Dynamics of Genomic, Epigenomic, and Transcriptomic Aberrations During Stepwise Hepatocarcinogenesis

Author Manuscript Published OnlineFirst on September 10, 2019; DOI: 10.1158/0008-5472.CAN-19-0991 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.

Dynamics of genomic, epigenomic, and transcriptomic aberrations during stepwise hepatocarcinogenesis

Byul A Jee1,2,*, Ji-Hye Choi1,2,*, Hyungjin Rhee3,*, Sarah Yoon1,2, So Mee Kwon4, Ji Hae Nahm5, Jeong Eun Yoo5, Youngsic Jeon5,6, Gi Hong Choi7, Hyun Goo Woo1,2,†, Young Nyun Park5,6†

1Department of Physiology, Ajou University School of Medicine, Suwon, Republic of Korea; 2Department of Biomedical Science, Graduate School, Ajou University, Suwon, Republic of Korea; 3Department of Radiology, Yonsei University College of Medicine, Seoul, Republic of Korea; 4Department of Biochemistry, Ajou University School of Medicine, Suwon, Republic of Korea; 5Department of Pathology, Yonsei University College of Medicine, Seoul, Republic of Korea; 6BK21 PLUS Project for Medical Science, Yonsei University College of Medicine, Seoul, Republic of Korea; 7Department of Surgery, Yonsei University College of Medicine, Seoul, Republic of Korea *BAJ, JHC, and HR contributed equally to this work.

Running Title Multi-layered genomic profiles during stepwise HCC

Keywords Multi-omics; stepwise HCC; hepatocarcinogenesis; early HCC

Financial Support This research was supported by grants from the National Research Foundation of Korea (NRF) funded by the Korea government (MSIP) (NRF-2017R1E1A1A01074733, NRF-2017M3A9B6061509, NRF-2017M3C9A6047620, NRF-2019R1A5A2026045, NRF-2017R1A2B4005871, and NRF- 2017M3A9B6061512).

†Co-corresponding authors: Hyun Goo Woo, M.D., Ph.D., Department of Physiology, Ajou University School of Medicine, 164 Worldcup-ro, Yeongtong-gu, Suwon, Korea; Tel: 82-31-219-5045, Fax number: 82-31-219-5049, E-mail address: [email protected]. Young Nyun Park, M.D., Ph.D., Department of Pathology, Yonsei University College of Medicine, 250 Seongsanno, Seodaemun-gu, Seoul, 120-752, Korea; Tel: 82-2-2228-1768, Fax number: 82-2-362- 0860, E-mail address: [email protected].

Downloaded from cancerres.aacrjournals.org on October 3, 2021. © 2019 American Association for Cancer Research. Author Manuscript Published OnlineFirst on September 10, 2019; DOI: 10.1158/0008-5472.CAN-19-0991 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited.

Conflict of interest statement Nothing to disclose

Data availability: Data from genomic profiles are available in the GEO database

(http://www.ncbi.nlm.nih.gov/projects/geo) under accession number GSE99036.

ABSTRACT

Hepatocellular carcinoma (HCC) undergoes a stepwise progression from liver cirrhosis (LC) to low- grade dysplastic nodule (LGDN), high-grade dysplastic nodule (HGDN), early HCC (eHCC), and progressed HCC (pHCC). Here, we profiled multi-layered genomic, epigenomic, and transcriptomic aberrations in the stepwise hepatocarcinogenesis. Initial DNA methylation was observed in eHCC (e.g., DKK3, SALL3, and SOX1) while more extensive methylation was observed in pHCC. In addition, eHCCs showed an initial loss of DNA copy numbers of tumor suppressor genes in the 4q and 13q regions, thereby conferring survival benefits to cancer cells. Transcriptome analysis revealed that HGDNs expressed endoplasmic reticulum (ER) stress-related genes, while eHCC started to express oncogenes. Furthermore, integrative analysis indicated that expression of the serine peptidase inhibitor, Kazal type 1 (SPINK1), played a pivotal role in eHCC development. Significant demethylation of SPINK1 was observed in eHCC compared to HGDN. The study also demonstrated that ER stress may induce SPINK1 demethylation and expression in liver cancer cells. In conclusion, these results reveal the dynamics of multiomic aberrations during malignant conversion of liver cancer, thus providing novel pathobiological insights into hepatocarcinogenesis.

Significance: Multiomics profiling and integrative analyses of stepwise hepatocarcinogenesis reveal novel mechanistic and clinical insights into hepatocarcinogenesis.

INTRODUCTION

Hepatocellular carcinoma (HCC) classically develops through a multistep process involving a series of pathologic states: liver cirrhotic lesion (LC), dysplastic nodule (DN), early HCC (eHCC), and progressed HCC (pHCC). Liver injury induced by viral hepatitis or other etiologic factors produces a chronic inflammatory milieu leading to the development of LCs. LCs subsequently develop into DNs, which are further classified as low-grade or high-grade DNs (LGDNs or HGDNs, respectively) based on the presence of cytological and architectural atypia. Dysplastic lesions eventually progress to eHCC and pHCC. Molecular alterations accompanying this stepwise pathological sequence have been elucidated to determine definitive markers for eHCC. At present, key cancer-associated pathways, such as Notch, Toll-like receptor, MYC, TGF-β, WNT, and epithelial–mesenchymal-transition (EMT)- related signaling, have been demonstrated as active during hepatocarcinogenesis (1-4). Additionally, high-throughput genome-wide studies have revealed genomic aberrations that occur during stepwise HCC development, including DNA methylation (5,6), DNA copy number changes, mutations [e.g., TP53, CTNNB1, or TERT promoter] (7-9), and transcriptional deregulation (10). Recently, 3

multiomic studies, such as those involving The Cancer Genome Atlas (TCGA), showed have enabled an integrated and systematic view of the genomic aberrations associated with liver cancer, revealing novel biomarkers, as well as therapeutic targets (11-13).

Recently, multiomic studies shifted their focus towards precancerous lesions, resulting in the creation of the Pre-Cancer Atlas (PCA), an integrated collection of molecular, structural, and functional maps that provide information regarding tumor initiation and progression (14). The PCA is expected to delineate mechanisms underlying tumor development or progression in order to enhance early detection, risk prediction, and the development of new therapeutic strategies.

In this study, we generated multiomic profiling data that demonstrated a molecular landscape of the multi-layered aberrations that occur during stepwise hepatocarcinogenesis from precancerous lesions to malignant development of eHCC and pHCC. By performing integrative analyses, we identified early events that may possibly promote malignant conversion of precancerous lesions, thereby providing novel mechanistic insights into HCC development. These data may facilitate upcoming studies focused on discovering novel biomarkers or therapeutic targets for HCC.

MATERIALS AND METHODS

Patients and tissue specimens A total of 131 tissue specimens, including those of liver cirrhosis (LC, n = 30), high-grade dysplastic nodule (HGDN, n = 28), early HCC (eHCC, n = 30), and progressed HCC (pHCC, n = 43), resected from 76 patients was obtained (YSHCC, HCC cohorts from Yonsei University). All lesions were classified according to the criteria stipulated by the ‘‘International Consensus Group for Hepatocellular Neoplasia’’ (15). Freshly frozen specimens were obtained from the Liver Cancer Specimen Bank (part of the National Research Bank Program, Korea Science and Engineering Foundation, Ministry of Science and Technology), and subjected to RNA-seq profiling. Tissues of low- grade dysplastic nodule (LGDN) were not included in this study due to limited availability of frozen tissues. Clinico-pathological features of the 76 patients are summarized (Supplementary Table S1). This study was approved by the Institutional Review Board of Severance Hospital, Yonsei University College of Medicine (4-2014-0423), and the requirement for informed consent was waived.

For validation via immunohistochemical staining, a total of 171 liver specimens from 90 patients was examined, which included LC (n = 30), LGDN (n = 30), HGDN (n = 40), eHCC (n = 40), and pHCC (n = 41) specimens. Patients were 34-76 years in age (57.7 ± 8.19, mean ± SD) and consisted of 75 males and 15 females. Most etiologies were HBV (n = 74), and the remainder included HCV (n = 5),

alcohol (n = 7), Nonalcoholic steatohepatitis (NASH, n = 1), occult HBV potential (n = 1), or of unknown etiology (n = 2). Clinico-pathological features of the 90 patients subjected to immunostaining are summarized in Supplementary Table S2.

