Transforming primary human hepatocytes into hepatocellular carcinoma with genetically defined factors

Abstract

Our understanding of human hepatocellular carcinoma (HCC) development and progression has been hampered by the lack of in vivo models. We performed a genetic screen of 10 oncogenes and genetic mutations in Fah‐ablated immunodeficient mice in which primary human hepatocytes (PHHs) are used to reconstitute a functional human liver. We identified that MYC, TP53^R249S, and KRAS^G12D are highly expressed in induced HCC (iHCC) samples. The overexpression of MYC and TP53^R249S transform PHHs into iHCC in situ, though the addition of KRAS^G12D significantly increases the tumorigenic efficiency. iHCC, which recapitulate the histological architecture and gene expression characteristics of clinical HCC samples, reconstituted HCC after serial transplantations. Transcriptomic analysis of iHCC and PHHs showed that MUC1 and FAP are expressed in iHCC but not in normal livers. Chimeric antigen receptor (CAR) T cells against these two surface markers efficiently lyse iHCC cells. The properties of iHCC model provide a biological basis for several clinical hallmarks of HCC, and iHCC may serve as a model to study HCC initiation and to identify diagnostic biomarkers and targets for cellular immunotherapy.

The understanding of hepatocellular carcinoma development and progression is hampered by the lack of in vivo models. This study presents an in situ mouse model of human induced HCC that allows studying HCC initiation and the identification of biomarkers.

Introduction

Primary liver cancer (PLC) has the second‐highest cancer mortality rate worldwide (Broutier et al, ²⁰¹⁷). The development of PLC is invariably associated with liver damage caused by hepatitis B virus (HBV) infection, alcohol abuse, or toxin exposure, sequentially resulting in liver cirrhosis, dysplastic lesions, and finally invasive liver carcinoma (Cancer Genome Atlas Research Network. Electronic address wbe & Cancer Genome Atlas Research N, 2017; Gao et al, ²⁰¹⁹). Based on its histopathological features, PLC may be mainly classified as HCC, cholangiocarcinoma (CC), hepatoblastoma (HB), and combined hepatocellular‐cholangiocarcinoma (CHC) (Broutier et al, ²⁰¹⁷). The cellular origin and molecular genetics of human PLC remain poorly understood due to the lack of suitable models.

Genomic analysis has identified frequently mutated genes, including TP53, KRAS, CTNNB1, IDH1, IDH2, and IL6ST in PLC samples (Totoki et al, ²⁰¹⁴; Schulze et al, ²⁰¹⁵; Cancer Genome Atlas Research Network. Electronic address wbe & Cancer Genome Atlas Research N, ²⁰¹⁷), indicating essential roles of these genes in PLC development. There are six hotspot mutations of TP53 at R175, G245, R248, R249, R273, and R282, identified in various primary and metastatic human cancers. Notably, these TP53 mutants exert their dominant‐negative effects of wild‐type TP53 and possess gains of new functions (Kern et al, ¹⁹⁹²; Brosh & Rotter, ²⁰⁰⁹; Zhao et al, ^{2019, 2020}). TP53^R249S is highly associated with HCC and the only hotspot mutant that has been identified among 30% of HCCs that harbor TP53 mutations (Hsu et al, ¹⁹⁹¹; Hussain et al, ²⁰⁰⁷; Sladky et al, ²⁰²⁰). Hotspot mutations of KRAS such as KRAS^G12D and KRAS^G13D are commonly detected in PLC samples (Hunter et al, ²⁰¹⁵). Activation of Kras^G12D and deletion of p53 induce the development of HCC and CC in a genetically engineered mouse model (O'Dell et al, ²⁰¹²). Genomic amplifications in MYC have been identified in 6–25% of primary human cancers (Poon et al, ²⁰⁰⁶; Schlaeger et al, ²⁰⁰⁸). Moreover, constitutive overexpression of c‐Myc in the liver of p53‐deficient mice promotes hepatocarcinogenesis (Klocke et al, ²⁰⁰¹). These studies reveal that PLC development results from crosstalk of multiple genetic alterations. However, whether these genetic lesions can induce tumorigenesis of PHHs in situ on mouse models has not been investigated.

