Background
Hepatocellular carcinoma (HCC) accounts for 85 - 90% of liver malignancies and is the second cause of cancer death in the world [
1]. Although many progresses have been achieved on HCC research, it still needs a better understanding of the molecular and regulatory mechanisms involved in HCC [
2]. HCC is often arisen by several risk factors including hepatitis virus B or C (HBV or HCV) infection, chemical damage and chronic excessive alcohol intake and so on [
3]. The occurrence of HBV induced HCC attributes to HBV proliferation and DNA integration into host genome to initiate the malignant proliferation and transformation [
4]. The HBx (HBV regulatory x protein) of hepatitis B virus plays a crucial role in hepatocarcinogenesis by transcriptional activation, driving deregulated cell cycle progression, modulation of apoptosis and inhibition of nucleotide excision repair of damaged cellular DNA [
5]. While chemical induced HCC leads to high levels of DNA damage and fails to block the cell cycle before the damaged DNA repaired [
6]. In recent years, chemical-induced and virus-induced HCC mouse models are widely used for the studies of pathogenesis and drug treatment of HCC [
6,
7]. However, the differences of gene expression and regulation in these two HCC models have not been compared, which is an important issue for model selection in HCC studies.
Diethylnitrosamine (DEN), a DNA alkylating agent, is widely used to induce liver cancer in rodent model with high success rate and similarity to human HCC [
8]. Recently, DEN-induced rodent HCC models have been used to investigate the pathogenesis, prevention and treatment of liver cancer, which include evaluating miRNA functions, exploring the antitumor effects of drugs and identifying biomarkers and therapeutic targets etc. [
9,
10]. HBx is a 154-amino acid hepatitis B viral protein which controls the level of HBV replication [
11]. As early as 1991, Kim, et al. constructed transgenic mouse model harboring HBx gene and firstly found this gene can specifically induce liver cancer [
12]. Lu, et al. used HBx to induce HCC in transgenic male C57 mice and identified some common regulators in HCC [
13]. Ye, et al. studied the synergistic function of Kras mutation and HBx in mice HCC [
14].
In recent years, microRNAs (miRNAs) and transcription factors (TFs) have drawn extensive attention in cancer research. They play pivotal roles in proliferation, differentiation, invasion and metastasis of tumors. Many miRNAs were reported as important regulators in HCC. For example, miR-221 promotes human HCC development and its silencing contributes to suppressing tumor properties [
15]. MiR-214 and miR-375 suppress the proliferation of HCC cells by directly targeting E2F3 and AEG-1 respectively [
16,
17]. TFs are paramount regulators in controlling gene expression in living organisms [
18]. For example, PPARs contribute to the pathogenesis in cell cycling, anti-inflammatory responses and apoptosis [
19]. Aberrant high expression of STAT3 may promote HCC migration and invasion [
20]. To further explore the molecular mechanisms in complex diseases, regulatory networks are widely studied. Our previous studies have revealed the miRNA and TF co-regulatory motifs are pervasive regulatory models in biological processes and diseases [
18,
21‐
23]. Through network analysis, the complex regulatory relationships in diseases will be illuminated on systematic level and the key regulators may be identified.
In this study, we applied bioinformatic approaches to analyze the high-throughput and microarray data of DEN and HBx HCC mice models. We aim to analyze the gene expression features and core regulatory factors in HCC from these two models, and reveal the similarities and differences between them as well as compare gene expression profiles in tumor, para-tumor and normal tissues in each model.
Methods
Sequencing of DEN model and data collection of HBx model
All of the experimental mice were male and purchased from Hubei Research Center of Laboratory Animals (Wuhan, China) and in the C57 genetic background. To induce HCC in mice, we used DEN as the inductive agent. In this study, all mice were divided into two groups: DEN group and control group. In DEN group, mice were fed with basal diet every day and given DEN treatment at a dose of 165 mg per kg body weight in sesame oil through oral once a week for 10 weeks, then drug treatment were stopped and the mice were fed with diet only until 30 weeks. In control group, mice were fed with basal diet only until 30 weeks. In order to determine whether the liver tissues have suffered cancerous changes, histopathologic examinations were conducted via microscope (Additional file
1: Figure S1). All of the animal procedures were performed in accordance with the Ethics Committee on Animal Experimentation of the Huazhong University of Science and Technology (Wuhan, China) and the NIH Guide for the Care and Use of Laboratory Animals (8th edition, 2011).
To detect gene and miRNA expression, high-throughput sequencing technologies were used: RNA-seq for detection of expressed transcripts and small RNA-seq for detection of miRNAs respectively. Samples of tumor and para-tumor liver tissues were excised from the same lobe of the liver. For control group, RNA was isolated from liver samples obtained from age-matched healthy mice. Three biological replicates were sequenced for each group. For RNA-seq, ribosome RNA was removed first and pair-end 150 bp sequencing were carried out through Illumina Hiseq 3000, while for small RNA-seq, single-end 50 bp sequencing were performed with Illumina Hiseq 2500. All of the sequencing and data filtering works were done by Ribobio company (Guangzhou, China).