RNA-seq profiling and variant profiling RNA-seq profiling was performed in the YSHCC specimens (n = 131), including LC (n = 30), HGDN (n = 28), eHCC (n = 30), and pHCC (n = 43). Sequencing data were obtained using an Illumina HiSeq2000 for 100 bp paired end reads with a coverage greater than 42 million reads per sample. Data processing for measuring RNA abundance and variant calling was performed using an R package ‘SEQprocess’ (16). RNA abundance was estimated by using Tophat2-Cufflinks and the sequence variants were identified using a Genome Analysis Toolkit (GATK, https://software.broadinstitute.org/gatk). Wild- and mutant-type variants were determined with the cutoff of read counts greater than 8, respectively. Otherwise, the variants were designated as missing values.

Profiling of DNA methylation and DNA copy number aberration DNA methylation profiling was performed in the subset of the YSHCC, including LC (n=6), HGDN (n=11), eHCC (n=9), and pHCC (n=6) using an Infinium Human Methylation 450K BeadChip. Data processing and analyses were performed using R/Bioconductor libraries. For DNA methylation profile, probe level β-values were imported using ‘RnBeads’ library. DNA copy numbers were estimated from the DNA methylation data using ‘ChAMP’. Batch effect was corrected by ‘combat’, and DNA copy number segments were calculated by using a circular binary segmentation algorithm with default parameters.

Validation of public data Publicly available DNA copy number and somatic mutation profiles of HCC were obtained from The Cancer Genome Atlas (TCGA-LIHC, https://cancergenome.nih.gov). The segmented DNA copy number values were mapped to gene level aberration. Gene lists with copy number gains and losses were obtained from a previous study by the International Cancer Genome Consortium (LICA-FR, https://dcc.icgc.org). Data were obtained from TCGA and GEO websites (accession No.; GSE89377, GSE6764, GSE25097, GSE14520, GSE44970, GSE60753, GSE73003, GSE4024, GSE87630, and GSE65373), to validate the expression, methylation, and survival analyses related to our finding. Gene sets obtained from literature or public data were used for data analyses (Supplementary Table 3).

Further details on the methods utilized for profiling of DNA methylation, DNA copy number, RNA- seq, variant calling, network analysis, and the molecular experiments including immunohistochemistry, RT-PCR, western blot, gene expression constructs, lentiviral vector transfection, and reagents are described (Supplementary Materials and Methods).

RESULTS

Stepwise aberration of DNA methylation in HCC Tissue specimens of LC, HGDN, eHCC, and pHCC were diagnosed as described in the Materials and Methods. Multiomic profiles of DNA methylation (n=32), DNA copy numbers (n=32), mRNA expression (n=131) were generated, and analysis results were validated using tissue microarrays (n=171) and public data from TCGA and GEO databases. The overall study design is summarized (Supplementary Fig. S1).

First, aberrations of DNA methylation during stepwise HCC progression in each group (LC, HGDN, eHCC, and pHCC, n=32) were evaluated. Overall, no significant differences in global methylation status were observed between the groups (Supplementary Fig. S2A, B). In order to determine methylome changes during hepatocarcinogenesis, we identified 595 differentially methylated probes (DMPs) for each group in comparison with the group preceding it (i.e., HGDN vs. LC, eHCC vs. HGDN, and pHCC vs. eHCC) (asymptotic one-way Fisher–Pitman permutation test P < 0.001 and fold difference > 0.1). Excluding the DMPs identified in earlier stage groups, we identified the DMPs of each group for HGDN (DMPHGDN, n = 50), eHCC (DMPeHCC, n = 56), and pHCC (DMPpHCC, n = 489), respectively (Supplementary Table S4). The DMPs showed predominant hypermethylation (n = 555) rather than hypomethylation (n = 40) during HCC development (Fig. 1A), including previously known methylated genes in HCCs such as TBX4, CCNA1, PENK, WT1, and TRIM58 (5,17). Additionally, we identified methylation of DKK3, SOX1, and SALL3 in eHCC, with each of these events having been previously reported in pHCC samples (18-20). The DMPpHCC was significantly enriched with development-related genes [enrichment score (ES) = 6.42; P = 1.65 × 10−10), indicating the dysregulation of developmental processes during pHCC progression (Fig. 1B). Comparison of each group revealed that the DMPpHCC were markedly methylated and showed mid-range beta values (0.25~0.75), whereas the other DMPs were hypomethylated and showed lower beta values (Fig. 1C). Moreover, the DMPs were preferentially located at 5′ untranslated regions, indicating their functional significance (Supplementary Fig. S3, top). Most hypermethylated DMPs resided in CpG-

island (CGI) loci, whereas this was not observed in hypomethylated DMPs (Supplementary Fig. S3, bottom).

DNA methylation at the flaking regions of CGIs has previously been known to play a critical role in cell differentiation and cancer development (21). Therefore, we explored methylation patterns in the flanking regions of CGIs, classified as Shore (within 2 kb of the CGI), Shelf (within 2–4 kb of the CGI), and Opensea regions (>4 kb from the CGI), and observed a distinctive pattern in the regional progression of DNA methylation among the groups (Fig. 1D). LC and HGDN exhibited lower DNA- methylation levels in CGI regions, which increased markedly during eHCC and pHCC development. Notably, we found that the DNA-methylation process initiated in the Shelf and Shore regions during HGDN development, with increased CGI-region methylation during eHCC and pHCC development. These findings indicated that sequential DNA methylation initiated in CGI-flanking regions and extended into CGI regions during HCC development.

We then evaluated the consistency of our data with previous DMPs (i.e., DMP36 and DMP100 signatures) that were identified as being methylated in HCCs (mostly pHCC; GSE44970; n = 102) (5). Although most DMP100 genes did not overlap with our DMPs, those genes did exhibit stepwise methylation during eHCC and pHCC development (Supplementary Fig. S4A, B). Additionally, we validated stepwise methylation at CGI regions, although the methylation pattern at CGI-flanking regions was not discernible during stepwise progression of HCC, which might be due to the low- resolution of the data platform (27K probes; Supplementary Fig. S4C).

DNA copy number aberration (CNA) during stepwise hepatocarcinogenesis. DNA copy numbers were obtained by inferring DNA-methylation data (for details see Materials and Methods). We observed that the number of CNA segments, particularly the deleted segments, was increased markedly during eHCC and pHCC development (Supplementary Fig. S5A). Moreover, the sizes of the CNA segments became larger, implying a stepwise increase in genomic instability and aberration during hepatocarcinogenesis (Supplementary Fig. S5B).

Group comparison indicated that LC and HGDN exhibited no significant alteration of DNA copy numbers; however, eHCC samples showed genome-wide deletion of DNA copy numbers, and pHCC samples showed more profound and typical chromosomal losses and gains, as described previously (22,23) (Fig. 2A). To determine the stepwise alteration of CNAs more precisely, differential copy number aberrations (DCNAs) were estimated (one-way Fisher–Pitman permutation test P < 0.05 and fold difference > 0.05) between each group and their respective precedent group: HGDN vs. LC (n =

182; DCNAHGDN), eHCC vs. HGDN (n = 2,056; DCNAeHCC), and pHCC vs. eHCC (n = 10,303; DCNApHCC).

Notably, we found that eHCC exhibited prominent copy number losses at 4q and 13q as compared with HGDN (Fig. 2B). Additionally, we observed that these regions harbored many tumor-suppressor genes (TSGs), such as retinoblastoma 1 (RB1) (Fig. 2C). In fact, oncogenes showed recurrent DNA copy number gains (e.g., MYC and CCND1), whereas TSG frequently lost their copy numbers (e.g., CDKN2A and PTEN) (24,25). In order to conduct a genome-wide evaluation of this hypothesis, we examined CNAs in the gene sets of oncogenes (n = 674) and TSGs (n = 1,088), and observed that oncogenes preferentially gained copy numbers (P = 0.001, Fisher’s exact test) while TSGs preferentially lost copy numbers with statistical significance (P = 0.005) (Fig. 2D, left). This finding was validated in TCGA-HCC data (LIHC), which showed frequent DNA copy number gains in oncogenes (P = 0.058) and losses in TSGs (P = 3.65 × 10−8) (Fig. 2D, right). Therefore, we suggest that the aberrations associated with DCNAeHCC are not random events, and instead may act as oncodrivers during malignant conversion. By contrast, DNA-methylation profiles did not show preferential alterations of oncogenes or TSGs (Supplementary Fig. S6). These results imply that aberrations in DNA copy numbers more than DNA methylation may play a pivotal role during HCC progression.