Humanized mouse models offer considerable advantages in the study of immunology, infectious diseases, regenerative medicine, and cancer (Shultz et al, ²⁰⁰⁷). Humanized livers have been reconstituted with PHHs in immunodeficient mice with transgenic urokinase plasminogen activator transgenic expression (Tateno et al, ²⁰⁰⁴) or fumarylacetoacetate hydrolase (Fah) deficiency (Azuma et al, ²⁰⁰⁷) for toxicity evaluation (Xu et al, ²⁰¹⁴). Patient‐derived xenografts (PDX) have been widely used to evaluate the efficacy of antitumor therapies (Kamel‐Reid et al, ¹⁹⁸⁹; Jiang et al, ²⁰¹⁶; Lai et al, ²⁰¹⁷; Bleijs et al, ²⁰¹⁹). However, these approaches have not allowed the cell types from which tumor‐initiating cells originate to be identified or the mechanism by which these tumor‐initiating cells phenotypically/genetically evolve during human disease progression to be assessed. Only a few in vivo models of cancers initiated from primary human cells have been developed. Upon transplantation into immunodeficient mice, primary human hematopoietic cells expressing mixed‐lineage leukemia (MLL) fusion gene generated acute leukemia (Barabe et al, ²⁰⁰⁷). In addition, both ectopic expression of oncogenes and ablation of tumor suppressor genes successfully induce reprogramming of human‐induced hepatocytes (hiHeps) and primary human intestinal stem cells into malignant cells in culture, although in vitro expansion and selection of organoids before transplantation were required (Hahn et al, ¹⁹⁹⁹; Drost et al, ²⁰¹⁵; Sun et al, ²⁰¹⁹). However, the direct conversion of human PHHs into HCC in situ has not been established. Here, we screened ten oncogenes and found that overexpression of MYC, TP53^R249S, and KRAS^G12D led to an efficient transformation of PHHs directly into HCC in a humanized liver mouse model. Further analyses confirmed that iHCC recapitulated the immunophenotypic histological architecture, tumor serum marker profile, phenotypes, and gene expression characteristics of HCC.

Results

Engraftment of PHHs in the liver of NSIF mice

To establish a humanized liver mouse model, we transplanted PHHs into NOD/SCID/IL2rg^‐/‐/Fah^‐/‐ (NSIF) mice, which were generated by targeting the exon 3 of Fah with embryo microinjection of TALEN mRNA (Fig 1A and B) into immunodeficient NOD/SCID/IL2rg^‐/‐ (NSI) mice (Ye et al, ²⁰¹⁵). Western blotting confirmed that fah was not expressed in NSIF mice (Fig 1C). NSIF mice exhibited liver injury upon withdrawal of nitisinone (NTBC) (Fig 1D), which promotes engraftment of PHHs (Azuma et al, ²⁰⁰⁷). NSIF mice that did not receive PHH transplantation experienced gradual body weight loss and died approximately 4 weeks after withdrawal of NTBC, whereas the PHH‐transplanted mice survived significantly longer (Fig 1E). The serum ALT and AST levels were significantly reduced in PHH‐transplanted NSIF mice (Fig 1E). The survival duration of PHH‐transplanted NSIF mice were positively correlated with the repopulation efficiencies of PHHs in the liver and the serum level of human ALB (Fig 1F and G, and Appendix Table S1). Next, immunohistochemistry (IHC) and immunofluorescence (IF) staining confirmed that the engrafted PHHs were morphologically similar to human hepatocytes and expressed human FAH, CK8/18, and ALB (Fig 1G and H). To trace PHHs in vivo, we overexpressed enhanced green fluorescent protein (EGFP) and luciferase (Luc) in PHHs and transplanted the cells into NSIF mice via an intrasplenic injection (Fig 2A and B). Bioluminescence imaging (BLI) using an IVIS Spectrum system showed that PHHs were specifically engrafted into the liver but not other organs in NSIF mice (Fig 2C and D). These data indicate NSIF mice could be used to generate humanized liver models.

Figure 1. Establishment of Fah knockout NSI mice for the humanized liver mouse model.