The microarray expression data of HBx model were downloaded from Gene Expression Omnibus (GEO) (
http://www.ncbi.nlm.nih.gov/geo/, GEO accession: GSE15251). This model used Hepatitis B virus X antigen (HBx) to induce HCC in transgenic male C57 mice. The microarray samples of tumor, para-tumor and normal tissues were from the 16-month-old mice and the experimental conditions were similar with our DEN model.
Gene expression and differential expression analysis among tumor, para-tumor, and normal tissues
In DEN model, RNA-seq reads were firstly quality-checked by fastqc software, then HISAT2 (version: 2.0.5) was used for mapping sequencing reads to mouse genome GRCm38, and StringTie (version: 1.2.2) was used to assemble the RNA-seq alignments into potential transcripts based on the reference sequences and calculate the abundance of transcripts [
24]. The expressed levels were estimated as FPKM (the number of Fragments (reads) per kilobase of transcript per million mapped reads). The differentially expressed genes (DEGs) were identified using NOISeq [
25] with thresholds FDR < 0.01 and |fold-change| ≥ 1.5. For small RNA-seq, firstly reads were aligned to the mouse genome, mouse miRNA and miRNA precursor data using Bowtie2 [
26], then the expression of miRNAs were estimated as RPM (reads per million mapped reads). The differential expression of miRNAs were calculated by DEGseq [
27] and edgeR [
28], the threshold was also set as |fold change| ≥ 1.5 and required their RPM ≥ 20 in at least one tissue. Results of two methods were pooled together subsequently. In HBx model, DEGs were identified by NOIseq with FDR < 0.01 and |fold-change| ≥ 4.0.
The expression characteristics of two models were displayed by cumulative distribution function plot. The definition of cumulative distribution function is: F
x(x) = P(X ≤ x), where the right-hand side represents the probability that the random variable X takes on a value less than or equal to x [
29]. And the top 5% expressed genes were extracted for function enrichment and comparison. To survey the functions of genes, we used DAVID (
https://david.ncifcrf.gov/tools.jsp) for GO (gene ontology) and KEGG (Kyoto Encyclopedia of Genes and Genomes) enrichment analysis. The top ranked enrichment results with high significant levels (
p-value <0.01) were selected for discussion and presented by bubble plots using R package ‘ggplot2’. In the bubble plot, rich factor is calculated as the ratio of gene counts that mapped to a certain pathway and the total gene number of that pathway.
To reveal the gene expression difference between tumor and non-tumor tissues in two models, the pairwise comparisons of DEGs were carried out among tumor, para-tumor and normal tissue. Venn diagrams of DEGs were drawn and pathway enrichment in each model were implemented as described above.
Gene function comparison between DEN model and HBx model
To further compare similarities and differences between two models, we firstly analyzed the similarities and differences between tumor tissue in DEN model and HBx model, and then focused on the para-tumor tissues of two models. The common pathways in two models were shown by bar charts, and the different pathways were exhibit by bubble plots. What’s more, heatmaps were used to recognize the gene expression patterns among three tissues.
Identification of miRNA and TF targets and construction of regulatory networks
Networks were constructed using the differentially expressed data by reference to the collected validated data from several databases. For miRNA, we collected experimentally verified targets as described in our previous review paper [
22]. The data mainly include the overlapped results from miRWalk2, miRecords4, miRTarbase6 and Tarbase7. TF targets were extracted from TRANSFAC database. The regulatory interactions between miRNA/TF and genes were obtained via Python script. Finally Cytoscape (Version 3.2.1) were used to visualize the networks.
Discussion
In this study, we investigated the highly expressed genes in both DEN and HBx mice models, and identified the DEGs through comparison among tumor, para-tumor and normal tissues. The gene expression profiles and function enrichment of different tissues demonstrate different characteristics and functional specificities of three tissues. Moreover, through comparison of two models, we found both similarities and differences exited in gene expression. Finally, we constructed miRNA and TF regulatory networks and identified several key regulators from the networks, as well as revealed some synergistic regulations in HCC models.
Through comparison of tumor, para-tumor and normal tissues, we suggest that para-tumor liver tissues should be regarded as inflammatory tissue rather than normal tissue. But in the study of human solid carcinomas, due to the sampling limitations, researchers usually use comparison of tumor and adjacent non-tumor tissues of patients to illustrate the molecular mechanism of cancers. In fact, from our study, the maximal effects of immune responses and inflammatory responses were detected in para-tumor tissues rather than tumor tissues or normal tissues. While the normal tissues were characterized with lipid mechanism and oxidation-reduction, and the tumor tissues enriched in cancer signal transduction, immune and inflammatory responses (Figs.
