Gastroenterology 2019;157:1383–1397 An Immune Gene Expression Signature Associated With Development of Human Hepatocellular Carcinoma Identifies Mice That Respond to Chemopreventive Agents Agrin Moeini,1 Sara Torrecilla,1 Victoria Tovar,1 Carla Montironi,1,2 Carmen Andreu-Oller,1 Judit Peix,1 Mónica Higuera,1,3 Dominik Pfister,4 Pierluigi Ramadori,4 Roser Pinyol,1 Manel Solé,1 Mathias Heikenwälder,4 Scott L. Friedman,2 Daniela Sia,2 and Josep M. Llovet1,2,5 1 Liver Cancer Translational Research Liver Cancer Translational Research Group, Institut d’Investigacions Biomèdiques August Pi i Sunyer (IDIBAPS)-Hospital Clínic, Liver Unit, Universitat de Barcelona, Barcelona, Catalonia, Spain; 2Mount Sinai Liver Cancer Program, Department of Liver Diseases, Icahn School of Medicine at Mount Sinai, New York, USA; 3Liver diseases, Vall d’Hebron Institut de Recerca (VHIR), Hospital Universitari Vall d’Hebron, Barcelona, Spain; 4Division of Chronic Inflammation and Cancer, German Cancer Research Center Heidelberg (DKFZ), Heidelberg, Germany; and 5Institució Catalana de Recerca i Estudis Avançats (ICREA), Barcelona, Catalonia, Spain immunosuppressive profile (10% of cases) had a higher risk of See Covering the Cover synopsis on page 1176. HCC (hazard ratio, 2.41; 95% confidence interval, 1.21–4.80). Mice with chemically induced fibrosis or diet-induced steato- BACKGROUND & AIMS: Cirrhosis and chronic inflammation hepatitis given nintedanib or aspirin and clopidogrel down- precede development of hepatocellular carcinoma (HCC) in regulated the ICF gene expression pattern in liver and devel- approximately 80% of cases. We investigated immune-related oped fewer and smaller tumors than mice given vehicle. CON- gene expression patterns in liver tissues surrounding early- CLUSIONS: We identified an immune-related gene expression stage HCCs and chemopreventive agents that might alter these pattern in liver tissues of patients with early-stage HCC, called patterns to prevent liver tumorigenesis. METHODS: We the ICF, that is associated with risk of HCC development in analyzed gene expression profiles of nontumor liver tissues from patients with cirrhosis. Administration of nintedanib or aspirin 392 patients with early-stage HCC (training set, N ¼ 167 and and clopidogrel to mice with chronic liver inflammation caused validation set, N ¼ 225) and liver tissue from patients with loss of this gene expression pattern and development of fewer cirrhosis without HCC (N ¼ 216, controls) to identify changes in and smaller liver tumors. Agents that alter immune regulatory TRANSLATIONAL LIVER expression of genes that regulate the immune response that gene expression patterns associated with carcinogenesis might BASIC AND could contribute to hepatocarcinogenesis. We defined 172 genes be tested as chemopreventive agents in patients with cirrhosis. as markers for this deregulated immune response, which we called the immune-mediated cancer field (ICF). We analyzed the expression data of liver tissues from 216 patients with cirrhosis Keywords: Cancer; Microenvironment; Cytokines; Lymphocytes. without HCC and investigated the association between this gene expression signature and development of HCC and outcomes of patients (median follow-up, 10 years). Human liver tissues were also analyzed by histology. C57BL/6J mice were given a single injection of diethylnitrosamine (DEN) followed by weekly doses L iver cancer is the fourth leading cause of cancer- related mortality worldwide.1 Hepatocellular carci- noma (HCC) accounts for more than 90% of liver cancers of carbon tetrachloride to induce liver fibrosis and tumorigen- and is the main cause of death in patients with cirrhosis.2,3 esis. Mice were then orally given the multiple tyrosine inhibitor HCC arises from chronic liver inflammation, fibrosis, and nintedanib or vehicle (controls); liver tissues were collected and eventually cirrhosis in 70%–80% of cases.2 In developed histology, transcriptome, and protein analyses were performed. countries, curative treatments are feasible in 30%–40% of We also analyzed transcriptomes of liver tissues collected from mice on a choline-deficient high-fat diet, which developed chronic liver inflammation and tumors, orally given aspirin and Abbreviations used in this paper: CI, confidence interval; CCl4, carbon clopidogrel or the anti-inflammatory agent sulindac vs mice on a tetrachloride; CD-HFD, choline-deficient high-fat diet; DEN, diethylni- chow (control) diet. RESULTS: We found the ICF gene expres- trosamine; EGFR, epidermal growth factor receptor; ERK, extracellular signal-regulated kinase; GSEA, gene set enrichment analysis; HCC, he- sion pattern in 50% of liver tissues from patients with cirrhosis patocellular carcinoma; HCV, hepatitis C virus; HSC, hepatic stellate cell; without HCC and in 60% of nontumor liver tissues from pa- ICF, immune-mediated cancer field; IFN, interferon; IL, interleukin; NK, tients with early-stage HCC. The liver tissues with the ICF gene natural killer; p, phosphorylated; ssGSEA, single sample gene set enrichment analysis; STAT, signal transducer and activator of transcrip- expression pattern had 3 different features: increased numbers tion; TGF, transforming growth factor; VEGFR, vascular endothelial growth of effector T cells; increased expression of genes that suppress factor receptor. the immune response and activation of transforming growth Most current article factor b signaling; or expression of genes that promote © 2019 by the AGA Institute inflammation and activation of interferon gamma signaling. 0016-5085/$36.00 Patients with cirrhosis and liver tissues with the https://doi.org/10.1053/j.gastro.2019.07.028 1384 Moeini et al Gastroenterology Vol. 157, No. 5 HCC tumors arise has led to the identification of an WHAT YOU NEED TO KNOW immune-mediated cancer field (ICF) in 60% of patients with BACKGROUND AND CONTEXT early HCC and 50% of patients with cirrhosis but without We investigated inflammatory and immune-related gene HCC. This ICF comprises 3 distinct molecular subtypes: the expression patterns in liver tissues surrounding early- high-infiltrate ICF subtype, with increased infiltration of stage HCCs and chemopreventive agents that might effector T cells; the immunosuppressive ICF subtype, with alter these patterns to prevent liver tumorigenesis. activation of stroma and transforming growth factor (TGF) b NEW FINDINGS signaling; and the proinflammatory ICF subtype, with up- regulation of interferon (IFN) gamma signaling. These im- We identified a gene signature that associated with risk of mune profiles, particularly the immunosuppressive cancer HCC development in patients with cirrhosis. Administration of immunomodulatory and anti- field, predict increased risk of HCC development in cirrhotic angiogenic agents to mice with chronic inflammation patients. Inhibition of this carcinogenic field significantly and fibrosis reduced this gene expression pattern and reduced HCC onset in 2 mouse models of chronic liver reduced development of liver tumors. damage and hepatocarcinogenesis. Overall, our study pro- LIMITATIONS vides the rationale for exploring chemopreventive strategies in patients with cirrhosis at high risk for HCC development. This was a retrospective analysis of human liver tissue samples and the chemopreventive agents were tested in mice with liver tumors due to chronic fibrosis or Materials and Methods steatosis. More human studies are needed. IMPACT Human Cohort Gene expression data from a cohort of 167 surgically resec- This gene expression signature might be used to identify ted fresh-frozen samples (Heptromic data set, GSE63898) with patients with cirrhosis who are at risk for HCC. matched tumor and adjacent nontumor tissue were analyzed. Pharmacologic agents that alter this gene expression pattern might be tested in chemoprevention studies. Samples were previously collected (1998–2008) in the setting of the HCC Genomic Consortium after institutional review board approval. Full descriptions of the cohort and RNA profiling data cases, but recurrence is high, and no effective adjuvant are available in previous publications.15,16 Supplementary therapies are available.2,4 In addition, in approximately Table 1 provides a summary of the clinicopathologic variables 40%–50% of patients, diagnosis is made at advanced stages, of the samples used in the current study (training cohort, when currently approved molecular therapies yield limited N ¼ 167). Validation of the identified molecular profiles was then survival benefits (approximately 1 year).3 Despite recent performed in an independent set of 225 adjacent nontumor liver advances in the management and clearance of hepatits C tissues previously characterized by our group (GSE10143).9 virus (HCV) infection, there is an unmet need for early Finally, to identify those patients without neoplasm at higher detection and application of chemopreventive approaches in risk of HCC development and most likely to benefit from TRANSLATIONAL LIVER patients at high risk for HCC development. chemopreventive strategies, our findings were evaluated in a BASIC AND To date, there are no established preventive strategies for previously characterized cohort of patients with early cirrhosis HCC in patients at risk beyond prevention with antiviral (N ¼ 216, GSE15654)14 and a publicly available data set of therapies.5 Once cirrhosis is established, antiviral therapies fibrotic liver tissues (N ¼ 124, GSE84044).17 reduce but do not eliminate the risk of HCC.4,6,7 Individual risk assessment is a key first step in the successful development of Modeling the Immune-Mediated Cancer Field any chemopreventive strategy. In this regard, increasing evi- Enrichment scores of 4872 gene sets that represent cell dence suggests the existence of the so-called cancer field effect states and perturbations of the immune system (collection C7 or field cancerization, which consists of predisposing onco- of MSigDB, Broad Institute)18 were calculated by single-sample genic and inflammatory signals occurring during chronic liver gene set enrichment analysis (ssGSEA) in the nontumor liver injury and ultimately leading to malignant transformation.8–10 tissue of the training