To further evaluate the functional roles of DCNAeHCC at 4q and 13q, we calculated the ES values of DCNAeHCC genes at 4q and 13q (n = 327) in DNA copy numbers and mRNA-expression levels, respectively. Previously, the genes showing DNA copy number-dependent transcriptional dysregulation have been shown to play functional driving roles in cancer progression (12,26). As expected, the ES values of CNAs (ESCNA) were correlated with the ES values of the corresponding gene expression (ESEXP; r = 0.37, P = 0.03; Pearson’s correlation test), indicating that the expression of the 4q and 13q region was dependent to its DNA copy number aberration (Fig. 2E). This finding was validated in the two independent datasets of GSE65373 (n = 38; r=0.51, P = 9.76 × 10−4) and TCGA-LIHC (n = 365; r = 0.73, P = 3.51 × 10−63) (Fig. 2F, G).

Next, to evaluate the clinical relevance of the DCNAeHCC at 4q and 13q, we selected patients who exhibited high correlation between ESCNA and ESmRNA,, and stratified them into High- and Low-groups.

The High group represented the patients who showed DCNAeHCC gains and transcriptional upregulation of their corresponding genes (ESEXP > 0 and ESCNA > 0; n =173), whereas the Low group represented the patients who showed DCNAeHCC losses with suppression of corresponding genes

(ESEXP < 0 and ESCNA < 0; n = 114). As 4q and 13q genes harbored many TSGs, we hypothesized that the High group may exhibit a better prognostic outcome compared that of the Low group. As expected, there was a clinical distinction between the groups, showing better clinical outcome of the High group compared to that of the Low group (hazard ratio = 0.32, P = 0.019, log-rank test; Fig. 2H).

Taken together, these results suggest that CNA losses at 4q and 13q along with concomitant transcriptional dysregulation may play a potential driving role in eHCC development.

Somatic mutations were acquired during stepwise hepatocarcinogenesis Next, we identified recurrent somatic nonsynonymous or nonsense mutations from the RNA-seq data (n = 1,824). No significant difference in the mutation frequencies were evident in group comparisons (Supplementary Fig. S7). Among the mutations, newly identified mutations in each group that were not present in the precedent group were designated as ‘differentially mutated genes’ (DMUTs) for each of HGDN (DMUTHGDN; n = 156), eHCC (DMUTeHCC; n = 34), and pHCC

(DMUTpHCC; n = 11), respectively (Fig. 3A, for details see the Materials and Methods). Next, we demonstrated the mutation status of the genes including mutations identified in oncogenes (e.g., ERBB2 and CTNNB1) and TSGs (e.g., RB1CC1 and TP53) among the DMUT genes, as well as the significant mutations identified in TCGA-LIHC data (n = 26), which may potentially play key roles in HCC development and progression (Fig. 3B). Genes harboring well-known trunk mutations, such as TP53 and CTNNB1, were detected in eHCC, as substantiated by the findings of previous studies. Additionally, we identified a new trunk mutation in BAP1 (BRCA1-associated protein-1) in eHCC, which was previously suggested as a driver mutation in HCC (12). These results consistently indicated a potential driver role for trunk mutations in HCC development.

We also examined mutations in the promoter region of TERT which have been observed to play key roles in hepatocarcinogenesis. Congruent with a previous study (27), mutation frequencies were increased along with the group for LC (2/30; 6.6%), HGDN (4/28; 14.2%), eHCC (5/28; 17.8%), and pHCC (12/43; 27.9%). However, the overall mutation frequency was much lower what was observed in the previous study (61% in eHCC and 64% in pHCC) (27), which may be due to cohort differences between studies. In fact, while the majority of patients in our cohort were patients with HBV etiological factor, the previous study cohort was mostly Caucasian and consists of etiological factors other than HBV.

In addition, we examined the driver-mutation pathways which have been shown to mutate frequently in liver cancers (Fig. 3C) (28). HGDN showed recurrent mutations in the oxidative-stress pathway (4/28, 14%). Compared with precancerous tissues, eHCC and pHCC tissues showed enriched mutations in the genes associated with cancer-driver pathways, such as chromatin regulators, TP53/cell cycle, WNT, and Ras/extracellular-signal-regulated kinase. Mutations in AKT/mTOR (mammalian target of rapamycin) pathway were found only in pHCC, implying a late event. These results demonstrated dynamic and sequential accumulations of pathway mutations during hepatocarcinogenesis. 9

Transcriptome alterations during stepwise hepatocarcinogenesis Epigenomic and genomic alterations may ultimately reprogram the transcriptomic architecture of cancers. To evaluate aberrations at the transcriptome-level, we constructed a pooled gene- expression profile (n = 293), including YSHCC (n = 131) and public datasets of multi-step hepatocarcinogenesis [i.e., GSE89377 (n = 87) and GSE6764 (n = 75)]. Pairwise comparison of each group identified differentially expressed genes (DEGs) for LC (n = 2,006), DN (n = 874), eHCC (n = 925), and pHCC (n = 1,736), respectively (permuted t test; P < 0.05 and fold difference > 0.3) (Fig. 4A). Gene ontology analysis revealed stepwise functional alterations among groups (Supplementary Fig. S8A), with the LC group exhibiting enriched expression of inflammation-related genes (e.g., those associated with type I interferon signaling; ES = 15.05; P = 1.81 × 10−18), whereas the HGDN group exhibited enriched expression of endoplasmic reticulum (ER)-stress-related genes (ES = 5.12; P = 9.85 × 10−10). Furthermore, eHCC and pHCC groups exhibited stepwise expression of cell cycle- related genes (eHCC: ES = 4.60 and P = 4.52 × 10−12; pHCC: ES = 33.06 and P = 8.57 × 10−66).

To further characterize the expression of immune-related genes, we examined previously defined immune-related gene signatures, such as “HCC immune” (n = 112), “Senescence” (n = 50), “Senescence-associated secretory phenotype” (SASP, n = 60), and “Cytokine pathway” (n = 187) (Supplementary Table S3). We found that the LC group was immune-active, showing the highest expression of immune-related and senescence-related genes, which may provide a microenvironmental milieu for carcinogenesis (Fig. 4A, top, and Supplementary Fig. S8B) (29). Additionally, we analyzed the immunotypes, “active-immune” and “exhausted-immune”, which were recently reported as playing distinct roles in HCC progression (30). Prediction of the immunotypes in each sample by using Nearest Template Prediction (NTP) algorithm (31) revealed that LC predominantly expressed the “active-immune” type (19/55; 34.54%), whereas eHCC and pHCC expressed the “exhausted-immune” type (17/53 of eHCC, 32.07%; 29/95 of pHCC, 30.52%) (Fig. 4B). This indicated that eHCC undergoes a transition in immune types, which may facilitate immune evasion of tumor cells. With respect to this finding, TGF‐β signaling regulates tumor-stroma interactions and suppresses host immune response by inducing T cell exhaustion (32). Moreover, TGF‐β acts as a tumor suppressor at the early stages of tumor development expressing the early TGF-β signature, whereas, at later stages, TGF‐β possesses tumor-promoting potential expressing the late TGF-β signature (33,34). Congruent with immunotype conversion, eHCCs showed a transition from early to late TGF‐β-related signaling, as demonstrated by the transition scores associated with TGF‐β signaling (i.e., ESlate-TGFB − ESearly-TGFB; Fig. 4A, middle). Such transition in TGF‐β

activity during hepatocarcinogenesis may indicate the induction of tumor-promoting activity, as well as exhausted-immune activity.

Gene set enrichment analysis validated the stepwise expression of ER-stress-related genes and oncogenes. HGDN exhibited prominent expression of ER-stress-related genes associated with “Response to ER stress” (n = 233) or the “Unfolded-protein response” (n = 113) (Fig. 4A, middle). Subsequently, eHCC showed enriched expression of oncogenes and counter-balanced suppression of TSGs. Calculation of onco-activity based on the differentially enriched activation of oncogenes and suppression of TSGs (i.e., ESoncogene − ESTSG) revealed that onco-activity transitioned from tumor- suppressive (negative onco-activity) to oncogenic (positive onco-activity) activity during eHCC development. A groupwise comparison of gene-set enrichment analysis confirmed the differential expression of ER-stress-related genes during DN development (ES = 0.45, P = 0.01; Fig. 4C, left) and oncogene activation during eHCC development (ES = 0.30, P = 0.05; Fig. 4C, right). These results suggested that stepwise expression of immune-related and ER-stress-related genes in LC and HGDN might provide a microenvironmental milieu that is favorable for eHCC development.

