Epigenetic modifications precede molecular alterations and drive human hepatocarcinogenesis

Development of primary liver cancer is a multistage process. Detailed understanding of sequential epigenetic alterations is largely missing. Here, we performed Infinium Human Methylation 450k BeadChips and RNA-Seq analyses for genome-wide methylome and transcriptome profiling of cirrhotic liver (n = 7), low- (n = 4) and high-grade (n = 9) dysplastic lesions, and early (n = 5) and progressed (n = 3) hepatocellular carcinomas (HCC) synchronously detected in 8 patients with HCC with chronic hepatitis B infection. Integrative analyses of epigenetically driven molecular changes were identified and validated in 2 independent cohorts comprising 887 HCCs. Mitochondrial DNA sequencing was further employed for clonality analyses, indicating multiclonal origin in the majority of investigated HCCs. Alterations in DNA methylation progressively increased from liver cirrhosis (CL) to dysplastic lesions and reached a maximum in early HCCs. Associated early alterations identified by Ingenuity Pathway Analysis (IPA) involved apoptosis, immune regulation, and stemness pathways, while late changes centered on cell survival, proliferation, and invasion. We further validated 23 putative epidrivers with concomitant expression changes and associated with overall survival. Functionally, Striatin 4 (STRN4) was demonstrated to be epigenetically regulated, and inhibition of STRN4 significantly suppressed tumorigenicity of HCC cell lines. Overall, application of integrative genomic analyses defines epigenetic driver alterations and provides promising targets for potentially novel therapeutic approaches.


Introduction
Hepatocellular carcinoma (HCC) is a hallmark of inflammation-induced cancers and ranks among the most common causes of cancer-related deaths worldwide (1). Herein, hepatocarcinogenesis is a multi-stage process that most frequently develops in the background of a chronic inflammatory liver disease and liver cirrhosis (CL) induced by chronic viral hepatitis (hepatitis B (HBV) or C viruses (HCV)), alcohol abuse or other metabolic and hereditary factors (2). Pre-neoplastic dysplastic lesions, i.e. low-(LGDN) and high-grade dysplastic nodules (HGDN), evolve into early hepatocellular carcinoma (eHCC) that, subsequently, progresses to advanced HCCs (pHCC) (3). This sequence is accelerated by genetic and epigenetic alterations that induce malignant transformation at early stages and promote progression into advanced stages. During the past decade, several molecular alterations as well as changes to the microenvironmental cellular contexture have been associated with increased risk of HCC in chronic liver diseases (3,4). Integrative transcriptome analysis of dysplastic lesions, eHCC and pHCC in patients with chronic HBV infection recently revealed that molecular profiles of early lesions are relatively uniform whereas a sharp increase in molecular heterogeneity is induced in pHCC (5). However, activation of prognostically adverse signaling pathways from DN to pHCC was only partially explained by observed genetic alterations, suggesting that complementary mechanisms might be operative and drive hepatocarcinogenesis. It is well established that epigenetic mechanisms in cancer cells are highly influenced by micro-environmental stimuli. In this context, changes in DNA methylation patterns are believed to be early events in tumor development in inflammatory cancers preceding allelic imbalances and ultimately leading to cancer progression thereby adding considerable complexity to the pathogenesis of liver and other cancers (6,7). In the liver, methylation patterns can be effectively used to classify patients according to different etiological factors (e.g. HBV, HCV, alcohol) (8). In addition to changes in global methylation patterns, distinct methylation profiles strongly correlated with clinical characteristics and survival of HCC patients (9)(10)(11)(12)(13). Recent studies also indicate that methylation signatures have a high prognostic value for HCC development and recurrence after curative resection or liver transplantation (12,14,15). Evidence for the importance of an DNA methylation dependent, multistep sequence of molecular alterations in hepatocarcinogenesis was further demonstrated in HBV-related liver cancers and indicated a major contribution of epigenetics in deregulation of key pro-oncogenic molecules from cirrhotic nodules over dysplastic nodules to eHCC and finally pHCC (16). However, our understanding of the molecular complexity is still limited and a detailed catalogue of key (epi)-genetic alterations commonly altered across the full spectrum of hepatocarcinogenesis remains to be defined. This lack of information represents a major challenge for preventive strategies as well as therapeutic approaches in HCC.
A significant drawback in the study of sequential evolution of liver cancer is the scarcity of tissues from early stages. In contrast to other cancers, detailed investigation of stage wise progression is also highly demanding due to shortage of the available lesions f rom the full spectrum of stages from livers of individual patients. To overcome this limitation, we here collected a cohort of unique HBV-infected patients with synchronous occurrence of the complete spectrum of early and advanced stages in the same patien t and performed multilevel sequencing analyses. Interestingly, our mitochondrial DNA sequencing analyses revealed a multi-clonal origin of co-developed lesions in the background of chronic liver inflammation. We further created a detailed landscape of epigenetic alterations and affected signaling pathways in HCC. New epigenetic drivers were identified by integrative approaches and subsequently validated in two Western cohorts of HCC patients comprising 887 human samples. Notably, 23 newly identified and validated epi-drivers have an impact on overall survival of HCC patients. Among them Striatin 4 (STRN4) was shown to be epigenetically regulated and highly activated in late stages of hepatocarcinogenesis with prognostically adverse implications for HCC patients. Here we demonstrate that targeting STRN4 resulted in decreased tumorigenicity of liver cancer cells.

