Inhibition of histone methyltransferase G9a attenuates liver cancer initiation by sensitizing DNA-damaged hepatocytes to p53-induced apoptosis

While the significance of acquired genetic abnormalities in the initiation of hepatocellular carcinoma (HCC) has been established, the role of epigenetic modification remains unknown. Here we identified the pivotal role of histone methyltransferase G9a in the DNA damage-triggered initiation of HCC. Using liver-specific G9a-deficient (G9aΔHep) mice, we revealed that loss of G9a significantly attenuated liver tumor initiation caused by diethylnitrosamine (DEN). In addition, pharmacological inhibition of G9a attenuated the DEN-induced initiation of HCC. After treatment with DEN, while the induction of γH2AX and p53 were comparable in the G9aΔHep and wild-type livers, more apoptotic hepatocytes were detected in the G9aΔHep liver. Transcriptome analysis identified Bcl-G, a pro-apoptotic Bcl-2 family member, to be markedly upregulated in the G9aΔHep liver. In human cultured hepatoma cells, a G9a inhibitor, UNC0638, upregulated BCL-G expression and enhanced the apoptotic response after treatment with hydrogen peroxide or irradiation, suggesting an essential role of the G9a-Bcl-G axis in DNA damage response in hepatocytes. The proposed mechanism was that DNA damage stimuli recruited G9a to the p53-responsive element of the Bcl-G gene, resulting in the impaired enrichment of p53 to the region and the attenuation of Bcl-G expression. G9a deletion allowed the recruitment of p53 and upregulated Bcl-G expression. These results demonstrate that G9a allows DNA-damaged hepatocytes to escape p53-induced apoptosis by silencing Bcl-G, which may contribute to the tumor initiation. Therefore, G9a inhibition can be a novel preventive strategy for HCC.


Introduction
Hepatocellular carcinoma (HCC) accounts for the majority of primary liver cancers and is the fourth most common cause of cancer-related deaths worldwide 1 . HCC occurs due to a variety of risk factors, including viral hepatitis, alcoholic and nonalcoholic fatty liver disease, carcinogen exposure, and metabolic liver diseases. Despite recent advances in the pathophysiology and treatment of HCC, the overall prognosis remains poor. A better understanding of the mechanisms of hepatocarcinogenesis will provide novel and efficacious targets for treatment of HCC.
HCC develops through a complex multistep process in which various genomic abnormalities are acquired. In addition to genetic alterations, accumulating evidence has highlighted the key role of epigenetic dysregulation in liver cancer pathogenesis 2 . Consistently, recent genome analysis has revealed a high number of genetic disturbances in the genes related to epigenetics in HCC 3,4 . Global DNA hypomethylation and promoter hypermethylation of specific tumor suppressor genes are wellcharacterized epigenetic changes in human carcinogenesis 5 . In addition to DNA methylation, aberrant expression of histone modifiers has also been implicated in the tumor development. Histone methylation critically determines the chromosomal structure as well as accessibility to transcription factors. 6 Disturbances in histone methylation associate with HCC development [7][8][9] . Therefore, modulating the enzymatic activities of histone modifiers might be a potential therapeutic strategy to prevent HCC.
G9a (also known as euchromatic histone-lysine Nmethyltransferase 2, EHMT2) is a SET domain-containing protein that catalyzes specifically the di-methylation of histone 3 lysine 9 (H3K9me2), which is a prominent epigenetic marker of transcriptional repression 10 . G9a plays important roles in diverse cellular processes, such as proliferation, differentiation, senescence, and autophagy 11 . The dysregulation of G9a and aberrant levels of H3K9me2 are involved in different types of human cancers [12][13][14][15] . As for HCC, G9a promotes tumor progression by silencing tumor suppressor genes or enhancing epithelial-mesenchymal transition 16,17 , and its inhibition reduces tumor aggressiveness 18,19 . In contrast, how G9a contributes to the development or initiation of HCC has not yet been investigated in vivo.
In chronic hepatic injury caused by viral infection or excessive fat accumulation, DNA damage from produced reactive oxygen species is a critical pathogenic factor for liver carcinogenesis 20 . Given the role of G9a for DNA damage repair and cell survival 21 , we hypothesized that G9a might be involved in DNA damage-induced liver cancer initiation. In this study, we demonstrate that liverspecific G9a-deficient (G9a ΔHep ) mice suppress HCC development triggered by diethylnitrosamine (DEN) and the pivotal role of G9a in DNA damaged hepatocytes.

