Characterization of HCC Mouse Models: Towards an Etiology-Oriented Subtyping Approach

This article is featured in Highlights of This Issue, p. 1415

Hepatocellular carcinoma (HCC) is the most common primary liver cancer and has become the third leading cause of cancer-related death worldwide (1, 2). The main risk factors for liver carcinogenesis are chronic liver diseases, such as viral hepatitis, alcoholic, or nonalcoholic steatohepatitis (ASH/NASH), exposure to aflatoxin or genetic disposition (e.g., α1-antitrypsin deficiency; ref. 3). Dietary-induced liver cancer is an emerging problem in developed as well as in developing countries (4, 5).

Strategies to improve the still poor survival of HCC patients rely on preclinical mouse models, such as cell line–derived models in immunocompromised mice (allografts and xenografts), genetically engineered mouse models (GEMM) and environmentally induced models. So far, the translational value of mouse models with respect to patient benefit has frequently fallen behind expectations. Besides the need for discovering new anti-HCC targets and compounds and testing them in vivo, it is of utmost importance to improve analyses and subtyping for preclinical mouse model research. First, distinct models may recapitulate only individual features of human HCC. Second, reporting of morphology, IHC profiles, genetic landscapes, sequencing of the key tumor suppressors/oncogenes, and growth monitoring of murine tumors is poorly standardized in mouse research (6). An important challenge in the comprehensive characterization of murine models is already to identify truly malignant lesions. Different criteria are used, such as atypia, increased proliferation, expansive growth, necrosis, or extracapsular invasion (7–9). Markers such as glutamine synthetase and collagen IV may serve as supplemental indicators for tumor diagnosis (10–12). Mutational profiles of murine liver tumors, with frequent CTNNB1 mutations and rare or absent TP53 alterations, were characterized in earlier studies (13–15).

Approaches to improve mouse model characterization and subtyping include (i) systematic assessment of human-like phenotypes including morphology, IHC profiles, and intratumor heterogeneity; (ii) evaluation of etiology-dependent models; and (iii) if possible, assignment to a clinically stratified patient subgroup. The murine models analyzed in this study comprise four GEMMs and one environmentally induced model, all with spontaneous, orthotopic tumor growth. Different genetic backgrounds were included, covering essential cancerogenesis pathways (8, 10, 11, 16–18): oncogene overexpression (c-Myc), chronic inflammation (TAK1^LPC-KO and AlbLTαβ), and liver cell loss with compensatory proliferation (Mcl-1^Δhep). The “classical” and widely used DEN model was included, because it is generally considered to mimic toxin-induced cancerogenesis (7).

The aim of our study was to subtype HCC mouse models with different cancerogenesis backgrounds, to increase the translational value of rodent models. On the basis of comparative genomic hybridization (CGH) analysis, we propose a novel strategy to quantify the similarity of murine and human tumors. The starting criterion was the percentage of genomic overlap in the synteny analysis of CGH data of murine tumors compared with HCC patients of The Cancer Genome Atlas (TCGA) database. Furthermore, we categorized histomorphologic features and IHC profiles to show different qualities and levels of overlap. Each set of murine tumors (DEN treatment, c-Myc induced, TAK1^LPC-KO knockout, AlbLTαβ transgenic mice, and Mcl-1^Δhep knockout) was compared with three clinically defined subsets of human HCC (alcohol, chronic viral hepatitis, NASH/cryptogenic) and molecular subclasses G1–6, in order to identify which rodent model recapitulates HCC carcinogenesis in specific etiologic backgrounds. Our approach might help future guidelines to stratify and compare preclinical mouse models—finally helping to increase the success rate in clinical trials.

Formalin-fixed, paraffin-embedded (FFPE) mouse liver tissues retrieved from previous studies as listed in Table 1 were used (7, 10, 11, 16, 17). Original experiments with mice had been conducted in concordance with local guidelines (approval: “Tierversuchsgenehmigung vom Kantonalen Veterinäramt Zürich 63/2011”). Five mouse models were used: In the DEN models, tumors were chemically induced in wild-type (C57BL/6 strain) mice (7). The c-Myc model is a transgenic model targeting the c-Myc proto-oncogene (16). In the TAK1^LPC-KO model, specific depletion of TAK1^LPC-KO in liver parenchymal cells leads to deregulated TNF signaling as well as defective AMPK activation resulting in chronic mTORC1 activation (19). Liver cells undergo uncontrolled proliferation and necroptosis/apoptosis leading to early, accelerated liver cancer formation in mice (17, 20). The transgenic AlbLTαβ/tg+ model reflects inflammation-induced carcinogenesis with aberrant expression of the cytokine lymphotoxin (10). The Mcl-1^Δhep model mimics chronic liver cell damage through hepatocyte-specific depletion of the antiapoptotic protein myeloid cell leukemia 1, which leads to continual hepatocyte apoptosis, increased cell turnover, compensatory proliferation, and spontaneous tumor formation (8, 11).

