Multiomics Integration Reveals the Landscape of Prometastasis Metabolism in Hepatocellular Carcinoma

The systematic investigation of gene mutation and expression is important to discover novel biomarkers and therapeutic targets in cancers. Here, we integrated genomics, transcriptomics, proteomics, and metabolomics to analyze three hepatocellular carcinoma (HCC) cell lines with differential metastatic potentials. The results revealed the profile of the prometastasis metabolism potentially associated with HCC metastasis. The multiomic analysis identified 12 genes with variations at multiple levels from three metabolic pathways, including glycolysis, starch, and sucrose metabolism, and glutathione metabolism. Furthermore, uridine diphosphate (UDP)-glucose pyrophosphorylase 2 (UGP2), was observed to be persistently up-regulated with increased metastatic potential. UGP2 overexpression promoted cell migration and invasion and enhanced glycogenesis in vitro. The role of UGP2 in metastasis was further confirmed using a tumor xenograft mouse model. Taken together, the compendium of multiomic data provides valuable insights in understanding the roles of shifted cellular metabolism in HCC metastasis. The systematic investigation of gene mutation and expression is important to discover novel biomarkers and therapeutic targets in cancers. Here, we integrated genomics, transcriptomics, proteomics, and metabolomics to analyze three hepatocellular carcinoma (HCC) cell lines with differential metastatic potentials. The results revealed the profile of the prometastasis metabolism potentially associated with HCC metastasis. The multiomic analysis identified 12 genes with variations at multiple levels from three metabolic pathways, including glycolysis, starch, and sucrose metabolism, and glutathione metabolism. Furthermore, uridine diphosphate (UDP)-glucose pyrophosphorylase 2 (UGP2), was observed to be persistently up-regulated with increased metastatic potential. UGP2 overexpression promoted cell migration and invasion and enhanced glycogenesis in vitro. The role of UGP2 in metastasis was further confirmed using a tumor xenograft mouse model. Taken together, the compendium of multiomic data provides valuable insights in understanding the roles of shifted cellular metabolism in HCC metastasis. Hepatocellular carcinoma (HCC) 1The abbreviations used are: HCC, hepatocellular carcinoma; CNAs, copy number alterations; DTT, dithiothreitol; FBS, fetal bovine serum; GCLM, glutamate-cysteine ligase regulatory subunit; GLUT, glucose transporters; GPI, glucose-6-phosphate isomerase; NCBI, National Center for Biotechnology Information; PFKP, ATP-dependent 6-phosphofructokinase; PKM2, pyruvate kinase M2; PYGB, glycogen phosphorylase; ROS, reactive oxygen species; SNP, single-nucleotide polymorphism; UDP, uridine diphosphate; UGP2, UDP-glucose pyrophosphorylase 2; WGCNA, weighted correlation network analysis. 1The abbreviations used are: HCC, hepatocellular carcinoma; CNAs, copy number alterations; DTT, dithiothreitol; FBS, fetal bovine serum; GCLM, glutamate-cysteine ligase regulatory subunit; GLUT, glucose transporters; GPI, glucose-6-phosphate isomerase; NCBI, National Center for Biotechnology Information; PFKP, ATP-dependent 6-phosphofructokinase; PKM2, pyruvate kinase M2; PYGB, glycogen phosphorylase; ROS, reactive oxygen species; SNP, single-nucleotide polymorphism; UDP, uridine diphosphate; UGP2, UDP-glucose pyrophosphorylase 2; WGCNA, weighted correlation network analysis. is the second leading cause of cancer-related death in the world (1.Ferlay J. Soerjomataram I. Dikshit R. Eser S. Mathers C. Rebelo M. Parkin D.M. Forman D. Bray F. Cancer incidence and mortality worldwide: Sources, methods and major patterns in GLOBOCAN 2012.Int. J. Cancer. 