Molecular characterization of hepatocarcinogenesis using mouse models

Study design for longitudinal analysis of hepatocarcinogenesis

We performed a systematic longitudinal analysis of molecular and physiological changes that occur during the process of HCC formation, under conditions that mimic the absence or presence of hepatitis, or upon exposure to the hepatotoxin AFB₁ at various stages in life. HBsAg transgenic mice and their matched wild-type (WT) counterparts were treated with AFB₁ on day 7 (D7) after birth, when the liver is at the proliferating stage (Shupe and Sell, 2004), or at adulthood at 6 or 12 months of age (Fig. 1A). Efficacy of AFB₁ injection to induce DNA adducts was confirmed by southwestern blotting using specific antibodies against the AFB₁-formamidopyrimidine adducts (supplementary material Fig. S1). Approximately 15-25 mice per group were collected periodically at 3-month intervals, up until 15 months, when all mice were finally sacrificed (Fig. 1A). A total of 727 mice, with about equal numbers of male and female mice, were used to determine the effect of gender on HCC formation (supplementary material Table S1A and B).

Livers and all other organs were collected, together with serum, from all animals at all time points (Fig. 1B). In the event that mice developed liver adenoma or carcinoma, distinct tumor nodules were collected separately from the surrounding liver tissue. Initial histological analysis was performed using livers and all organs from at least six mice per group. Liver tissues were stored frozen in nine separate parts as a reservoir bank for subsequent RNA and genomic DNA analysis.

AFB₁ exposure synergizes with hepatitis in inducing cancers only when the liver is proliferating

Time-course analysis indicated that corn oil-treated control HBsAg mice generally developed overt liver lesions beginning from 9-12 months, unlike WT mice, which were without any observable lesions until 15 months (Fig. 1C). By contrast, WT mice injected with AFB₁ on D7 developed fewer liver lesions. However, HBsAg mice injected with AFB₁ developed the highest number of liver lesions, macroscopically obvious from around 9 months of age (beginning from 6 months), consistent with epidemiological data (Wild and Montesano, 2009). Analysis of WT mice injected with AFB₁ at 6 or 12 months did not show any obvious liver lesions until 15 months (Fig. 1D). However, HBsAg mice injected with AFB₁ at adulthood showed an increase in liver lesions compared with untreated mice (compare Fig. 1C and D), indicating that the quiescent non-proliferating liver (in the case of WT mice) does not provide a permissive environment for liver lesions to be induced by AFB₁ exposure.

To obtain a detailed understanding on the type of lesions induced, we scored the liver nodules by size. Analysis of large nodules that are ≥0.5 cm indicated that the corn oil-treated control HBsAg transgenic mice themselves developed nodules at 9 months of age, in contrast to WT mice injected with AFB₁ (Fig. 2A). However, such lesions were visible even at 6 months in HBsAg mice injected with AFB₁. In general, the average numbers of nodules/mouse were much greater in HBsAg mice injected with AFB₁, followed by control HBsAg mice and WT mice injected with AFB₁ [mean nodule numbers (≥0.5 cm)/mouse at 15 months were 2, 1 and 0.2, respectively] (Fig. 2A, top panels). Consistently, the size of nodules mirrored the pattern of nodule abundance in each category. For ease of enumeration, we qualified all small nodules smaller than 0.5 cm as 0.2 cm. Control HBsAg mice developed some large nodules (ranging from 0.2 cm to 2.2 cm, average of 0.6 cm, at 15 months), with most being small (Fig. 2A, bottom panels). AFB₁ injection alone of WT mice led to fewer nodules that averaged 0.8 cm. By contrast, the bulk of the nodules in HBsAg mice injected with AFB₁ were larger, ranging from 0.2 to 2.5 cm, with an average size of 1.0 cm (Fig. 2A, bottom panels). Detailed statistical analysis of the number of mice that contained liver nodules confirmed the synergistic effects of HBsAg and AFB₁ over the individual factors, especially at 9 months after treatment (supplementary material Table S2A). Moreover, the hepatitis component provided by the HBsAg mice was more tumorigenic than the AFB₁ exposure, as there was a significant difference at 9 months in the HBsAg mice when compared with control mice, whereas it took longer for the AFB₁-treated WT mice to develop tumors (supplementary material Table S2B).