cohort. Next, unsupervised clustering Gene signatures derived from the cirrhotic tissue adjacent to analysis by the nonnegative matrix factorization19 method was HCC tumors have been designed to predict poor outcome, performed to identify the presence of an ICF. To characterize particularly in patients with cirrhosis and HCV infection who the samples presenting an ICF and to identify different ICF are at higher risk of HCC development.9,11–14 Overall, these subtypes, a second unsupervised clustering was performed studies support the feasibility of using molecular scores of the with ssGSEA scores obtained for a curated set of gene signa- carcinogenic field to identify patients at high risk for HCC tures representative of individual cell types,20,21 cancer development. However, the carcinogenic roles of inflamma- immune-related signaling pathways,22 and inflammation- or tion and immune response in the context of field cancerization immune-specific biological processes (Hallmark collection of MSigDB, Broad Institute). have been poorly explored. Understanding the immune fea- tures governing the unresolved cancer field effect is crucial for identifying potential therapeutic targets in patients at high Generation of an Immune-Mediated Field Gene risk for HCC development. Signature In this study, the analysis of the inflammatory milieu An ICF field gene signature was generated by using the top that characterizes the underlying liver disease in which differentially expressed genes in each molecular group (false November 2019 Immune-Mediated Field Cancerization as Target for HCC Prevention 1385 discover rate, <0.05; fold-change 2) and was then validated in regression models. In in vivo studies, the Mann-Whitney U test an independent data set with nearest template prediction was used to compare differences in body weights, liver func- analysis (P < .05) (Gene Pattern modules).23 tion, tumor number, tumor size, and CD4/CD8-stained area in human samples. The Fisher exact test was performed for analysis of HCC incidence and phosphorylated extracellular Molecular Characterization of the Immune- signal-regulated kinase (pERK) staining. The Student t test was Mediated Cancer Field Subtypes and used to compare the differences in Sirius Red quantification, Identification of Candidate Therapies CD31 staining, CD4/CD8 staining proportion of immune cell To characterize the ICF subtypes, gene expression signa- infiltrate in mice, and relative gene expression. tures (available in MSigDB [Broad Institute] or previously re- ported [Supplementary Table 2]) were assessed by GSEA, ssGSEA, nearest template prediction, and ingenuity pathway Transcript Profiling analyses. CIBERSORT20 was used to estimate the relative frac- Gene Expression Omnibus accession numbers are from tion of 22 immune cell types within the leukocyte compartment previously deposited data from our group (GSE63898, of nontumor liver tissues. The immunophenoscore algorithm24 GSE10143, GSE15654) and others (GSE84044). Newly profiled was used to analyze the major immunogenic determinants. An mice samples are in the Gene Expression Omnibus under in silico analysis based on ssGSEA scores of approximately accession number GSE125975 and GSE133969. 1230 gene sets (DSigDB) recapitulating targets of approved therapies was also performed for the screening of candidate targeted therapies. Results Identification of a Novel Immune-Mediated Histologic Evaluation of Infiltrating Inflammation Cancer Field Effect in Nontumor Liver Tissue of Histopathologic analysis was performed in 98 of 167 cases. Patients With Early Hepatocellular Carcinoma Specifically, H&E staining of formalin-fixed, paraffin-embedded To characterize the immune features governing the un- tissue section of HCCs and their matched adjacent nontumor livers were evaluated by 2 expert pathologists (Carla Montironi resolved cancer field in which new cancers arise, and Manel Sole). The presence of inflammation (portal/septal, transcriptome-based analysis of a compendium of approxi- interface, pericentral, and lobular) as well as the lymphoid mately 5000 annotated immunology-specific gene sets18 aggregates were assessed in the nontumor liver tissue sections. was performed in the nontumor liver tissue of patients More details on the histologic evaluation of the samples have with early-stage HCC. This analysis showed the presence of been included in Supplementary Material. an ICF in approximately 60% (98/167) of samples (Figure 1A and Supplementary Figure 1). Specifically, these samples were characterized by enrichment of several gene Animal Models sets recapitulating the presence of activated immune cells, We generated a chemically induced model of HCC and TRANSLATIONAL LIVER as well as up-regulation of core signaling pathways involved fibrosis in male C57BL/6J mice (Harlan Laboratories, N ¼ 55) in immune response (both innate and adaptive) and those BASIC AND by a single injection of diethylnitrosamine (DEN) followed by involved in the modulation of inflammatory response (ie, weekly dosing with carbon tetrachloride (CCl4), as previously described.25 Once fibrosis was established, mice were randomly interleukin [IL] 2–signal transducer and activator of tran- assigned to receive vehicle or nintedanib (50 mg/kg; scription [STAT] 5, IL-6/STAT3, IL-17, IFN gamma, colony- Boehringer Ingelheim, Rhein, Germany). Mice were killed at stimulating factor, tumor necrosis factor [TNF] a, and different time points, and liver and tumor tissue samples were TGF-b signaling) (Figure 1A and B and Supplementary collected and processed for histologic, transcriptomic, and Figure 1). Moreover, histologic evaluation confirmed that protein expression analyses (see Supplementary Material). All liver tissues with ICF contained a higher frequency of experimental procedures were carried out with the approval of moderate to marked inflammatory infiltrate (74% in ICF vs the institutional ethical committee of the University of Barce- 52% in non-ICF, P ¼ .034) and lymphoid aggregates (80% in lona and Hospital Clinic of Barcelona. Additionally, liver ICF vs 55% in non-ICF, P ¼ .009) (Figure 1C and D and samples of a choline-deficient high-fat diet (CD-HFD)–fed Supplementary Table 3). Immunostaining for CD4þ and mouse model reported in a recent study26 were collected. A CD8þ further confirmed significantly higher levels of T-cell total of 25 samples were processed for transcriptomic profiling, infiltrates in the adjacent livers of patients with the ICF including mice fed a chow diet (n ¼ 5) or CD-HFD for (Supplementary Figure 2A). In contrast, histologic evalua- 12 months and given vehicle (n ¼ 4), aspirin/clopidogrel tion of the tumor showed no significant correlation between (n ¼ 6), or sulindac (n ¼ 10). presence of the ICF and detection of intratumoral or peri- tumoral infiltration (Supplementary Table 3). This is in Statistical Analysis accordance with our recent publication,15 where the tumor All analyses were performed with SPSS software, version 23 immune-based profile did not correlate with the presence or (IBM, Armonk, NY) or GraphPad (San Diego, CA) Prism, version absence of immune gene signatures in the surrounding 5.00. Correlations for categorical and continuous variables nontumor tissue. were analyzed by Fisher exact test and Wilcoxon rank-sum test, While characterizing the ICF we detected that, in addi- respectively. The prognostic value of the signatures was tion to immunogenic features, several well-known carcino- assessed with Kaplan-Meier estimates, log-rank test, and Cox genic signals, such as epithelial-to-mesenchymal transition, 1386 Moeini et al Gastroenterology Vol. 157, No. 5 KRAS, epidermal growth factor receptor (EGFR), and (Supplementary Table 4). In line with these oncogenic sig- vascular endothelial growth factor (VEGF) signaling, were nals, a significant enrichment of previously reported prog- also significantly enriched in liver tissues containing the ICF nostic signatures derived from the adjacent nontumoral TRANSLATIONAL LIVER BASIC AND Figure 1. Identification of an ICF effect in nontumoral liver tissue adjacent to early HCCs. (A) Heatmap representation of the ICF present in 60% of patients with HCC. High and low ssGSEA scores are represented in red and blue, respectively. (B) Top predicted upstream cytokine and transcription factors activated in liver tissues of ICF patients. (C) Representative images of degree of portal/septal infiltrating inflammation. Original magnifications are given. (D) Representative images depicting presence or absence of lymphoid aggregates. Original magnifications are given. HIR, hepatic injury and regeneration; NTP, nearest template prediction; PD-1, programmed cell death protein 1. November 2019 Immune-Mediated Field Cancerization as Target for HCC Prevention 1387 liver were also detected. These signatures included the (P < .0001); it is called the proinflammatory ICF subtype 186-gene cancer-field signature,9 activated hepatic stellate (Figure 2A–C). The high-infiltrate and immunosuppressive cells (HSCs),11 hepatic injury and regeneration,13 and mul- subtypes shared several molecular features, including the ticentric occurrence of HCCs27 (Figure 1A). The presence of enrichment of key signaling pathways involved in the ICF significantly correlated with HCV infection; features modulating the immune response (ie, IL-2 and TNF indicative of liver dysfunction such as high bilirubin, low signaling), proliferation (ie, KRAS signaling), and angiogen- platelet count, and albumin levels (Supplementary Table 5); esis (P < .001) (Figure 2A). and poor survival (median overall survival, 43.4 months To further confirm the presence and molecular traits of in the ICF group vs 94.8 months in the non-ICF group; the identified ICF, we generated a transcriptome-based gene P ¼ .001], (Supplementary Table 6 and Supplementary signature able to capture the 3 ICF subtypes. This signature Figure 1B). Altogether, our data highlight the presence of showed only minimal overlap (0%–5%) with previously an ICF in 60% of patients with early HCC. This ICF is reported gene signatures of field cancerization in HCC characterized by activation of immunomodulatory signaling (Supplementary Figure 3).9,12,14,32 