is a deadly disease. The current classification system for HCC patients used to determine the therapeutic strategy relies mostly on a collection of defined criteria that discriminate patients based on tumor burden, liver function and health status (Villanueva and Llovet, 2011). Transplantations and surgical resection remain the most effective treatment for early localized tumors, but due to the lack of symptoms or biomarkers only a minority of patients is identified at that stage. Most of the patients are diagnosed at an advanced stage with an aggressive disease and an almost systematic concomitant failure in liver function (Nordenstedt et al., 2012; Villanueva and Llovet, 2011). To date, systemic chemotherapy is ineffective against HCC, and no single drug or drug combination efficiently prolongs survival. With the current medication, the 5‐years survival rate remains below 5% in the developed world and Sorafenib, the standard therapy for patients with advanced HCC, only results in modest and transient benefits (Siegel et al., 2010; Villanueva and Llovet, 2011).

One of the most frequently cited reasons for the failure of therapeutic agents when transposed to clinical settings are the lack of pre‐clinical models faithfully recapitulating the disease. This is particularly true for liver cancer for two main reasons: the intrinsic complexity of the disease, which likely requires the development of more than one pre‐clinical model, and the gap between the animal models used to experimentally dissect the disease and those widely used in high‐throughput translational research and drug testing. In the years to come, efforts should be directed toward further improving the modeling of liver tumorigenesis, maximizing the information obtained from such models, improving the interaction between experimental and (pre‐)clinical research and increasing the involvement of industry and private sector, as well as other scientific disciplines in the rational design and exploitation of animal models. Only then will we achieve sufficient improvement in liver cancer diagnostic, prevention and eventually treatment.

As illustrated earlier, GEMMs allow the investigation of liver cancer in the context of a controlled, intact or altered environment. Therefore, mouse models have a good potential to facilitate the discovery of biomarkers that will serve as diagnostic and prognostic indicators or to identify potential systemic factors that can modulate tumor development. Cross‐species comparative studies indicated that such models are indeed useful to extract information allowing tumor and patient stratification beyond morphological criteria, an important step toward personalized treatment. The next step will be to apply such large‐scale comparative analyses to earlier stage of disease, where the preventive and therapeutic window is most favorable, but where the limiting factor at the moment is the availability of primary patient material. The current technological advances have already allowed the inclusion of patient peri‐tumoral tissues and dysplastic nodules in large‐scale analyses (Hoshida et al., 2008; Kaposi‐Novak et al., 2009) and we anticipate that such studies will become more informative, in particular if performed across species. Undertaking such comparative studies present higher technical, logistical and financial challenges and a better interaction across disciplines is certainly needed. For example, basic researchers would highly benefit from a standardized and generally agreed staging system for murine liver cancers similar to the one used in the clinic. Likewise, concerted transnational initiatives for integrated mouse/human biobanking and resource‐sharing might facilitate a fruitful crosstalk between basic and clinical science and increase the informative value of GEMMs.

Xenograft models, where human tumors are “cultivated” in immune‐compromised mice offer a fast and cheap solution for therapeutic testing. Such models are part of all standard drug development pipelines and are a pre‐requisite to clinical trials. For historical reasons these “artificial” models are preferred over GEMMs, as they have been used for over 50 years (Newell et al., 2008). More recently arguments, such as cost, speed and intellectual property issues are invoked as major reasons why GEMMs are not extensively used in pre‐clinical settings and/or in industry. Additionally, the lack of consistent genetic background, health status eg. Helicobacter species between different laboratories, the need for extensive phenotypic characterization for each model, the cost and logistics of maintaining sufficient numbers of animals with a slow progressing disease and biological cross‐species incompatibilities in drug metabolism, ligands etc. have been additional obstacles impeding the broad implementation of GEMMs in drug discovery and development processes (Singh and Johnson, 2006).

Xenograft models have shown a poor predictive value for the therapeutic effect of cancer drugs. The number of anti‐cancer agents that fail in clinical trials is far greater than the ones considered effective, which indicate that the drug selection procedure needs to be improved. In addition, molecular staging parameters are still scarce and mortality from liver dysfunction is a critical issue in HCC, therefore variation in trial design and interpretation complicates the evaluation of therapeutic agents, when transposed to the clinic (Villanueva and Llovet, 2011). Several limitations of xenografts models have been recognized. Transplant models do not replicate tumor–stroma interactions, either because tumor cells are not implanted at their native site or because human cells may fail to respond to signals derived from the mouse stroma. For example, mouse Hgf binds human MET with low affinity and does not potently activate human MET signaling. The first shortcoming can be mitigated using orthotopic models in which the tumor is implanted in the liver, provided that non‐invasive methods are available to assess tumor development and therapeutic efficacy. Biological cross‐species incompatibilities can also be dampened by generating “humanized” recipient mouse strains expressing human ligands, such as HGF (Zhang et al., 2005) to better recapitulate the tumors' paracrine environment. Similarly, several mouse lines have been generated that express the human version of essential liver drug metabolism genes to improve drug and carcinogen toxicology studies (Cheung and Gonzalez, 2008).

A majority of HCCs develop in the context of a cirrhotic liver and/or a systemic metabolic condition that complicate treatment. Both GEMMs and association studies have demonstrated the importance of liver and/or systemic dysfunction for tumor development, yet transplantation models usually assess the tumor's therapeutic response in the context of a rather healthy animal. Furthermore, given the overwhelming evidence for an important role of inflammation in HCC, the immune deficiency of the transplant host is another important aspect to consider. Mice that are pre‐conditioned to harbor a HCC‐relevant underlying condition by for example, high fat feeding (Tang et al., 2012) or further humanized by engraftment of a functional human immune system (Shultz et al., 2007) will certainly provide the opportunity to carry out translational research in a setting closer to the reality of the human disease. There has been a recent increase in the use of patient‐derived tumor xenografts for drug‐testing, including in HCC research (Tentler et al., 2012) and a future where patient‐derived tumor and immune cells are engrafted in rodents is likely a reachable horizon. Strikingly, most advances in the ability to derive better xeno‐transplant models have been made possible by parallel progress in mouse genetic manipulation.

While no model is ideal for all purposes, GEMMs have unraveled the complexity of HCC and demonstrated that liver cancer is the result of the interaction of multiple cellular and molecular pathways that have to be considered collectively. Additional work is needed to understand HCC pathogenesis and identify stage‐specific events that can be utilized for better and earlier intervention. Intelligent, knowledge‐based therapy design will likely be the key to success and animal models, whether GEMMs or transplant models are certainly an essential tool to increase our understanding of HCC. We have just begun to harvest the fruits of years of animal modeling research and further concerted efforts should be made to maximize the clinical benefits that can be obtained from mouse models.

Conflict of interest

The authors declare no conflict of interest.

Bakiri Latifa, Wagner Erwin F., (2013), Mouse models for liver cancer, Molecular Oncology, 7, doi: 10.1016/j.molonc.2013.01.005.