Clonal diversity of (pre)neoplastic lesions.
Several recent studies indicate substantial intratumoral heterogeneity in HCC (17). The presented cohort comprising low-(n=4) and high-grade (n=9) dysplastic lesions, eHCC (n=5) and pHCC (n=3) as well as cirrhotic liver (n=7) from 8 individual HCC patients offers a unique background to delineate if synchronously co-existing pre-neoplastic and cancerous lesions are derived from the same clonal origin within individual livers. Clinico -pathological characteristics of the patients are displayed in Table 1. Mitochondria are highly exposed to reactive oxygen species (ROS) (18). Consequently, mitochondrial genome integrity is significantly disrupted during tumor development, leading to clonal expansion or loss of mutated mtDNA copies.
Thus, tracking of mt-genome variants, in particular heteroplasmic mt-variants, effectively defines clonal origin of different lesions (19).
We applied mt-DNA sequencing to the entire cohort and identified a total of 830 mt -variants.
A median of 29.5 mt-variants were detected in individual samples (CL: 30±4.7; LGDN: 32±2.6; HGDN: 30±2.9; eHCC: 28±4.0; pHCC: 28±2.1). The overall number of mt-variants was comparable across all stages of the disease indicating that a high mt-mutation rates is present already in the HBV-infected, diseased cirrhotic livers ( Figure 1A). As expected, majority of variants were homoplasmic (735/830) and highest frequency of variants in pre-neoplastic and cancer lesions was observed in the D-LOOP region, i.e. involved in genes important for replication and expression of mt-DNA ( Figure 1B, upper graph) (19,20). Most of the alterations were single nucleotide variations with G>A base transitions ( Figure 1B, lower graph). Notably, several mt-variants have been associated with other cancer types ( Supplemental Table 1).
Interestingly, LGDN and HGDN of PT1 had two common heteroplasmic mt-variants (310T>C; 8701A>G) suggesting that lesions could either have evolved from clonal expansion and share the same cellular origin or might be associated with malignant transformation in general (21,22). However, we detected several heteroplasmic mt-variants that occurred only in the LGDN and, consistently, mutational profiles of pre-neoplastic and cancer lesions of PT2, PT5 and PT8 were highly heterogeneous indicating a multi-clonal origin in the majority of lesions ( Figure   1C). We further confirmed that 20% (±8.52) of variants were present in more than one lesion whereas 80% (±8.52) were unique variants (p<0.0001; Figure 1D) overall suggesting that the mutational profiles of the different lesions is driven by de novo emergence in individual lesions ( Figure 1C). These observations suggest that the diseased hepatic microenvironment might induce a field effect that predisposes induction of epigenetic and genetic changes throughout the liver resulting in multiple pre-neoplastic lesions gradually progressing to advanced HCC (17,23). These findings prompted us to next dissect the epigenetic signature and the resulting transcriptomic changes during sequential evolution of liver cancer.