Abundance and size of liver lesions in adult HBsAg mice exposed to AFB₁ also increased over time after treatment (Fig. 2B). Importantly, comparison of HBsAg mice treated with AFB₁ at 6 months and the control (corn oil) group indicated that AFB₁ treatment indeed accelerated the liver lesions in the HBsAg background of older mice (percentage of mice with lesions ≥0.5 cm at 15 months: untreated versus AFB₁ at d7 versus AFB₁ at 6 m: 58.8 versus 73.0 versus 72.0) (Fig. 2B; supplementary material Fig. S2). Although AFB₁ injection at 12 months did not significantly accelerate the number of nodules present (percentage of mice with lesions ≥0.5 cm at 15 months – untreated versus AFB₁ at 12 m: 58.5 versus 51.4), the number of mice having ≥2 nodules was higher (4/19, 21% for D7 injections versus 13/35, 37% for 12 M injection) (supplementary material Fig. S2).

Altogether, these data indicate that AFB₁ exposure requires livers to be at a proliferative state for lesions to form, as adult WT mice without any perturbations were more resistant to liver lesions than HBsAg mice. Moreover, synergistic effect of hepatitis and AFB₁ was evident regardless of age of exposure.

Gender bias in formation of liver lesions

Gender bias has been noted in HCC formation in humans, with males being predisposed to the disease (Li et al., 2012). Similarly, liver lesions in our model also showed a gender bias in all cases. Male WT neonatal mice exposed to AFB₁ had a clear tendency to develop lesions earlier than female mice (Fig. 2C). Similarly, lesions in HBsAg mice developed much earlier in males than in females (P=0.0443 at 12 months). However, by 15 months, female mice had an about equal frequency of liver lesions as male mice (P=0.226). A similar trend was seen in adult HBsAg mice injected with AFB₁ at 6 months, with male mice being more susceptible to liver lesion much earlier than female mice; although by 15 months, almost all mice of both genders had similar incidence rates (P=0.004 and P=0.003 for 9 and 12 months, respectively, and P=0.424 at 15 months) (Fig. 2D).

Histological analysis of mouse and human HCCs

Histological analysis was performed using 6-14 mice per group to determine the status of the liver lesions classified as regenerative foci (F), adenoma (A) or carcinoma (H). Control WT mice showed signs of degenerative changes only at 15 months of age (Fig. 3A). However, livers from control HBsAg mice were composed of large hepatocytes which varied in size and shape, with nuclear pleomorphism and vacuolated cell foci (Fig. 3B). AFB₁ injection did not alter the morphology of the normal livers of mice significantly as compared with the effects of HBsAg (compare Fig. 3A with C, and Fig. 3B with D). HCCs arising in the various groups of mice at 15 months indicated that they were histologically similar. Comparison of human HCC with mice HCCs from this study also indicated that they were strikingly similar (Fig. 3E).

Based on the histological analysis, control HBsAg mice generally developed only foci or adenomas (2 and 4 out of 12, respectively, at 15 months of age) (Table 1A), as has been previously described (Ghebranious and Sell, 1998). AFB₁ exposure at D7 generally led to foci and carcinoma formation in WT mice (5 and 1 out of 10, respectively, at 15 months of age). By contrast, HBsAg mice injected with AFB₁ had more cases of carcinoma, together with adenoma and foci of alteration (3 out of 11 mice each), indicating that the severity of liver lesions correlates with the synergistic effects of hepatitis and AFB₁, concurring with the gross morphological analysis (Fig. 1C,D). Similarly, adult HBsAg mice injected with AFB₁ also developed carcinomas in addition to adenomas and foci, much more frequently compared with the group without AFB₁ injection (Table 1B). By contrast, WT mice did not develop any lesions, again highlighting the ability of AFB₁ to induce lesions only on an HBsAg background at adulthood.