Serine protease inhibitor Kazal type 1 (SPINK1) is a potential key driver of eHCC development To delineate key regulators modulating malignant conversion, we constructed a genetic network of DEGs in HGDN (DEGHGDN; n = 143) and eHCC (DEGeHCC; n = 277) to determine their functional enrichment of ER-stress-related genes and oncogenes. Our analysis revealed 5 genes (i.e., SPINK1,

RRAGD, CLDN15, CAP2, and ARK1C3) that potentially bridge DEGHGDN and DEGeHCC subsets (Fig. 5A). Among them, we focused in SPINK1, CAP2, and RRAGD, which were differentially expressed during multi-step hepatocarcinogenesis (permuted t test, P < 0.05 and fold difference > 0.3) (Supplementary Fig. S9A). Stepwise expression of these genes was validated in the pooled data (n = 293) and each data set of YSHCC (n = 131), GSE89377 (n = 87), and GSE6764 (n = 75), respectively (Fig. 5B; Supplementary Fig. S9B). Moreover, these genes showed tumor-specific expression as compared to the levels in non-tumor tissues in TCGA-LIHC (n = 421), GSE14520 (n = 486), GSE25097 (n = 557), and GSE87630 (n = 94) (Supplementary Fig. S9C). Our findings suggested that SPINK1, CAP2, and RRAGD might play potential oncodriver roles during hepatocarcinogenesis.

We then evaluated functional and clinical relevance of the candidate genes. SPINK1 expression was consistently correlated with clinical outcomes of overall survival (OS) or recurrence-free survival (RFS) in independent studies (GSE14520 and TCGA-LIHC; Fig. 5C). Moreover, univariate and multivariate analyses of TCGA data also revealed the prognostic significance of SPINK1 expression in TCGA-LIHC data (hazard ratio = 1.52, 95% C.I = 1.07-2.15, P = 0.018, univariate analysis; hazard ratio = 1.64, 95% C.I = 1.11-2.43, P = 0.013, multivariate analysis) (Supplementary Table S5). Cell culture 11

experiments also indicated that the overexpression of SPINK1 promoted proliferation as well as invasion activities in liver cancer cells (Supplementary Fig. S10A-C). Accordingly, we prioritized SPINK1 as a potential driver in the following analyses.

Clinical utility of SPINK1 was also evaluated via immunohistochemical staining of the 171 tissues, which included LC (n = 30), LGDN (n = 30), HGDN (n = 40), eHCC (n = 40), and pHCC (n = 41) specimens (Supplementary Table S2). We observed that SPINK1 protein levels gradually increased in eHCC (6/40; 15%) and pHCC (22/41; 54%) but were not detectable in precancerous lesions (Fig. 5D). This finding suggested that SPINK1 exhibits potential clinical utility as a differential diagnostic marker for distinguishing eHCC from noncancerous lesions.

SPINK1 expression is regulated by DNA methylation Previous integrative multiomic studies demonstrate that genes exhibiting transcriptional dysregulation and concomitant epigenomic or genomic aberrations play pivotal roles in cancer progression. Therefore, we considered genes which were inversely correlated with DNA methylation to be indicative of DNA-methylation-dependent transcriptional dysregulation (METcor, n = 48; Z- transformed correlation coefficient, r < −1.96; one-way Fisher–Pitman permutation test, P < 0.05) (Supplementary Fig. S11A). Interestingly, SPINK1 was identified again as a DNA methylation- dependent gene (Fig. 6A). Additionally, stepwise demethylation of SPINK1 at exon 1 (cg04577715) during hepatocarcinogenesis was observed in the YSHCC and GSE44970 datasets, respectively (Fig. 6B), and HCC-specific demethylation of SPINK1 relative to its status in non-tumor tissues was confirmed in multiple data sets [i.e., GSE60753 (n = 143), GSE73003 (n = 40), SNUHCC (HCC data from Seoul National University, n = 64), and TCGA-LIHC (n = 427)] (Supplementary Fig. S11B). Moreover, DNA methylation-dependent regulation of SPINK1 expression was validated by the inverse correlation seen between DNA methylation and SPINK1 expression in the pooled dataset (n = 467; r = −0.49, P < 0.001), which included TCGA-LIHC and GSE87630 (Fig. 6C). Furthermore, we confirmed these findings via cell-culture experiments, wherein transfection of DNA methyltransferase 1 (DNMT1) into liver cancer cell lines significantly suppressed SPINK1 expression (Fig. 6D; Supplementary Fig. S12). By contrast, treatment of these cells with a demethylating agent (5-aza-deoxycytidine) markedly upregulated SPINK1 expression (Fig. 6E). These findings strongly indicated that SPINK1 expression was regulated by DNMT1-dependent DNA methylation.

As shown in Fig. 5A, ER stress in HGDN may promote oncogene activation in eHCC. In order to test this hypothesis, we evaluated whether ER stress affected DNA-methylation-mediated SPINK1 expression. Upon treating liver cancer cells with ER-stress inducers, such as thapsigargin, tunicamycin, or dithiothreitol (DTT), SPINK1 expression was upregulated while DNMT1 levels were 12

significantly downregulated (Figs 6F, G). Induction of ER stress in the cells was confirmed by examining the expression levels of GRP78 (glucose-regulated protein, 78 kDa), an ER-stress marker (Supplementary Fig. S13). These results suggested that SPINK1 expression was regulated by epigenetic DNA methylation, which was promoted by induced ER stress.

Additionally, nuclear factor-kappaB (NF-B) is reported to be a regulator of SPINK1 expression (35), while ER stress acts as an activator of NF-B signaling (36,37). Therefore, we examined the role of NF-B in ER-stress-induced SPINK1 expression. However, we observed that neither ER-stress induction nor DNMT1 overexpression affected NF-B signaling (Fig. 6H, I). Moreover, short-hairpin RNA-mediated knockdown of NF-B p65 levels showed no effect on SPINK1 expression, indicating that NF-B signaling was not involved in the ER stress-induced SPINK1 expression (Supplementary Fig. S14A-C).

DISCUSSION

In this study, we performed multiomics profiling and integrative analyses to comprehensively demonstrate the sequential dynamics underlying multi-layered aberrations in DNA methylation, DNA copy numbers, mutations, and transcriptional activities during hepatocarcinogenesis. Our findings are summarized in Fig. 7.

Our results demonstrated that aberrant DNA methylation occurred during eHCC and pHCC development. In particular, we observed initial DNA methylation in eHCC (e.g., DKK3, SALL3, and SOX1) and further profound methylation in pHCC (e.g., CCNA1, HOXA9, WT1, PENK, TRIM58, and TBX4). Remarkably, we observed that DNA methylation process was initiated from CGI-flanking regions to CGIs during hepatocarcinogenesis.

DNA copy number analysis revealed that loss in DNA copy numbers initiated during eHCC, particularly in the TSGs (e.g., RB1) of 4q and 13q regions. This finding contrasts with previous results showing that small HCCs exhibit copy number gains at 1q and 8q (6.9%) (9). Indeed, we also observed marginal levels of copy number gains at 1q and 8q (3/9 of eHCC samples) (Fig. 2C); however, small HCC is defined as a tumor <2.0 cm in diameter, whereas eHCC is a histologically well- differentiated subgroup of small HCC that shows indistinct margins and represents the earliest tumor lesions (38). Hence, it is plausible that the subtle alterations at 4q and 13q in eHCC may not be discernible in the extended group of small HCCs. We are currently investigating this issue in order to validate our findings via large-scale studies of whole-genome sequencing data.

In-depth transcriptome analysis revealed stepwise reprogramming that induced increased immune activity in precancerous lesions. Increasingly, evidence indicates that the immune system affects tumor initiation as well as progression, and that complex interactions that occur between immune cells and cancer cells are capable of both inhibiting or enhancing tumor growth (39). The active-immune reactions taking place in precancerous tissues may participate in cancer-specific immune editing to eliminate tumor cells. During malignant conversion, eHCC was reprogrammed to express exhausted-immune and late-TGF-β signaling, which may be considered as a hallmark of eHCC development. Because TGF-β activity represents a primary mechanism that enables immune evasion in tumor microenvironment (40), we suggest that the transition of expression from an early TGF‐β signature to a late one may promote immune evasion by suppressing microenvironmental immune surveillance in eHCC.

In addition to immune-related activities, precancerous HGDN expressed ER-stress-related genes. In fact, a hepatitis viral infection in the liver tissues activates the immune system and alters mitochondrial function, which enhances oxidative DNA damage, promoting ER stress in cells (41). The resulting inflamed and damaged cells activate cell-proliferation‐related signals and promote tumor formation (42). Indeed, sustained ER stress in cancers fosters an immunosuppressive and pro- tumorigenic microenvironment (43). Moreover, our unsupervised analysis suggested that the induction of ER stress in precancerous tissues activates oncogenes, thereby promotes eHCC progression. SPINK1 was found as a potential driver gene linking the functional domains of ER stress to oncogene activation during eHCC development (Fig. 5A). As the limited data availability of LGDN and HGDN samples, we did not further analyze the detailed difference between them. Moreover, LGDN and HGDN showed similar expression pattern to each other compared to those of malignant lesions of eHCC or pHCC (see Fig. 4, and Supplementary Fig S7, S8).