Epigenetic landscape during sequential evolution of HCC
We first assessed global transcriptome changes as well as significantly deregulated signaling pathways of the different lesions. The results confirmed our previous findings that activation of key oncogenic pathways occurred late i.e. in eHCC and pHCC lesions (5). However, when analyzing networks related to epigenetic modifications, we detected a significant activation of 'genes related to DNA methylation and transcriptional repression signaling' that occurred early during malignant transformation with a peak in dysplastic nodules and eHCC . Interestingly, pathways associated with epigenetic changes were largely inactivated in pHCC suggesting that mechanisms beyond epigenetics might be operative at advan ced stages ( Table 3). Next, we sought to define and quantify global methylation patterns affected during hepatocarcinogenesis. We applied Infinium Human Methylation 450k BeadChips to all lesions. As already demonstrated in previous studies, beta-value density during hepatocarcinogenesis displayed a trend to global hypomethylation ( Figure 2B). To further identify differentially methylated genomic regions (DMGR) associated with HCC development and progression, we compared epigenetic alterations of dysplastic and cancer lesions to non-infected and non-cirrhotic liver (NL) (DMGRNL: n=10; Figure 2C). Consistent with the observed pathway activation of epigenetic modifications, we detected an increase and maximum peak of DMGR in eHCC lesions (DMGReHCC: n=4965; DMGRpHCC: n=1702) in comparison to other stages of disease ( Figure 2D). While hypomethylated marks were mainly located in open sea regions, hypermethylated marks that progressively increased during hepatocarcinogenesis, occurred mainly in CpG Island regions suggesting regulatory importance and potential impact on gene expression ( Figure 2E and Supplemental Figure 2B).
Unsupervised cluster analyses based on the identified DMGRs effectively subdivided normal liver from cirrhotic liver as well as from pre-neoplastic and malignant lesions. Interestingly, we did not achieve a sharp distinction between cirrhotic parts, pre -neoplastic, and malignant lesions by DNA methylation profiling alone ( Figure 2C Table 4).
We next analyzed functional networks and signaling pathway regulation of identified DMGR.
While early epigenetic alterations from CL to LGDN centered on signaling pathways related to cell death, apoptosis and immune regulation, late changes involved cell survival, growth, and migration. We further detected a common regulation of stem cell-associated pathways including Wnt/b-catenin signaling only in dysplastic nodules as well as eHCC (Figure 3, Supplemental Table 5).
We next dissected the immune cell composition based on the gene expression profiles using the cibersort tool that revealed early changes of the immune compartment in the d iseased microenvironment ( Figure 4A). Interestingly, we saw an increase in M0 macrophages during the transition from LGDN to HGDN and eHCC, whereas B cell content were considerably reduced in premalignant and malignant lesions and increased in pHCC ( Figure 4A). In order to evaluate if these early changes of the immune compartment during malignant transformation are valid and applicable independently of the underlying etiology, we explored the presence of B cells (CD20+), T cells (CD3+, CD8+) and macrophages (CD68+, CD163+) in an independent cohort of HCV-infected patients. Consistently, we were able to confirm an increase of macrophages (CD68+) during malignant transformation. Interestingly, we did not observe changes in the CD163+ macrophage (M2) population ( Figure 4B). Furthermore, we consistently detected a significant decrease of B cells already in dysplastic lesions and HCC lesions, whereas population of T cells did not significantly change ( Figure 4B).