For mutation analysis and CGH (n = 75), DNA was extracted from FFPE tissues (Kit, GE-healthcare). PCR was performed with following the manufacturer's protocols (AmpliTaq Gold, Applied Biosystems), conducting 40 cycles (TP53 exons 5–8, CTNNB1) or 35 cycles (BRAF, HRAS). Primers were used as previously described: CTNNB1 exon 2 (21), TP53 exons 5–8 (22), HRAS, and BRAF (23). Annealing temperatures were 56°C (TP53 exons 6 and 8, BRAF, HRAS) or 60°C (TP53 exons 5 and 7). PCR amplification and sequencing of the mTERT core promoter fragment (24, 25) was performed on a subset of tumors (n = 31 tumor samples) and 10 unaffected tissues. We used (−279 to +14 coverage) two primer pairs, forward 1/2: TTA CTC CAA CAC ATC CAG CAA and CCT TCC GCT ACA ACG CTT; reverse 1/2: AAA GAT GAG GCT GGG AAC G and GAG CGC GGG TCA TTG TG, at 58°C annealing temperature. Sequencing was performed using a commercial service (Microsynth) with dropouts (1%–10%) due to poor DNA quality. Mutation analysis was performed using BioEdit freeware, GRCm38/mm9 served as the reference genome. For CGH analysis, commercially available kits (Oligonucleotide Array-Based CGH for Genomic DNA Analysis, Agilent) were used (26), and CGH results were matched with results from the TCGA cohort (TCGA-LIHC; http://cancergenome.nih.gov/. Synteny analysis of CGH data was performed according to the theory of eutherian chromosome evolution (27). Contingency tables were constructed with etiology-dependent HCC subsets such as alcohol-related, hepatitis B/C-induced, or NASH-induced/cryptogenic HCC. Cryptogenic HCC were used for the analysis because these tumors are likely caused by burned out NASH even in the absence of cirrhosis (28–30). Fisher's exact test was used for statistical analysis, adjusted for multiple testing (Benjamini–Hochberg correction). A significance level of 5% was set to detect significant correlations between human and mouse chromosomal losses and gains controlling for the alpha error. The classification of HCC proposed by Boyault and colleagues (31) into the subgroups G1–6 was applied to TCGA cohort. The data set E-TABM-36 was retrieved from ArrayExpress (ebi.ac.uk/arrayexpress) and class labels were extracted from Fig. 1 of Boyault and colleagues. An additional data set, GSE62232, was retrieved from GEO (ncbi.nlm.nih.gov/geo), and class labels were kindly provided by the authors. A list of top overexpressed genes per subgroup was produced by comparing patients in each subgroup with patients in all other subgroups. These lists were then used to classify patients from the TCGA cohort into the six subgroups using the Nearest Template Prediction algorithm (32).

A systematic review of the documented murine models was performed (10, 11, 16, 17). For virtual microscopy, we digitalized slides of murine liver lesions with available IHC results (n = 149) using a NanoZoomer C9600 Virtual Slide Light microscope scanner by Hamamatsu using NDP, View Software, version 1.2.36.

Murine liver lesions were classified as tumors based on morphologic criteria reported by Thoolen and colleagues (9) and five markers of liver pathology (Supplementary Fig. S1). Briefly, overgrowth compressing the normal tissue and/or distortion of the lobular architecture was considered as the main criteria for tumors in contrast to dysplastic nodules. Collagen IV loss or broadening of trabecular structures was regarded as neoplastic growth. Cytological features considered as indicators for malignancy were cell polymorphism, atypia, increased nucleus–cytoplasm ratio, inclusion bodies, or basophilia. Sizes of cells and nuclei were measured using digitalized histologic pictures and dichotomized by the median. The tumor grading was based on a combination of nuclei sizes (<10 μm = 1, 10–15 μm = 2, 15–20 μm = 3), presence of nucleoli and cell–plasma ratio (decreased/normal). Proliferation in tumors was assessed as a 4-point scale (none, few, many, and abundant).

IHC on mouse tissues (glutamine synthetase, A6, GP73, collagen IV, Ki-67) was performed as described (10, 17). For human liver tissues, stainings of glutamine synthetase, CK7 and CK19, were conducted and scored as reported (33). Positivity for a marker was defined as follows: A6 and CK7/19: >10% of tumor cells, glutamine synthetase: diffuse strong staining of >50% cells, GP73: weak or strong positivity. For statistical analysis of morphologic features, IHC and mutational profiles, SPSS software was used (IBM SPSS, Version 21).