SPINK1 exhibits tumor-promoting functions, revealing a clinical association in diverse cancer types, including breast, prostate and pancreatic cancer (44,45). In liver cancer, SPINK1 expression enhances aggressive phenotypes, such as tumor growth, portal vein invasion, and early recurrence (46). A recent study showed that stromal cells which produced SPINK1 following genotoxic treatment subsequently reprogrammed cancer cells and promoted EMT (47). However, the mechanisms associated with SPINK1 expression and its involvement in cancer initiation and clinical implications remain unclear. We found that SPINK1 expression was regulated by DNMT1-mediated DNA methylation. Therefore, we suggest that the elevated ER stress present in precancerous lesions may induce SPINK1 expression, which in turn promotes HCC development and progression. Furthermore, we demonstrated the clinical utility of SPINK1 as a potential immunostaining marker

for differential diagnosis of eHCC from precancerous lesions. These results imply that the modulation of ER stress, SPINK1 expression, or SPINK1 methylation can be new therapeutic targets for HCC treatment, although further extended studies might be required.

In conclusion, our comprehensive analysis of genomic and epigenomic regulation during the stepwise development of HCC provides novel mechanistic and clinical insights into the development of novel diagnostics and therapeutics for HCC.

REFERENCES

1. Nam SW, Park JY, Ramasamy A, Shevade S, Islam A, Long PM, et al. Molecular changes from dysplastic nodule to hepatocellular carcinoma through gene expression profiling. Hepatology 2005;42:809-18 2. Wurmbach E, Chen YB, Khitrov G, Zhang W, Roayaie S, Schwartz M, et al. Genome-wide molecular profiles of HCV-induced dysplasia and hepatocellular carcinoma. Hepatology 2007;45:938-47 3. Kaposi-Novak P, Libbrecht L, Woo HG, Lee YH, Sears NC, Coulouarn C, et al. Central role of c- Myc during malignant conversion in human hepatocarcinogenesis. Cancer Res 2009;69:2775-82 4. Marquardt JU, Seo D, Andersen JB, Gillen MC, Kim MS, Conner EA, et al. Sequential transcriptome analysis of human liver cancer indicates late stage acquisition of malignant traits. J Hepatol 2014;60:346-53 5. Villanueva A, Portela A, Sayols S, Battiston C, Hoshida Y, Mendez-Gonzalez J, et al. DNA methylation-based prognosis and epidrivers in hepatocellular carcinoma. Hepatology 2015;61:1945-56 6. Hlady RA, Zhou D, Puszyk W, Roberts LR, Liu C, Robertson KD. Initiation of aberrant DNA methylation patterns and heterogeneity in precancerous lesions of human hepatocellular cancer. Epigenetics 2017;12:215-25 7. Nault JC. High frequency of telomerase reverse-transcriptase promoter somatic mutations in hepatocellular carcinoma and preneoplastic lesions. Nat Commun 2013;4:2218 8. Qiu J, Peng B, Tang Y, Qian Y, Guo P, Li M, et al. CpG Methylation Signature Predicts Recurrence in Early-Stage Hepatocellular Carcinoma: Results From a Multicenter Study. J Clin Oncol 2017;35:734-42 9. Torrecilla S, Sia D, Harrington AN, Zhang Z, Cabellos L, Cornella H, et al. Trunk events present minimal intra- and inter-tumoral heterogeneity in hepatocellular carcinoma. J Hepatol 2017 10. Qingyu S, Woo EJ, Kyungbun L, Seok KH, Doo YH, Yean KS, et al. Barrier to autointegration factor 1, procollagen-lysine, 2-oxoglutarate 5-dioxygenase 3, and splicing factor 3b subunit 4 as early-stage cancer decision markers and drivers of hepatocellular carcinoma. Hepatology 2018;67:1360-77 11. Cancer Genome Atlas Research Network. Electronic address wbe, Cancer Genome Atlas Research N. Comprehensive and Integrative Genomic Characterization of Hepatocellular Carcinoma. Cell 2017;169:1327-41 e23 12. Woo HG, Choi JH, Yoon S, Jee BA, Cho EJ, Lee JH, et al. Integrative analysis of genomic and epigenomic regulation of the transcriptome in liver cancer. Nat Commun 2017;8:839 13. Chaisaingmongkol J, Budhu A, Dang H, Rabibhadana S, Pupacdi B, Kwon SM, et al. Common Molecular Subtypes Among Asian Hepatocellular Carcinoma and Cholangiocarcinoma. Cancer Cell 2017

14. Srivastava S, Ghosh S, Kagan J, Mazurchuk R, Boja E, Chuaqui R, et al. The Making of a PreCancer Atlas: Promises, Challenges, and Opportunities. Trends in Cancer 2018;4:523-36 15. International Consensus Group for Hepatocellular NeoplasiaThe International Consensus Group for Hepatocellular N. Pathologic diagnosis of early hepatocellular carcinoma: a report of the international consensus group for hepatocellular neoplasia. Hepatology 2009;49:658- 64 16. Joo T, Choi JH, Lee JH, Park SE, Jeon Y, Jung SH, et al. SEQprocess: a modularized and customizable pipeline framework for NGS processing in R package. BMC Bioinformatics 2019;20:90 17. Mah WC, Lee CG. DNA methylation: potential biomarker in Hepatocellular Carcinoma. Biomark Res 2014;2:5 18. Ding Z, Qian YB, Zhu LX, Xiong QR. Promoter methylation and mRNA expression of DKK-3 and WIF-1 in hepatocellular carcinoma. World J Gastroenterol 2009;15:2595-601 19. Shikauchi Y, Saiura A, Kubo T, Niwa Y, Yamamoto J, Murase Y, et al. SALL3 interacts with DNMT3A and shows the ability to inhibit CpG island methylation in hepatocellular carcinoma. Mol Cell Biol 2009;29:1944-58 20. Tsao CM, Yan MD, Shih YL, Yu PN, Kuo CC, Lin WC, et al. SOX1 functions as a tumor suppressor by antagonizing the WNT/beta-catenin signaling pathway in hepatocellular carcinoma. Hepatology 2012;56:2277-87 21. Doi A, Park IH, Wen B, Murakami P, Aryee MJ, Irizarry R, et al. Differential methylation of tissue- and cancer-specific CpG island shores distinguishes human induced pluripotent stem cells, embryonic stem cells and fibroblasts. Nat Genet 2009;41:1350-3 22. Jia D, Wei L, Guo W, Zha R, Bao M, Chen Z, et al. Genome-wide copy number analyses identified novel cancer genes in hepatocellular carcinoma. Hepatology 2011;54:1227-36 23. Wang K, Lim HY, Shi S, Lee J, Deng S, Xie T, et al. Genomic landscape of copy number aberrations enables the identification of oncogenic drivers in hepatocellular carcinoma. Hepatology 2013;58:706-17 24. Kaposi-Novak P, Libbrecht L, Woo HG, Lee YH, Sears NC, Conner EA, et al. Central role of c- Myc during malignant conversion in human hepatocarcinogenesis. Cancer Res 2009;69:2775-82 25. Schulze K, Imbeaud S, Letouze E, Alexandrov LB, Calderaro J, Rebouissou S, et al. Exome sequencing of hepatocellular carcinomas identifies new mutational signatures and potential therapeutic targets. Nature genetics 2015;47:505-11 26. Woo HG, Park ES, Lee JS, Lee YH, Ishikawa T, Kim YJ, et al. Identification of potential driver genes in human liver carcinoma by genomewide screening. Cancer Res 2009;69:4059-66 27. Nault JC, Calderaro J, Di Tommaso L, Balabaud C, Zafrani ES, Bioulac-Sage P, et al. Telomerase reverse transcriptase promoter mutation is an early somatic genetic alteration in the transformation of premalignant nodules in hepatocellular carcinoma on cirrhosis. Hepatology 2014;60:1983-92 28. Nakagawa H, Fujita M, Fujimoto A. Genome sequencing analysis of liver cancer for precision medicine. Semin Cancer Biol 2018 29. Eggert T, Wolter K, Ji J, Ma C, Yevsa T, Klotz S, et al. Distinct Functions of Senescence- Associated Immune Responses in Liver Tumor Surveillance and Tumor Progression. Cancer Cell 2016;30:533-47 30. Sia D, Jiao Y, Martinez-Quetglas I, Kuchuk O, Villacorta-Martin C, Castro de Moura M, et al. Identification of an Immune-specific Class of Hepatocellular Carcinoma, Based on Molecular Features. Gastroenterology 2017 31. Reich M, Liefeld T, Gould J, Lerner J, Tamayo P, Mesirov JP. GenePattern 2.0. Nat Genet 2006;38:500-1 32. Jiang Y, Li Y, Zhu B. T-cell exhaustion in the tumor microenvironment. Cell Death Dis 2015;6:e1792 16