Identification and validation of Epi-Drivers in HCC development and progression.
In order to identify epigenetic alterations with high oncogenic potential, so called epi-drivers, we defined three signatures of (1) early DMGR common in all lesions from LGDN to pHCC (2) late DMGR common in all lesions from HGDN to pHCC and (3) Table 6).
Several DMGR have been previously described in the context of HCC including NKX6-2, NSD1, TBX15, ZIC1 (9). Next, we performed integrative analyses of our RNAseq data in order to define those DMGR that lead to a concomitant progressive gene expression alteration in cancer tissue (eHCC and/or pHCC). We detected 24 (20.5%) DMGR out of signature 1, 24  Table 8). We further confirmed a significant enrichment in signaling pathways associated with stem cell activation, immune regulation as well as oncogenic traits such as cell growth, survival and migration/invasion (Supplemental Figure 3B). To next investigate whether the 162 identified DMGRs have potential impact on biological traits of tumors, we integrated our results with an independent cohort of HCC patients (24) and the cohort of advanced HCC from the TCGAdatabase (TCGA-LIHC cohort (n=366 HCC)) and assessed clinical outcomes by sub-clustering the tumors based on the expression profiles of the 162 DMGR.
Notably, significant association to overall survival of patients could be revealed in our independent patient cohort as well as in the TCGA-LIHC cohort ( Figure Figure 4).
Next, we validated methylation and expression changes of each gene of the 162 gene signature as well as their prognostic association and relevance for cancer progression using the TCGA-database. We compared differentially methylated probes (DMP) in cancer tissues to non-cirrhotic livers (n=75) and validated differential DNA methylation in 121 genes out of our 162 identified epigenetic oncogenic marks ( Figure 5C, Supplemental Table 9). Among the validated 121 DMGR we further confirmed expression changes in 92 genes, of which expression of 23 genes had a significant impact on overall survival of HCC p atients ( Figure   5D, Supplemental Table 10 and 11). Multivariate analyses revealed 14 out of 23 genes with significant prognostic implications in HCC (Supplemental Table 12).
Functional network analyses of these epi-drivers confirmed relevance of the genes for cancer, organismal injury and abnormalities, gene expression, cell cycle as well as connective tissue development and function (Supplemental Figure 5A). Signaling pathway analyses of the putative epi-drivers further centered on molecular mechanisms of cancer, cell survival, proliferation and invasion as well as stem cell activation and immune regulation (Supplemental Figure 5B). While some genes (ATG4B, CCR5, MCM6, UCN) were detected in the context hepatocarcinogenesis before, most of the epi-drivers were result of the investigation of our unique cohort (26)(27)(28)(29).

Striatin 4 (STRN4) is a novel oncogenic epi-driver in HCC progression.
The application of integrative genomic analyses enabled us to define putative epigenetic driver alterations with relevance to malignant transformation and progression in the liver . Next, we evaluated if a newly identified molecule could be a new target for novel therapeutic approaches as a proof-of-concept. Among the 23 putative epi-drivers associated with overall survival, we identified Striatin 4 as a putative pro-oncogenic molecule specifically activated in late stages of hepatocarcinogenesis. Our DNA methylation analyses revealed an early , progressive hypomethylation of the body region of STRN4 (cg12254611, Chromosome 19: 47,249,193).
Consistent with a stepwise activation, activation of gene expression occurred late and dependent on the degree of hypomethylation ( Figure 6A). Using public available databases, we confirmed up-regulated expression of STRN4 in several tumor types including HCC (Supplemental Figure 6A). Importantly, high expression of STRN4 was significantly associated with poor prognosis of HCC patients (Supplemental Figure 6B). Up-regulation of STRN4 in HCC in comparison to corresponding non-tumor tissue as well as dysplastic nodules was further validated by immunohistochemistry using an independent cohort of 521 patients of the University Medical Center Mainz with confirmed HCC involving different etiologies ( Figure 6B and C). Clinico-pathological characteristics of the patients are displayed in Supplemental Table   13. HCC patients with high expression of STRN4 had a significantly worse outcome compared to HCC patients with low expression ( Figure 6D). Next, we functionally explored tumorigenic potential of STRN4. We silenced STRN4 expression in hepatoma cell lines Hu h7 and Hep3B by siRNA (Huh7: Figure 6E; Hep3B: Supplemental Figure 7A and B). Consistently, decreased STRN4 expression resulted in impaired ability to form colonies and spher es implicating impairment of their oncogenic potential (Huh7: Figure 6F; Hep3B: Supplemental Figure 7C).
Finally, to confirm the epigenetic regulation of STRN4, we employed an epigenetic unmasking approach using treatment of cells with 5-AZA. We observed a significant downregulation of STRN4 by 5-AZA treatment ( Figure 6G). These investigations establish the importance of STRN4 as a novel oncogenic epi-driver in HCC.

