A tumorigenic index for quantitative analysis of liver cancer initiation and progression Gaowei Wanga, Xiaolin Luoa, Yan Lianga, Kota Kanekoa, Hairi Lib, Xiang-Dong Fub, and Gen-Sheng Fenga,1 aDepartment of Pathology, Division of Biological Sciences, Moores Cancer Center, University of California San Diego, La Jolla, CA 92093; and bDepartment of Cellular and Molecular Medicine, University of California San Diego, La Jolla, CA 92093 Edited by Roger J. Davis, University of Massachusetts Medical School, Worcester, MA, and approved November 13, 2019 (received for review June 29, 2019) Primary liver cancer develops from multifactorial etiologies, result- prognosis of liver cancer patients with diverse etiologies and also in ing in extensive genomic heterogeneity. To probe the common diagnosis of precancer patients with steatosis, fibrosis, or cirrhosis. mechanism of hepatocarcinogenesis, we interrogated temporal Upon further development and optimization, this TI platform may gene expression profiles in a group of mouse models with hepatic become a quantitative means for early detection of liver tumor steatosis, fibrosis, inflammation, and, consequently, tumorigenesis. initiation or risk assessment of chronic disease patients, based on Instead of anticipated progressive changes, we observed a sudden comprehensive analysis of the whole transcriptome rather than a molecular switch at a critical precancer stage, by developing analyt- few biomarker molecules. We believe that the principle and ra- ical platform that focuses on transcription factor (TF) clusters. tionale of this TI derivation method can be extended from HCC to Coarse-grained network modeling demonstrated that an abrupt other types of cancer, to develop a quantitative analytical tool for transcriptomic transition occurred once changes were accumulated cancer prediction and prognosis in a given organ or tissue. to reach a threshold. Based on the experimental and bioinformatic data analyses as well as mathematical modeling, we derived a tu- Results morigenic index (TI) to quantify tumorigenic signal strengths. The TI Temporal Gene Expression Patterns in Liver Tumorigenesis. Pten is a is powerful in predicting the disease status of patients with meta- tumor suppressor that counteracts the PI3K/Akt signaling path- bolic disorders and also the tumor stages and prognosis of liver way, and Pten deficiency has often been detected in liver can- cancer patients with diverse backgrounds. This work establishes a cer patients (5, 6). Targeted deletion of Pten in hepatocytes quantitative tool for triage of liver cancer patients and also for can- induced NAFLD and, subsequently, NASH, followed by tumor cer risk assessment of chronic liver disease patients. development in mice (7, 8). The NASH-driven pathogenic pro- MEDICAL SCIENCES cess in Pten-deficient liver was accelerated and aggravated by liver cancer | tumorigenic index | quantitative analysis | additional deletion of Shp2/Ptpn11 (9), a newly identified liver transcription factor clusters tumor suppressor (10, 11). These mutant mouse lines with de- fined genetic background and characterized tumor phenotype he incidences and mortality of liver cancer, mainly hepato- constitute an ideal group of animal models to dissect stepwise Tcellular carcinoma (HCC), are increasing rapidly worldwide mechanisms underlying NASH-HCC development. Toward this (1). Diverse risk factors for primary liver cancer have been goal, we isolated liver samples at multiple time points from mice identified, including infection of hepatitis B virus and hepatitis C with hepatocyte-specific deletion of Pten (PKO), Shp2 (SKO), virus, alcohol abuse, nonalcoholic fatty liver disease (NAFLD) or nonalcoholic steatohepatitis (NASH), and intake of aflatoxin B1. Significance Consistent with the complex and multifactorial etiologies, multiomics analyses of human liver tumor samples have identified The mechanisms of hepatocarcinogenesis are poorly understood. vast genomic heterogeneity, molecular and cellular defects, In this study, we interrogated the temporal gene expression metabolic reprogramming, and subtypes of tumors as well as profiles in mouse livers during progression of chronic fatty liver altered tumor microenvironment in the liver (2–4). diseases to carcinogenesis. By establishing an analytical system However, it remains to be determined if any common molecular that focuses on transcription factor (TF) clusters, we identified a signatures in the transcriptomes exist for liver cancer, despite their sudden switch from healthy liver to tumor tissues in the tran- considerable genomic heterogeneity. Furthermore, little is known scriptomes at precancer stage, prior to detection of any tumor about the kinetics and fashions, either gradual accumulation or