2015; 136: E359-E386Crossref PubMed Scopus (21243) Google Scholar). Liver transplantation and resection are believed to be the best approaches to treat HCC and to deliver long-term survival for HCC patients (2.Deng G.L. Zeng S. Shen H. Chemotherapy and target therapy for hepatocellular carcinoma: New advances and challenges.World J. Hepatol. 2015; 7: 787-798Crossref PubMed Scopus (124) Google Scholar, 3.Morise Z. Kawabe N. Tomishige H. Nagata H. Kawase J. Arakawa S. Yoshida R. Isetani M. Recent advances in liver resection for hepatocellular carcinoma.Front Surg. 2014; 1: 21Crossref PubMed Scopus (31) Google Scholar). However, these methods are not applicable for advanced HCC, and the overall prognosis for HCC remains poor mostly due to tumor metastasis and recurrence (4.Desai J.R. Ochoa S. Prins P.A. He A.R. Systemic therapy for advanced hepatocellular carcinoma: An update.J. Gastrointest. Oncol. 2017; 8: 243-255Crossref PubMed Scopus (53) Google Scholar). Metastasis, the spread of tumor from its primary site to other parts of the body, defines the switch between benign tumor and malignant cancer. Metastasis is estimated to be responsible for ∼90% of cancer-associated deaths (5.Valastyan S. Weinberg R.A. Tumor metastasis: Molecular insights and evolving paradigms.Cell. 2011; 147: 275-292Abstract Full Text Full Text PDF PubMed Scopus (2646) Google Scholar). To explore the underlying mechanisms of HCC metastasis is crucial for developing novel and more efficient HCC treatments, especially for advanced HCCs. The reprogramming of cellular metabolism has been recognized as a key hallmark of cancer (6.Hanahan D. Weinberg R.A. Hallmarks of cancer: The next generation.Cell. 2011; 144: 646-674Abstract Full Text Full Text PDF PubMed Scopus (42713) Google Scholar). Since Otto Warburg first reported that some cancer cells use the glycolysis pathway for energy production even in the presence of oxygen, it has become more and more clear that cancer cells rewire the metabolic fluxes to cope with various microenvironmental situations in order to sustain proliferation and invasion (7.Pavlova N.N. Thompson C.B. The emerging hallmarks of cancer metabolism.Cell Metab. 2016; 23: 27-47Abstract Full Text Full Text PDF PubMed Scopus (2917) Google Scholar). Increasing evidence suggests that metabolism is also a major driver for cancer metastasis (8.Payen V.L. Porporato P.E. Baselet B. Sonveaux P. Metabolic changes associated with tumor metastasis, part 1: Tumor pH, glycolysis and the pentose phosphate pathway.Cell Mol. Life Sci. 2016; 73: 1333-1348Crossref PubMed Scopus (149) Google Scholar, 9.Porporato P.E. Payen V.L. Baselet B. Sonveaux P. Metabolic changes associated with tumor metastasis, part 2: Mitochondria, lipid and amino acid metabolism.Cell Mol. Life Sci. 2016; 73: 1349-1363Crossref PubMed Scopus (72) Google Scholar). For example, a glycolytic enzyme phosphoglucoseisomerase has long been known as the autocrine motility factor that promotes tumor cell migration and invasion (10.Kho D.H. Zhang T. Balan V. Wang Y. Ha S.W. Xie Y. Raz A. Autocrine motility factor modulates EGF-mediated invasion signaling.Cancer Res. 2014; 74: 2229-2237Crossref PubMed Scopus (17) Google Scholar). The glycolytic end-product lactate is reported to be positively associated with metastasis in many types of cancer (11.Doherty J.R. Cleveland J.L. Targeting lactate metabolism for cancer therapeutics.J. Clin. Invest. 2013; 123: 3685-3692Crossref PubMed Scopus (696) Google Scholar). Lipid metabolism has also been implicated in tumor metastasis (12.Luo X. Cheng C. Tan Z. Li N. Tang M. Yang L. Cao Y. Emerging roles of lipid metabolism in cancer metastasis.Mol. Cancer. 