Human liver tissues were retrieved from the archives and biobank of the Department of Pathology and Molecular Pathology, University Hospital Zurich. Tissue microarrays (TMA) with duplicates of a total of 61 HCC patients and 60 matched controls were used for IHC analysis as described (34). Follow-up data for all patients were available. The study was reviewed and approved by the Cantonal Ethics Committee of Zurich, Switzerland, according to guidelines (KEK-ZH-Nr. 2013-0382).

First, we aimed to perform a systematic histopathologic characterization comparing murine livers of all models (Table 2). Analysis of a total 49 mouse livers with an average of ∼3 tumors per mouse (mean 3.04 ± 0.81) revealed several differences among the five mouse models. Smaller, rather monomorphic tumors and numerous dysplastic lesions were found in the DEN-treated and TAK1^LPC-KO models, compared with larger, less-abundant tumors in the c-Myc, the Mcl-1^Δhep, and the AlbLTαβ models. In the Mcl-1^Δhep model, subnodules were observed, which were reminiscent of those observable in human HCC (33). The size of cells and nuclei in individual tumors was higher in the TAK1^LPC-KO and AlbLTαβ models compared with those in other models. High-grade tumors, defined by a combination of large nuclei, presence of nucleoli, and an increased nuclear/cytoplasmatic ratio were found in the TAK1^LPC-KO and the AlbLTαβ models (67% and 83% of tumors, respectively). High proliferation was found in tumors of the Mcl-1^Δhep model (21 tumors of total 38) and the c-Myc model (15 tumors of total 22).

Analysis of chromosomal gains and losses per mouse and tumor revealed patterns with predominant chromosomal gains (DEN, TAK1^LPC-KO) and patterns with predominant chromosomal losses (AlbLTαβ; Fig. 1A). In comparison with the unstratified TCGA reference HCC cohort, the percentage of combined aberrations (amplification and deletions) that overlapped between the murine tumors and human HCCs ranged from 56% in the TAK1^LPC-KO model to 71% (mean = 61%) in the AlbLTαβ model (Table 2).

Targeted mutational analysis of commonly affected genes (TP53, HRAS, NRAS, CTNNB1, and TERT) showed that murine liver tumors across all mouse models were TP53 wild-type (Fig. 1B). Thirty-three percent of analyzed DEN-induced tumors showed BRAF mutations, lower than previously reported (35). CTNNB1 were found in the c-Myc model (10%) and in four liver tumors (21%) of the same animal within the Mcl-1^Δhep group. At low frequency (max. 11%), HRAS mutations (DEN, Mcl-1^Δhep, and c-Myc models) were detected. TERT promoter mutations of the transcription factor binding sites were analyzed in a subset of murine tumor samples (n = 31) and were not detected.

By comparing CGH profiles of distinct HCC patient subsets from the TCGA database with profiles of each mouse model, the mean overlap further increased by maximally 14% (Fig. 2A and B). Murine tumors of the DEN and c-Myc models shared genomic changes predominantly with alcohol-induced HCCs (63%–69%; P < 0.01) and the G5 molecular subclass. AlbLTαβ and TAK1^LPC-KO most closely resembled NASH-HCC (57%–67%; P < 0.01) and the G3 molecular subclass. Mcl-1^Δhep showed highest overlap with viral hepatitis-induced HCC (60%, P < 0.01) and the G3 molecular subclass.

We next tested whether morphologic findings support the CGH-based classification of murine tumors (Fig. 3A). In the DEN model, abundant cellular inclusions mimic Mallory–Denk bodies found in toxin-damaged liver cells. The presence of fatty change and clear cell cytology supports chronic nutritive-toxic liver cell damage. The c-Myc model was the second closest match for alcohol-induced cancer, based on CGH analysis. Histopathologic findings comprised clear cell features and pale inclusion bodies in combination with lymphocyte infiltration and distorted lobular architecture (Fig. 3B).

HCC of NASH/cryptogenic background matched closest with tumors from the TAK1^LPC-KO model and the AlbLTαβ model (CGH analysis). Histopathology showed steatosis, massive lymphocyte infiltration, and tumor necrosis in the AlbLTαβ model. NASH-typical morphologic findings were less frequent in the TAK1^LPC-KO model, possibly due to the early onset of carcinogenesis in this model. In contrast, a frequent finding was the mixed-cell phenotype, consisting of side-by-side eosinophilic and basophilic cells typically found in rodents and indicative of liver damage.