Discussion
Oncogenesis in the liver involves a multi-stage process that is fueled by chronic inflammatory liver diseases (3). The early dysplastic lesions emerge in the disrupted tissue microenvironment and subsequently progress to early and advanced HCC lesions (16).
Here we addressed clonal evolution of individual lesions as well as inter-tumoral heterogeneity in HCC and further defined a detailed catalogue of epigenetic alterations that promote human hepatocarcinogenesis.
Our unique cohort of HBV-infected patients included the complete spectrum of early and advanced stages synchronously arisen in the same patient. To date, few data are available that address whether multifocal HCC result by an intrahepatic metastatic process or by multicentric carcinogenesis. We here investigated the evolutionary background of dysplastic and cancerous lesions synchronously detected in the same patient by mitochondrial DNA profiling and addressed degree of corresponding inter -tumoral heterogeneity (19).
Heteroplasmic mt-variants were highly heterogeneous across pre-neoplastic and cancer lesions ( Figure 1C-D). Therefore, our results indicate that multifocal co-existing pre-neoplastic and cancerous lesions might not regularly derive from the same clonal origin within individual livers but rather emerge as de novo clones and, potentially, as a consequence of the ubiquitous inflammatory cell death. These observations are in accordance with previous findings, which employed multi-omic approaches and revealed a profound intra-and intertumoral heterogeneity in HCC indicative of multi-clonal origins in multifocal HCCs (17,30,31). A limitation of this study is the limited sample size of the patient cohort. However, our investigations on mitochondrial DNA profiling for intertumoral heterogeneity represent relevant findings in this rare and unique cohort of patients with different stages of HCC disease in the same liver and our warrant further investigations in larger collectives. Interestingly, we observed that the amount of variants per sample was similar across all stages (~30%) of disease reflecting high mutation rates already in the diseased cirrhotic livers ( Figure 1A).
Consistently, we have detected cellular alterations within the diseased livers including large (LLCC: 87,5% grade ≥2) and small liver cell changes (SLCC: 62.5% grade ≥2; Supplemental Table 14), representing pre-neoplastic dysplastic lesions <1mm in diameter without circumscribed nodular appearance. Detection and extent of LLCC and SLCC have been related to hepatocarcinogenesis in several studies (32,33). These results confirm that severe pre-neoplastic changes in the diseased microenvironment potentially predispose malignant transformation even before defined lesions emerge. Recent investigations suggest that the pronounced hepatic field effect might be induced by methylation abnormalities in early preneoplastic phases of HCC that precede genomic instability in advanced stages (17). In contrast to previous studies that focused on either single genes, stage-specific or on late stages of hepatocarcinogenesis (4,16,38,39), we here systematically examined epigenetic changes that commonly occurred across the different stages