The resulting 172-gene cascades (ie, IFN-g, TNF-a, TGF-b, and IL-6), along with signature (Supplementary Table 7) was then validated in cancer-promoting signaling pathways (ie, epithelial- the adjacent nontumor tissue of 225 patients with early mesenchymal transition, EGFR, and vascular endothelial HCC, previously characterized by our group9,33 growth factor receptor [VEGFR]), and is associated with (Supplementary Figure 4A). Similar to what was previ- HCV infection and poor prognosis. ously observed in the training cohort, 58% (130/225) of patients belonged to the ICF. Moreover, in this cohort, the presence of the ICF was an independent predictor of poor The Immune-Mediated Cancer Field Contains 3 survival (hazard ratio, 2.73; 95% confidence interval [CI], Distinct Molecular Subtypes 1.1–6.8; P ¼ .03) (Supplementary Figure 4B and Further dissection of the key immune-modulating Supplementary Table 8). Within the ICF group, approxi- signaling pathways and immune-cell infiltrates in those mately 31% (40/130) presented the high-infiltrate ICF, samples harboring the ICF showed the existence of 3 approximately 27% (35/130) the immunosuppressive ICF, distinct molecular subtypes. The first molecular subtype, the and approximately 42% (55/130) the proinflammatory ICF high-infiltrate ICF subtype (23% of the ICF), showed a sig- subtype (Supplementary Figure 4A). Subsequent molecular nificant enrichment of several previously established gene characterization further confirmed the ability of the signatures mirroring the presence and/or activation of im- signature to capture the main molecular traits defining mune cell infiltrates such as lymphocytes (T and B cells)22,28 each subtype, such as increased infiltration of effector T or macrophages29 (Figure 2A and B). Consistently, immu- cells in the high-infiltrate subtype, activation of stroma nogenicity, herein captured either by the cytolytic activity and TGF-b signaling in the immunosuppressive subtype, score (Figure 2A)30 or by using the immunophenoscore al- and up-regulation of IFN gamma signaling in the proin- TRANSLATIONAL LIVER gorithm24 (Figure 2B), was also significantly higher in these flammatory subtype (Supplementary Figure 4A). Overall, samples (P < .001). Specifically, nontumor liver samples our results highlight the presence of a poor prognosis- BASIC AND belonging to the high-infiltrate ICF subtype showed signifi- related ICF comprising 3 molecular subtypes with a cant infiltration of effector T cells (Figure 2B) (P  .001), high degree of lymphocyte infiltration (overall, 16% of including increased levels of cytotoxic CD8þ T cells assessed patients with HCC) or predominance of either immuno- both by transcriptomic (P ¼ .03) and immunohistochem- suppressive (overall, 20% of patients with HCC) or istry (P ¼ .0002) (Figure 2C and Supplementary Figure 2B). proinflammatory (overall, 24% of patients with HCC) This subtype also was characterized by enrichment of the signaling cascades. previously reported ectopic lymphoid structures signa- ture31 (Figure 2A). In addition, the high-infiltrate ICF was significantly associated with poor survival compared with The Immune-Mediated Cancer Field, Particularly the rest of the patients (Supplementary Figure 1C), although the Immunosuppressive Subtype, Predicts a High there were no significant differences among the distinct ICF Risk of Hepatocellular Carcinoma Development subtypes (Supplementary Figure 1D). The second subtype, in Patients With Cirrhosis the immunosuppressive ICF (36% of the ICF), was charac- After the identification of an ICF in the livers of 60% of terized primarily by activation of stroma and HSCs, patients with early HCC, we next sought to assess its role in increased TGF-b signaling, and T-cell exhaustion liver disease progression and HCC primary occurrence. (Figure 2A). Moreover, several immune checkpoints (ie, To this end, the 172-gene signature was analyzed in a cohort CTLA-4, TIGIT, LAG3) were significantly overexpressed of 216 patients with nonmalignant cirrhosis with a median (immunophenoscore, P < .01) in this class, along with follow-up of 10 years in the context of an HCC surveillance higher levels of M2 macrophages (P ¼ .04) and CD4þ program.14 Overall, 51% (110/216) of patients with memory resting cells (P ¼ .005), which are among the main cirrhosis harbored the ICF, including the high-infiltrate ICF mediators of immune tolerance and inhibition (Figure 2B subtype in 28% (31/110), the immunosuppressive and C). The third subtype (41% of the ICF) showed a clear ICF subtype in 19% (21/110), and the proinflammatory ICF predominance of IFN gamma signaling (P < .001) subtype in 53% (58/110) of patients with cirrhosis and enrichment of the inflammatory M1 macrophages harboring the ICF. 1388 Moeini et al Gastroenterology Vol. 157, No. 5 TRANSLATIONAL LIVER BASIC AND November 2019 Immune-Mediated Field Cancerization as Target for HCC Prevention 1389 Next, we tested the capacity of the ICF subtypes to predict the risk of HCC development in patients with cirrhosis. The presence of the immunosuppressive ICF subtype (10% of all patients with cirrhosis) was signifi- cantly associated with a higher risk of HCC development (median time to HCC development of 7.4 years [95% CI, 3.2– 11.7] vs 17.1 years [95% CI, 10.6–23.7] in Rest, P < .0001] and was found to be an independent predictor of HCC occurrence in patients with cirrhosis in a multivariate analysis (hazard ratio, 2.41; 95% CI, 1.2–4.8; P ¼ .012) (Figure 3A and Table 1). In addition, the immunosuppres- sive ICF was also significantly associated with poor survival (median overall survival, 7.1 years [95% CI, 4.5–9.6] vs 16.3 years [95% CI, 9.1–23.5] in Rest; P < .0001) and higher risk of hepatic decompensation (median time to hepatic decompensation, 6.5 years [95% CI, 4.3–8.6] vs >15 years in Rest; P < .0001) (Figure 3B and C). Patients with cirrhosis harboring the high-infiltrate and proinflammatory ICF sub- types also showed a nonsignificant trend toward a higher risk of HCC development compared with those patients lacking the ICF (mean time to HCC development, 12.8 years [95% CI, 11.5–14.2] in other ICF subtypes vs 16.3 years [95% CI, 14.2–18.5] in non ICF, P ¼ .06) (Supplementary Figure 5A). Moreover, the analysis of an additional cohort of 124 patients without neoplasia with liver fibrosis17 showed that the ICF may occur as a progressive event because it was significantly correlated with increasing levels of fibrosis stage and degree of inflammation (Supplementary Figure 5B). Particularly, the presence of the immunosup- pressive ICF was significantly correlated with the presence of advanced liver fibrosis (Scheuer fibrosis S3-4 score,17 P ¼ .034) (Supplementary Figure 5B). TRANSLATIONAL LIVER In conclusion, the ICF detected in patients with early HCC is also present in the livers of approximately 50% of BASIC AND patients with cirrhosis and captures the presence of a damaging and continuous inflammatory response in the underlying liver disease. Furthermore, our results under- score the critical role of an immunosuppressive ICF (overall, Figure 3. Association of the presence of the immunnosup- 10% of patients with cirrhosis) in defining a 2.4-times risk pressive ICF with HCC occurrence and prognostic variables of HCC development and, to a smaller extent, of the high- in patients with cirrhosis. (A) Kaplan-Meier estimates of infiltrate and proinflammatory subtypes. HCC development, (B) overall survival, and (C) hepatic decompensation, according to the presence of the immunosuppressive ICF subtype (orange). The Immune-Mediated Field as a Target for Chemoprevention in a Mouse Model immunosuppressive ICF subtype, may represent an ideal Recapitulating Chronic Liver Inflammation and target for chemopreventive strategies in patients with Hepatocellular Carcinoma Development cirrhosis at high risk for HCC development. To this pur- Based on the compelling results described, we hy- pose, an in silico–based analysis was performed with our pothesized that the ICR, particularly the training cohort to identify those candidate therapies most = Figure 2. The ICF contains 3 distinct molecular subtypes. (A) Heatmap representation of the 3 ICF subtypes. Statistical sig- nificance is highlighted. (B) Immunophenogram representing the enrichment of immunogenic determinants in the distinct ICF subtypes (C) Comparison of estimated proportion of immune cells (CIBERSORT method) between the ICF subtypes, repre- senting those immune populations with estimated average fraction >5% and significant differences between the ICF subtypes. Significant statistical differences observed among the different ICF subtypes are highlighted (high infiltrate, purple; immunosuppressive, orange; proinflammatory, green; and both high infiltrate and immunosuppressive, black). *P < .05, **P < .01, and ***P < .001. CP, checkpoint, EC, effector cell; MDSC, myeloid-derived suppressor cell; MHC, major histocompatibility complex in antigen presenting cells; NTP, nearest template prediction; SC, suppressor cells; Treg, regulatory T cell. 1390 Moeini et al Gastroenterology Vol. 157, No. 5 Table 1.Univariate and Multivariate Analysis of Risk of Hepatocellular Carcinoma Development in Patients With Cirrhosis Including Gene Signatures and Clinicopathologic Variables (N ¼ 216) Univariate analysis Multivariate analysis (Cox regression) Variable P value Hazard ratio 95% CI P value Nontumoral liver tissue–based transcriptomic profiles lmmunosuppressive-lCF .03 2.41 1.21–4.80 .01 186-gene poor prognosis signature <.0001 1.56 0.89–2.7 .12 Clinicopathologic variables Age (>median) .87 Sex .22 Diabetes .48 HCV genotype 1b .18 Alcohol consumption (>80 g/day) .68 HCV etiology plus alcohol consumption .68 History of antiviral treatment (IFN based) .65 Varices .02 1.49 0.85–2.6 .17 Spleen .13 lshak score of 6 vs 5 .24 Platelet count (<100,000/mm3) .02 1.51 0.91–2.64 .15 Bilirubin (>1 mg/dL) <.01 1.85 1.07–3.2 .03 Alpha-fetoprotein (>10 ng/mL) .87 Prothrombin time (international normalized ratio >1.2) .38 NOTE. Text in bold font indicates statistical significance. likely to modulate the identified ICF. This analysis was was significantly reduced in treated mice (22% vs 77%, P based on the enrichment of a compendium of approxi- < .001) (Figure 4B). In addition, nintedanib significantly mately 1230 gene sets (DSigDB collections D1 and D2)34 reduced the overall tumor number and size both at 17 and recapitulating the main targets of 1202 approved drugs. 