2 and
3). Till to now, only a few articles have reported the comparisons among three tissues. Chandran et al. investigated the gene expression differences among prostate cancer tissue, adjacent prostate cancer tissue and normal prostate tissue from cancer free organ donors. They found that the tumor vs normal expression profile was more extensive than the tumor vs adjacent normal profile, when using normal as the baseline, similar genes were up-regulated in both tumor and adjacent tissue, and these genes can not be identified from tumor vs adjacent normal profile [
38]. In other words, para-tumor tissue is somewhat similar to tumor tissue. Also in this study, the tumor and para-tumor tissues emerged with higher response of inflammatory and immune than normal tissues, which was in agreement with previous report that inflammation was closely related to cancer. Grivennikov and Karin reported that about 20% of all cancers generated in association with chronic inflammation, and most of the solid tumors contain proinflammatory cytokines which can mediate immune responses and effect tumor initiation, growth and progression [
39].
In previous studies, both DEN- and HBx-induced mice liver cancer models have been used to investigate the pathogenesis of HCC, they have become the two major concerns in liver cancer research [
9,
10,
13]. It has been verified that the cancers of DEN and HBx induced in mice are very similar to human HCC [
8,
40]. Comparison of these two representative models can accurately reveal the mechanism of HCC carcinogenesis. But so far, no studies have investigated the difference of molecular mechanisms between drug-induced and virus-induced HCC. Through comparison of the same tissues between two models, we found that immune response, inflammatory responses, metabolic process, and oxidation-reduction process are the common activities in both two models. There are also several different enriched pathways in two models. The most significant different pathway in DEN model is chemical carcinogenesis while in HBx model they are retinol metabolism, wound healing, positive regulation of gene expression and positive regulation of protein kinase B signaling pathway. These differences may due to the different induction of two model (DEN and HBx). Another obviously difference is that the HBx model appears with more cancer related activities, for example, cell signal transduction and communication, cell cycle and programmed cell death. This may due to the long/short period of induction. DEN induction is more acute and can resulted in serious injury even death before all the cancer pathways significantly emerged, while HBx induction is chronic and mice can live longer until all the cancer pathways obviously emerged. It was reported that signaling molecules such as EGFR, Ras, PKC, AKT/PKB and mTOR, were found to play important roles in human cancer cells and were involved in cell proliferation, differentiation and survival [
41]. PDCD4 (Programmed Cell Death Protein4) gene can be inhibited by miRNA-21 and results in the failure of programmed cell death, and finally enhances cell proliferation in HCC and breast cancer [
42,
43]. These molecules or their subtypes were also found highly expressed in the models in this study.
At the same time, we also identified differentially expressed miRNAs and TFs from the data, and constructed two major regulatory networks. From network analysis, miR-381-3p, miR-142a-3p, miR-214-3p and Egr1, Atf3, Klf4 were identified to be the most important regulators in HCC. These are important tumor suppressors which can be considered as diagnosis biomarkers and treatment targets. What’s more, the synergistic regulation analysis indicated that Cdkn1a is the most import target gene of multiple TFs in HCC (Fig.
6c). It can encode a potent cyclin-dependent kinase inhibitor and inhibits the activities of several cyclin-cyclin-dependent kinases, not only functions as a regulator of cell cycle progression at the G1 pahse, but also involved in the regulation of transcription, apoptosis, DNA repair and cell motility [
44]. Previous study has reported that the expression of this gene is tightly controlled by the tumor suppressor protein p53 [
45]. In this study, we found it was also synchronously regulated by the TFs Egr1, Atf3, Klf4, Vdr and Ar, this result enriched the recognition of regulatory mechanisms of this gene. Besides, the TSG Tgfbr2 is also a top ranked gene that was synchronously regulated by multiple TFs. Tgfbr2 is a key molecule in
TGFbeta signaling pathway and was found to prevent the formation of hepatocellular carcinomas and cholangiocarcinoma development [
46]. Taken together, through network analysis we identified several pivotal regulators and cancer related genes, this work provided comprehensive information for gene regulation involved in HCC.
Conclusions
Through gene expression and function analyses, we found that both DEN and HBx model shared common pathways including lipid metabolism, oxidation-reduction process, immune process and inflammatory responses. Through tissue comparison, we suggest that the para-tumor should be recognized as inflammatory tissue. Network analysis revealed that miRNAs such as miR-381-3p, miR-142a-3p, miR-214-3p and TFs such as Egr1, Atf3 and Klf4 are the hub regulators in HCC. These results will help to increase understanding of molecular mechanisms of HCC.