1. Schematic outline of the pronuclear microinjection procedure. In brief, female NSI mice were superovulated with PMSG and hCG and then mated with male NSI mice. Fertilized eggs were collected from NSI mice with a vaginal plug. TALEN mRNA pairs were injected into the cytoplasm of pronuclear‐stage mouse embryos. The injected embryos were then transferred into pseudopregnant surrogate mothers. The mouse pups were genotyped. The mouse was drawn by the authors (Z. J. and P. Li) using Adobe Photoshop.
2. DNA‐binding sequences and the spacer region (in blue) of TALENs targeting the exon 3 of Fah.
3. NSI mice homozygous for the Fah knockout displayed a lethal neonatal phenotype (left). Fah expression was detected in the livers of homozygous NSI mice with Fah knockout and Fah intact NSI mice by Western blotting (right). Cropped blots are shown.
4. Histological staining of livers from Fah‐deficient (NSIF) with and without NTBC treatment and Fah intact (NSI) mice without NTBC treatment. Scale bars, 50 μm.
5. (Left) Kaplan–Meier survival curves of PHH‐infused NSIF mice (in blue, n = 15) and NSIF mice without PHHs (in red, n = 9) after NTBC withdrawal and NTBC‐treated NSIF mice (in green, n = 10); (Right) Serum ALT and AST levels in NSIF mice (+NTBC, n = 5), moribund NSIF mice (−NTBC, n = 5) and PHH‐engrafted (8‐week‐old) mice (−NTBC, n = 5) at indicated time points post‐NTBC withdrawal; data are presented as the mean ± SD, and statistical significance was determined by two‐way ANOVA with Sidak's multiple comparisons test and defined as ***P ≤ 0.001.
6. Human ALB levels in the sera of PHH‐engrafted NSIF mice (n = 5) were measured by ELISA at indicated time points post‐NTBC withdrawal, data are presented as the mean ± SD. N‐numbers refer to biological replicates.
7. (Left) Quantification of the repopulation efficiencies of PHHs in livers estimated by hALB‐positive staining at indicated time points post‐transplantation. Data are presented as the mean ± SD. N‐numbers refer to biological replicates (n = 5); (Right) Representative immunohistochemical staining of hALB in the livers of NSIF mice at indicated time points post‐PHH transplantation. Scale bars, 1,000 μm.
8. Engraftment of PHHs into NSIF mouse livers was detected immunostaining for human ALB, CK8/18, and FAH. Scale bars, 50 μm.

Source data are available online for this figure.

Figure 2. Bioluminescence imaging of humanized liver mice.

1. Schematic illustration of PHHs labeled via EGFP and luciferase (Lentivirus‐GL) and intrasplenically injected into NSIF mice.
2. PHHs were transduced with/without lentiviruses containing EGFP and luciferase. Scale bar, 100 μm.
3. Bioluminescence signals were monitored in NSI and NSIF mice (n = 3) infused with EGFP‐Luc transduced PHHs at the indicated time points post‐PHH infusion. Representative images of a NSI and a NSIF mouse transduced with PHHs are shown.
4. Bioluminescence signals were detected in the livers but not in other tissues of NSIF mice that were transplanted with EGFP‐Luc‐labeled PHHs (n = 3).