18 weeks of age (Figure 4D and E). Overall, nintedanib Among the top 10 most significantly enriched drugs was well tolerated with no significant induction of (Supplementary Figure 6), nintedanib was the only ther- body weight loss or hepatotoxicity measured by serum TRANSLATIONAL LIVER apy approved by the US Food and Drug Administration alanine aminotransferase and aspartate aminotransferase indicated for a noncancer condition. Specifically, ninteda- levels (Supplementary Figure 7B and C). Taken together, BASIC AND nib is the first molecular targeted therapy with clinical our data suggest that nintedanib is safe and efficacious efficacy in patients with idiopathic pulmonary fibrosis as in preventing HCC development in our experimental both an anti-fibrogenic and anti-inflammatory agent.35 model. Given these considerations, the efficacy of nintedanib in reverting the pro-tumorigenic ICF was tested in a mouse model of HCC development in the setting of chronic Nintedanib Treatment Reverts the Immune- inflammation and liver fibrosis (Supplementary Mediated Cancer Field Effect Figure 7A). In this model, the macroscopic evaluation of Next, we sought to assess the impact of nintedanib explanted livers in DEN/CCl4 mice killed at the ages of 15, treatment on the ICF. For this purpose, we analyzed gene 17, and 18 weeks confirmed the development of numerous expression profiling of nontumor liver samples from hepatic tumors (Figure 4A). Tumor penetrance and num- 17-week-old DEN/CCl4 mice given nintedanib (n ¼ 10) or ber of tumors progressively increased, ultimately reaching vehicle (n ¼ 9) and 3 healthy control mice. First, the com- a 100% incidence at 18 weeks of age (Figure 4A and B). At parison between the healthy control and vehicle groups all time points, histologic evaluation of the liver sections showed a profile of activated pathways compatible with showed that a portion of the tumors were preneoplastic HCC development within a fibrotic and inflammatory (dysplastic) nodules (Figure 4C). In mice killed at 15 background. In this regard, functional analysis of differen- weeks of age (Supplementary Figure 7A), nintedanib tially expressed genes (Supplementary Table 9) highlighted showed a clear trend toward reducing HCC incidence, the activation of hepatic stellate cells and fibrogenesis, as number, and size of tumors (Figure 4B, D, and E). These well as immune system activation (inflammatory response, differences reached significance at 17 weeks of age chemotaxis, binding of myeloid and leukocytes) in vehicle- (Figure 4B), having a marked decrease in both overall treated DEN/CCl4 livers (Supplementary Table 10). tumor burden (30% in nintedanib vs 89% in vehicle Notably, our model presented a significant enrichment of group, P ¼ .019) and, specifically, in HCC incidence (7% vs the gene set representing the ICF identified in humans 33%, P ¼ .04). Similarly, at 18 weeks of age, HCC incidence (P ¼ .001) and faithfully recapitulated the human ICF November 2019 Immune-Mediated Field Cancerization as Target for HCC Prevention 1391 TRANSLATIONAL LIVER BASIC AND Figure 4. Nintedanib reduces HCC onset in mice. (A) Representative pictures of macroscopic evaluation of hepatic tumors in mice given vehicle or nintedanib and killed at 15, 17, and 18 weeks of age. Arrows indicate macroscopically visible tumors. (B) Evaluation of overall tumor burden and HCC incidence. #Statistical significance for overall tumor burden. *Statistical signifi- cance for HCC incidence. (C) Microscopic evaluation of the number of tumors per mouse in each group. (D) Number of macroscopic tumors per mouse given vehicle or nintedanib at the different time points. (E) Diameter size of the largest tumor per mouse given vehicle or nintedanib at the 3 different time points. # or *P < .05, ## or **P < .01, and ### or ***P < .001. Nin, nintedanib; Veh, vehicle; w, weeks. 1392 Moeini et al Gastroenterology Vol. 157, No. 5 TRANSLATIONAL LIVER BASIC AND November 2019 Immune-Mediated Field Cancerization as Target for HCC Prevention 1393 subtypes described (Figure 5A). The comparison of the development of HCC associated with advanced chronic liver gene expression profiles of adjacent nontumor liver disease. from mice treated with vehicle or nintedanib showed that nintedanib significantly down-regulated the ICF sub- types and, more specifically, the proinflammatory and Immunomodulatory Effects of Aspirin/ immunosuppressive ICF phenotypes, which predict risk of Clopidogrel Treatment Revert the Immune- HCC development in patients with cirrhosis (P ¼ .02) Mediated Cancer Field Effect and Prevent (Figure 5A). A nonsignificant trend was also observed for Hepatocarcinogenesis In Vivo the high-infiltration subtype (Figure 5A). Treatment with To further support the concept of an ICF in promoting nintedanib led to a significant down-regulation of inflam- HCC development and its therapeutic immunomodulation matory cues (IL-6/STAT3, IFN alfa, and IFN gamma) and as a candidate strategy for chemoprevention, we per- immune-related signaling (IL-2/STAT5 activation, allograft formed gene expression profiling in nontumor liver rejection) (Figure 5A). Among the infiltrating immune cells, derived from the recently described mouse model of CD- nintedanib significantly reduced the presence of B and HFD treated either with the immunomodulatory combi- T cells, activated macrophages, helper T cells, and regula- nation aspirin/clopidogrel or the anti-inflammatory tory T cells along with associated immune modulators sulindac.26 Of particular interest, in this model, which (ie, IL1, CCL5, and PDL1) (Figure 5A). Despite exhibiting presents nonalcoholic fatty liver disease–related liver similar global levels of inflammatory infiltrates, quantifi- inflammation with various degrees of fibrosis and HCC cation of CD4- and CD8-positive infiltrating lymphocytes development after 12 months of diet regimen,26,36,37 HCC by immunohistochemistry showed a significant decrease prevention was achieved only through the combination of of CD4þ T cells in nintedanib-treated mice compared the anti-inflammatory drug, aspirin, and clopidogrel (an with controls (P < .05) (Figure 5B). P2Y12 inhibitor) (25% vs 0% 12-month HCC incidence, Next, to further characterize the chemopreventive control vs combination, respectively; P ¼ .01)26 and not effects of nintedanib, we assessed the activation status of sulindac alone (25% vs 20% 12-month HCC incidence, the main nintedanib targets (ie, VEGFR2 and platelet- control vs sulindac, respectively; P ¼ not significant; data derived growth factor receptor b). Western blot of not shown). nontumor liver tissue confirmed that nintedanib blocked In the context of our study, comparative analysis be- the activation of VEGFR2 (Figure 5C) and its downstream tween the nontumor liver of healthy control and CD-HFD effectors protein kinase B and ERK (Supplementary mice showed a significant enrichment of the ICF signa- Figure 8A). Consistently, both liver parenchyma and ture in CD-HFD mice (P ¼ .002) (Figure 5E). All 3 ICF liver tumors were pERK positive in vehicle-treated mice subtypes were significantly up-regulated in CD-HFD mice and pERK negative in nintedanib-treated mice (P < .05) compared with healthy controls (Figure 5E), along with the (Supplementary Figure 8B), indicating an antiproliferative enrichment of signaling pathways regulating inflammation TRANSLATIONAL LIVER effect of nintedanib as well. Given the strong inhibition of (ie, IL6/STAT3, TNF-a), immune infiltration and activa- VEGFR signaling observed, we next assessed the anti- tion,22,28 and epithelial-to-mesenchymal transition (ie, BASIC AND angiogenic effect of nintedanib in DEN/CCl4 mice. In this TGF-b) (P < .05). These data were consistent with the high model, reduced CD31 staining was associated with intrahepatic influx of metabolically activated CD8þ T cells diminished blood vessel area in both liver parenchyma and natural killer (NK) cells (CD3þNK1.1þ) measured in and liver tumors of nintedanib-treated mice (Figure 5D). CD-HFD–fed mice by flow cytometry.26 Overall, these data Together, these data suggest that nintedanib exhibits its confirm the existence of an ICF in an independent model of chemopreventive effects in part by inducing vascular chronic liver disease, further suggesting a role in normalization and inhibiting hepatic proliferation. In hepatocarcinogenesis. contrast, no reduction of fibrosis degree, the profibrogenic Next, we compared the expression profiles of liver signaling pathway platelet-derived growth factor samples from CD-HFD vehicle-treated mice with CD-HFD receptor b, or collagen markers were detected in the mice treated with the combination aspirin/clopidogrel livers of nintedanib-treated mice (Supplementary (n ¼ 6) or sulindac alone (n ¼ 10). Only aspirin/clopidogrel, Figure 8C–E). but not sulindac alone, was able to prevent HCC and revert Overall, our data confirm that therapeutic targeting of the ICF within the liver microenvironment (P ¼ .05), the ICF, accompanied by liver vascular normalization and with the proinflammatory ICF subtype being the most suppression of hepatic proliferation, can prevent the significantly down-regulated upon treatment (Figure 5B). = Figure 5. Nintedanib and aspirin/clopidogrel reduce the ICF in animal models of chronic inflammation and HCC development. (A) Heatmap representation of high and low ssGSEA scores for the 172-gene signature and gene sets recapitulating the ICF subtypes. (B) Representative images and quantification of CD4þ and CD8þ infiltrating lymphocytes in the livers of 17-week-old mice given vehicle (n ¼ 5) or nintedanib (n ¼ 5). (C) Western blot analysis of VEGFR2 activation in the nontumor liver pa- renchyma of 17-week-old mice given vehicle (n ¼ 6) or nintedanib (n ¼ 6). (D) Morphometric quantification of blood vessel area by CD31 immunostaining in 5 randomly selected, low-magnification fields in mice given vehicle (n ¼ 5) or nintedanib (n ¼ 5). (E) Single sample GSEA analysis of the ICF signature in the different treatment arms of the CD-HFD model. **P < .01 and ***P < .001. AspClo, aspirin/clopidogrel; Nin., nintedanib; Sulin, sulindac; Veh., vehicle. 