G9a is frequently upregulated in human HCC
First, we analyzed the expression levels of histone modifiers from 373 HCC cases identified in The Cancer Genome Atlas (TCGA) dataset. From the 82 identified histone modifiers, many histone methyltransferases were upregulated; consistent with previous papers, SETDB1 was the most significantly upregulated 9,22 (Figs. 1A, S1). G9a was also identified as one of the most upregulated genes, with high expression levels (Z-score > 2) in 16.6% (62/373) of the cases (Fig. 1A). Expression levels of G9a were not significantly different among etiologies but rather tended to increase according to the pathological malignancy grade of HCC (Fig. 1B). G9a expression was also higher in many HCC cell lines than in normal hepatocytes (Fig. 1C). In addition, in an original cohort consisting of 40 pairs of HCC and adjacent NT livers in our institution, we successfully validated the upregulation of G9a in HCC compared with NT livers in almost half of the cases (Fig. 1D). In the TCGA cohort, copy number gain of G9a gene was detected in 4.29% (16/373) of the cases, and it was positively correlated with higher G9a mRNA expression levels (Fig. 1E). These data confirm the hypothesis that accumulated genomic alterations in advanced HCC are linked to G9a upregulation as well as SETDB1 16,22 , and highlight the pathological significance of G9a in HCC. Of note, given the enzymatic roles of G9a and SETDB1 related to H3K9 methylation, the above findings suggest the molecular significance of H3K9 methylation in HCC pathogenesis. In contrast, however, the involvement of G9a and its H3K9 methylation activity in hepatocarcinogenesis in vivo remains unclear. Thus, we next examined the significance of G9a in liver tumor development using liverspecific G9a-deficient (G9a ΔHep ) mice.