Transcriptomic analysis identifies HCC-specific molecular signature independent of etiological factors

Given that the tumors that arise in the WT or HBsAg mice treated with AFB₁ appear histologically similar, we next evaluated whether the molecular profiles of tumors or normal tissues from the various etiologies would also be similar. To this end, we performed an mRNA profiling experiment using samples from the following categories of 15-month-old mice: histologically normal liver tissues from control or AFB₁-treated WT mice and control HBsAg mice; normal tissue adjacent to the tumor nodules from AFB₁-treated HBsAg mice; and, finally, tumor nodules from control HBsAg mice, or from AFB₁-treated WT or HBsAg mice [total of seven categories: WT_Oil_N (n=3), WT_AFB_N (n=2), HBsAg_Oil_N (n=3), WT_AFB_T (n=3), HBsAg_Oil_T (n=2), HBsAg_AFB_AN (n=3) and HBsAg_AFB_T (n=4)].

A principal component analysis (PCA) indicated that normal livers from the three groups (etiologies) clustered together (n=8), distinct from the tumor group (n=9) (Fig. 4A, left panel). Normal tissue adjacent to the tumor nodules from AFB₁-treated HBsAg mice formed a distinct intermediate category, indicating a progressive nature of the changes from normal liver to nodules. Unsupervised hierarchical clustering based on 256 probes (s.d.≥1) showed that tumors and normal livers were clustered separately, with subtle differences within each group (Fig. 4A, right panel). Closer inspection of the histologically normal liver cluster indicated subtle clustering based on genotype, attributable to the presence of the HBsAg transgene. Similarly, tumor samples from all three categories also clustered together, being broadly similar. However, closer inspection revealed that the tumor samples were more distinctly clustered, based not only on genotype but also on AFB₁ exposure (supplementary material Fig. S3). The AFB₁ tumor signature appeared to be distinct from the HBsAg tumor signature, and tumors from HBsAg mice treated with AFB₁ had a mix of both signatures (supplementary material Fig. S3), suggesting that treatment of AFB₁ can accentuate a distinct pathogenic process in tumors.

Differential analyses between HBsAg_Oil_T and HBsAg_Oil_N revealed 11 genes that passed Bonferroni correction, constituting an HCC signature, being able to distinctly segregate the tumors from normal tissues regardless of etiology and inducing factors (Fig. 4B). Two genes, TTC39B (Ttc39b – Mouse Genome Informatics Database) and CYP2C37 (Cyp2c37), were downregulated in tumors, whereas nine genes (CNIH4/Cnih4, CAV2/Cav2, CDC63, ELOVL7/Elovl7, NUCB2/Nucb2, SYT14/Syt14, FKBP11/Fkbp11, 10339559 and CAV1/Cav1) were upregulated. Seven of the genes (all except one were overexpressed in the mouse model) were mappable to human microarray transcripts.

We therefore examined whether the HCC gene signature identified in the mouse models here is capable of segregating human HCCs of various etiologies. To this end, we examined three published HCC datasets: one from a European population (that lacks AFB₁ exposure and HBV; GEO dataset 54236); one from the hepatitis-based Chinese population (GSE25097), and one from the Singaporean Chinese hepatitis-based population lacking AFB₁ (E-MEXP-84). Interestingly, the gene expression signature was able to distinctly segregate the HCCs from normal tissues in all three cases (Fig. 5A,B; supplementary material Fig. S4A), indicating the robustness of the signature that was not limited by the etiological factors. Importantly, this expression signature was unable to clearly segregate other human cancers such as breast (TCGA dataset) (Fig. 5C) or colon adenocarcinoma (TCGA dataset) (Fig. 5D). To further assess the specificity of this mouse-derived gene expression signature for human HCCs, a support vector machines (SVM) model was generated using the mouse HCC data and then tested on the other human cancer datasets. As shown in Fig. 5E, the mouse HCC model could predict well for human HCCs but not for other cancers, such as breast and colon. For instance, the mouse gene expression signature was able to discern out all 246 HCC tumors from one dataset and 78 of 81 HCCs from another, resulting in over 96% sensitivity. By contrast, the signature only predicted 12 out of 155 colon cancers and 147 out of 582 breast cancer samples, indicative of ∼7.7-27.8% specificity, markedly lower than the values obtained with the HCC datasets.