nodule. We further developed a platform to calculate tumorigenic dramatic transition, in generation of cell-intrinsic and cell-extrinsic index (TI) based on the transcriptome, especially the TF clusters, to signals that are intertwined to drive malignant transformation of measure tumorigenic signal strengths. This quantitative analytical hepatocytes and tumor initiation. To dissect the stepwise onco- tool of TI is powerful in assessing cancer risk of chronic liver dis- genic signals and mechanisms at the precancer stages, we chose ease patients and in predicting tumor stages and prognosis of to work on mouse models that recapitulate key pathogenic fea- liver cancer patients. tures in human liver cancer. By pathological examination, RNA- sequencing (RNA-seq), and bioinformatic data analysis, we iden- Author contributions: G.-S.F. designed research; G.W., X.L., and G.-S.F. performed re- search; G.W., X.L., K.K., H.L., X.-D.F., and G.-S.F. contributed new reagents/analytic tools; tified a sudden switch in transcription factor (TF) clusters at a G.W., X.L., Y.L., and G.-S.F. analyzed data; and G.W. and G.-S.F. wrote the paper. precancer stage of hepatocarcinogenesis. Based on a multilayer The authors declare no competing interest. analysis of the TF clusters and mathematic modeling, we de- This article is a PNAS Direct Submission. veloped a tumorigenic index (TI) calculation system for quantita- tive measurement of liver tumorigenesis. Although this platform Published under the PNAS license. was established based on the transcriptomic data derived from Data deposition: The RNA-seq data have been deposited in the National Center for Bio- technology Information Gene Expression Omnibus (NCBI GEO) database under ID code genetically modified mouse tumor models, we have found appli- GEO: GSE123427. Codes have been deposited in GitHub (https://github.com/wanyewang1/ cable effectiveness of the derived TI values in determination of Index_model). disease status in other mouse models of chronic or malignant liver 1To whom correspondence may be addressed. Email: [email protected]. diseases with diverse backgrounds. Furthermore, using the TI This article contains supporting information online at https://www.pnas.org/lookup/suppl/ method developed from animal models to interrogate the hu- doi:10.1073/pnas.1911193116/-/DCSupplemental. man patients’ data, we demonstrated the power of the TI tool in www.pnas.org/cgi/doi/10.1073/pnas.1911193116 PNAS Latest Articles | 1of8 Downloaded by guest on September 28, 2021 Pten and Shp2 (DKO), and wild-type (WT) control (Fig. 1A). point, we identified genes that were significantly changed in Based on the developmental or pathological features, these liver mutant livers at different stages (Dataset S1). The resulting samples were divided into 4 groups: 1) youth, 2) adult, 3) pre- heatmap depicts the progressive changes in liver transcriptomes cancer, and 4) cancer (Fig. 1A). Representative histological or of the 3 mutant mouse lines (Fig. 1B), which correlated well with pathological properties of the specimens were shown (SI Ap- the kinetics and severity of tumor progression in SKO, PKO, and pendix, Fig. S1) and were also described in detail previously (9). DKO livers (SI Appendix, Fig. S1). Highlighted in Fig. 1B are RNA-seq was performed for all liver samples, and the data some significantly changed genes related to the MAPK pathway, quality analysis showed high levels of correlation for biological fatty acid, glucose or bile acid metabolism, extracellular matrix, replicates in each group (SI Appendix, Fig. S2A), with reduced epigenetic machinery, and inflammatory response. expression of Shp2 and/or Pten in corresponding mutants as By comparing the transcriptomes between PKO and WT livers expected (SI Appendix, Fig. S2 B and C). By comparing expres- at multiple time points, we identified significantly changed bi- sion levels in each mutant with WT control at the same time ological processes in the mutant liver using gene set enrichment A B 1 Pkm Jun Time (month) 0.5 0 Mapk3 1 2 3 4 5 7 912 16 −0.5 Mmp2 WT −1 Abhd5 PKO Tnf SKO 1M 2M 3M 4M 5M 7M_T 12M_T 16M_T 16M_T 1M 2M 7M 12M 1M 2M 7M_T 12M_T T T T DKO Kmt2c Kat6a Slc27a2 Youth Adult Pre-cancer Cancer Cyp7b1 PKO SKO DKO CDE F 1 1 1 Extracellular JAG2_NOTCH1 1 TRIGLYCERIDE_BIOSYNTHESIS BMP4_ACVR2B BMP7_ACVR2B Hdac5 EXTRACELLULAR_MATRIX_ORGANIZATION Structure CSF2_CSF2RA 0.5 EFNB3_EPHB2 INTEGRIN1_PATHWAY PDGFB_PDGFRB 0.5 0.5 0.5 ECM_RECEPTOR_INTERACTION Fatty acid FGF9_FGFR3 EFNA5_EPHA8 Mta3 ECM_AFFILIATED 0 TGFB3_TGFBR2 metabolism CSF1_CSF1R AVB3_INTEGRIN_PATHWAY IGF2_IGF2R Top2a JAG1_NOTCH4 −0.5 0 0 0 COLLAGEN_FORMATION Inflammatory MDK_PTPRB TNFSF12_TNFRSF12 CORE_MATRISOME NTF5_NTRK2 Dnmt1 response ECM_GLYCOPROTEINS −1 IL5_CSF2RB IFNG_IFNGR1