Loss of G9a attenuates liver tumorigenesis in a carcinogen-induced HCC model
G9a ΔHep mice were established by crossing G9a flox/flox mice with albumin-Cre transgenic mice. To confirm efficient deletion of G9a in hepatocytes, the expression levels of G9a protein were measured in whole liver and isolated primary hepatocytes. Deletion of G9a was confirmed in G9a ΔHep mouse hepatocyte, while residual G9a expression was detected in the whole liver lysate of G9a ΔHep mice, probably from nonhepatocyte fractions ( Fig. 2A). Immunohistochemistry revealed that H3K9me2 levels in G9a ΔHep mouse hepatocytes nuclei decreased significantly, suggesting the critical role of G9a in the H3K9me2 modification (Fig. 2B). G9a ΔHep mice lineage was consistent with Mendel's laws of inheritance, and there were no significant issues with their growth. As for the liver, there were no differences in liver weight or histology between G9a ΔHep and wild-type (WT) mice (Fig. 2B).
To address the role of G9a in liver carcinogenesis, we injected a combination of DEN and hepatotoxin carbontetrachloride (CCl 4 ) solution into WT and G9a ΔHep mice. The chemical carcinogenic model mimics chronic inflammation and fibrosis of the human liver, and the developed tumor has the characteristics of human HCC 23 . In the inflammation-related liver-tumorigenic setting, G9a ΔHep mice displayed a profound reduction in tumor number and size (Fig. 2C). G9a ΔHep mice did not show significant changes in the pathology, proliferation, and apoptosis of the barely formed tumors (Fig. S2A), which leads us to suggest that G9a is critical in the initial stage of liver tumorigenesis. Supportive of this notion, decreased liver tumor initiation in G9a ΔHep mice was also observed Official journal of the Cell Death Differentiation Association under a purely genotoxic carcinogenic protocol without inflammation, namely, simple DEN administration without CCl 4 injection (Fig. S2B).
To confirm the tumor-initiating role for G9a, we investigated the effects of pharmacological inhibition of G9a in the above in vivo settings. Treatment with the G9a   inhibitor UNC0642 effectively reduced H3K9me2 levels in liver cells (Fig. 2D). Notably, treatment with UNC0642 resulted in significant attenuation of DEN-induced liver tumor development, regardless of subsequent CCl 4 administration (Fig. 2E). In contrast, the size of largest tumors was not efficiently affected by the administration of UNC0642 for 10 days, suggesting that the inhibition of G9a may predominantly affect the tumor initiation. These findings support our hypothesis that the enzymatic function of G9a is involved in the liver tumor initiation following DEN administration.  pericentral hepatocytes 24 . DNA damage caused by DEN is a significant factor for subsequent HCC development 25 . To assess the implication of G9a in DNA damage-triggered tumor initiation, both WT and G9a ΔHep mice were injected with a higher dose of DEN and were euthanized 48 h postinjection. Expression of the G9a protein was predominantly observed around the central vein in WT hepatocytes after DEN administration (Fig. 3A), suggesting that G9a is upregulated in DNA-damaged hepatocytes. Although the levels of DNA damage, detected by γH2AX staining and p53 induction, were identical in WT and G9a ΔHep mice livers, higher numbers of cleaved-Caspase 3-positive hepatocytes were observed around the central vein in G9a ΔHep mice than in WT mice (Fig. 3A). The serum levels of alanine aminotransferase (ALT) after DEN injection were markedly higher in G9a ΔHep mice than in WT mice (Fig. 3B). To exclude the possibility that the chemical activity of DEN was enhanced by G9a deletion, we examined the expression of Cyp2e1. As reported previously, Cyp2e1 was downregulated after DEN treatment in WT mice liver 26 . As we did not observe any significant difference in Cyp2e1 levels between WT and G9a ΔHep liver, we suggest that G9a deletion affects the cellular response against DEN-induced DNA damage, but not the chemical activity of DEN, in pericentral hepatocytes (Fig. 3C).
Given the epigenetic function of G9a as a transcriptional repressor, we hypothesized that G9a might regulate the expression of genes which determine cellular responses under DNA damage. To identify such genes regulated by G9a, comprehensive gene expression profiles of NT areas were analyzed in WT and G9a ΔHep livers. Based on the function of G9a as a transcriptional repressor, we focused on the genes upregulated in the G9a ΔHep liver. The top ten genes with the greatest fold inductions in G9a ΔHep liver are shown in Fig. 3D. Among them, Bcl2-like 14 (also known as Bcl-G), a pro-apoptotic Bcl-2 family member, was significantly upregulated in the G9a ΔHep liver. Quantitative real-time PCR confirmed that Bcl-G mRNA expression levels increased significantly in the NT areas of G9a ΔHep liver (Fig. 3E). Immunoblotting revealed that Bcl-G protein, together with cleaved-PARP and cleaved-Caspase 3, accumulate after DEN administration in G9a ΔHep liver but not in WT liver (Fig. 3F). In addition, immunohistochemistry showed that Bcl-G expression is markedly elevated in pericentral hepatocytes of G9a ΔHep mice (Fig. 3A). These findings suggest that G9a-dependent transcriptional regulation of Bcl-G gene suppresses apoptosis under DNA damage in murine liver. Furthermore, to study the relationships in the expression of G9a and Bcl-G gene, we measured the levels of G9a and BCL-G mRNA in human NT liver samples. Supporting the findings in murine liver, the expression levels of BCL-G were negatively correlated to that of G9a in human NT liver (Fig. 3G).