1394 Moeini et al Gastroenterology Vol. 157, No. 5 Particularly, based on previous molecular characteriza- stages, in addition to curative treatment options for early tion,26 the inhibition of the ICF seemed to be accompanied lesions. Universal immunization against hepatitis B virus by a significant reduction in the degree of liver damage and and antiviral therapies against hepatitis B virus and HCV a significantly reduced number of CD8þ and NK T cells in have been associated with greatly reduced HCC risk.2,41,42 the liver. Once cirrhosis is established, the risk of HCC development Overall, these data support the role of the ICF in pro- remains despite achievement of a sustained virologic moting carcinogenesis and suggest that only those drugs response in patients with HCV.6,7 In addition, the incidence able to simultaneously inhibit several components of the ICF of other risk factors, such as nonalcoholic steatohepatitis, is by targeting mitogenic, angiogenic, and immunomodulatory dramatically increasing.2 Thus, alternative HCC preventive kinases (ie, nintedanib and aspirin/clopidogrel) present a strategies capable of interfering with molecular hep- more efficacious therapeutic index for HCC prevention. atocarcinogenesis are an unmet need. Furthermore, identi- fying those patients at high risk of HCC development should enable a cost-effective selection of patients most likely to Discussion benefit from chemopreventive approaches. In this scenario, This study represents an in-depth analysis of the in- our results are of clinical relevance because the ICF, and flammatory milieu associated with the field cancerization in specifically the immunosuppressive subtype, may provide a the chronically injured liver and investigates its clinical novel companion biomarker to enrich at-risk patients in implications for the prediction and prevention of HCC chemoprevention clinical trials. Given these observations, occurrence in patients with cirrhosis. we then sought to investigate whether the molecular forces The role of the cancer field effect in promoting driving such cancer fields could serve as targets for che- neoplastic transformation has gained much interest in mopreventive strategies. Hence, we first verified that the recent years, and currently an altered microenvironment is molecular profiles observed in human cirrhosis were faith- considered a promoter of cancer.8,10 Although under phys- fully reproduced in 2 animal models of chronic liver injury. iologic conditions inflammation is an adaptive response to The DEN/CCl4 chemically induced mouse model and the tissue injury, when the inflammatory stimuli persist, the recently described nonalcoholic steatohepatitis/HCC nonresolved inflammation contributes to carcinogen- model26 reliably recapitulated the presence of a carcino- esis.38,39 Along these lines, activation of HSC and certain genic phenotype observed in liver tissues from patients pathways, such as nuclear factor kB and TGF-b signaling, belonging to the ICR. have been previously associated with liver fibrogenesis and, To identify the most promising candidate therapies eventually, neoplastic transformation.9,12 With this study, for novel chemopreventive strategies, we conducted an we move beyond the limits of current knowledge and pro- in silico analysis using a large compendium of gene vide a detailed description of the immune microenviron- sets34 recapitulating the main targets of 1202 approved ment underlying the field cancerization in the liver. To this drugs. Among the top 10 most significantly enriched TRANSLATIONAL LIVER end, we first characterized the immune profile of the non- drugs, we selected nintedanib, the only therapy tumor liver parenchyma of 392 early HCCs and then approved by the US Food and Drug Administration for BASIC AND investigated its role in predicting HCC development in 216 non-neoplastic conditions. In the DEN/CCl4 animal patients with cirrhosis having long-term surveillance for model, oral administration of nintedanib reduced the HCC (median, 10 years).14 The analysis showed that up to ICF, including the immunosuppressive ICF subtype, ul- 60% of HCCs and 50% of cirrhotic patients showed a timately reducing HCC incidence and growth. Reversion deleterious immune-mediated response in the surrounding of the ICF induced by treatment with nintedanib was tissue, which was associated with impaired liver function, accompanied by reduction of CD4þ lymphocytes, which activation of specific oncogenic loops, angiogenesis, and could be due to its mechanism of action inhibiting the poor survival. Further characterization identified 3 distinct src family of kinases (ie, LCK, FLT3, and SRC). These subtypes with different levels of lymphocyte infiltration and findings are in line with previous reports suggesting activation of either immunosuppressive or proinflammatory that CD4þ cells propagate immune-mediated liver injury traits. In particular, the immunosuppressive ICF subtype in models of chronic liver inflammation or autoimmune (approximately 10% of patients with cirrhosis) was an in- liver disease.43,44 Pretreatment with T-cell–specific anti- dependent predictor of HCC development, increasing by 2.4 bodies or immunosuppressive agents, such as anti-CD4 times the risk of cancer development, whereas both the monoclonal antibody, FK506 (tacrolimus), or cyclo- high-infiltrate and the proinflammatory subtypes showed a sporine A, have been shown to ameliorate hepatitis in trend toward higher risk of HCC occurrence in cirrhosis. The these models, further supporting the role of CD4þ T cells identification of distinct immune subtypes reflects the in inducing liver damage.43 Results in a second animal complex role of the immune system in hepatocarcino- model treated with the combination of the anti- genesis, with both an activated immune response and an inflammatory drug, aspirin, and clopidogrel (a P2Y12 exhausted immune microenvironment contributing to inhibitor) confirmed the therapeutic potential of immu- create a pro-tumorigenic environment and increase the risk nomodulating the ICF and supported the pro-tumorigenic of HCC development.40 role exerted by the immune response. Indeed, only the Reducing the incidence and mortality of HCC requires treatment able to modulate the ICF, as indicated by the advances in chemopreventive approaches at preneoplastic reduction of immune cells (ie, CD8þ and NK T cells) and November 2019 Immune-Mediated Field Cancerization as Target for HCC Prevention 1395 the reversion of the ICF signature, successfully reduced mortality worldwide for 36 cancers in 185 countries. CA liver damage and prevented HCC development. Overall, Cancer J Clin 2018;68:394–424. our study identifies a novel promising chemopreventive 2. Llovet JM, Zucman-Rossi J, Pikarsky E, et al. strategy for HCC and confirms the validity of using the Hepatocellular carcinoma. Nat Rev Dis Prim 2016; reversion of the ICF as reliable readout of efficacy. This 2:16018. is of great clinical importance because there is currently 3. Llovet JM, Montal R, Sia D, et al. Molecular therapies and no effective method to monitor the short-term effects of precision medicine for hepatocellular carcinoma. Nat Rev chemopreventive drugs.5 Nintedanib belongs to a new Clin Oncol 2018;15:599–616. generation of tyrosine kinase inhibitors that, in addition 4. Galle PR, Forner A, Llovet JM, et al. EASL clinical to exerting immune modulation, block the activation of practice guidelines: management of hepatocellular main angiogenic receptors.45 Many cytokines and growth carcinoma. J Hepatol 2018;69:182–236. factors are involved in modulating the formation of new 5. Fujiwara N, Friedman SL, Goossens N, et al. Risk factors vessels. Expression of VEGF and its receptors is elevated and prevention of hepatocellular carcinoma in the era of in HCC cell lines and tissues, as well as in the blood precision medicine. J Hepatol 2018;68:526–549. circulation of patients with HCC.33,46–48 In our model, 6. Calvaruso V, Cabibbo G, Cacciola I, et al. Incidence of nintedanib exerted its chemopreventive mechanisms in hepatocellular carcinoma in patients with HCV- part through the inhibition of VEGF signaling, a major associated cirrhosis treated with direct-acting antiviral driver of angiogenesis.49 Thus far, independent studies agents. Gastroenterology 2018;155:411–421. had reported that HCC prevention can be achieved in 7. Kanwal F, Kramer J, Asch SM, et al. Risk of hepato- animal models by attenuating liver fibrosis through the cellular cancer in HCV patients treated with direct- inhibition of EGFR50,51 or lysophosphatidic acid32 acting antiviral agents. Gastroenterology 2017; 153:996–1005. signaling. With the current study, we show that modu- lation of the liver microenvironment by molecular tar- 8. Lochhead P, Chan AT, Nishihara R, et al. Etiologic field effect: reappraisal of the field effect concept in cancer geted drugs, which simultaneously block liver predisposition and progression. Mod Pathol 2015;28: inflammation and angiogenesis, might represent a 14–29. powerful alternative strategy. 