We also examined the ability of this gene expression signature to distinguish between transformed and untransformed hepatocytes, by analyzing the datasets from several hepatic cell types. Interestingly, our HCC signature was indeed able to segregate out the HepG2 transformed cells from the primary HepRG cells or the embryonic stem cell-derived primary hepatocytes, regardless of AFB₁ treatment (GSE40117) (supplementary material Fig. S4B). Altogether, the transcriptomic data analyses demonstrate the identification of a gene expression signature from the mouse models that is able to distinguish HCCs from untransformed primary hepatocytes, regardless of the etiological factors, supporting the specificity and relevance of the mouse model to human HCC.

Distinct pathways lead to hepatocarcinogenesis based on etiological factors

Although we have noted the presence of a common gene signature that is able to distinguish HCCs regardless of etiological factors, suggesting a convergence of signaling cascades in HCC formation, we also observed that AFB₁ exposure or the presence of HBsAg has some molecularly distinguishing features (supplementary material Fig. S3). This highlights that the pathways perturbed by these etiological factors are different, alluding to the existence of varying pathogenic processes that lead to HCC formation. We therefore investigated this phenomenon further by analyzing the differences in gene expression from tumors from the three groups (AFB₁ or HBsAg alone or in combination). Tumors arising in the presence of HBsAg alone led to the lowest number of differentially expressed genes (compared with normal livers), as opposed to those induced by AFB₁ (203 versus 381 genes) (Fig. 6A). Unsurprisingly, presence of both etiological factors led to the highest number of changes (716 genes altered). Furthermore, the tumors arising due to AFB₁ or HBsAg exposure alone were more dissimilar from each other than from their common counterparts, further alluding to different pathways perturbed during the carcinogenic process. Nonetheless, 80 genes were commonly perturbed in all tumors, which contained some of the genes that constituted the HCC gene expression signature.

Using these differentially expressed genes, we tested for biological pathways that were affected. As expected, the profiles from the three different groups revealed that upstream regulators of the differentially expressed genes identified by the pathway analysis program (Fig. 6B), and biological pathways they affect (Fig. 6C) could be distinguished in each group. Top upstream regulators deregulated included the cytokines TgfB1, IL-13, TNF, Tp53 and FoxM1, which were regulated mainly by the combined presence of HBsAg and AFB₁; CDKN1A, estrogen receptor, HDAC1 and ETS1, which were deregulated by the presence of HBsAg alone; and UXT (regulating androgen receptor), ESR2 (estrogen receptor 2 signaling), ERBB1 and P300, which were affected by AFB₁ alone. An interesting observation was that the tumors arising from HBsAg+AFB₁ also had different signaling pathways activated compared with individual risk factors, suggesting that the molecular physiology of the tumors is different. Moreover, gross analysis of top canonical pathways affected in these three groups of tumors indicated that there were common as well as distinct pathways affected, based on the etiological factors. There were more commonalities between tumors arising from both HBsAg/AFB₁ or from AFB₁ alone, whereas the HBsAg-derived tumors were very distinct (Fig. 6C; supplementary material Fig. S5). Future in-depth analyses of larger sample size will provide further details based on the etiological factors.

Finally, to evaluate the overlap between the biological pathways and processes affected in mouse tumors with human HCCs samples, we compared the transcriptome profiles of human hepatitis-based HCC samples (Wang et al., 2007) with the HBsAg_AFB_T versus WT_Oil_N, which revealed over 50% common upstream regulator pathways (Fig. 6D), highlighting the commonality in pathogenesis of the liver lesions. Consistently, most of the top canonical pathways, such as LPS/IL-1-mediated inhibition of RXR function, nicotine, acetone and buproporin degradation pathways, and PXR/RXR activation, were similar in both mouse and human tumors (supplementary material Fig. S6). Similarly, gene ontology analysis indicated that the biological processes (metabolic and cellular processes), molecular functions (electron carrier activity, binding and catalytic activity) and cellular components (extracellular region and cell junctions) affected by tumor development were similar in all mouse tumors and the human HCCs (supplementary material Fig. S7). This corroborated further that liver tumors from these mouse models are indeed molecularly similar and resemble the human HCCs. Taken together, these data collectively indicate that whereas the biological processes affected due to the various etiological agents en route to hepatocarcinogenesis do differ, highlighting specificity of the inducing agent, there is also a significant degree of overlap, indicative of culmination in a common mode for the transformation of hepatocytes.