G9a inhibits DNA damage-induced apoptosis by regulating BCL-G expression
We hypothesized that the regulation of Bcl-G expression by G9a might be a gatekeeper of apoptosis in DNAdamaged hepatocytes. To prove this hypothesis, we first examined the induction of apoptosis after BCL-G overexpression in a normal human hepatocyte cell line. BCL-G overexpression induced the cleavage of PARP and Caspase3, suggesting the pro-apoptotic role of BCL-G in human hepatocytes (Fig. 4A).
Next, we examined if the G9a-BCL-G axis is involved in apoptosis caused by DNA damage in the hepatocytes. For this purpose, the hepatocyte lines were treated with hydrogen peroxide (H 2 O 2 , 2 mM) or UVB irradiation (300 J/m 2 ), which are well-known DNA damage inducers. As expected, both H 2 O 2 administration and UVB irradiation increased total p53 and phosphorylated p53 levels, the active form often induced by DNA damage 27 , followed by the upregulation of cleaved-PARP and cleaved-Caspase3 (Fig. 4B, D). Treatment with 5 µM of a G9a inhibitor UNC0638 significantly attenuated the global levels of H3K9me2 in the hepatocyte lines ( Fig.  S3A), without affecting cell cycle status (Fig. S3B). In order to examine the role of G9a in DNA-damaged hepatocyte, the hepatocyte lines were treated with 5 µM of UNC0638 for 24 h before treatment with DNA damage inducers. In contrast to the slight increase of BCL-G after the introduction of DNA damage inducers in the control hepatocytes, the G9a inhibitor UNC0638 strikingly enhanced the expression of BCL-G (Fig. 4B, D). UNC0638 also induced PARP or Cas-pase3 cleavage while it did not always affect p53 activation, and it was suppressed in BCL-G knockdown cells (Fig. 4B, D). The enhancement of apoptotic changes by G9a inhibitor and their attenuation by BCL-G knockdown were also confirmed by flow cytometric analysis (Fig. 4C, E). These findings indicate that BCL-G, induced by G9a inhibition, is involved in the induction of apoptosis in DNA-damaged hepatocytes. The global levels of H3K9me2 increased after both treatments, and decreased, at least in part, by UNC0638, indicating that G9a contributes to DNA damage-induced H3K9me2 modification in hepatocytes. The residual H3K9me2 marks imply the involvement of other histone modifiers for H3K9me2.
G9a regulates Bcl-G expression by mediating p53 recruitment to its response element Bcl-G is a p53 direct target gene. 28 Notably, UNC0638 altered BCL-G expression without affecting p53 activity in DNA-damaged hepatocytes (Fig. 4B, C). In addition, although p53 activation was comparable between WT and G9a ΔHep mice livers after administering a higher dose of DEN (Fig. 3A), Bcl-G protein and gene expression levels were significantly higher in the liver of G9a ΔHep mice (Figs. 3G and 5A). These findings suggest that G9a regulates Bcl-G expression without affecting p53 status in DNA-damaged hepatocytes. Histone modifiers regulate gene expression by mediating the recruitment of transcriptional machinery to their target genes. [29][30][31] Therefore, to determine the mechanism through which G9a regulates the expression of Bcl-G, we first examined the binding of G9a to the p53   response element (RE) on the Bcl-G gene using the chromatin immunoprecipitation (ChIP) assay in mice livers. G9a was significantly enriched in the p53 RE of the Bcl-G gene in the WT liver after administration of a high dose of DEN, while it did not bind to the p53 RE in non-treated WT liver cells (Fig. 5B). H3K9me2 levels at the p53 RE were elevated in the WT liver after DEN administration, but it remained unchanged in the G9a ΔHep liver (Fig. 5C). Importantly, while p53 binding to the RE of the Bcl-G gene was not detected in the WT liver, G9a deletion increased significantly the recruitment of p53 to the RE of the Bcl-G gene after DEN administration (Fig. 5D). Thus, G9a might regulate Bcl-G expression by interfering with the p53 recruitment in an H3K9 methylase-dependent manner. Next, we examined whether the regulation of p53 recruitment by G9a is a common phenomenon for p53 target genes. We focused on p21, one of the major target genes of p53, because it is well known that DEN induces p53-dependent p21 upregulation. 32 High doses of DEN significantly upregulated the expression of p21, and importantly, the upregulation was independent of G9a status (Fig. 5E). The ChIP assay revealed that G9a did not bound to the p53 RE of the p21 gene after DEN administration (Fig. 5F). Furthermore, the levels of H3k9me2 were not elevated following DEN administration (Fig. 5G), and p53 was enriched in the p53 RE of the p21 gene in both WT and G9a ΔHep liver (Fig. 5H). These findings indicate that G9a is not involved in the regulation of p21 expression. Thus, G9a individually regulates p53 target genes expression through selective control of p53 recruitment (Fig. 5I).