2017; 16: 76Crossref PubMed Scopus (312) Google Scholar). A recent study shows that blocking lipid synthesis can overcome tumor metastasis after antiangiogenic therapy (13.Sounni N.E. Cimino J. Biacher S. Primac I. Truong A. Mazzucchelli G. Paye A. Calligaris D. Debois D. De Tullio P. Mari B. de Pauw E. Noel A. Blocking lipid synthesis overcomes tumor regrowth and metastasis after antiangiogenic therapy withdrawal.Cell Metab. 2014; 20: 280-294Abstract Full Text Full Text PDF PubMed Scopus (133) Google Scholar). However, the detailed metabolism shift associated with metastasis is still largely unclear. As for HCC, most studies have focused on the investigation of glucose metabolism (14.Shang R.Z. Qu S.B. Wang D.S. Reprogramming of glucose metabolism in hepatocellular carcinoma: Progress and prospects.World J. Gastroenterol. 2016; 22: 9933-9943Crossref PubMed Scopus (66) Google Scholar). For instance, the up-regulation of several enzymes in the glycolysis pathway, including pyruvate kinase M2 (PKM2), glucose transporters (GLUTs), lactate dehydrogenase, etc., have been reported to be associated with HCC progression and poor prognosis (15.Amann T. Maegdefrau U. Hartmann A. Agaimy A. Marienhagen J. Weiss T.S. Stoeltzing O. Warnecke C. Schölmerich J. Oefner P.J. Kreutz M. Bosserhoff A.K. Hellerbrand C. GLUT1 expression is increased in hepatocellular carcinoma and promotes tumorigenesis.Am. J. Pathol. 2009; 174: 1544-1552Abstract Full Text Full Text PDF PubMed Scopus (245) Google Scholar, 16.Chen Z. Lu X. Wang Z. Jin G. Wang Q. Chen D. Chen T. Li J. Fan J. Cong W. Gao Q. He X. Co-expression of PKM2 and TRIM35 predicts survival and recurrence in hepatocellular carcinoma.Oncotarget. 2015; 6: 2538-2548PubMed Google Scholar, 17.Sheng S.L. Liu J.J. Dai Y.H. Sun X.G. Xiong X.P. Huang G. Knockdown of lactate dehydrogenase A suppresses tumor growth and metastasis of human hepatocellular carcinoma.FEBS J. 2012; 279: 3898-3910Crossref PubMed Scopus (183) Google Scholar). However, less is known about the other metabolic pathways. More importantly, the link between metabolism and HCC metastasis is still missing. The metabolic reprogramming in cancer is a highly complicated process that requires the coordination of diverse intertwined metabolic pathways. These pathways form a dynamic network that is regulated by multiple levels of gene expression. Therefore, a large-scale and comprehensive analysis of cancer cell metabolism is required to understand the mechanisms and functional consequences of metabolic alterations associated with metastasis. In this study, we integrated data of genomics, transcriptomics, proteomics, and metabolomics from three HCC cell lines, including a low-metastatic cell line, Huh7; a medium-metastatic cell line, MHCC97L; and a highly metastatic cell line, HCCLM3, to mine potential genes and pathways contributing to HCC metastasis. Based on the multiomic analysis and functional study, UDP-glucose pyrophosphorylase 2 (UGP2), an enzyme critical for glycogen synthesis, was found to play an essential role in promoting HCC cell migration and tumor metastasis. Overall, our study described a systematic view of the cellular metabolism associated with HCC metastasis, providing valuable information for developing novel prognostic tools and therapeutic strategies for HCC. Dithiothreitol (DTT), iodoacetamide, urea, formaldehyde, deuterated-formaldehyde, C13-labeled deuterated-formaldehyde, sodium cyanoborohydride, and deuterated sodium borocyanohydride were purchased from Sigma Aldrich (St. Louis, MO, USA); Mouse monoclonal antibody against β-actin was purchased from Santa Cruz (Santa Cruz, CA, USA); rabbit polyclonal antibody against UGP2, rabbit monoclonal antibodies against ATP-dependent 6-phosphofructokinase (PFKP), glutamate-cysteine ligase