Genotype–phenotype relationships in hepatocellular tumors from mice and man


  • Potential conflict of interest: Nothing to report.


Experimentally induced liver tumors in mice harbor activating mutations in either Catnb (β-catenin) or Ha-ras, according to the carcinogenic treatment. We have now investigated by microarray analysis the gene expression profiles in tumors of the two genotypes. In total, 364 genes or expressed sequences with aberrant expression relative to normal liver were identified, but only 30 of these demonstrated unidirectional changes in both tumor types. Several functional clusters were identified that involve changes in amino acid utilization and ammonia disposition in Catnb-mutated tumors as opposed to alterations in lipid and cholesterol metabolism in Ha-ras–mutated tumors. Moreover, several genes coding for inhibitory molecules within the Wnt-signaling pathway were upregulated in Catnb-mutated tumors, suggesting induction of a negative feedback loop, whereas Ha-ras–mutated tumors showed alterations in the expression of several genes functional in monomeric G-protein signaling. We conclude that mouse hepatoma cells adopt different evolutionary strategies that allow for their selective outgrowth under variable environmental conditions. Human hepatocellular cancers (HCC) lack RAS mutations but are frequently mutated in CTNNB1, the human Catnb ortholog. The set of genes aberrantly expressed in Catnb-mutated mouse tumors was used to screen, by expression profiling, for dysregulation of orthologous genes within a panel of 25 HCCs, of which 10 were CTNNB1-mutated. HCCs with activated β-catenin displayed a gene expression profile that was similar to Catnb-mutated mouse tumors but distinct from the other human HCCs. In conclusion, expression fingerprints may be used for diagnostic purposes and potential new therapeutic intervention strategies. Supplementary material for this article can be found on the HEPATOLOGY website ( (HEPATOLOGY 2005.)

Rodent hepatocarcinogenesis is a powerful experimental system to analyze genetic changes relevant for tumor formation in the liver. Mouse liver tumors induced by a single injection of a liver carcinogen such as N-nitrosodiethylamine (DEN) frequently harbor activating mutations in the Ha-ras proto-oncogene.1, 2 If, however, DEN treatment is combined with subsequent chronic administration of the liver tumor promoter phenobarbital (PB) according to a classical initiation–promotion protocol, tumors predominate that lack ras mutations but show activating mutations in the Catnb (β-catenin) proto-oncogene instead.3 On histological examination, liver tumors generated in the absence or presence of the tumor promoter PB demonstrate considerable differences in hematoxylin-eosin–stained sections: the former are often basophilic and are generally composed of comparatively small cells, whereas the latter are often eosinophilic and contain larger cells with enlarged nuclei.4, 5 Several additional differences have been described if individual markers were used for discrimination of tumor types including glutamine synthetase (GS), which is strongly increased in expression in Catnb-mutated but undetectable in ras-mutated mouse hepatocytes.6 This suggests that mutation in either of the two genes produces divergent phenotypes; however, comparative genome-wide expression profiling employing Ha-ras and Catnb-mutated mouse liver tumors has not been performed. Human hepatocellular carcinomas (HCCs) are very infrequently mutated in RAS genes7; however, approximately 30% of tumors contain an activated version of β-catenin by mutation of the CTNNB1 gene, the human ortholog to mouse Catnb. We therefore screened microarray expression data of a panel of 25 human HCCs for corresponding alterations in a subset of genes identified to be dysregulated in Catnb-mutated mouse liver tumors. Based on this two-step procedure, we could establish a mutation-specific genotype/phenotype relationship for a subset of HCCs that showed activating mutations in CTNNB1 and strong overexpression of the marker enzyme GS.


DEN, N-nitrosodiethylamine; PB, phenobarbital; GS/glul, glutamine synthetase (protein/gene); HCC, hepatocellular carcinoma; PCR, polymerase chain reaction; HCV, hepatitis C virus; CAR, constitutive androstane receptor.

Materials and Methods

Animal Experiment.

The tumors analyzed in the current study were available to us from a previous experiment conducted in our laboratory.3, 4 In brief, liver tumors were induced in 6-week-old male C3H/He mice by a single intraperitoneal injection of 90 μg/g body weight of DEN. One group of mice (hereafter referred to as DEN/PB) was thereafter kept on a diet containing 0.05% of PB, and the second group received control diet (hereafter referred to as DEN/0). All mice were sacrificed at 45 weeks after DEN injection. Liver tumors and normal-appearing liver tissue were isolated and frozen in liquid nitrogen. Aliquots of tissues were used for preparation of total RNA (for details see Stahl et al.8) and analysis of mutations in Ha-ras and Catnb, respectively, as recently described.3

Human Hepatocellular Carcinomas.