Screening oncogenes contribute to the transformation of PHHs into HCC

To identify oncogenes that were able to induce reconstituted PHHs to transform into tumor cells in situ, we selected 10 oncogenes or oncogenic mutants as oncogenic candidates (OC), including MYC, TP53^R249S, KRAS^G12D, NRAS^G12D, CTNNB1^S45F, BRAF^V600E, AXIN1^G652S, IL6, PIK3CA^E542K , and CSF1R^Y969C , based on their contributions to HCC development identified in previous studies (Schulze et al, ²⁰¹⁵; Bailey et al, ²⁰¹⁸; Llovet et al, ²⁰¹⁸). The expression levels of KRAS, NRAS, and AXIN1 were positively correlated with poor prognosis in HCC patients according to The Cancer Genome Atlas (TCGA) database analysis (Appendix Fig S1A). We cloned these oncogenes and oncogenic mutants individually into a lentiviral vector containing the EGFP sequence (Appendix Fig S1B). PHHs from healthy donors highly expressed Albumin (ALB) but barely expressed CK19, a bile duct/progenitor marker (Zhang et al, ²⁰¹⁸) (Fig EV1A). PHHs were transduced with a cocktail of lentiviruses containing 10 OC and lentiviruses encoding EGFP and luciferase as a reporter with similar transduction efficiencies higher than 97% (Fig EV1B and Appendix Table S2) and were then transplanted into NSIF mice via splenic injection on day 2 (Fig 3A). The qPCR results showed that the copy numbers of these 10 oncogenic lentiviruses in PHHs were similar (Fig EV1C). The functions of PHHs were not severely affected by the transduction of oncogenes, as both OC‐transduced PHHs and unmodified PHHs produced Albumin and APOA, stored glycogen, and expressed hepatocyte‐associated genes, including ALB, AAT, CYP2B6, CYP1A2, ARG1, and TAT at similar levels (Fig EV1D–G). Notably, ectopic expression of mutant TP53 was significantly higher than endogenous TP53 (WT) (Fig EV1H). After 4 months of transplantation, genetically modified PHHs significantly engrafted in the liver of NSIF mice (Fig 3B). The serum levels of AFP, an HCC marker (Sun et al, ²⁰¹⁹), increased, while ALB concentrations did not change 3 months post‐OC transduced‐PHHs infusion (Fig 3C). In contrast, AFP was not detectable in NSIF mice that were transplanted with EGFP only‐transduced PHHs (mock group) (Fig 3C). Three to 6 months post‐transplantation, 8 out of the 30 NSIF mice transplanted with OC‐transduced PHHs showed sporadic sickness and were necropsied. Tumors were detected in the recipients of OC‐transduced PHHs but not in the mock group (Figs 3D and EV1I, and Appendix Table S2). These induced HCC (iHCC) were derived from human cells, as they expressed EGFP and HLA (Fig 3E). Single‐cell suspensions from these tumors reconstituted tumors in immunodeficient NSI mice after both serial subcutaneous transplantation (Fig EV1J) and orthotopic injection (Fig EV1K and L). To identify the oncogenic candidates that contributed to the tumorigenicity of PHHs in vivo, we performed PCR to amplify lentiviral sequences that integrated into the genomic DNA of the primary iHCC from the OC group mice. We found that all eight tumor samples contained copies of MYC and TP53^R249 , and seven out of eight (87.5%) tumor samples contained copies of KRAS^G12D but not any other candidates (Fig 3F and Appendix Table S2). The DNA sequencing results confirmed that mutant forms of TP53^R249S and KRAS^G12D were integrated into the iHCC genomes (Fig 3G) and were expressed (Fig 3H). Consistently, MYC, p53, and KRAS^G12D were highly expressed in iHCC samples but not in humanized livers reconstituted with mock‐PHHs from NSIF mice (Fig 3I and J). In addition, none of the 9 NSIF mice that were transplanted with mock‐PHHs from three different donors developed tumors (Fig 3J and Appendix Table S2). Those data indicate that humanized liver models are suitable for inducing tumorigenesis.

Figure EV1. iHCC derived from PHHs by oncogenic candidate transduction.