regulatory subunit (GCLM), glutathione S-transferase omega-1, and thioredoxin domain-containing protein 12 were purchased from Abcam (Cambridge, MA, USA). Rabbit polyclonal antibody against glycogen phosphorylase (PYGB) and PKM2 were bought from Proteintech (Chicago, IL, USA). BCA reagents were purchased from Invitrogen (Grand Island, NY, USA). Enhanced chemiluminescence reagents were purchased from Pierce Biotechnology (Rockford, IL, USA). Protease Inhibitor Mixture tablets were purchased from Roche Diagnostics (Indianapolis, IN, USA). Sequencing-grade modified trypsin was purchased from Promega (Madison, WI, USA). Acetonitrile was from Merck (Whitehouse Station, NJ, USA). Water used in this study was deionized using a Milli-Q purification system (Millipore, Billerica, MA, USA). LO2 cells and HCC cell lines, including Huh7, MHCC97L, HCCLM3, PLC, HepG2, MHCC97H, and Hep3B, were cultured in DMEM supplemented with 10% fetal bovine serum and 1% penicillin-streptomycin (100 μg/ml) at 37 °C in a humidified atmosphere with 5% CO2. Genomic DNA was extracted from Huh7, MHCC97L, and HCCLM3 cells using a TIANamp Genomic DNA Kit (Tiangen, Beijing, China) according to the manufacturer instructions. Deep-coverage exome sequencing for HCC cell lines of Huh7 (111×), MHCC97L (131×), and HCCLM3 (124×) were performed at Shanghai Biotechnology Corporation (Shanghai, China) using the illumina 2500 platform (2 × 125 bp). Adapters and low-quality sequences were cleaned by using Trimmomatics (18.Bolger A.M. Lohse M. Usadel B. Trimmomatic: A flexible trimmer for Illumina sequence data.Bioinformatics. 2014; 30: 2114-2120Crossref PubMed Scopus (27908) Google Scholar). Cleaned reads were mapped to the reference genome (GRCh38) with Burrows-Wheeler Aligner (BWA) (19.Li H. Durbin R. Fast and accurate long-read alignment with Burrows-Wheeler transform.Bioinformatics. 2010; 26: 589-595Crossref PubMed Scopus (7012) Google Scholar). On average, 99.9% of the exon positions in the reference genome were covered by the studied samples. We then removed duplicated reads and sorted remaining reads with SAMtools (20.Li H. A statistical framework for SNP calling, mutation discovery, association mapping and population genetical parameter estimation from sequencing data.Bioinformatics. 2011; 27: 2987-2993Crossref PubMed Scopus (3124) Google Scholar). VarScan 2 was used to call candidate single-nucleotide polymorphisms (SNPs) (21.Koboldt D.C. Zhang Q. Larson D.E. Shen D. McLellan M.D. Lin L. Miller C.A. Mardis E.R. Ding L. Wilson R.K. VarScan 2: Somatic mutation and copy number alteration discovery in cancer by exome sequencing.Genome Res. 2012; 22: 568-576Crossref PubMed Scopus (2937) Google Scholar). The putative SNPs in cell lines were determined using the additional steps: (1) the SNPs showing read depth < 20 and phred score < 15 were removed. (2) We required that a SNP at a certain nucleotide position should be supported by two or more reads on forward strand and two or more reads on reverse strand, as well as be supported by at least 50% of the reads covering this position. Finally, the specific SNPs in MHCC97L and HCCLM3 were identified as these distinguishing from the Huh7 cell line. We removed SNPs from the Single Nucleotide Polymorphism database (dbSNP147). A sliding window approach (CNV-seq) (22.Xie C. Tammi M.T. CNV-seq, a new method to detect copy number variation using high-throughput sequencing.BMC Bioinformatics. 