Primary HCCs were obtained from patients undergoing surgical resection of HCC. Control liver tissues were from nondiseased liver as well as liver with viral hepatitis and cirrhosis. No patient received preoperative or adjuvant chemoembolization or radioablative therapy. Samples were from the University Hospitals of Berlin, Essen, and Tuebingen, all in Germany. Written informed consent was obtained from each patient. After resection, the tissue was rinsed in sterile phosphate-buffered saline solution and immediately stored in liquid nitrogen until RNA isolation. The study protocol conformed to the ethical guidelines of the 1975 Declaration of Helsinki as reflected in a prior approval by the various universities human research review committees. Pathology reports with tumor typing, staging (performed using UICC criteria), and grading as well as clinical data were supplied for each tissue sample.

Mutation Analyses.

For analysis of Catnb and Ha-ras mutations in mouse liver tumors, see Aydinlik et al.3 For screening of CTNNB1 mutations in human HCCs, a 1,688-bp genomic fragment flanking exon 3 was amplified by polymerase chain reaction (PCR), using the following primers: forward: 5′-GCGTGGACAATGGCTACTCAAG-3′; reverse: 5′-ACGCACTGCCATTTTAGCTCCT-3′. Point mutations at hotspot sites within exon 3 were analyzed by direct sequencing of both strands of the PCR-products using the primers 5′-AGCTGATTTGATGGAGTTGGAC-3′ and 5′-ACCAGCTACTTGTTCTTGAGTG-3′, respectively. If the annealing sites of these primers were eliminated by larger deletions, PCR-products were sequenced with the forward PCR primer and an internal reverse primer (5′-AGAACGCATGATAGCGTGTCTG-3′).

Microarray Analysis and Statistical Evaluation of Data.

The Affymetrix GeneChip MOE-430A (mouse tissues) and the GeneChip HG U133A (human tissues) were used for RNA expression profiling. In the mouse study, 9 chips were hybridized with cRNA from 3 Ha-ras and 3 Catnb-mutated tumors, as well as from 3 tissues isolated from DEN/0 mice (normal liver). In the human study, cRNA from 25 HCCs and 15 control livers was used for hybridization. Mouse RNA quality was controlled by Lab-on-Chip-System Bioanalyzer 2100 (Agilent, Palo Alto, CA), whereas for human tissues the 3′ to 5′ proportions of 2 control genes (GAPDH and HAS) were determined. The original datasets from the microarray analysis can be obtained on request.

The statistical analysis was carried out using the software package R, version 1.9.1,9 together with the libraries gcrma and limma of the Bioconductor Project, version 1.4.10 The data preprocessing steps and computation of GCRMA gene expression measures were performed according to Wu et al.11 for the MOE-430A chips.

Empirical Bayes inference for linear models with factor tissue type (normal tissue, Ha-ras-, Catnb-mutated tumors) was used for the statistical analysis of the mouse data.12 From there moderated t statistics based on shrinkage of the estimated sample variance toward a pooled estimate and corresponding P values were calculated for the comparisons Ha-ras–mutated tumors vs. normal tissue, and Catnb-mutated tumors vs. normal tissue. P values were adjusted according to Benjamini and Hochberg13 to control the false discovery rate. We used a threshold of .02 for the adjusted P values and selected only those probe sets that showed |log2 expression ratios| ≥ 2 for the related effects corresponding to a ≥ 4-fold change. A hierarchical clustering (manhattan metric + complete linkage) for the centered expression profiles over all experimental groups was performed for the presentation of probe sets selected by moderated t tests.

For the subset of probe sets satisfying the conditions mentioned for the comparison Catnb-mutated tumors versus normal tissue in mice, the human orthologs were identified using the NetAffx-annotation file for the MOE-430A chip. MAS 5 expression values of these probe sets were compared by exact Wilcoxon rank sum tests between CTNNB1-mutated HCCs and normal tissue. P values were adjusted to control the false discovery rate at a level of 10%. A minimal effect size of |log2 expression ratio| ≥ 1 was chosen.