9. Hoshida Y, Villanueva A, Kobayashi M, et al. Gene We recently defined the immune class of HCC15 and the expression in fixed tissues and outcome in hepatocel- immune exclusion class (characterized by active Wnt/ lular carcinoma. N Engl J Med 2008;359:1995–2004. CTNNB1),3,15,52 which might predict response and primary 10. Hernandez-Gea V, Toffanin S, Friedman SL, et al. Role of resistance to checkpoint inhibitors, respectively.3,52 Here, the microenvironment in the pathogenesis and treatment we explored the immunomodulatory mechanisms under- of hepatocellular carcinoma. Gastroenterology 2013; lying HCC occurrence by defining an immune-mediated 144:512–527. field effect that conforms to a cancer-permissive milieu, 11. Ji J, Eggert T, Budhu A, et al. Hepatic stellate cell and TRANSLATIONAL LIVER thus placing these patients at the highest risk of HCC monocyte interaction contributes to poor prognosis development. In addition, our preclinical data with a drug BASIC AND in hepatocellular carcinoma. Hepatology 2015;62:481– approved for pulmonology and non–small-cell lung cancer 495. treatment suggest that the permissive microenvironment 12. Zhang DY, Goossens N, Guo J, et al. A hepatic stellate cell can be reverted, leading to a reduction in HCC occurrence. gene expression signature associated with outcomes in These data provide the rationale for testing this strategy in hepatitis C cirrhosis and hepatocellular carcinoma after early chemoprevention trials targeting patients with curative resection. Gut 2016;65:1754–1764. cirrhosis who are at high risk of HCC development. In 13. Kim JH, Sohn BH, Lee H-S, et al. Genomic predictors addition, this strategy could also be further explored in the for recurrence patterns of hepatocellular carcinoma: adjuvant setting considering that 60% of HCC undergoing model derivation and validation. PLoS Med 2014;11(12): resection also present this permissive milieu in the adja- e1001770. cent nontumoral tissue. 14. Hoshida Y, Villanueva A, Sangiovanni A, et al. Prognostic gene expression signature for patients with hepatitis C- related early-stage cirrhosis. Gastroenterology 2013; Supplementary Material 144:1024–1030. Note: To access the supplementary material accompanying 15. Sia D, Jiao Y, Martinez-Quetglas I, et al. Identification of this article, visit the online version of Gastroenterology at an immune-specific class of hepatocellular carcinoma, www.gastrojournal.org, and at https://doi.org/10.1053/ based on molecular features. Gastroenterology 2017; 153:812–826. j.gastro.2019.07.028. 16. Villanueva A, Portela A, Sayols S, et al. DNA methylation-based prognosis and epidrivers in hepato- References cellular carcinoma. Hepatology 2015;61:1945–1956. 1. Bray F, Ferlay J, Soerjomataram I, et al. Global cancer 17. Wang M, Gong Q, Zhang J, et al. Characterization of statistics 2018: GLOBOCAN estimates of incidence and gene expression profiles in HBV-related liver fibrosis 1396 Moeini et al Gastroenterology Vol. 157, No. 5 patients and identification of ITGBL1 as a key regulator 34. Yoo M, Shin J, Kim J, et al. DSigDB: drug signatures of fibrogenesis. Sci Rep 2017;7:43446. database for gene set analysis. Bioinformatics 2015; 18. Godec J, Tan Y, Liberzon A, et al. Compendium of im- 31:3069–3071. mune signatures identifies conserved and species- 35. Richeldi L, du Bois RM, Raghu G, et al. Efficacy and specific biology in response to inflammation. Immunity safety of nintedanib in idiopathic pulmonary fibrosis. 2016;44:194–206. N Engl J Med 2014;370:2071–2082. 19. Brunet J-P, Tamayo P, Golub TR, et al. Metagenes 36. Wolf MJ, Adili A, Piotrowitz K, et al. Metabolic activation and molecular pattern discovery using matrix factor- of intrahepatic CD8þ T cells and NKT cells causes ization. Proc Natl Acad Sci U S A 2004;101:4164– nonalcoholic steatohepatitis and liver cancer via 4169. cross-talk with hepatocytes. Cancer Cell 2014;26: 20. Newman AM, Liu CL, Green MR, et al. Robust 549–564. enumeration of cell subsets from tissue expression 37. Brown ZJ, Heinrich B, Greten TF. Mouse models of profiles. Nat Methods 2015;12:453–457. hepatocellular carcinoma: an overview and highlights for 21. Bindea G, Mlecnik B, Tosolini M, et al. Spatiotemporal immunotherapy research. Nat Rev Gastroenterol Hepatol dynamics of intratumoral immune cells reveal the im- 2018;15:536–554. mune landscape in human cancer. Immunity 2013; 38. Coussens LM, Werb Z. Inflammation and cancer. Nature 39:782–795. 2002;420(6917):860–867. 22. Thorsson V, Gibbs DL, Brown SD, et al. The immune 39. Hanahan D, Weinberg RA. Hallmarks of cancer: the next landscape of cancer. Immunity 2018;48:812–830. generation. Cell 2011;144:646–674. 23. Reich M, Liefeld T, Gould J, et al. GenePattern 2.0. Nat 40. Makarova-Rusher OV, Medina-Echeverz J, Duffy AG, Genet 2006;38:500–501. et al. The yin and yang of evasion and immune activation 24. Charoentong P, Finotello F, Angelova M, et al. Pan- in HCC. J Hepatol 2015;62:1420–1429. cancer immunogenomic analyses reveal genotype- 41. Papatheodoridis GV, Dalekos GN, Yurdaydin C, et al. immunophenotype relationships and predictors of Incidence and predictors of hepatocellular carcinoma response to checkpoint blockade. Cell Rep 2017; in Caucasian chronic hepatitis B patients 18:248–262. receiving entecavir or tenofovir. J Hepatol 2015; 25. Dapito DH, Mencin A, Gwak G-Y, et al. Promotion of 62:363–370. hepatocellular carcinoma by the intestinal microbiota 42. Morgan RL, Baack B, Smith BD, et al. Eradication of and TLR4. Cancer Cell 2012;21:504–516. hepatitis C virus infection and the development of he- 26. Malehmir M, Pfister D, Gallage S, et al. Platelet GPIba patocellular carcinoma: a meta-analysis of observational is a mediator and potential interventional target for studies. Ann Intern Med 2013;158:329–337. NASH and subsequent liver cancer. Nat Med 2019; 43. Tiegs G, Hentschel J, Wendel A. A T cell-dependent 25:641–655. experimental liver injury in mice inducible by concanav- 27. Okamoto M, Utsunomiya T, Wakiyama S, et al. Specific alin A. J Clin Invest 1992;90:196–203. TRANSLATIONAL LIVER gene-expression profiles of noncancerous liver tissue 44. Omenetti S, Brogi M, Goodman WA, et al. Dysregulated intrahepatic CD4þ T-cell activation drives liver inflam- BASIC AND predict the risk for multicentric occurrence of hepato- cellular carcinoma in hepatitis C virus-positive patients. mation in ileitis-prone SAMP1/YitFc mice. Cell Mol Ann Surg Oncol 2006;13:947–954. Gastroenterol Hepatol 2015;1:406–419. 28. Wolf DM, Lenburg ME, Yau C, et al. Gene co- 45. Hilberg F, Roth GJ, Krssak M, et al. BIBF 1120: expression modules as clinically relevant hallmarks triple angiokinase inhibitor with sustained receptor of breast cancer diversity. PLoS One 2014;9(2): blockade and good antitumor efficacy. Cancer Res 2008; e88309. 68:4774–4782. 29. Beck AH, Espinosa I, Edris B, et al. The macrophage 46. Mas VR, Maluf DG, Archer KJ, et al. Angiogenesis sol- colony-stimulating factor 1 response signature in breast uble factors as hepatocellular carcinoma noninvasive carcinoma. Clin Cancer Res 2009;15:778–787. markers for monitoring hepatitis C virus cirrhotic patients 30. Rooney MS, Shukla SA, Wu CJ, et al. Molecular awaiting liver transplantation. Transplantation 2007; and genetic properties of tumors associated with local 84:1262–1271. immune cytolytic activity. Cell 2015;160:48–61. 47. Poon RTP, Ho JWY, Tong CSW, et al. Prognostic sig- 31. Finkin S, Yuan D, Stein I, et al. Ectopic lymphoid struc- nificance of serum vascular endothelial growth factor tures function as microniches for tumor progenitor and endostatin in patients with hepatocellular carcinoma. cells in hepatocellular carcinoma. Nat Immunol 2015; Br J Surg 2004;91:1354–1360. 16:1235–1244. 48. Horwitz E, Stein I, Andreozzi M, et al. Human and mouse 32. Nakagawa S, Wei L, Song WM, et al. Molecular liver VEGFA-amplified hepatocellular carcinomas are highly cancer prevention in cirrhosis by organ transcriptome sensitive to sorafenib treatment. Cancer Discov 2014; analysis and lysophosphatidic acid pathway inhibition. 4:730–743. Cancer Cell 2016;30:879–890. 49. Carmeliet P. VEGF as a key mediator of angiogenesis in 33. Chiang DY, Villanueva A, Hoshida Y, et al. Focal gains of cancer. Oncology 2005;69(Suppl 3):4–10. VEGFA and molecular classification of hepatocellular 50. Schiffer E, Housset C, Cacheux W, et al. Gefitinib, an carcinoma. Cancer Res 2008;68:6779–6788. EGFR inhibitor, prevents hepatocellular carcinoma November 2019 Immune-Mediated Field Cancerization as Target for HCC Prevention 1397 development in the rat liver with cirrhosis. Hepatology Conflicts of interest These authors disclose the following: Part of the study was supported with an 2005;41:307–314. investigator-initiated research grant by Boehringer Ingelheim. Josep M. Llovet 51. Fuchs BC, Hoshida Y, Fujii T, et al. Epidermal growth has been a consultant and advisory board member and has received research factor receptor inhibition attenuates liver fibrosis and funding from Boehringer Ingelheim and is receiving research support from Bayer HealthCare Pharmaceuticals, Eisai Inc, Bristol-Myers Squibb, and development of hepatocellular carcinoma. Hepatology Ipsen and consulting fees from Bayer HealthCare Pharmaceuticals, Bristol- 2014;59:1577–1590. Myers Squibb, Eisai Inc, Celsion Corporation, Eli Lilly, Exelixis, Merck, Ipsen, Glycotest, Navigant, Leerink Swann LLC, Midatech Ltd, Fortress Biotech, 52. Pinyol R, Sia D, Llovet JM. Immune exclusion-Wnt/ Sprink Pharmaceuticals, and Nucleix. Scott L. Friedman has been a CTNNB1 class predicts resistance to immunotherapies consultant for Abide Therapeutics, Allergan Pharmaceuticals, Angion in HCC. Clin Cancer Res 2019;25:2021–2023. Biomedica, Blade Therapeutics, Can-Fite Biopharma, Enanta Pharmaceuticals, Escient Therapeutics, Forbion, Galmed, Genfit, Glycotest, Glympse Bio, Metacrine Inc, Mistral Biosciences, Morphic Rock Therapeutics, North Sea Therapeutics, Novartis, Novo Nordisk, Pfizer Pharmaceuticals, Salix Pharmaceuticals, Scholar Rock, Seal Rock Author names in bold designate shared co-first authorship. Therapeutics, Second Genome, Surrozen, Symic Bio, Viking Therapeutics, and Kintai; has received research funding from Blade Therapeutics, Can-Fite Received January 28, 2019. Accepted July 17, 2019. Biopharma, Ferring Research Institute, Galmed; and has stock options from Intercept, Exalenz, Madrigal, Akarna Therapeutics, BirdRock Bio, Blade Reprint requests Therapeutics, Conatus, DeuteRx, Exalenz, Galectin, Galmed, Genfit, Address requests for reprints to Josep M. Llovet, MD, Liver Cancer Glympse. The remaining authors disclose no conflicts. Translational Research Laboratory, IDIBAPS-Hospital Clinic, Rosselló 153, 08039, Barcelona, Catalonia, Spain. e-mail: jmllovet@clinic.cat; fax: Funding þ34933129406; or Daniela Sia, PhD, Mount Sinai Liver Cancer Program, Josep M. Llovet is supported by the European Commission (EC)/Horizon 2020 Division of Liver Diseases, Tisch Cancer Institute, Icahn School of Medicine Program (HEPCAR, reference no. 667273-2); US Department of Defense at Mount Sinai, Madison Ave 1425. 