2009; 10: 80Crossref PubMed Scopus (432) Google Scholar) was adopted by using the 100 kb windows sliding in 50 kb increments to estimate read number (coverage) of each window in the three cell lines. The coverage of each window was normalized by the mean coverage of windows for each cell line. Copy numbers of MHCC97L and HCCLM3 were estimated by comparing the normalized coverage of the paired window between an MHCC97L cell and Huh7 cell, as well as between an HCCLM3 cell and Huh7 cell. Total RNAs from Huh7, MHCC97L, and HCCLM3 cells were isolated using Trizol reagent (Life Technologies, Carlsbad, CA, USA). A hybridization-based microarray assay was performed at Shanghai Biotechnology Corporation using the Human lncRNA expression microarray (4 × 180K, Agilent). We used three biological replicates for each cell line. Over 20,000 coding genes were covered by Agilent probes in each cell sample. The raw data of nine samples were normalized using the R package limma (quantile algorithm). Cells were lysed with 8 m urea, and the protein concentration was measured using BCA assay. The samples were reduced by incubating with 10 mm DTT at 37 °C for 1 h. The reduced proteins were alkylated for 1 h in darkness with 40 mm iodoacetamide. The alkylation reaction was quenched by adding DTT to a final concentration of 50 mm. The urea in the solution was exchanged to 50 mm sodium bicarbonate buffer by centrifugation using 3 kDa ultrafiltration devices (Millipore). The samples were incubated with trypsin at 37 °C overnight for the digestion to complete. For triplex dimethylation isotopic labeling, sodium cyanoborohydride was added to the protein digest for a final concentration of 50 mm, and the deuterated sodium borocyanohydride was used for the heavy labeled samples. Samples were incubated with 0.2 mm formaldehyde, deuterated-formaldehyde or C13-labeled deuterated-formaldehyde, respectively, at 37 °C for 1 h. The reaction was quenched with 2 m NH4OH, and the samples were mixed and separated using high pH RP-HPLC. Tryptic digested samples were injected onto an HPLC system (Waters, Milford, MA, USA) coupled with a high pH stable C18 column (Phenomenex Gemini C18, 150 × 2.1 mm, 3 μm) at a flow rate of 150 μl/min. The peptides were eluted with a 40-min gradient 5–45% buffer B (Buffer A: 50 mm ammonium formate, pH 10; Buffer B: acetonitrile). Fractions were collected every 3 min for 60 min. Collected fractions were dried by SpeedVac (ThermoFisher Scientific, Waltham, MA, USA) and reconstituted in 20 μl of 0.1% formic acid for the downstream LC-MS/MS analyses. The tryptic peptide samples eluted from the first-dimensional HPLC were desalted using C18 ziptip and loaded on a nanoUPLC system (Waters) equipped with a self-packed C18 column (C18, 150 × 0.075 mm, 1.7 μm). The peptides were eluted using a 5–40% B gradient (0.1% formic acid in acetonitrile) over 90 min into a nano-electrospray ionization Q Exactive mass spectrometer (ThermoFisher Scientific). The mass spectrometer was operated in data-dependent mode in which an initial Fourier transform (FT) scan recorded the mass range of m/z 350–1,500. The spray voltage was set between 1.8 and 2.0 kV, and the mass resolution used for MS scan was 70,000. The dynamic exclusion was set to 45 s. The top 20 most intense masses were selected for higher-energy collision dissociation fragmentations. MS/MS spectra were acquired starting at m/z 200 with a resolution of 17,500. The automatic gain control (AGC) target value and maximum injection time were set as 1e6 and 100 ms, respectively, for MS scans, as well as 5e4 and 110 ms for MS/MS scans. Raw data were searched against the Uniprot (release December 2016) human protein database containing 129,499 sequence entries using the SEQUEST database search algorithm embedded in the Protein Discoverer 1.4 Software (ThermoFisher Scientific). The following parameters were applied during the database search: 10 ppm precursor and fragment mass error tolerance, static modifications of carbamidomethylation for all cysteine residues, dimethylation for