Characteristics of Mouse Liver Tumors.

All 3 Ha-ras–mutated tumors were from DEN/0 mice (treated with DEN followed by control diet) and showed a heterozygous A>G base transition at the second base of codon 61 (c.182A>G). The 3 Catnb-mutated tumors were from DEN/PB mice and demonstrated heterozygous base substitutions in codons 32 (1 tumor, c95A>G) and 33 (2 tumors, c97T>C). Histological examination of tumors was not accessible, because the entire material was used up for DNA and RNA preparation. We know, however, from studies of our own and those of other groups,4, 5 that mouse liver tumors generated in the absence of PB often showed a basophilic appearance in hematoxylin-eosin–stained sections, whereas mostly eosinophilic tumors were observed in the presence of PB. In addition, approximately 80% of neoplasia in DEN/PB-treated mice showed strong overexpression of glutamine synthetase (GS), whereas lesions from mice of the DEN/0 group were all GS negative.6

Global Gene Expression in Mouse Tumors.

RNA expression patterns of normal liver tissue and Ha-ras– and Catnb-mutated mouse liver tumors were analyzed by use of the Affymetrix GeneChip MOE-430A, which contains approximately 22,600 probe sets, including more than 14,000 well-characterized mouse genes. Tumor-specific changes in RNA expression levels relative to normal tissue were identified using an adjusted P value ≤ .02 and a |log2 expression ratio| ≥ 2 as discriminators. In total, we identified 364 genes and expressed sequences that were dysregulated in liver tumors, 159 in Catnb-mutated and 205 in Ha-ras–mutated tumors. A detailed list of aberrantly expressed genes in the two tumor types along with their postulated functions is given in Supplementary Tables 1 and 2. (Please note that the number of probe sets with significant alteration was actually somewhat higher because several genes are represented multiple times on the array.) Only 30 of the 364 genes (approximately 8%) showed concomitant unidirectional alterations in both Catnb- and Ha-ras–mutated tumors (Supplementary Table 3). For illustration of global expression changes in tumors relative to normal liver, data are shown in Fig. 1 in the form of so-called volcano plots, where the adjusted P value of each probe set is plotted against the corresponding log2 expression ratio.14 Probe sets meeting the criteria of significance are represented in the gray areas of the plots. Catnb-mutated tumors showed symmetrical distribution of upregulated and downregulated genes; 10 genes had a fold change ≤1/64, and 12 genes showed a fold change ≥ 64. By contrast, genes with a fold change ≤ 1/64 were absent in Ha-ras–mutated tumors, but 14 genes showed fold changes ≥ 64 (Fig. 1).

Figure 1.

Volcano plots demonstrating changes in gene expression in Catnb (A) and Ha-ras (B) mutated mouse liver tumors relative to normal liver tissue. Each of the approximately 22,600 transcripts is represented by a single point. Probe sets meeting the criteria of discrimination (adjusted P value ≤ .02 and |log2 expression ratio| ≥ 2) are represented in the gray areas of the plots. Vertical broken lines separate probe sets with very strong dysregulation (|log2 expression ratio| ≥ 6).

A hierarchical cluster analysis of RNA expression data of Catnb- and Ha-ras–mutated tumors relative to normal liver was undertaken next (Fig. 2). Expression values of each probe set were centered before clustering to compensate for differences in basal transcription rates. The blue and red scales in Fig. 2 indicate decreased or increased expression (on a logarithmic scale) relative to the mean log2 expression in all 9 tissues. We are aware of the limitations of a cluster analysis based on only 9 samples from 3 different groups; we simply used this kind of analysis to stratify genes into subgroups of comparatively homogeneous response without necessarily asking for a biological meaning behind the clustering. The advantage of this procedure is that the resulting heatmaps are ideally suited to clearly visualize the strongly different gene expression patterns of Catnb- and Ha-ras–mutated tumors relative to normal tissue. Figure 2A demonstrates the results for those transcripts that showed significant alterations in expression in Catnb-mutated tumors when compared with normal liver. The respective expression values for each of these transcripts in Ha-ras–mutated tumors is also indicated, thus allowing direct comparison of expression data between tumor types. In analogy, the subset of transcripts with significant alteration in expression in Ha-ras–mutated tumors is visualized in Fig. 2B, along with the corresponding expression values of this second subset of genes in Catnb-mutated tumors. It becomes obvious that genes upregulated in one tumor type often demonstrated unchanged or even downregulated expression in the other, and vice versa.