Mcl-1^Δhep tumors most closely matched with virus-induced HCC than other etiologies based on CGH analysis (60%, P < 0.004), and the greatest overlap was seen with patients with hepatitis B (P < 0.025). Morphologically, tumors of the Mcl-1^Δhep model showed apoptotic hepatocytes, highly proliferative tumors, tumor necrosis, steatosis, and moderate lymphocyte infiltration. Of note, fibrosis was rare in non-tumorous liver tissue throughout all models (Fig. 3C).

Given that human HCC are mostly well-demarcated tumors with variable growth patterns and cytology, we next analyzed tumor growth including inter- and intratumor heterogeneity (32, 36). Intertumor heterogeneity refers to the diversity of tumors within each model, and is defined by the number of histology patterns per mouse cohort. Intratumor heterogeneity refers to the heterogeneity within each tumor, and is defined by histology patterns per individual tumor. As for the intertumor heterogeneity, we analyzed it on a morphological, IHC, and CGH level. Two of the models (DEN and TAK1^LPC-KO) scarcely showed any tumor heterogeneity, whereas three models (Mcl-1^Δhep, c-Myc, and AlbLTαβ) display tumor heterogeneity on a CGH level and histologically. In detail, a single major growth pattern and maximum two cytological features were found in the DEN model and the TAK1^LPC-KO model. In contrast, two (c-Myc model) or more than three growth patterns (Mcl-1^Δhep and AlbLTαβ models) in combination with cytological features were observed (intertumor heterogeneity). Intratumor heterogeneity (>2 different growth patterns and/or cytological features within the same tumor) was present in ∼50% of Mcl-1^Δhep and AlbLTαβ tumors. In three of the models (DEN, Mcl-1^Δhep, AlbLTαβ), tumors were clearly demarcated compared with a diffuse intrahepatic growth in the other two models, i.e., c-Myc and TAK1^LPC-KO (Fig. 3D).

Next, we were wondering whether IHC profiles in murine liver tumors mimicked the profiles of human HCC. Homogeneous IHC profiles were observed in the TAK1^LPC-KO model and the DEN-treated model. Tumors of the TAK1^LPC-KO model were nearly exclusively negative for A6 (biliary/progenitor phenotype) and glutamine synthetase (a marker of β-catenin activation) (36). Tumors of the DEN-treated mice were A6 positive in 85%. Heterogeneous, more human-like profiles were present in the c-Myc and Mcl-1^Δhep models, including positivity for glutamine synthetase and/or A6. The closest resemblance to the HCC cohort IHC profile was found in the AlbLTαβ model (Fig. 4A and B).

In summary, murine tumors segregate into mainly biliary/progenitor-like phenotypes (DEN-treated, Mcl-1^Δhep), β-catenin-activated phenotype (c-Myc), and mixed (AlbLTαβ). Table 3 shows a summary of IHC, genetic and morphologic subtyping results.

The challenge for future liver cancer models is to account for heterogeneity of disease, etiology-dependent pathogenesis, and therapeutic targets. Our approach suggests that these aspects should be considered to improve the clinical relevance and translational value of preclinical cancer research models.

Taking advantage of CGH, we were able to discriminate between preclinical models recapitulating alcohol-induced, virus-related and NASH-HCC as well as molecular subclasses G1–6. By matching chromosomal aberrations of mouse and human tumors, it is possible to construct an algorithm to measure the concordance (P value) of a model and a specific patient subgroup. As previously reported, synteny studies efficiently compare homologue mouse and human chromosomal aberrations (37). Sequencing of tumor suppressors and oncogenes might also be helpful to identify molecular markers, though mutational profiles of murine liver tumors largely differ from human HCC. For example, we found BRAF and HRAS mutations (rare in humans) as was reported in previous studies (23, 35). The absence of TP53 mutations in murine tumors is in line with earlier findings (15) and stands in contrast to human HCC. CTNNB1 mutations, found in 27% of human HCC regardless of their etiology (38), were present only in two models (c-Myc and Mcl-1^Δhep). Recently, a study with design similar to ours was performed by Dow and colleagues based on genomic and transcriptomic profiles in mouse versus human tumor tissues (39). The authors claim that distinct mouse models reflect aspects of low-grade human tumors, whereas, e.g., DEN tumors carry a high mutational burden similar to poorly differentiated tumors. Going beyond the molecular level, we have aimed to perform a comprehensive approach integrating morphologic, IHC, and CGH analysis to assess human–mouse similarities.