the formaldehyde labeling (Δ28 Da), deuterated-formaldehyde labeling (Δ32 Da), or C13-labeled deuterated-formaldehyde labeling (Δ36 Da) on lysines and the N terminus, flexible modification of oxidation modifications for methionine residues, and one missed cleavage site of trypsin were allowed. To determine the confidence of identification, false discovery rate was calculated by searching a decoy database generated by reversing all the protein sequences, and false discovery rate <0.01 was used as filtering criterion for all identified peptides. In addition, proteins identified with two or more peptides were considered, and proteins identified with the same set of peptides were grouped and treated as one. Quantification analysis was conducted using the Protein Discoverer 1.4 software. Ratios between the three HCC cell lysates were calculated based on the extracted chromatography areas and normalized using the median. To analyze the metabolites in the three HCC cell lines, we employed a quantitative polar metabolomics profiling platform by using selected reaction monitoring with a 5500 QTRAP hybrid triple quadrupole mass spectrometer (AB/SCIEX, Framingham, MA) that covered all major metabolic pathways by using a protocol reported by Yuan et al. (23.Yuan M. Breitkopf S.B. Yang X. Asara J.M. A positive/negative ion-switching, targeted mass spectrometry-based metabolomics platform for bodily fluids, cells, and fresh and fixed tissue.Nat. Protoc. 2012; 7: 872-881Crossref PubMed Scopus (640) Google Scholar). Briefly, we extracted the metabolites from 107 cells with 80% (v/v) methanol (cooled to −80 °C). The metabolites were separated with hydrophilic interaction liquid chromatography (3.5 μm; 4.6 mm inner diameter × 100 mm length; Waters) and detected with positive/negative ion switching to analyze 287 metabolites (315 Q1/Q3 transitions) from a single 15-min LC-MS acquisition with a 3-ms dwell time and a 1.55 s duty cycle time. Once the selected reaction monitoring data were acquired, peaks were integrated to generate chromatographic peak areas used for quantification across the sample set by using MultiQuant 2.0 (AB/SCIEX). For the raw data of transcriptome, proteome, and metabolome, the redundant data were merged, and then each sample was scaled and centered. We marked outliers with normalized values larger than three. There were three, six, and three replicates in each cell line for transcriptome, proteome, and metabolome. Here, we allowed one, three, and two abnormal or missed replicates for each cell line in the transcriptome, proteome, and metabolome, respectively. The mRNAs, proteins, and metabolites that did not fit the criteria in replicate counts were removed. All the outliers and missed replicates were supplied by multiple interpolation method using R package mice. To investigate the association between metastatic capabilities and the levels of mRNA, protein, and metabolite, weighted correlation network analysis (WGCNA) (24.Langfelder P. Horvath S. WGCNA: An R package for weighted correlation network analysis.BMC Bioinformatics. 2008; 9: 559Crossref PubMed Scopus (10250) Google Scholar) was applied for finding clusters (modules) of highly correlated genes or metabolites. We quantitated the metastatic capability as 1, 2, 3 for Huh7, MHCC97L, and HCCLM3 (trait 1) or as 1, 2, 2 for Huh7, MHCC97L, and HCCLM3 (trait 2). Soft threshold powers were set as 10, 10, 20 for the interpolated data of transcriptome, proteome, and metabolome. Here, we kept modules that significantly correlated to traits (r > 0.8 or r < -0.8; p < 0.05), and then kept mRNAs, proteins or metabolites that significantly correlated to the traits in these modules (r > 0.8 or r < -0.8; p < 0.05). For mRNAs and proteins, R package