Figure 2.

Hierarchical cluster analysis of transcripts significantly modified in expression in mouse liver tumors. (A) Subset of genes with significant alteration in expression in Catnb-mutated tumors versus normal liver (n.l.); (B) Subset of genes with significant alteration in Ha-ras–mutated tumors versus normal liver (n.l.). For each of the transcripts in A and B, changes in expression (log2 ratios) relative to the mean of values from all 9 tissues of the study (6 tumors, 3 Catnb- and 3 Ha-ras–mutated, and 3 normal liver tissues from non-PB-treated mice) are scaled by differences in color (blue, low expression; red, high expression).

The phenotypical alterations observed in Catnb- and Ha-ras–mutated tumors affect various cellular signaling and metabolic pathways in a characteristically different manner. In an attempt to better understand the biology of the two tumor types, the lists of altered genes (Supplementary Tables 1 and 2) were screened for patterns of deviations that uniformly affect fundamental cellular functions. Out of the many different pathways that may be disturbed, we briefly describe those that appear biologically meaningful and, in addition, are highly characteristic for either of the two tumor genotypes.

Changes in Genes Encoding Signaling Molecules.

Catnb-mutated mouse liver tumors showed alterations in several genes coding for proteins with an inhibitory activity within the Wnt/β-catenin/TCF pathway (Fig. 3A). These include Wnt inhibitory protein 1, Axin 2 (see also Lustig et al.15), naked cuticle 1, and Nemo-like kinase, which were significantly increased in expression relative to normal liver tissue. By contrast, Ha-ras–mutated tumors showed upregulation of a considerable number of RNAs encoding proteins involved in signaling cascades via small monomeric G-proteins (Fig. 3B). Each pattern of changes was specific for either Catnb- or Ha-ras–mutated mouse liver tumors without significant overlap between the two tumor types.

Figure 3.

Aberrant expression of genes coding for cell signaling molecules in Catnb- and Ha-ras–mutated mouse liver tumors. (A) Catnb-mutated tumors display characteristic increases in the expression of genes encoding negative regulations of Wnt-signaling. (B) Ha-ras–mutated tumors demonstrate alterations in other small GTPases, proteins directly associated with the monomeric G-proteins or proteins that signal through monomeric G-proteins. Abbreviations: (A) APC, Adenomatosis polyposis coli susceptibility protein; GSK, Glycogen synthetase kinase 3β; CKI, Casein kinase I; TCF, T-cell factor (also known as Lef1). Note that Axin2 (also known as Conductin) was upregulated, whereas Axin1 was unchanged. (B) Arf, ADP-ribosylation factor; Arfgap, Arf GTPase activating protein; Blnk, B-cell linker; Hrb, HIV-1 Rev binding protein; Nrg, neuregulin. Rab, Ras-related in brain; Ran, Ras-related nuclear protein; Rho, Ras homolog. Arrows indicate upregulation or downregulation of respective genes.

Changes in Genes Encoding Enzymes of the Intermediary Metabolism.

Several genes encoding enzymes of the intermediary metabolism were altered in Catnb- and Ha-ras–mutated tumors, but in a very different way. As shown in Fig. 4, Catnb-mutated tumors demonstrated a decrease in several RNAs for enzymes involved in the degradation of amino acids as well as two key enzymes of the urea cycle, whereas RNAs for a neutral amino acid and an ammonia transport protein were upregulated. In addition, Catnb-mutated tumors showed a strongly increased expression of GS. A completely different picture emerged with respect to Ha-ras–mutated tumors, which displayed a variety of changes in genes encoding enzymes involved in cholesterol and fatty acid metabolism and in serine and phosphocholine biosynthesis (Table 1). Again, the metabolic patterns were characteristic for either Catnb- or Ha-ras–mutated liver tumors, respectively.

Figure 4.

Catnb-mutated mouse liver tumors demonstrate significant changes in the expression of a cluster of genes encoding proteins functional in amino acid degradation and ammonia detoxification. Changes in expression are indicated by arrows. Amino acid degradation: EC, Glycine decarboxylase; EC, Serine dehydratase; EC, Cystathionine ligase; EC, Formiminotransferase cyclodeaminase; EC, Tyrosine aminotransferase; EC, Histidine ammonia lyase; EC, Arginase 1; Urea cycle: EC, Argininosuccinate synthetase 1; EC, Arginase 1. EC, Glutamine synthetase. Carriers: Solute carrier 1A4 (neutral amino acids); Rhbg, (ammonia transport).