Mice and treatment

HBsAg transgenic mice, C57BL/6J-Tg(Alb1HBV)44Bri/J (Jackson Laboratory), were crossed to C57BL/6J wild-type (WT) mice to obtain HBsAg and WT littermate control mice, which were given a single intraperitoneal injection of AFB₁ [6 µg/g body weight (19 nmol/g) in trioctanoin containing 6% DMSO or corn oil] on day 7 (D7) after birth, as described (Tong et al., 2006), or at 6 or 12 months (M) of age.

Liver samples (nodules and un-inflamed normal areas), all other major organs and serum (by cardiac puncture) were collected at 3-month intervals until 15 months, for histological analysis, RNA extraction, genomic DNA and serum extraction, and were kept as frozen samples or fixed in 4% formalin. All sacrifice was carried out between 11 am and 1 pm for all collections, livers were photographed after sacrifice, and nodules ≥0.2 cm in diameter were counted. All animal experiments were approved by and performed in accordance with the guidelines of the SingHealth Animal Care and Use Committee. Histological analysis is described in the supplementary material.

Microarray analysis

Between two and four samples per group were used for transcriptome analysis, using the Affymetrix GeneChip Mouse 1.0 ST arrays, as described (Lee et al., 2012). A nomenclature of genotype/treatment/tissue was used to denote the 7 groups: WT_Oil_N (n=3), WT_AFB_N (n=2), HBsAg_Oil_N (n=3), WT_AFB_T (n=3), HBsAg_Oil_T (n=2), HBsAg_AFB_AN (n=3), HBsAg_AFB_T (n=4), in which N, AN and T refer to normal, adjacent normal, or tumor tissue, respectively.

PCA and heat map analyses: microarray data were processed with default preprocessing settings using RMA and quantile normalization. Principal component analysis (PCA) was performed using all 35,556 probes. Unsupervised hierarchical clustering was generated using probes with standard deviation (s.d.)≥1. Differential analyses were performed between two groups using ANOVA, and probes were selected based on P<1e⁻⁰⁶ (Bonferroni correction). All analyses were performed on the Partek Genomic Suite (Partek).

Pathway analyses: Ingenuity Pathway Analysis Software (www.ingenuity.com) was used to determine canonical pathways and upstream regulators enriched. Differentially expressed genes in each group were selected using a fold change>2 and FDR<0.05. Upstream regulators of these genes were subsequently elucidated from IPA and compared. Pathways analyses were performed on the upstream regulators.

Human genomics data: Human HCC transcriptome profiles were obtained from three sets of HCCs and matched with normal sets as indicated. Human breast and colorectal datasets were from TCGA. Probes between mouse and human transcripts were mapped based on Refseq gene symbol, resulting in 13,359 common transcripts. Differentially expressed genes were elucidated using ANOVA between tumor and normal samples in each genomics dataset. Our mouse microarray data have been deposited at NCBI/GEO, with the GEOSET number GSE54054.

For analysis of specificity, we used the support vector machine (SVM) (www.chibi.ubc.ca/gist) as modelling method with default settings.

Statistical analysis

Fisher's exact test was used to analyze the difference in incidences of liver nodules in different experimental groups. Chi-square test was used to analyze the biggest nodule size of each mouse in different experimental groups.

Histology

Histological analysis was performed on 3-µm-thick sections stained with hematoxylin and eosin (H&E), as described (Tong et al., 2006). Foci of cell alteration have no obvious disruption of the liver architecture, and the affected hepatocytes merge with adjacent hepatocytes without compressing adjacent normal parenchyma. Hepatocellular adenomas are small, formed from closely packed cells, generally larger than the surrounding liver cells without great variability in size. HCC is represented either as a trabecular or solid form compressing adjacent normal parenchyma, with pleomorphic nuclei and aberrant mitotic figures. Human HCC slides were obtained from Peking Union Medical College (Beijing, China).