KEGGprofile was used for human Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis (p < 0.05), and protein analysis through evolutionary relationships (PANTHER) database search (http://pantherdb.org/) was used for GO analysis. Mbrole 2.0 (http://csbg.cnb.csic.es/mbrole2/) was used for KEGG pathway analysis for the metabolites (p < 0.05) (25.López-Ibáñez J. Pazos F. Chagoyen M. MBROLE 2.0-functional enrichment of chemical compounds.Nucleic Acids Res. 2016; 44: W201-W204Crossref PubMed Scopus (123) Google Scholar). Total RNA was transcripted to complementary DNA using a FastQuant RT kit (TianGen) according to the manufacturer's protocols. Quantitative mRNA expression analysis was performed on a 7500 Fast Real-Time PCR System (ABI, Foster City, CA, USA) using the SuperReal SYBR Green PreMix (TianGen) following the manufacturer's protocols. The mean Ct for each sample was normalized to 18s-rRNA as the reference gene (for primer sequences, see Supplemental Table S1). Protein concentration of each cell extracts was assayed using BCA protein assay kit (ThermoFisher Scientific), and the same amount of proteins was separated by SDS-PAGE and transferred onto the polyvinylidene fluoride membranes using a wet electro-blotter. The membranes were incubated with primary antibodies at 4 °C overnight, followed by incubation with secondary antibodies at room temperature for 1 h. Bound antibodies were detected by the enhanced chemiluminescence (ECL) immunoblotting detection reagent. DNA with the complete coding sequence of UGP2 was amplified by PCR using the Premix TaqDNA Polymerase (TaKaRa, Otsu, Japan). The flanking NheI and BamHI restriction sites were created, and the UGP2 DNA was cloned in the pCDH-GFP lentivector expression vector (System Biosciences, Palo Alto, CA, USA). The UGP2 construct was transfected into cells using Lipofectamine 2000 (Life Technologies, Paisley, Scotland). The overexpression efficiency of UGP2 was measured by qRT-PCR and Western blotting. The primer sequences used for cloning the full-length UGP2 are listed in Supplemental Table S2. In a wound-healing assay, 5 × 105 cells/well were seeded in six-well plate, allowed to grow for 24 h to 90–100% confluence, and starved overnight. A scratch was created through the confluent monolayer using a sterile pipette tip. The floating cells were removed with serum-free medium. Then, the cells were cultured with medium containing reduced fetal bovine serum (FBS, 2%) for another 24 h. The remaining width of the scratch was recorded from five randomly selected fields. Migration and invasion assays were performed using 24-well transwell chamber filters (Millicell Hanging Cell Culture Insert, polyethylene terephthalate 8.0 μm, Millipore). For the invasion assays, the membrane was prepared with Matrigel (BD Biosciences, San Jose, CA, USA) following the manufacturer's protocols. After starvation overnight, 1 × 105 cells in 200 μl of serum-free medium were added to the upper chamber for incubation at 37 °C. Next, 600 μl DMEM with 10% FBS were added to the lower chamber. Then, nonmigrated or noninvaded cells on the upper membrane surface were removed with a cotton swab, and the migrated and invasive cells on the lower membrane surface were fixed, stained with 0.01% crystal violet solution for 10 min, imaged, and counted in five random 200× microscopic fields. The glycogen was quantified using a glycogen detection kit (Jiancheng, Nanjing, China) according to the manufacturer's protocol. First, 50 mg of tumor tissues or cells were washed with normal saline, mixed with alkali solution, and boiled for 20 min. Then, the solution was mixed with detection solution, vortexed, and boiled for 5 min before the absorbance was measured at 620 nm wave length. The contents of glycogen were determined by using the standard curve measured at the same time. In vivo metastasis assays were performed using five-week-old male BALB/c-nude mice (Chinese Academy of Sciences, Beijing, China) (26.Yao J. Liang L. Huang S. Ding J. Tan N. Zhao Y. Yan M. Ge C. Zhang Z. Chen T. Wan D. Yao M. Li J. Gu J. He X. MicroRNA-30d promotes tumor invasion and metastasis by targeting Galphai2 in hepatocellular carcinoma.Hepatology. 