Table 1. Selection of Genes Dysregulated in Ha-ras-Mutated Tumors With Relation to Lipid Metabolism
Accession No.Gene (EC-No.)Change
Cholesterol metabolism  
 BC010973CYP8B1 (EC 1.14.-.-)Down
 NM_007824CYP7A1 (EC
 NM_007825CYP7B1 (EC 1.14.13.-)Down
 AK014742Lanosterol synthase (EC
 NM_009286Sulfotransferase, hydroxysteroid preferring 2 (EC
 BC026757Hydroxysteroid dehydrogenase 2 (EC
 NM_134066Aldo-keto reductase 1C18 (EC
Fatty acid metabolism  
 NM_054094Butyryl-CoA synthase 1 (EC
 NM_008280Lipase, hepatic (EC
 NM_009128Stearyl-CoA desaturase 2 (EC
 NM_024406Fatty acid binding protein 4 (−)Up
 BC002008Fatty acid binding protein 5, epidermal (−)Up
 BB430611Fatty acid desaturase 2 (−)Up
Serine and phosphocholine biosynthesis  
 AA5617263-Phosphoglycerate dehydrogenase (EC
 BC004827Phosphoserine aminotransferase 1 (EC
 NM_133900Phosphoserine phosphatase (EC
 NM_013490Choline kinase (EC

Expression Profiling of Human Hepatocellular Carcinomas.

The subset of genes dysregulated in Catnb-mutated mouse liver tumors served as the basis for the subsequent analysis of gene expression data obtained from a panel of 25 human HCCs and 15 normal human liver tissues. We first screened 22 of the HCCs from which sufficient material was available for the presence of mutations in the human Catnb ortholog CTNNB1. Ten (45%) of the 22 HCCs demonstrated single base substitutions in 1 of the known mutational hot spot positions of CTNNB1 or showed partial deletions of sequences coding for the N-terminal domain of the protein (for mutation types, see Table 2). Total RNAs were then isolated from the human tissues and used for cDNA microarray analysis employing the Affymetrix HG-U133A chip containing approximately 22,000 probe sets. The NetAffx-annotation file for the mouse MOE-430A chip was used to allocate the human orthologs to those mouse probe sets that were aberrantly expressed in the Catnb-mutated mouse liver tumors and present on the human HG-U133A chip (Supplementary Table 4). In total, 48 genes and expressed sequences were identified as significantly dysregulated in the human CTNNB1-mutated HCCs when compared with normal liver, of which 18 were upregulated and 30 downregulated (Supplementary Table 5). The heatmap in Fig. 5 illustrates changes in expression of these probe sets in CTNNB1-mutated, CTNNB1-wild-type HCCs, and normal liver. The figure shows that the CTNNB1-mutated HCCs demonstrate a characteristic expression profile that is different from that of their wild-type cousins and from normal liver. Genes such as glutamine synthetase (glul), which was upregulated in 9 of the 10 CTNNB1-mutated HCCs, may serve as a marker for the presence of mutation of the oncogene. Please note that the NetAffx-annotation file used did not contain all human genes/expressed sequence tags (ESTs) orthologous to the mouse genes in Supplementary Table 1. Therefore, absence of a particular gene in Supplementary Table 5 does not necessarily mean that it was not dysregulated in the human HCCs. For example, axin 2 and naked cuticle 1 were absent from the HCC screen for this reason. Remarkably, 3 (33%) of the 10 CTNNB1-mutated HCCs were hepatitis C virus (HCV)-positive, whereas the prevalence of HCV in the remaining samples was only 13% (2 of 15). Moreover, none of the 10 CTNNB1-mutated HCCs was hepatitis B virus positive (Fig. 5).

Table 2. Some Characteristics of Human HCCs of the Current Study
Code No.Virus Status (B/C Virus)CTNNB1 Mutation
  • *

    Not analyzed for presence of mutations.

  • Partial or complete deletion of exon 3.