Discussion
Accumulating evidence suggests that epigenetic dysregulation plays an important role in human carcinogenesis. 5 Histone modifiers play crucial roles in regulating oncogenes and tumor suppressor genes in various cancer types, including HCC. The establishment of HCC is a complex process which involves the accumulation of multiple gene mutations. However, it is unclear how mutated hepatocytes escape p53-dependent genome surveillance and progress to HCC. In this study, we reveal the role of histone methyltransferase G9a in hepatocarcinogenesis in vivo. In a pro-carcinogen DEN-induced murine HCC model, G9a prevented DNA-damaged hepatocytes from undergoing apoptosis by modulating p53 transactivation (Fig. 6). This is a novel finding of the role of epigenetic regulation to liver cancer initiation.
We showed, by analyzing a TCGA data set, that several histone methyltransferases, including SETDB1 and G9a, are overexpressed in HCC. Interestingly, the loci of both genes, chromosome 6p21 of G9a and chromosome 1q21 of SETDB1, are amplified in human HCC 22,33 , suggesting that upregulation of G9a may be due to copy number amplification, as in the case of SETDB1. H3K9 methyltransferases including G9a, GLp, SETDB1, and Suv39h1, functionally cooperate as components of the transcriptional megacomplex 34 . Since they are important in liver carcinogenesis 8,9,16 , we suggest that the coordinated activity of H3K9 methyltransferase may play a role in liver tumor initiation. Indeed, we observed a residual increase in H3K9 levels regardless of G9a inhibition in DNAdamaged human hepatocytes, suggesting that other H3K9 methyltransferases may be involved.
Surviving DNA damage, such as that induced by reactive oxygen species or radiation exposure, involves the cells; ability to adapt to DNA damage. However, these cellular adaptations may enhance the viability of premalignant cells and their transformation to tumor cells. One group reported that G9a is required for DNA damage repair and allowed cancer cells to survive under DNA damage 21 . Another group demonstrated that G9a is recruited to DNA damage sites with other repressive chromatin proteins such as enhancer of zeste homolog 2 (EZH2) and DNA methyltransferase (DNMT) and maintained transcriptional silencing of tumor suppressor genes in colorectal cancer 35 . These reports indicate the important roles for G9a in transcriptional repression and tumor initiation under DNA damage. Consistent with this notion, G9a ablation or pharmacological inhibition resulted in enhanced apoptosis in DNA-damaged (see figure on previous page) Fig. 3 G9a protects hepatocytes against apoptosis after a high dose of DEN. A Immunohistochemical analysis of the mice liver 48 h after administering a high dose of diethylnitrosamine (DEN) (100 mg/kg i.p.). G9a is upregulated after DEN administration, and cleaved-Caspase 3 induction occurs only in the G9a ΔHep liver, whereas γH2AX and p53 levels are comparable between wild-type (WT) and G9a ΔHep livers. Bcl-G expression is elevated in the pericentral hepatocytes of G9a ΔHep mice. Scale bars, 50 µm. The number of positive hepatocytes is shown in the graph below (**p < 0.01, NS not significant, Student's t test). B Serum ALT levels are significantly elevated in G9a ΔHep mice 48 h after administering a high dose of DEN. Data are represented as mean ± SEM (WT/DEN0h, n = 4; G9a ΔHep /DEN0h, n = 3; WT/DEN48h, n = 5; G9a ΔHep /DEN48h, n = 3). C Cyp2e1 expression, as assessed by qRT-PCR, is comparable between WT and G9a ΔHep livers (each, n = 3; NS not significant, Student's t test). D Top 10 upregulated genes in the non-tumorous area of G9a ΔHep liver are listed. E qRT-PCR analysis validates Bcl-G upregulation in the G9a ΔHep liver (*p < 0.05, Student's t test). F Immunoblots reveal increased Bcl-G, cleaved-PARP, and cleaved-Caspase3 protein levels in the G9a ΔHep liver 48 h after high-dose DEN administration. Right graph shows relative protein level of Bcl-G (each, n = 3; **p < 0.01, NS not significant, Student's t test). G Relative expression levels of G9a and BCL-G in human non-tumorous (NT) liver obtained in our institution. BCL-G expression is negatively correlated with G9a expression. (rho = −0.37, p = 0.02, Spearman's rank correlation coefficient).   hepatocytes and attenuated liver tumor initiation. Our data emphasize the pro-tumorigenic role of G9a by suppressing DNA damage-induced apoptosis in mammalian cells, including hepatocytes. Transcriptional analysis revealed that G9a silenced Bcl-G expression, a pro-apoptotic member of the Bcl-2 family. In humans, there are two BCL-G isoforms, BCL-G S and BCL-G L , generated by alternative splicing 36 . Both isoforms are related to the induction of apoptosis, with BCL-G L being widely expressed in adult human tissues, whereas BCL-G S is found only in the testicles. In mice, only one form of Bcl-G exists, which is homologous to human BCL-G L and is expressed in similar tissues 37 . Studies on breast and colorectal cancers suggest that BCL-G has a tumorsuppressive function 38,39 . Bcl-G expression in HCC was less than half of that in the surrounding NT liver in our cohort (data not shown). Moreover, we found that overexpression of BCL-G in hepatoma cell lines in vitro induces apoptotic changes, suggesting its role as a tumor suppressor in hepatocarcinogenesis. Taken together, our study demonstrates that G9a allows cell survival and future malignant transformation by silencing the expression of tumor-suppressing Bcl-G in DNA-damaged hepatocytes.
Recent papers have reported the functional link of G9a with p53 activity. G9a specifically methylates p53 at lysine 373, resulting in its inactivation 40 . Loss of G9a delayed malignant transition in progenitors of a murine chemical mutagen-induced squamous tumor through p53 overactivation 15 . These reports suggest that G9a promotes tumorigenesis by impairing p53 transactivation. Moreover, G9a and HDAC1 epigenetically repress p53 target genes by maintaining H3K9 methylation and histone deacetylation of the p53 target genes promoter 41 . Consistently, our ChIP experiments revealed that G9a regulates p53 recruitment onto its target pro-apoptotic gene, Bcl-G, in an H3K9methylase dependent manner. In contrast, the regulation was not in the case of p21 promoter. It remains unclear how the specificity of G9a binding to p53 target genes is determined. Since histone modifiers regulate gene expression through interaction with chromatin remodeling complexes 29,42 , a chromatin remodeling mechanism or chromatin conformational changes may be involved in the processes. Further studies are needed to elucidate this point.
Finally, G9a inhibitors could provide additional effects when combined with other agents such as DNMT inhibitors 19,43 or immune checkpoint inhibitors 43,44 , which are expected to be effective against HCC 45 . Our results might point to a new therapeutic strategy of combinatory therapies, including G9a inhibitors, against HCC.