2010; 51: 846-856Crossref PubMed Scopus (202) Google Scholar). Briefly, 2 × 106 cells were mixed with 20 μl of serum-free DMEM and 20 μl of Matrigel, then orthotopically inoculated in the left hepatic lobe by a microsyringe through an 8-mm midline incision in the upper abdomen under anesthesia. Four mice were used for each cell line. Mice were sacrificed after 6 weeks. The livers were dissected, tumor nodes were counted, and the tumor tissues were used for biochemical assays. The experimental protocols were evaluated and approved by Tianjin Medical University Animal Care and Use Committee. For proteomics, three biological replicates and two technical replicates were investigated for each type of cell, resulting in six data points for each cell line for quantification. Statistical and correlation analysis was performed in the same way as the transcriptomics and metabolomics data by WGCNA as described in details in the sections of "Raw Data Processing" and "Differential Modules Analysis." All sequencing data that support this study have been deposited in the NCBI (Bioproject: PRJNA399198). These raw data of cell lines are available under the following experiment accessions: Huh7 (SRX3108375), MHCC97L (SRX3108376), and HCCLM3 (SRX3108377). Microarray data have been deposited in the NCBI (Bioproject: PRJNA382487, GEO accessions: GSE97626). These raw data of cell lines are available under the following experiment accessions: Huh7 (GSM2573327, GSM2573328, GSM2573329), MHCC97L (GSM2573324, GSM2573325, GSM2573326), and HCCLM3 (GSM2573330, GSM2573331, GSM2573332). The proteomic data have been deposited to the ProteomeXchange Consortium via the proteomics identifications database PRIDE (27.Vizcaíno J.A. Csordas A. Del-Toro N. Dianes J.A. Griss J. Lavidas I. Mayer G. Perez-Riverol Y. Reisinger F. Ternent T. Xu Q.W. Wang R. Hermjakob H. 2016 update of the PRIDE database and related tools.Nucleic Acids Res. 2016; 44: 11033Crossref PubMed Scopus (21) Google Scholar) partner repository with the dataset identifier PXD005647. To understand the relationship between cellular metabolism and metastasis in HCC, we employed a multiomic strategy to compare the genome, transcriptome, proteome, and metabolome in three different HCC cell lines with increasing metastatic capabilities, including Huh7, MHCC97L, and HCCLM3 cells. Huh7 is a well differentiated hepatocyte-derived cellular carcinoma cell line that was originally taken from a primary liver tumor, and the MHCC97L and HCCLM3 cell lines were both derived from a metastatic tumor with differential metastatic capabilities (28.Li Y. Tang Z.Y. Ye S.L. Liu Y.K. Chen J. Xue Q. Chen J. Gao D.M. Bao W.H. Establishment of cell clones with different metastatic potential from the metastatic hepatocellular carcinoma cell line MHCC97.World J. Gastroenterol. 2001; 7: 630-636Crossref PubMed Scopus (318) Google Scholar, 29.Sun F.X. Tang Z.Y. Lui K.D. Ye S.L. Xue Q. Gao D.M. Ma Z.C. Establishment of a metastatic model of human hepatocellular carcinoma in nude mice via orthotopic implantation of histologically intact tissues.Int. J. Cancer. 1996; 66: 239-243Crossref PubMed Scopus (164) Google Scholar). First, whole-exome sequencing was performed to analyze the SNPs and copy number alterations (CNAs); then, microarray analysis was performed t