10NegativeCodon 41A/GCC
20NegativeDeletion548 bp
40HCVCodon 33TC/TT
50NegativeCodon 34GG/AA
110NegativeCodon 41A/GCC
160HCVCodon 32G/AAC
210NegativeCodon 41AC/TC
230HCVCodon 32G/AAC
250NegativeCodon 45Deletion
310NegativeDeletion766 bp
Figure 5.

Gene expression profiles of CTNNB1-mutated and CTNNB1-wild-type HCCs in comparison with normal liver. Changes in expression (log2 ratios) relative to the weighted mean of the three groups (10 CTNNB1-mutated and 15 CTNNB1 wild-type tumors, and 15 normal liver tissues) are scaled by differences in color (blue, low expression; red, high expression). Tumors are separated by CTNNB1 genotype (mut., mutated; w.t., wild-type). Tissues are sorted according to their expression of the marker glutamine synthetase (Glul). Association with HCV and HBV are indicated by black and gray colors, respectively; other tissues were B/C virus negative.


The mouse liver tumors investigated in the current study were all induced by a single injection of the hepatocarcinogen DEN into 6-week-old mice. Subsequent treatment of animals with the tumor promoter PB resulted in a selective outgrowth of tumors with Catnb mutations, whereas mutations of Ha-ras predominated in tumors developing in the absence of the tumor promoter.3, 16 Because PB was administered through the diet until the end of the experiment, mice of this group were acutely exposed to the barbiturate at the time of their sacrifice. PB is known to modify the expression of a variety of genes, particularly those encoding enzymes of xenobiotic metabolism, mostly through activation of the constitutive androstane receptor (CAR). In fact, cytochrome P450 2B10, P450 reductase, and the glutathione-S-transferases m2 and m3, all of which represent well-known CAR targets, were overexpressed in liver tumors from PB-exposed mice. Importantly, however, all other genes aberrantly expressed in Catnb-mutated liver tumors were not identified as overexpressed in normal liver tissue from mice acutely exposed to PB.8 This evidence shows that the remaining genes are dysregulated in a tumor-specific way but not directly affected by the enzyme inducer PB. The same cytochrome P450-isoforms were strongly repressed in the Ha-ras–mutated tumors in comparison with normal liver (Supplementary Table 2), which may be due to the more than 8-fold reduction in expression of CAR in these tumors (Supplementary Table 2). We also propose that the lack of proliferative response of Ha-ras–mutated hepatocytes to PB, as opposed to effects seen in their Catnb-mutated counterparts, may result from their reduced CAR expression levels.

The gene expression profiles of the two types of tumors, Ha-ras– as opposed to Catnb-mutated, differed dramatically. Only 30 of 364 genes demonstrated concomitant aberrant expression in both tumor types (Supplementary Table 3). The observed expression fingerprints give some interesting hints on biological strategies of liver tumors. For example, we have previously shown that the intracellular concentration of β-catenin is not increased in Catnb-mutated mouse hepatocytes,3, 6 which is different from what is seen in other tumors.18 This effect could be the result of compensatory regulation by a negative feedback loop aimed at inhibiting β-catenin–mediated gene transcription (for a review, see Lustig and Behrens19), as evidenced by our current data (Fig. 3A). By contrast, a completely different set of signaling molecules was activated in Ha-ras–mutated tumors. Ha-ras is a prototype of the large family of small monomeric GTPases, which represent signaling molecules involved in various cellular processes including proliferation and differentiation (Ras-subfamily), regulation of the active cytoskeleton (Rho-subfamily), and regulation of directed intracellular transport mediated by membrane trafficking (Rab-subfamily), vesicular trafficking (Arf-subfamily), and nuclear transport (Ran-subfamily) (for a review, see Macara et al.20). Small monomeric GTPases represent molecular switches regulated by guanyl nucleotide exchange factors and GTPase activating proteins. We now identified a considerable number of GTPases and associated proteins that showed altered expression in the Ha-ras–mutated tumors (Fig. 3B), demonstrating that constitutive activation of Ras-dependent signaling affects various other monomeric G-proteins, which are coupled to intracellular processes that may be linked to Ha-ras–triggered tumorigenesis.