33. Achyut BR, Yang L. Transforming growth factor-beta in the gastrointestinal and hepatic tumor microenvironment. Gastroenterology 2011;141:1167-78 34. Bachman KE, Park BH. Duel nature of TGF-beta signaling: tumor suppressor vs. tumor promoter. Curr Opin Oncol 2005;17:49-54 35. Chen F, Long Q, Fu D, Zhu D, Ji Y, Han L, et al. Targeting SPINK1 in the damaged tumour microenvironment alleviates therapeutic resistance. Nat Commun 2018;9:4315 36. Lin WC, Chuang YC, Chang YS, Lai MD, Teng YN, Su IJ, et al. Endoplasmic reticulum stress stimulates p53 expression through NF-kappaB activation. PLoS One 2012;7:e39120 37. Garg AD, Kaczmarek A, Krysko O, Vandenabeele P, Krysko DV, Agostinis P. ER stress-induced inflammation: does it aid or impede disease progression? Trends Mol Med 2012;18:589-98 38. Park YN. Update on Precursor and Early Lesions of Hepatocellular Carcinomas. Archives of Pathology & Laboratory Medicine 2011;135:704-15 39. Vinay DS, Ryan EP, Pawelec G, Talib WH, Stagg J, Elkord E, et al. Immune evasion in cancer: Mechanistic basis and therapeutic strategies. Seminars in Cancer Biology 2015;35:S185-S98 40. Tauriello DVF, Palomo-Ponce S, Stork D, Berenguer-Llergo A, Badia-Ramentol J, Iglesias M, et al. TGFβ drives immune evasion in genetically reconstituted colon cancer metastasis. Nature 2018;554:538 41. Kim SY, Kyaw YY, Cheong J. Functional interaction of endoplasmic reticulum stress and hepatitis B virus in the pathogenesis of liver diseases. World journal of gastroenterology 2017;23:7657-65 42. Hui-Ching W, Wenya H, Ming-Der L, Ih-Jen S. Hepatitis B virus pre-S mutants, endoplasmic reticulum stress and hepatocarcinogenesis. Cancer Science 2006;97:683-8 43. Cubillos-Ruiz JR, Bettigole SE, Glimcher LH. Tumorigenic and Immunosuppressive Effects of Endoplasmic Reticulum Stress in Cancer. Cell 2017;168:692-706 44. Ateeq B, Tomlins SA, Laxman B, Asangani IA, Cao Q, Cao X, et al. Therapeutic targeting of SPINK1-positive prostate cancer. Sci Transl Med 2011;3:72ra17 45. Soon WW, Miller LD, Black MA, Dalmasso C, Chan XB, Pang B, et al. Combined genomic and phenotype screening reveals secretory factor SPINK1 as an invasion and survival factor associated with patient prognosis in breast cancer. EMBO Mol Med 2011;3:451-64 46. Lee YC, Pan HW, Peng SY, Lai PL, Kuo WS, Ou YH, et al. Overexpression of tumour-associated trypsin inhibitor (TATI) enhances tumour growth and is associated with portal vein invasion, early recurrence and a stage-independent prognostic factor of hepatocellular carcinoma. Eur J Cancer 2007;43:736-44 47. Chen F, Long Q, Fu D, Zhu D, Ji Y, Han L, et al. Targeting SPINK1 in the damaged tumour microenvironment alleviates therapeutic resistance. Nature Communications 2018;9:4315

FIGURE LEGENDS

Figure 1. Stepwise aberration of DNA methylation during HCC development (A) Heatmap showing DNA methylation levels of 595 DMPs. Eight top-ranked genes among the hypomethylated genes (n = 40) and the hypermethylated genes (n = 555) in eHCC or pHCC are shown. Previously known DNA methylated genes in HCCs are indicated (blue). (B) A plot showing the functional enrichment scores of the genes mapped to DMPpHCC. Sphere size indicates the number of the genes in each gene set. (C) Distribution of the average DMP beta-values in each group is shown.

(D) Average methylation levels in each group according to the genomic coordinates in relation to CGI are shown.

Figure 2. DNA copy number aberration during stepwise hepatocarcinogenesis (A) Circos plots showing average CNA values according to the genomic coordinates in each group. (B) Fold differences for the DCNAs for DN vs. LC (top), eHCC vs. DN (middle), and pHCC vs. eHCC (bottom) are plotted, and the chromosomal regional CNA gains (red lines) and losses (blue lines) are indicated, respectively. Early CNA alterations in eHCC at 4q and 13q are indicated (cyan lines) (C) A heatmap shows the DCNAs according to the chromosomal locations. DCNA gains and losses from LICA-FR and TCGA-LIHC, and oncogenes and TSGs are indicated in sidebars (right). 4q and 13q regions are indicated (yellow box), and oncogenes and TSGs in this region are shown in zoomed color bars (rightmost). (D) Pie-plots showing the proportion of the overlap between CNA genes and oncogenes (n = 674, top) or the TSGs (n = 1,088, bottom) in each group of YSHCC and TCGA-LIHC data, respectively. (E, F, G) Correlations between CNAs enrichment scores (ESCNA) and the corresponding gene expression levels (ESEXP) of the DCNAeHCC (4q and 13q) genes are shown for YSHCC (n = 32, E), GSE65373 (n = 38, F), or TCGA-LIHC (n = 365, G), respectively. (H) A Kaplan-Meier plot analysis of overall survival (OS) between the patients with gain (ESEXP> 0 and ESCNA >0, n = 173) and loss (ESEXP < 0 and ESCNA < 0, n = 114) of the DCNAeHCC (4q and 13q) genes is shown for TCGA-LIHC data (bottom right).

Figure 3. Mutation alterations during stepwise hepatocarcinogenesis (A) A diagram showing the number of differentially mutated genes (DMUT) in each group. (B) A heatmap showing mutations including the oncogenes and TSGs in DMUT and the significantly mutated genes from TCGA-LIHC. Mutations at TERT promoter region are shown (top). (C) Mutation frequencies in the driver mutation pathways are shown.

Figure 4. Transcriptomic alterations during stepwise hepatocarcinogenesis (A) A heatmap showing the expression of the DEGs among the groups (top). Enrichment scores for each of the gene signatures of immune, TGF-β, ER stress, oncogene, and TSG are calculated in each sample. Active and exhausted immune types were determined by applying the NTP algorithm based on the expression of ‘Activated stroma’ and ‘Normal stroma’ signatures (false discovery rate < 0.05). TGF-β activity is estimated with early- and late- TGF-β signatures, respectively, and the expression transition from early- to late-TGF-β signatures (ESlate-TGFB - ESearly-TGFB) is plotted. Onco-activity is calculated as ESoncogenes - ESTSGs. (B) A bar plot shows the proportion of immunotypes in each group. (C) Gene set enrichment analysis for the ER-stress genes between HGDN and LC groups (left) and oncogenes between eHCC and HGDN groups (right) are shown. 18

Figure 5. SPINK1 expression is a putative marker of eHCC development

(A) Genetic networks of DEGHGDN and DEGeHCC genes were constructed with the interactions of physical, genetic, and pathways by using GeneMania software. ER stress-related genes in DEGHGDN, and the oncogenes in DEGeHCC are indicated, and their connectors are shown. (B) A heatmap showing the stepwise expression of SPINK1, CAP2, and RRAGD during hepatocarcinogenesis (top). Boxplots showing the distribution of expression levels of SPINK1, CAP2, and RRAGD in the pooled data, YSHCC, GSE89377, and GSE6764, respectively (bottom). (C) Kaplan-Meir plot analyses of overall survival (OS) and recurrence-free survival (RFS) for the patient groups stratified by above or below the average expression levels of SPINK1 in each HCC cohort of GSE14520 and LIHC-TCGA, respectively. Follow-up time for OS or RFS is truncated to 5 years. (D) Immunohistochemical staining of H&E (top) and SPINK1 (bottom) in LC, LGDN, HGDN, eHCC, and pHCC specimens is shown (left). A bar plot shows the semi-quantitative positivity of SPINK1 expression in LC, LGDN, HGDN, eHCC, and pHCC specimens (right).