of disease and controlled gene expression, i.e. bona fide epi-drivers. As expected, early changes commonly disrupted showed lowest amount of DMGRs whereas DMGRs associated with eHCC and pHCC involved almost 500 alterations. Integration with our transcriptomic dataset revealed a total of 162 differentially methylated genes with concomitant expression changes in oncogenic lesions. We further confirmed that the gene expression signature of the 162 DMGRs showed significant association to overall survival in two independent HCC cohorts ( Figure 5B and C) (24) Moreover, we validated a total of 121 (75%) DMGRs in the TCGA-LIHC dataset with concomitant expression changes in 92 DMGRs (57%; Figure 4D) out of our 162 identified DMGR and proved therefore, that validated DMGRs were independent from underlying HCC etiology. Among them, we confirmed that gene expression changes in 23 genes were significantly associated with survival of HCC patients. Among those, >60% (14/23) genes showed an independent impact on overall survival in the TCGA-LIHC cohort as revealed by multivariate analyses (Figure 4D, Supplemental Table 11 and 12). Importantly, only few genes were previously implicated to liver cancer development (ATG4B, CCR5, MCM6, UCN) (26-29), whereas most of the epi-drivers were newly identified.
In line with recent findings that suggest robust prognostic impact of DNA methylation driven genes 40 , our integrative epigenetic analyses provide a powerful approach to identify novel drivers of hepatocarcinogenesis.
Among newly identified epi-drivers, we identified STRN4 activated in late stages of hepatocarcinogenesis. STRN4 belongs to the striatin protein family, which are part of the striatin interaction phosphatases and kinases complex (41). Recent studies have revealed metastatic and pro-tumorigenic properties of STRN4 in several tumor etiologies including colorectal, prostate cancer as well as NSCLC (42)(43)(44)(45). Indeed, we detected up-regulated gene expression of STRN4 in several tumor types including HCC in the TCGA database. We further confirmed up-regulated protein expression by immunohistochemistry using an independent cohort of 521 HCC patients ( Figure 6B and C). Importantly, validations of the findings in independent cohorts compromising 887 HCC patients confirmed that high STRN4 expression was significantly associated with poor prognosis ( Figure 6D and Supplemental Figure 6B). Thus, our study provides first evidence that STRN4 indeed possess oncogenic properties in liver cancer. Consistently, silencing of STRN4 significantly affected tumorigenic properties of human hepatoma cells ( Figure 6E-F) and might provide a new promising target for therapeutic applications in HCC.
In conclusion, we analyzed the epigenetic landscape during sequential evolution of hepatocarcinogenesis. The study provides evidence that early epigenetic alterations promote immune escape and induces stemness properties in pre-neoplastic lesions, thus, enhancing malignant properties during liver cancer development and progression. We subsequently defined and validated new epigenetic driver alterations including STRN4 that might provide new predictive and therapeutic opportunities for HCC patients.