Apart from signaling pathways, several characteristic differences in metabolic capacities between Catnb- and Ha-ras–mutated liver tumors were evident. Catnb-mutated tumors demonstrated alterations in several RNAs for enzymes involved in amino acid utilization and ammonia disposition (Fig. 4). In accordance with previous findings,6, 17Catnb-mutated tumors also demonstrated increased expression of GS, which catalyzes synthesis of glutamine from glutamate and ammonia. In synopsis, these observations suggest that Catnb-mutated hepatocytes adapt their intermediary metabolism to preserve their amino acid pool and minimize nitrogen excretion, which may instead be used for glutamine synthesis. This is advantageous in liver because more than 95% of hepatocytes lack GS but have high levels of glutaminase instead, which lowers their glutamine levels.21 Glutamine, however, serves as an amino group donor in pyrimidine and purine synthesis, a prerequisite for DNA-synthesis. Ha-ras–mutated tumors, on the contrary, displayed a variety of changes in genes encoding enzymes involved in cholesterol and fatty acid metabolism/utilization (Table 1), which may lead to increased cholesterol concentrations because of a decreased efflux of cholesterol to bile acid synthesis. In addition, our data suggest that degradation of long-chain fatty acids via mitochondrial β-oxidation may be decreased, resulting in elevation in intracellular concentration of unsaturated long-chain fatty acids required for membrane synthesis. Finally, Ha-ras–mutated tumors showed increased expression of all 3 genes that encode the enzymes involved in the formation of serine from 3-phosphoglycerate and overexpressed RNA for choline kinase α, suggesting increased production of phosphocholine, which is a component of membrane structures. Thus, Ha-ras–mutated tumors demonstrated dysregulation in metabolic pathways that are, directly or indirectly, connected to membrane turnover.

The divergence in gene expression patterns of Ha-ras– and Catnb-mutated tumors appears to reflect the fact that the oncoproteins encoded by the activated genes act in two divergent signaling pathways, suggesting that the observed changes are a direct consequence of alterations in the activity of transcriptional modifiers that are under the control of Ha-ras– and β-catenin–dependent signaling cascades. Consequently, the respective phenotype would directly reflect the respective genetic alteration, given that Ha-ras and Catnb mutations were the only rate-limiting genetic alterations present in the mouse liver tumors investigated. In contrast to human HCCs, which are genetically very unstable, particularly if associated with hepatitis B virus infection,22, 23 mouse liver tumors demonstrate a remarkable chromosomal stability.24, 25 The idea that only one genetic alteration is relevant for malignant transformation of DEN-induced mouse liver tumors is indirectly supported by quantitative analysis of the number of tumors induced in liver after a single injection of different doses of DEN, which revealed a linear dose–response relationship.26 This observation is compatible with the hypothesis that a single rate-limiting mutation induced by the carcinogen is sufficient for liver tumor formation in mice.

Ras mutations are rare in human HCCs; however, approximately 30% of these cancers harbor activating mutations in the CTNNB1 gene.22 Global gene expression in human HCCs has been investigated in several different studies, and a large number of genes with tumor-specific aberrant expression have been identified.27, 28 By use of our two-step screen employing mutation-type–specific gene expression patterns in mouse liver tumors, we were able to identify a cluster of genes that are most likely dysregulated because of a specific molecular alteration in the oncoprotein β-catenin. The results of our study substantiate the notion that overexpression of GS is a good predictor for the presence of an activated version of β-catenin in hepatocellular tumors of both mice and man. The expression of GS can be simply demonstrated by immunohistochemistry, allowing for indirect molecular diagnosis of the underlying genotype of the tumor. Our results further show that Catnb/CTNNB1-mutated tumors display a very characteristic fingerprint of aberrant gene expression that includes strongly upregulated genes that may be suitable targets for therapeutic interventions, such as the G-protein coupled receptor 49 (Gpr49)29 and Rhbg (coding for an ammonia transporter).

In agreement with literature data,30 we observed that CTNNB1-mutated HCCs are frequently associated with chronic HCV but not with HBV infection. Most CTNNB1-mutated HCCs of the current study developed on the basis of a C virus–induced cirrhosis that appears to create a promotional stimulus similar to the one delivered by the tumor promoter PB during selection of Catnb-mutated hepatocytes in mouse liver. In the experimental system, withdrawal of the tumor promoter leads to regression of the tumors, at least in part.31 We therefore suspect that interference with the promotional stimulus delivered by the C virus infection or the resultant cirrhosis would be most effective in treatment of HCV-associated HCCs.


The authors acknowledge the excellent technical assistance of Elke Zabinsky.