Figure 6. SPINK1 expression is regulated by DNA hypomethylation (A) Venn-diagram showing the overlap of SPINK1 between the DMPs and METcor genes. (B) Stepwise methylation of SPINK1 during multi-step hepatocarcinogenesis is shown in YSHCC and GSE44970, respectively. (C) Inverse correlation of SPINK1 expression with its DNA methylation levels is shown in pooled HCC including YSHCC, LIHC-TCGA, and GSE87630. (D) SPINK1 mRNA expression levels in the indicated cells with stable transfection of Vector or DNMT1 were measured by quantitative RT-PCR. (E) 5-aza-DC (5-aza-deoxycytidine, 10 µM) is treated on the liver cancer cells (Huh7, Hep3B, and HepG2), and the SPINK1 mRNA expression levels are measured. (F-H) Liver cancer cell lines (Huh7, Hep3B, HepG2) are treated with three ER stress inducers including thapsigargin (Thap, 1 µM) for 72 h, tunicamycin (Tuni, 1 µg/ml) for 24 h, and dithiothreitol (DTT, 1mM) for 24 h. SPINK1 expression level was measured by quantitative RT-PCR (F), and the protein levels of DNMT1 (G) and NF-kB signaling pathway were measured by immunoblot (H). (I) NF-kB signaling in the indicated cells with stable transfection of Vector or DNMT1 were measured by immunoblot. Statistical significance is indicated (*P < 0.05, **P < 0.01, and ***P < 0.001, Student’s T-test).

Figure 7. Graphical summary of the dynamics of genomic, epigenomic, and transcriptomic aberrations during stepwise hepatocarcinogenesis

Figure 1

A LC HGDN eHCC pHCC B METhypo (40) Nervous system development # of SPINK1 DKK4 genes SLC44A4 KBTBD11 Embryonic skeletal DCPEY2 SCAPER system development 20 CBX2 CADM1 40 Cell-cell signaling 60 Pattern speicification 80 process METhyper (555) Central nervous eHCC pHCC system delopment DKK3 DKK3 0 2 4 6 SALL3 SALL3 Enrichment score SOX1 CCNA1 JDP2 HOXA9 C DUOX1 WT1 LC HGDN eHCC pHCC DMPs (595) IRX4 PENK TCERG1L TRIM58 3 CYB561 TBX4

Density 1

0 low high Beta values 0.00 0.25 0.50 0.75 1.00 Beta values D LC eHCC 0.04 HGDN pHCC

0.02

0.00

Beta values -0.02

-4kb -2kb 2kb 4kb -0.04 Opensea Shelf Shore CGIShore Shelf Opensea

Figure 2

22 22 A 21 B 21 20 20 1 1 19 19

18 18 2 2 17 17

16 16

3 3 15 15

14 14 CNA 4 CNA (fold) 4

13 13

5 5 12 12

11 6 11 6

10 7 10 7

8 8 9 9

LC HGDN eHCC pHCC HGDN vs. LC eHCC vs.HGDN pHCC vs. eHCC

C D gain loss none LC HGDN eHCC pHCC LC HGDN eHCC pHCC LIHC Oncogene TMPRSS1 SMR3B MOB1B FGF5 UNC5C 1p EIF4E NFKB1 TSG CXXC4 TET2 AIMP1 LEF1 PITX2 NDST4 1q SYNPO2 E F PRDM5 YSHCC GSE65373 FAT4 0.2 2p PGRMC2 0.25 2q PCDH10 0.1 3q INPP4B

0.00 EXP EXP 0.0 4p FBXW7 ES

4q ES HPGD -0.25 -0.1 5p GMP6A 6p -0.2 6q SPATA4 -0.50 7p -0.3 7q 8p 9p RB1 -0.5 0.0 0.5 1.0 -0.6 -0.3 0.0 0.3 9q 8q INTS6 ESCNA 10p ESCNA 10q PCDH17 11p 13q PCDH9 G H 14q TCGA-LIHC TCGA-LIHC 11q DACH1 12p 15q 1.0 12q 16p DIS3 High (173) 0.25 0.8 0.6 16q LICA-FR gain 0.00 LICA-FR loss EXP 0.4 17q Low (114) ES LIHC gain -0.25 Probability 0.2 19q LIHC loss 20p Oncogene p=0.019 20q -0.50 0.0 TSG -0.5 0.0 0.5 0 10 20 30 40 50 60 21q ES OS(months) low high CNA CNA

Figure 3 A

Differentially Mutated Genes LC HGDN (156) eHCC (34) pHCC (11) (DMUT)

B LC HGDN eHCC pHCC Frequency (%) TERT promoter 0 5 10 15 20 25 30 APOB HIST1H1C CREB3L3 NFE2L2 AHCTF1 ARID1A GPATCH4 LZTR1 RB1 EPB41 LIMA1 RB1CC1 TCGA mutation ITGAV FAM189B DMUT ZNF292 ERBB2 TSG PTPRJ Oncogene DUSP6 PINX1 PELP1 Not determined MT1G Wild type DAB2 DDHD2 RBM5 AZIN1 TP53 CTNNB1 TPR KDM5A HLTF BAP1 TSC2 CREBBP KEAP1 KRAS C Wnt/b-Catenin Ras/ERK AKT/mTOR

15 7.5 10 10 5.0 5 2.5 5 Frequency (%) Frequency 0 0.0 0 LC LC LC eHCC eHCC pHCC pHCC eHCC HGDN pHCC HGDN HGDN

Liver metabolic Oxidative stress Chromatin regulators TP53/cell-cycle 20 30 7.5 10 15 20 5.0 10 5 10 2.5 Frequency (%) Frequency 5

0 0 0.0 0 LC LC LC LC eHCC eHCC eHCC eHCC pHCC pHCC pHCC pHCC HGDN HGDN HGDN HGDN

Figure 4

A HL LC LGDN HGDN eHCC pHCC YSHCC (n=131) GSE89377 (n=87) GSE6764 (n=75)

HCC immune (112) Senescence (50) SASP (60) Immune signatures Onco_SASP (38) Cytokine pathway (187) Cytokine genes (456)

Activated stroma (25) Normal stroma (23) Immunotype Exhausted immune Active immune Not determined

early TGFβ (138) late TGFβ (111) 0.6 TGFB 0.3 0.0 TGFβ activity -0.3 -0.6

ER stress ER stress (233) UPR (113)

Oncogenes (674) TSGs (1,088) Oncogenes 0.2 & TSGs 0.1 Onco-activity 0.0 -0.1 -0.2 low high low high Expression levels Enrichment Scores

B Immunotype C Response to ER stress Oncogene 100 (n=233) (n=674)

50 0.0 0.2 0.0 0.5

25 Percentage (%) 0 nrichmen t Sc ore s nrichmen t Sc ore s E E NL LC eHCC pHCC LGDN −0.8 −0.4 HGDN ES=0.452, P=0.01 ES=0.306, P=0.05 −1.0 −0.5 Exhausted immune 0 5000 10000 15000 0 5000 10000 15000 Active immune Not determined HGDN vs. LC eHCC vs. HGDN