11F-70. Box 1123, New York, NY 10029. (CA150272P3); an Accelerator Award (CRUCK, AECC, AIRC) (HUNTER, e-mail: daniela.sia@mssm.edu; fax: þ1212.849.2574. reference no. C9380/A26813), NCI Cancer Center Support Grant, National Cancer Institute; Tisch Cancer Institute (P30-CA196521); Samuel Waxman Acknowledgements Cancer Research Foundation; Spanish National Health Institute (SAF2016- The authors thank Juan José Lozano for technical assistance in the 76390); and the Generalitat de Catalunya/AGAUR (SGR-1358). Agrin Moeini normalization of the transcriptomic array of the animal model. This study was is supported by Spanish National Health Institute. Sara Torrecilla and Judit developed at the building of Centre Esther Koplowitz from IDIBAPS/CERCA Peix are funded by Centro de Investigación Biomedica en Red de Programme/Generalitat de Catalunya. The authors also acknowledge Angelo Enfermedades Hepáticas y Digestivas (Ciberehd-ISCIII). Carla Montironi is a Sangiovanni and Massimo Colombo for providing the seminal cohort of recipient of Josep Font grant. Carmen Andreu-Oller is supported by “la patients with cirrhotic in our previous studies.12,14 Caixa” INPhINIT Fellowship Grant (LCF/BQ/IN17/11620024). Roser Pinyol is Author contributions: Study concept and design: Agrin Moeini, Victoria supported by HEPCAR and AECC. Daniela Sia is supported by the Gilead Tovar, Daniela Sia, and Josep M. Llovet; acquisition of data: Agrin Moeini, Sciences Research Scholar Program in Liver Disease. Scott L. Friedman is Sara Torrecilla, Carla Montironi, Judit Peix, Mónica Higuera, and Manel Sole; supported by the National Institutes of Health Research project grant analysis and interpretation of data: Agrin Moeini, Sara Torrecilla, Victoria (R01,DK5662) and US Department of Defense (CA150272P3). Mathias Tovar, Carla Montironi, Carmen Andreu-Oller, and Manel Sole; drafting of the Heikenwälder was supported by an ERC Consolidator grant manuscript: Agrin Moeini, Daniela Sia, and Josep M. Llovet; critical revision (HepatoMetaboPath), the SFBTR 209, 1335 and SFBTR179. This project has of the manuscript for important intellectual content: Roser Pinyol, Scott L. received funding from the European Union’s Horizon 2020 research and Friedman, Daniela Sia, and Josep M. Llovet; obtained funding: Josep M. innovation program under grant agreement no. 667273 and the Helmholtz Llovet; study supervision: Josep M. Llovet and Daniela Sia. future topic (Zukunftsthema) Immunology and Inflammation. TRANSLATIONAL LIVER BASIC AND 1397.e1 Moeini et al Gastroenterology Vol. 157, No. 5 Supplementary Methods sodium chloride solution) given at day 15 postpartum fol- lowed by 11–14 weekly injections of CCl4 (0.5 mL/kg intraperitoneally, dissolved in corn oil) starting at 4 weeks Histologic Analysis of Human Samples of age (Supplementary Figure 7A). Mice (N ¼ 55) were Histologic evaluation was performed in 98 tissue sam- randomly assigned at 12 weeks of age to receive 50 mg/kg/ ples obtained from patients with early HCC included in the day of nintedanib (Boehringer Ingelheim) (n ¼ 29) or study cohort. The portal/septal, pericentral, and lobular vehicle (n ¼ 26). The vehicle solution was formulated as inflammation were assessed as follows: 0, absent; 1, mild; 2, follows: 1.8% hydroxypropyl b–cyclodextrin, 5% acetic acid moderate; and 3, marked. The presence or absence (1 or 0, (10%), and Natrosol (0.5%) (Sigma-Aldrich (#54290), Saint respectively) of interface inflammation and lymphoid ag- Louis, MO). At 15, 17, and 18 weeks of age, mice were killed gregates was also evaluated. The latter structures were 48 hours after the last dose of CCl4, having been treated with mainly found in the periportal/periseptal areas. An in- nintedanib for 3, 5, and 6 weeks, respectively (n ¼ 15–21 per flammatory infiltrate score was created by summing the time point). Immediately after mice were killed, livers were values given by the portal/septal, lobular, and interface explanted, digitally photographed, and weighed. The evalu- inflammation. Pericentral inflammation was not considered ation of macroscopic malignant nodules was assessed by for the scoring system because 57% of patients had 2 independent investigators. Based on visual criteria, a he- cirrhosis. According to the final score, we defined 2 in- patic lesion with a diameter >0.5 mm and with dysmorphic flammatory categories: the absent/mild category if the and/or dyschromic surface was considered a hepatic tumor. score was <3 and the moderate/marked category if the The diameter of tumors was measured with a hand caliper. score was 3. The presence or absence of cirrhosis was The biological endpoints for chemopreventive efficacy were defined according to the METAVIR algorithm1 (F0-1-2-3/ (1) incidence of hepatic tumors; (2) number of tumors; F4). Ductular proliferation was also considered. Steatosis and (3) size, considering the largest diameter of all counted was assessed based on the size of the fat vesicles (macro- tumors. The largest liver lobe was fixed in buffered 4% vesicular or microvesicular) and the localization of the fat paraformaldehyde for 24 hours for posterior histologic droplets in the liver parenchyma (periseptal/periportal, and immunohistochemical analysis. In addition, samples of pericentral, or lobular). The presence or absence of adjacent nontumor liver were snap frozen at –80 C for ballooning, apoptotic bodies, and oncocytic change was also subsequent RNA and protein analysis. Potential treatment- evaluated. Immunohistochemistry for CD4 and CD8 was related toxicity was evaluated by monitoring body weight performed in a subset of 70 patients of the training cohort. losses and quantitative determination of serum aspartate Staining was carried out on 3-mm-thick formalin-fixed aminotransferase and alanine aminotransferase. paraffin-embedded tissue sections collected from a 40 – 44 C flotation bath containing deionized water and moun- Histologic and Immunohistochemical Analysis of ted on 25  75 mm positively (þ) charged slides. The slides the Mouse Samples were dried at 60 C in a convection oven for 30 minutes. FFPE mouse liver samples were cut in 4-mm sections Deparaffinization was performed, followed by standard cell and stained with H&E for further histologic examination. conditioning 1 (ULTRA CC1 from Ventana Medical Systems, The samples were assessed, with nontumor and tumor Tucson, AZ). Staining was performed with the Ventana tissue considered, by 2 expert liver pathologists (CM, MS) BenchMark ULTRA system with the primary antibody for blinded to the treatments. In the nontumor tissue, the approximately 20 minutes at 36 C. Signal was captured number of lymphoid aggregates was determined. These using the Ultraview Universal diaminobenzidine detection were defined as polarized aggregates composed mainly of kit and blocked with antibody diluent (REF 251-018) for 8 lymphocytes with scant plasma cells, neutrophils, and minutes. Samples were counterstained with hematoxylin for macrophages localized in the periportal and pericentral 4 minutes and post-counterstained with Bluing Reagent for areas with a measure no bigger than 0.1 mm. The presence 4 minutes. The primary antibodies used were anti-CD4 of inflammatory infiltrates was also determined for the (clone SP35; Roche, Basel, Switzerland) and anti-CD8 periportal, pericentral, and lobular areas using a scoring (clone SP57, Roche). The quantification was done auto- system as follows: 0, absence or rare; 1, mild; 2, mild/ matically (Image J, National Institutes of Health, Bethesda, moderate; 3, moderate; 4, moderate/marked; and 5, MD) by calculating the positive areas (mm2)—considering 5 marked. Other variables, such as the presence of micro- random areas within the lobular and portal/septal in- abscesses, ductular proliferation (0, absence; 1, <25% of filtrates of the nontumor liver tissue—at 200 times portal tracts involved; 2, 25%–50% of portal tracts magnification and using the same threshold for all samples. involved; 3, >50% of portal tracts involved), necrosis, steatosis, and apoptotic bodies, were also determined. Generation of the Diethylnitrosamine/Carbon Ballooning was defined according to the following scale: 0, Tetrachloride Animal Model absence; 1, focal; 2, multiple foci; 3, diffuse. Hepatic fibrosis The chemically induced (DEN/CCl4) model was gener- was assessed by Sirius Red staining. The METAVIR algo- ated as previously described.2,3 HCC was induced by a rithm1 was used to grade hepatic fibrosis from F0 (no single intraperitoneal injection of DEN (Sigma-Aldrich, St. scarring) to F4 (cirrhosis or severe fibrosis). Also, peri- Louis, MO; 25 mg/kg intraperitoneally, dissolved in 0.9% sinusoidal fibrosis was evaluated as follows: 1, 0%–5%; 2, November 2019 Immune-Mediated Field Cancerization as Target for HCC Prevention 1397.e2 5%–33%; 3, 33%–66%, and 4, >60%. Histologically, he- expression microarray studies were conducted using the patic tumors presented a high cellular density composed of Gene Chip HT MG-430 (Affymetrix, Santa Clara, CA). The small cells with altered nuclear/cytoplasmic ratio and a raw .cel files were background corrected and normalized clear cytoplasm. Among these hepatic tumors, those that with the robust multiarray averaging procedure,4 with a were circular, well-defined lesions with pushing margins custom chip definition file (.cdf) from the Custom CDF were diagnosed as HCCs, whereas a dysplastic nodule was project (HTMG430PM_Mm_ENTREZG v18.0).5 To assess considered when the lesion had an ill-defined shape within ICF-signature enrichment through ssGSEA (GenePattern), the liver parenchyma. The presence of vascular invasion mice genes were humanized, and intensity values were log was a characteristic attributed to HCC. The size of both le- transformed. For quantitative reverse-transcription sions was not used to make a distinction between them, polymerase chain reaction analysis, complementary DNA although HCCs tend to be larger than dysplastic nodules. was synthesized from 1 mg purified total RNA using Su- Finally, for the evaluation of the chemopreventive effects of perScript III reverse transcriptase (Invitrogen). Relative a tyrosine kinase inhibitor in adjacent liver tissue, FFPE gene expression levels were measured by