Samples
A total of 28 samples were collected including 7 surrounding liver tissues, i.e. cirrhotic liver (CL), 4 low-grade dysplastic nodules (LGDN), 9 high-grade dysplastic nodules (HGDN), 5 early HCC (eHCC) and 3 progressed HCC (pHCC) from 8 HCC patients with chronic hepatitis B infection. Nodules were resected from explanted cirrhotic livers. All lesions were classified according to the criteria of 'International Consensus Group for Hepatocellular Neoplasia' by two independent expert pathologists (46). All procedures were approved by the local authorities and prior patient consent was obtained. Demographic and clinic -pathological data of the patients can be found in Table 1.

Nucleic acid extraction
Total RNA was extracted using the Qiagen RNeasy mini Kit (Qiagen GMBH, Hilden, Germany) following the manufacturer's instructions. RNA quantity and purity were estimated using a Nanodrop ND-1000 Spectrophotometer (NanoDrop Technologies, Wilmington, DE), and integrity was assessed by Agilent 2100 Bioanalyzer (Agilent, Palo Alto, CA). DNA was extracted using Qiagen Qiamp DNA Kit (Qiagen GMBH, Hilden, Germany) following the manufacturer's instructions.

Whole mitochondrial DNA ultra-deep sequencing
Multiplex PCR-based ultra-deep sequencing analysis of the whole mt-genome was performed and analyzed as previously described (19). In brief, PCR amplicable DNA was quantified by  subsequently filtered by manual analysis. Variants, which occur in different sample sets but with a similar frequency as well as variants which were located in repetitive or highly homologous regions of the mt-genome, in high background noise regions, or at the end of the amplicons were considered as putative false variants. Potential false positive variants were either deleted when they were clearly recognizable as artifacts or were further re -assessed by Sanger sequencing. In addition, whenever DNA was still available, the mt -DNA regions carrying a variant in one lesion sample but not in another of the same patient sample set, were subsequently re-analyzed by conventional Sanger sequencing (Supplemental Table 15 and Supplemental Figure 1B). Frequency of variants represent amount of variants normalized to sample size (19).

RNA sequencing
RNA sequencing was performed using Illumina HiSeq2000 and Illumina HiSeq4000. Raw reads were filtered by removing adapter sequences, contamination and low-quality reads. The reads were then mapped with human genome reference sequence (GRCh37.82) using HISAT2 (hisat2-2.0.2-beta) followed by read summarization with featureCounts (subread-1.5.0-p1) (47)(48)(49). All data analysis was performed using R programing language and related packages. The output matrix from feature counts was input into the Bioconductor package DESeq2 for differential expression analysis (50). Significance testing was performed using Wald Test statistics. Ingenuity Pathway Analysis (IPA) online tool provided by Qiagen was used for functional classification and pathways analyses. Significantly differentially expressed genes of each stage of disease were uploaded and a comparison analyses based on log ratio was performed. The analysis determines the most significantly affected pathways displayed by a -log(p-value). P-values ≤0.05 were considered statistically significant. The significance of each network, function and pathway was determined by the scoring system provided by Ingenuity Pathway Analysis tool.
Cluster analysis of methylome and transcriptome data was performed using the R package  Table 4). Estimation of the immune cell composition based on gene expression data was performed using the Cibersort-Tool (51).

Real-time PCR
A two-step RT-qPCR, cDNA synthesis using Superscript III (Invitrogen), SYBR Green MasterMix (Bio-Rad) and iQ5 or CFX Connect System was performed. Oligonucleotide primers were designed using Primer3 v.0.4.0 (http://frodo.wi.mit.edu/primer3/) as described before (27). The amplification protocol was as follows: 95°C for 3 min, followed by 40 cycles of 95°C for 15 seconds and 1 minute at 60°C, completed by a dissociation curve to identify false positive amplicons. Glyceraldehyde-3-phosphate dehydrogenase (GAPDH) was used as a reference. The relative expression level of each gene was normalized to untreated cells and calculated using the formula 2 (−ΔΔCt) . All experiments were performed in three independent replicates.

Western blotting
Monolayer cultures of each cell line were exposed to siRNA as described. replicates.

Colony and sphere formation assays
Cells were treated with siRNA as described. After treatment, we seeded 1x10 3 cells per plated on 6-well plates for colony formation assay and 1x10 3 cells on 48-well plates for sphere formation assay on agarose at 2%. Colonies and spheres were calculated at day 14 and represented as percentage of colonies/spheres of control. Hep3B cells did not form spheres.
All experiments were performed in three independent replicates.

Statistics
Statistical analysis was performed using Student's t-test or Mann-Whitney U test, for multiple group comparisons One Way ANOVA (Bonferroni Correction) as indicated. P-values ≤0.05 were considered statistically significant. Results are presented as means ± SD.
For integration of patients, publically available expression data sets were used (24).
Hierarchical cluster analyses were performed using Euclidea n distance by Bioconductor

Availability of data and materials
The dataset supporting the conclusions of this article is available in the Bioproject database (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE146286).

Study approval
All analyses were approved by the local authorities and prior patient consent was obtained.  Shown is the occurrence of mt-variants in the lesions per patient in % divided in unique occurrence vs. presence in more than one (>1) lesion, students t-test; ****p<0.0001.  Signaling pathway regulation during sequential evolution of HCC analyzed by ingenuity pathway analyses based on detected stage-specific DMGR to NL. Significance of each pathway was determined by scoring system provided by Ingenuity Pathway Analysis tool.