Human clinical samples
Surgically resected HCC samples were obtained from patients who underwent hepatectomy at the University of Tokyo between November 2013 and October 2014. These procedures were approved by the Ethical Committee for Clinical Research of our institution and written informed consent was obtained from each patient. The clinical diagnosis of all samples as HCC was confirmed by the Department of Pathology at the University of Tokyo Hospital.
The cancer genome atlas (TCGA) dataset 373 HCC samples with both mutation and mRNA data were obtained from TCGA via cBioportal (http://www. cbioportal.org/). Among them, putative copy-number alterations data calculated by GISTIC was available in 360 cases.

Quantitative real-time PCR (qRT-PCR)
Total RNA was extracted from frozen clinical liver specimens, HCC cell lines, and mouse livers, using NucleoSpin RNAII (Takara, Shiga, Japan) or ISOGEN-II (Nippon Gene, Tokyo, Japan). The purified RNA was reverse transcribed using the ImProm_II Reverse Transcription System (Promega, Madison, WI, USA) and amplified by reverse transcription PCR. The qRT-PCR analysis was performed using a PCR mixture containing a complementary DNA sample, forward and reverse primers, and THUNDERBIRD SYBR qPCR Mix (Toyobo, Osaka, Japan), using the StepOnePlus Real-Time PCR System (Applied Biosystems, Foster City, CA, USA) according to the manufacturer's instructions. Values were internally normalized against β-actin mRNA expression. The primer sequences are listed in Supplementary Table 1.