TaqMan sections from mice killed at 17 weeks of age were analyzed Gene Expression Assays (Applied Biosystems, Foster City, by immunohistochemistry (n ¼ 5 mice per treatment arm). CA) by using specific probes for Col1a1 (Mm00801666_g1), Heat-induced antigen retrieval was performed with 10 Col1a2 (Mm00483888_m1), Acta2 (Mm01546133_m1), mmol/L sodium citrate buffer (pH 6.0) or 0.5 mol/L Tris and Pdgfrb (Mm00435546_m1). The housekeeping gene buffer (pH 10.0) for 15 minutes (5 minutes, 3 times), and GAPDH (Mm99999915_g1) was used for normalization. the reaction was quenched using hydrogen peroxide 3%. Microarray data of these newly profiled samples are in the After washing with phosphate-buffered saline, samples Gene Expression Omnibus under accession number were incubated with anti-pERK (phosphoThr202/Tyr204) GSE125975. (Cell Signaling, Danvers, MA), anti-CD31 (Abcam, Cam- bridge, UK), and anti-CD4 and anti-CD8 (both from Cell Western Blot Analysis Signaling). Diaminobenzidine (3,30 -diaminobenzidine) was Whole-cell extracts were collected in lysis buffer (50 used as detection system (EnVisionþ System-HRP; Dako, mmol/L Tris pH 7.4, 150 mmol/L NaCl, 1% Triton X-100, Santa Clara, CA). Morphometric quantification of the 0.1% sodium dodecyl sulfate, 0.25 mmol/L EDTA, 1% so- percent area of liver vasculature (n ¼ 5 mice per treatment dium deoxycholate) containing phosphatases and proteases arm) was performed by evaluating the mean area of CD31 inhibitors (Roche). Next, 30–70 mg of total protein were staining in randomly selected low-magnification fields (n ¼ resolved in polyacrylamide gels and transferred to nitro- 5, 10 fields for CD31 staining; n ¼ 10, 20 fields for cellulose membranes (Pierce, Rockford, IL). Membranes collagen) with ImageJ software. were blocked with bovine serum albumin and hybridized overnight at 4 C with primary antibodies against VEGFR-2, RNA Extraction and Gene Expression Profiling of pVEGFR-2 (Tyr951), Akt, pAkt (Ser473), ERK1/2, pERK1/2 Mice Livers (Thr202/Tyr204), Bcl-xL, PARP (all from Cell Signaling), Total RNA from the 2 mouse models described was and tubulin (Sigma-Aldrich). Appropriate horseradish extracted from 20 mg fresh-frozen nontumoral liver tissue peroxidase–conjugated secondary antibodies (Dako) were with Trizol reagent (Invitrogen, Waltham, MA) and purified incubated for 1 hour at room temperature, and immuno- with RNeasy columns (Qiagen, Valencia, CA). RNA sample reactivity was detected with a LAS-3000 imaging system concentration and quality were assessed by NanoDrop ND- (Fujifilm, Tokyo, Japan) by using Amersham ECL Prime 1000 spectrometer (NanoDrop, Wilmington, DE) and bio- Western Blotting Detection Reagent (GE Healthcare, analyzer (Agilent, Palo Alto, CA), respectively. Gene Chicago, IL). 1397.e3 Moeini et al Gastroenterology Vol. 157, No. 5 6. Nakagawa S, Wei L, Song WM, et al. Molecular liver Supplementary References cancer prevention in cirrhosis by organ transcriptome 1. Bedossa P, Poynard T. An algorithm for the grading of analysis and lysophosphatidic acid pathway inhibition. activity in chronic hepatitis C. The METAVIR Cooperative Cancer Cell 2016;30:879–890. Study Group. Hepatology 1996;24:289–293. 7. Hoshida Y, Villanueva A, Kobayashi M, et al. Gene 2. Uehara T, Ainslie GR, Kutanzi K, et al. Molecular expression in fixed tissues and outcome in hepatocel- mechanisms of fibrosis-associated promotion of liver lular carcinoma. N Engl J Med 2008;359:1995–2004. carcinogenesis. Toxicol Sci 2013;132:53–63. 8. Kim JH, Sohn BH, Lee H-S, et al. Genomic predictors for 3. Dapito DH, Mencin A, Gwak G-Y, et al. Promotion of recurrence patterns of hepatocellular carcinoma: model hepatocellular carcinoma by the intestinal microbiota derivation and validation. PLoS Med 2014;11(12): and TLR4. Cancer Cell 2012;21:504–516. e1001770. 4. Irizarry RA, Hobbs B, Collin F, et al. Exploration, 9. Ji J, Eggert T, Budhu A, et al. Hepatic stellate cell and normalization, and summaries of high density oligonu- monocyte interaction contributes to poor prognosis in cleotide array probe level data. Biostatistics 2003;4:249– hepatocellular carcinoma. Hepatology 2015;62:481–495. 264. 10. Zhang DY, Goossens N, Guo J, et al. A hepatic stellate 5. Dai M, Wang P, Boyd AD, et al. Evolving gene/ cell gene expression signature associated with transcript definitions significantly alter the interpretation outcomes in hepatitis C cirrhosis and hepatocellular of GeneChip data. Nucleic Acids Res 2005;33(20): carcinoma after curative resection. Gut 2016;65:1754– e175. 1764. November 2019 Immune-Mediated Field Cancerization as Target for HCC Prevention 1397.e4 Supplementary Figure 1. Identification of an ICF in the nontumor liver tissues and its association with overall survival in patients with early HCC. (A) Consensus-clustered analysis of gene sets recapitulating the different cell states and perturba- tions within the immune system. The figure includes a subgroup of representative gene sets among the approximately 5000 gene sets evaluated (Collection C7 of MsigDB). (B–D) Kaplan-Meier estimates of overall survival according to (B) the presence of the ICF, (C) the presence of the high-infiltrate subtype, and (C) the different ICF subtypes. ns, not significant. 1397.e5 Moeini et al Gastroenterology Vol. 157, No. 5 Supplementary Figure 2. Immunohistochemistry assessment of CD4 and CD8 positivity in nontumor liver tissues. Average stained area was automatically quantified and compared in (A) samples from patients belonging to the ICF and non-ICF subgroups, and (B) between ICF subtypes. n ¼ 15–19 samples per group. ***P < .001. Supplementary Figure 3. Gene overlap between the 172 ICF gene-expression signature and other previously reported poor-prognosis expression signatures in nontumor liver tissue. Signatures are denoted by the name of the first author of each of the publications.6–10 Numbers in brackets indicate the total number of genes that constitute each signature, and bold numbers represent the number of genes that overlap between each signature and the ICF signature (here, referred to as Moeini). November 2019 Immune-Mediated Field Cancerization as Target for HCC Prevention 1397.e6 Supplementary Figure 4. External validation of the molecular and clinical features of the ICF and its subtypes. (A) The main molecular characteristics of the ICF and the 3 distinct molecular subtypes were validated in a previously reported cohort that included 225 FFPE nontumor liver tissue samples from patients with early HCC profiled by cDNA-mediated Annealing, Selection, Extension, and Ligation (DASL) array. (B) Kaplan-Meier estimates of overall survival according to the presence of the ICF. NTP, nearest template prediction. 1397.e7 Moeini et al Gastroenterology Vol. 157, No. 5 Supplementary Figure 5. Association of the presence of ICF subtypes with risk of HCC development and advanced liver disease in patients with fibrosis. (A) Kaplan-Meier analysis of the correlation of the ICF subtypes with risk of HCC development. (B) Correlation of the prediction of ICF subtypes with fibrosis and inflammation degree. Fibrosis degree was classified as none (S0), low (S1/S2), and high (S3/S4). Inflammation degree was classified as none (G0), low (G1/G2), and high (G3/G4). Non ICF, non-ICF subtype and unclassified cases; other ICF: high-infiltrate and proinflammatory ICFs. November 2019 Immune-Mediated Field Cancerization as Target for HCC Prevention 1397.e8 Supplementary Figure 6. In silico enrich- ment analysis of gene sets recapitulating the targets of approved molecular therapies in nontumor liver of patients with early HCC. (A) Heatmap representing the enrichment scores of gene sets recapitulating the molecular targets of top enriched therapies in nontumor liver of HCC pa- tients harboring the ICF compared with non-ICF/ other. (B) Constellation map representation of enrichment of the gene set of the main targets of top- scored therapies centered around the ICF phenotype. The blue lines denote presence of overlapping genes among the different gene sets. AUC.pval, area under the curve P value; NMI, normalized mutual information; t.pval, t sta- tistic P value; t.stat, t statistic. 1397.e9 Moeini et al Gastroenterology Vol. 157, No. 5 Supplementary Figure 7. Prolonged treatment with nintedanib does not induce significant toxicity in DEN/CCl4-induced mouse models. (A) Experimental design of the murine model of DEN/CCl4-induced hepatocarcinogenesis in the context of liver fibrosis. (B) Monitoring of body weight in all mice from each experimental group during the administration of nintedanib or vehicle at the longest time point, 18 weeks of age. (C) Evaluation of the serum levels of alanine transaminase (ALT) and aspartate transaminase (AST) in each experimental group. Nin., nintedanib; Veh., vehicle. November 2019 Immune-Mediated Field Cancerization as Target for HCC Prevention 1397.e10 1397.e11 Moeini et al Gastroenterology Vol. 157, No. 5 = Supplementary Figure 8. Nintedanib inhibits the downstream mitogen-activated protein kinase pathway but has no effect on reverting DEN/CCl4-induced platelet-derived growth factor signaling or hepatic fibrosis. (A) Western-blot analysis of down- stream mitogen-activated protein kinase (protein kinase B and ERK) signaling. (B) Representative images and quantification of pERK staining in 17-week-old mice treated with vehicle or nintedanib. In the vehicle arm, the tumors and adjacent nontumor tissue are positive with patchy nuclear and cytoplasmic staining, whereas in the nintedanib arm, both are negative. (C) Western blot analysis of the profibrogenic platelet-derived growth factor receptor signaling. (D) Representative images of Sirius Red staining in mice treated with vehicle or nintedanib. Nintedanib did not exert any significant effect on liver fibrosis. (E) Gene expression levels of profibrogenic marker genes by quantitative reverse-transcription polymerase chain reaction in livers of mice killed at 15, 17, and 18 weeks of age treated with vehicle or nintedanib. The GAPDH gene was used as a housekeeper for normalization. Significant statistical differences are defined as follows: *P < .05 and ***P < .001. Nin., nintedanib; NT, non- tumor; T, tumor; Veh., vehicle.