Figure 5 Oncogenes A ER stress PDGFRB SEL1L S100A4 SDF2L1 LEF1

TARBP1 GPR158 SULT1C2 COX6C LPL OTUD6B AFP TPM2 MGP ESM1 THBS2 RPL7 HMGCR PBK RRM2 PIK3C2B F13A1 ZBED3 MMP9 UBD

DCK CEP350

HEYL CCNA2

COL8A1 TOP2A

SNAI2 NOTCH2NL

TSNAX IL7R

GNPAT ZHX1 PDIA4 VWF PSTPIP2 RAP2A FAM20B

PPP2R5A TTK

FLVCR1 EIF3H NRAS GGPS1 TP53BP2 CCDC170 CTGF

CCDC80 ARHGEF3

TOMM40L THY1

CD3D PRC1

MDK TNFSF4

UBE2T CNN1

ZNF281 SFN

RHOD ERO1B SERINC2 WNK3 BUB1B ALPK2 SLC5A6 LAPTM4B SLC28A1 UGGT1 CHRNA4 FICD MLEC TMEM64 MSH2 PLA2G2A GSKIP

TOMM20 RPL30 CSTA GPR37

TNFSF14 ZNF165 MTR IDI1

C1QTNF3 CNDP1 GLUL SLC30A10

SLC2A10 SLC35B4 RBBP5 CYP7A1

COL5A3 CPN2 MEF2C SDCBP FGF13 RND1 ECHDC3 MBNL3 TBC1D16

IRX3 NDC80 MAMDC4 HMGB3

TTC13 TRAF5

EPO CYP17A1 NUSAP1 TXNIP DEG-HGDN ANGPTL4 ABCB4 MAP3K5

PRIM1 SGCE XK BBOX1 SPINK1 CDK1 COL1A1 ADGRV1 NCAPG CYP4F22

ASCL1 KCNB1

KIF20A ASPH

TM4SF20 SOCS1

SERPINI1 ALDH3A1

SF3B4 FAM213A TRIM9 SDS

NEK2 WWP1 DEG-eHCC ECHDC2 MANF STMN1 HYOU1 PLVAP DPP4 CRP TMEM45B ACTG2 CD34

CPQ COL6A3

PIR EFNA2

PCOLCE2 GSTA4 GK AS3MT PTTG1 RPS4Y1 RHOBTB1 RRAGD HMGN4 NSMCE2 TRPM4 SCUBE1 ER stress

HLTF NPC2

APOA4 IFI27 KIF4A ARID4B

RFX5 RACGAP1

RASD1 DBNDD1 HSPA5 DEG CENPW KNTC1 DEG TRIM33 TBX15 NEDD4 eHCC HGDN SFRP4 MAP2

ACTA2 SMOC2

TLE6 PTH2R

DLGAP5 IGF2BP3 Oncogenes

SATB2 HSPA13 MPC2 FOXN2

CNIH4 ANLN UBXN10 ATF7IP2 CLDN15 CCNB2 LYZ SLC44A5

FNDC4 IGSF9 GFPT1 ATAD2 NUAK1

SLC38A6 ENAH AKR1B10 STEAP1 (n=277) SRXN1 PABPC1

ENPP7 (n=143) CDC6 FDPS HOXA13 LBR Connectors

SLC35C1 LPGAT1 ASPM NEK7

ITM2A COL5A1

LSMEM1 G0S2

FBLN1 VCAM1

CLPTM1L GREM2 FGF21 NOTCH3 NCOA2 LOXL4 SLC39A14 CAP2 CCNB1 PAGE4 NDRG1 HLA-DMA PRKAA2

CRELD2 DCPS

PEA15 ITPR2

CLGN NNMT Genetic interaction

PLEKHO1 EFR3A

MYBPH ELL3

GGH RBM24

MUC13 RAMP1

RGS16 CENPF

CRYM FAM20A ARID3A ZBTB41 NUF2

RAD54B BCHE

SMPX CTNNBL1

GPX2 GMDS

EDEM3 KLF13 OLFML2A CGA LEPR

ARMC6 FGGY LGR5 CUX2

CFHR3 MYOM1 Pathway SLC41A1 SPARCL1 SLC17A2 MAB21L3

CHI3L1 ETV1 RABEPK EIF3E HHIPL2 IRS1 TRIM55 SLC1A1 NEB FAM83D AKR1C3 FAM83H SLC1A3

JADE2 CPNE3

TBCE CDC20

HMMR VRK1

TMEM98 EDIL3

HMGCS1 PALLD Physical interaction

DNAJC3 UBE2C ANKRD27 ACSL4

SOAT2 CALCRL

APOBEC3B SESTD1

GPR88 SPARC

RAB29 LAMC1

RHOBTB3 STK39

MYH11 GPAM

ADAMDEC1 FBN1

CDKN2C WASF2 CXCR4 PTGDS COL15A1 CELSR3 IGF2BP2

TMEM97 RAI14

CABYR SMYD3

CAPG PODXL

COL5A2 ROBO1 DNAJB9 SMAD5 EBF1 LRP11 CEBPA

RBP7 FHIT PNMA6A JAZF1 GLMP DHRS2 SOWAHA GPC3 CTHRC1 PHYHIPL PEG10 IQSEC1 S100P COL1A2 POSTN WDYHV1 CRIP1 FOXN3 TCF4 SVIL DNAJB11 ZIC2 CDKN3 CREB3L3 SMO

B HL (n=23) YSHCC (n=131) LC (n=55) GSE89377 (n=87) SPINK1 LGDN (n=21) GSE6764 (n=75) CAP2 HGDN (n=46) eHCC (n=53) low high RRAGD pHCC (n=95) Expression levels

pooled data (293) YSHCC (131) GSE6764 (75) GSE89377(87) 8 8 6 6 6 6 4 4 4 2 4 2 2 0 0 2 0

(SPINK1) (SPINK1) -2 (SPINK1) (SPINK1) -2 -2 0 -4 -4 mRNA expression mRNA expression mRNA expression

mRNA expression -4 -2 -6 -6 NL LC LC LC LC eHCC pHCC eHCC pHCC LGDN HGDN eHCC pHCC eHCC pHCC HGDN LGDN LGDN HGDN HGDN

C GSE14520 (n=247) GSE14520 (n=247) TCGA-LIHC (n=371) 1.0 1.0 1.0 0.8 0.8 0.8 0.6 0.6 0.6

0.4 0.4 0.4 Probablity Probablity Probablity 0.2 0.2 0.2 P=0.024 P=0.006 P=0.023 0.0 0.0 0.0 0 10 30 50 70 0 10 30 50 70 0 10 20 30 40 50 60 OS (Months) RFS (Months) OS (Months)

D LC LGDN HGDN eHCC pHCC n=171, P < 0.001 100 Histoscore 80

H&E 0 60 1 ositivity(%)

p 40 2 3 NK 1 20 SP I 0

SPINK1 LC LGDN HGDN eHCC pHCC (30) (30) (30) (40) (41)

Figure 6

A B YSHCC (n=32) C pooled HCC (n=467) 1.0 15 DMP METcor 0.8 10 0.6 5 5941 47 0.4 (SPINK1) (SPINK1) 0

DNA Methylation DNA 0.2 mRNA Expression 0.0 cg04577715 r=-0.49 -5 p<0.001

SPINK1 LC 0.25 0.50 0.75 1.00 eHCC pHCC DNA methylation HGDN (SPINK1)

D E Huh7 Hep3B HepG2 1.5 Huh7 Hep3B HepG2 * ** ** ** **** 3 40 * ** 15 ** 1.0 30 2 10 20 0.5 1

(SPINK1) 5

(SPINK1) 10 mRNA Expression mRNA Expression 0 0 0 0.0 0 2 4 0 2 4 0 2 4 Vec DNMT1 Vec DNMT1 Vec DNMT1 5-aza-dC (day) 5-aza-dC (day) 5-aza-dC (day)

Huh7 Hep3B HepG2 ** ** ** F *** ** * G *** *** 40 *** 600 25 Huh7 Hep3B HepG2 20 30 DNMT1 400 15 20 ß-actin 10 (SPINK1) 200 + - - - 10 CTL + - - - + - - - 5 Thapsigargin - + - - - + - - - + - - mRNA Expression 0 0 0 Tunicamycin - - + - - - + - - - + - CTL + - - - + - - - + - - - DTT - - - + - - - + - - - + Thapsigargin - + - - - + - - - + - - Tunicamycin - - + - - - + - - - + - DTT - - - + - - - + - - - +

Huh7 Hep3B HepG2 H I Huh7 Hep3B HepG2 p65 p65 pIκBα pIκBα IκBα IκBα β-actin β CTL + - - - + - - - + - - - -actin Thapsigargin - + - - - + - - - + - - DNMT1 - + - + - + Tunicamycin - - + - - - + - - - + - DTT - - - + - - - + - - - +

Figure 7

LC HGDN eHCC pHCC

CGI SALL3, SOX1, DKK3 DNA Shelf/ Shore methylation Hypo SPINK1 methylation 1q, 8q, 19q, 20q, oncogenes Gain/loss CNA 1p, 8p, 16p, 16q, TSGs

eHCC loss 4q, 13q, TSGs

Trunk mutation TP53, CTNNB1, BAP1

Liver metabolic Mutation Pathway Oxidative stress mutation Chromatin regulators, TP53/Cell-cycle, Wnt/ß-Catenin, Ras/ARK AKT/mTOR

Immune Immune Active-immune Exhausted-immune

TGFB Late-TGFB Early-TGFB Transcription ER stress

Onco- Oncogenic activity Tumor-suppressive

Stemness

Dynamics of genomic, epigenomic, and transcriptomic aberrations during stepwise hepatocarcinogenesis

Byul A Jee, Ji-Hye Choi, Hyungjin Rhee, et al.

Cancer Res Published OnlineFirst September 10, 2019.

Updated version Access the most recent version of this article at: doi:10.1158/0008-5472.CAN-19-0991

Supplementary Access the most recent supplemental material at: Material http://cancerres.aacrjournals.org/content/suppl/2019/09/10/0008-5472.CAN-19-0991.DC1

Author Author manuscripts have been peer reviewed and accepted for publication but have not yet Manuscript been edited.

E-mail alerts Sign up to receive free email-alerts related to this article or journal.

Reprints and To order reprints of this article or to subscribe to the journal, contact the AACR Publications Subscriptions Department at [email protected].

Permissions To request permission to re-use all or part of this article, use this link http://cancerres.aacrjournals.org/content/early/2019/09/10/0008-5472.CAN-19-0991. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.