Animal experiments
Floxed G9a (G9a flox/flox ) mice were generated as described previously 10,46 and Alb-Cre mice were purchased from The Jackson Laboratory 47 . Mice were maintained in a temperature-and light-controlled facility, and permitted ad libitum regular chow diet and autoclaved water. All mice were backcrossed with the C57BL/ 6 strain and the male progeny were analyzed. All the experiments were performed in accordance with protocols approved by the Animal Ethics Committee of the University of Tokyo.
HCC was induced as previously described 23

Primary hepatocyte isolation
Primary hepatocytes were isolated from mouse livers as previously described 49 . Briefly, 40 μg/ml Liberase (Roche, Grenzacherstrasse, Basel, Switzerland) was perfused through the liver. Digested liver was passed through a 40 μm cell strainer, and centrifuged several times at 50 g for 1 min to remove nonparenchymal cells.

Serum alanine aminotransferase (ALT) measurement
Serum samples for ALT measurement were collected after a 16 h starvation (SRL, Tokyo, Japan).

Microarray analysis
GeneChip Mouse Genome 430 2.0 Arrays were performed according to the manufacturer's protocol (Affymetrix, Santa Clara, CA, USA). Total RNA was extracted from non-tumorous (NT) liver of 30-week-old WT and G9a ΔHep mice treated with DEN followed by 22 repeated CCl 4 administration (each, n = 2). Probe design files and microarray data have been submitted to the National Center for Biotechnology Information Gene Expression Omnibus database under accession number GSE147061.

Plasmids, short hairpin RNA (shRNA), and transfection
The human BCL-G expression plasmid, pCAGGS human flag-BCL-G, was kindly provided by Dr. Toyomasa Katagiri (Division of Genome Medicine, Institute of Advanced Medical Sciences, Tokushima University, Japan) 38 The lentiviral-based knockdown plasmid expressing the specific shRNA of BCL-G was purchased from Open Biosystems (Huntsville, AL, USA). The cells were transfected using the Effectene Transfection Reagent (Qiagen, Hilden, Germany) according to the manufacturer's instructions.  Fig. 6 Graphical scheme. Schematic representation of the G9a-p53-Bcl-G circuit that controls liver tumor initiation from DNA-damaged hepatocytes. DEN-exposed pericentral hepatocytes undergo DNA damage and mutagenesis. G9a allows DNA-damaged hepatocytes to escape p53-induced apoptosis, the potent genome surveillance checkpoint, via Bcl-G silencing, which results in the fixation of mutations and promotion of future HCC development. In summary, G9a determines whether hepatocytes undergo apoptosis or survive during DNA damage response via the regulation of p53 transactivation.

Flow cytometry
For cell cycle distribution analysis, the hepatocyte lines were collected after 24 h of treatment with a G9a inhibitor UNC0638 or mock. The cells were fixed in 1 mL ethanol (70 %) at 4°C for 30 min and were resuspended in phosphate buffered saline containing 50 μg/mL propidium iodide (PI) and incubated at room temperature for 30 min before analysis.
For apoptosis detection assay, the hepatocyte lines were collected 24 h after treatment with ultraviolet B (UVB) irradiation (300 J/m 2 ) by a UVB lamp (UVP, Upland, CA, USA) or hydrogen peroxide (2 mM, lasting 24 h). The cells were stained with 5 μL of FITC Annexin V (BD Biosciences, Franklin Lakes, NJ, USA) and PI (50 μg/mL) for 15 min in the dark. PI negative and Annexin V positive cells were considered early apoptotic; PI and Annexin V positive cells were considered to be in late apoptosis.
Data analysis and acquisition were performed using the Guava® EasyCyte™ Plus Flow Cytometry System and the Guava Express Pro Software (Guava Technologies, Hayward, CA, USA).
The ChIPed DNA was purified using QIA quick PCR purification kit (Qiagen). Results from each immunoprecipitation were presented as input percentage. The primer sequences for the target regions are available in Supplementary Table 2.

Statistical analyses
All statistical analyses were performed using R software (version 3.6.1; R Development Core Team, Vienna, Austria). All the results are expressed as mean ± standard error of the mean (SEM) of at least three independent experiments. Statistical significance was determined by the two-tailed Student's t test and Fisher's exact probability test. p values < 0.05 were considered as statistically significant.