The correlation we observed between histopathologic characteristics and CGH results supports the etiology-oriented subtyping of HCC mouse models accounting for heterogeneity of disease (33, 40, 41). A recent study by Calderaro and colleagues reported the relationship between heterogeneous histologic subtypes and associated oncogenic pathways in HCC (42). As was demonstrated in our analysis, intracellular hyaline bodies are abundant in DEN-induced tumors that most closely matched alcohol-induced HCC. The cellular inclusions are reminiscent of Mallory–Denk bodies, typical for human alcoholic steatohepatitis. The rounder hyaline bodies and irregular, keratin 8 containing Mallory bodies (43) coexisted in a study on 174 human HCC in 7.5% of cases (44). A crucial finding is also the presence of steatosis, indicative of metabolic deregulation (9) characteristic for NASH patients. NASH/cryptogenic HCC were best recapitulated by tumors of the AlbLTαβ and TAK1^LPC-KO model. Particularly the AlbLTαβ model showed features diagnostic for NASH such as steatosis and inflammatory infiltration(45, 46) criteria for diagnosing NAFLD/NASH. Even though the model was originally developed to mimic human chronic viral hepatitis (10), the current analysis found more similarities with NASH-induced HCC. Fibrosis, an important feature of human chronic liver disease, was rare in murine models, as has been documented before (47).

One of the key observations of this study is that inter- and intratumor heterogeneity is present in varying degrees in HCC mouse models, which could be considered as an indicator of the appropriateness of murine models. Although human HCC typically show inter- and intratumor heterogeneity (33), this feature is recapitulated only by particular liver cancer models (Mcl-1^Δhep, c-Myc, and AlbLTαβ). Taking into account inter- and intratumor heterogeneity in preclinical models is crucial for many solid cancer models, especially for testing systemic treatments in advanced disease. Suitable preclinical animal models recapitulating diverse histopathology, IHC profiles, and associated oncogenic pathways of human HCC subtypes can be expected to better recapitulate reponsiveness to treatment.

A limitation of our study is that intratumor heterogeneity was analyzed only on the level of morphology and IHC. In line with earlier findings, phenotype–genotype correlation studies have shown that genetic heterogeneity frequently goes along with morphologic and immune-phenotypic heterogeneity (33). Because we could not find morphologic and/or immune-phenotypical intratumor heterogeneity except for nodule-in-nodule growth in one model (Mcl-1^Δhep), we did not follow up on microdissection of the lesions. Another limitation of our study regards imaging and treatment responses in preclinical models, which were performed in a study by Gross and colleagues (12). This study compared the DEN model with the allograft model McA looking at tumor imaging in conjunction with histopathology, CGH, and treatment response.

Regarding the recent interest in the immune microenviroment (48, 49) with focus on T cells in NASH (37, 50), mainly the AlbLTαβ model seems to have the potential for consecutive subtyping of lymphocytes and PD-L1 expression analysis. A response rate of 20% for PD-1 (anti-programmed cell death-1 antibody) monotherapy in phase I/II trials has been attributed to the refractory immune-suppressive status in liver cancer patients (51), which needs further investigation.

In summary, contemporary preclinical models may be assigned to etiology-dependent patient groups and should account for inter- and intratumor heterogeneity. This holds implications for the preclinical testing of targeted treatments and could improve patient management.

No potential conflicts of interest were disclosed.

Conception and design: J. Friemel, L. Frick, M. Heikenwalder, A. Weber

Development of methodology: L. Frick, A. Weber

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J. Friemel, L. Frick, K. Unger, M. Egger, R. Parrotta, A. Adili, T. Luedde, M. Heikenwalder, A. Weber

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Friemel, L. Frick, K. Unger, R. Parrotta, Y.T. Böge, M. Karin, M. Heikenwalder, A. Weber

Writing, review, and/or revision of the manuscript: J. Friemel, L. Frick, M. Karin, T. Luedde, M. Heikenwalder, A. Weber

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J. Friemel, A. Weber

Study supervision: M. Heikenwalder, A. Weber

Other (ran the experiment (aCGH array) and analyzed the acquired data): A. Adili

A. Weber was supported by a grant from the Swiss Cancer League (Oncosuisse). M. Heikenwalder was supported by an ERC Consolidator grant (HepatoMetaboPath); the DKFZ MOST program, the SFB 209, and the HepCAR 2020. T. Luedde was supported by a Mildred-Scheel Endowed Professorship from the German Cancer Aid (Deutsche Krebshilfe) and the German Research Foundation (DFG) (LU 1360/3-1 and SFB-TRR57/P06). The results published here are, in whole or part, based upon data generated by the TCGA Research Network: http://cancergenome.nih.gov/.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.