• A
  Commercial cryopreserved PHHs (PHH1, PHH4, and PHH5) were seeded in 6‐well plates and immunoassayed with antibodies that recognize ALB and KRT19. Scale bars, 20 μm.
• B
  Representative FACS analysis of PHHs that were transduced with a cocktail of 10 OC lentiviruses (right) or mock lentivirus (left).
• C
  qPCR assays measuring the transduction efficiencies of individual oncogenes using the genomic DNA samples of OC‐transduced PHHs. Relative copy numbers of DNA of oncogenes to the DNA of GFP. Data are presented as the mean ± SD (n = 3). N‐numbers refer to biological replicates.
• D
  The levels of albumin secretion in PHHs and OC‐transduced PHHs were measured by ELISA (n = 3), P = 0.0378 with unpaired t‐test, *P ≤ 0.05. Data are presented as the mean ± SD. N‐numbers refer to biological replicates.
• E
  The levels of APOA1 secretion in PHHs and OC‐transduced PHHs were measured by ELISA (n = 3), P > 0.05 with unpaired t‐test, data are presented as the mean ± SD, N.S. indicates non‐statistical significance. N‐numbers refer to biological replicates.
• F
  Glycogen storage was quantified by colorimetric in PHHs and OC‐transduced PHHs (n = 3 ), P > 0.05 with unpaired t‐test, data are presented as the mean ± SD. N.S. indicates non‐statistical significance. N‐numbers refer to biological replicates.
• G
  qPCR results of expression levels of hepatic genes in PHHs, OC‐transduced PHHs, and HepG2 cells on day 2 post‐transduction in culture (n = 3), data are presented as the mean ± SD. N‐numbers refer to biological replicates.
• H
  The expression levels of endogenous (WT) and mutant (Mut) TP53 in OC‐transduced PHHs (n = 5, P < 0.0001 with one‐way ANOVA by Tukey's multiple comparison test), ***P ≤ 0.001, data are presented as the mean ± SD. N‐numbers refer to biological replicates.
• I
  Mice with advanced intrahepatic tumor exhibited swollen and ascites‐containing abdomens, allowing detection of tumor burden by palpation. Tumors were pointed with white arrows.
• J–L
  Primary iHCC samples were harvested from mice from the OC group and were transplanted. 1 × 10⁶ iHCC cells were subjected for serial transplantations into NSI mice subcutaneously or splenically. Tumors were reconstituted post‐three rounds of subcutaneous transplantations of iHCC (J); Bright‐field and EGFP imaging of an explanted liver from a tumor‐bearing NSI mouse post splenic injection of iHCC cells. Scale bars, 0.5 cm (K); The anti‐GFP immunofluorescence and H&E staining of tumor dissection from a tumor‐bearing NSI mouse post splenic transplantation of iHCC. Scale bars, 20 μm (L).

Figure 3. Screening of oncogenes involved in the transformation of PHHs into HCC cells.

• A
  Schematic representation showing the process for induction of in situ HCC after lentiviral transduction of PHHs with a cocktail of 10 OC lentiviruses.
• B
  Bioluminescence imaging of three recipient NSIF mice (n = 3) that had been transplanted with OC‐transduced PHHs 4 months ago. The colors from blue to red indicate intrahepatic bioluminescence signal strength 4 months post‐transplantation of OC transduced PHHs.
• C
  ALB and AFP levels in sera from NSIF mice that were infused with OC‐transduced PHHs (OC‐PHHs) or EGFP‐transduced PHHs (mock‐PHHs) were measured by ELISA at indicated time points. (n = 5, the data are presented as the mean ± SD values. Two out of five mice died before 120 days post‐transplantation in the OC‐PHHs group. The levels of ALB and AFP in the rest three mice were measured.).
• D
  Bright and EGFP imaging of an explanted liver of a mouse from the OC group revealed an advanced intrahepatic tumor, Scale bars, 0.5 cm.
• E
  Anti‐GFP immunofluorescence (left), H&E staining (center), anti‐HLA‐I IHC staining (right) images of tumor dissection from the OC group. Scale bars, 50 μm.
• F
  Representative PCR amplification of MYC, TP53^R249S, KRAS^G12D, NRAS^G12D, CTNNB1^S45F, BRAF^V600E, AXIN1^G652S, IL6, PIK3CA^E542K and CSF1R^Y969C in the genomic DNA of dissected primary tumors harvested from NSIF mice transplanted with OC‐PHHs (Tumor) and their corresponding overexpression vector plasmids as positive controls (OCs).
• G, H
  Genomic DNA (gDNA) and cDNA of tumor samples were obtained for PCR amplification using forward primers that recognize TP53^R249S or KRAS^G12D and reverse primers specific for EGFP. DNA sequencing confirmed mutations of TP53 and KRAS in the genomic DNA (G) and mRNA (H) of dissected primary tumors harvested from tumor‐bearing mice. Red squares indicate mutant codes in TP53^R249S and KRAS^G12D .
• I
  Expression of MYC, p53, and KRAS^G12D was detected in the iHCC samples (n = 3, biological replicates) by Western blotting. Cropped blots are shown.
• J
  Immunofluorescence analysis of MYC, p53, and KRAS expression in iHCC samples derived from OC‐PHHs and normal humanized livers that were reconstituted with mock‐PHHs. Scale bars, 20 μm.

Source data are available online for this figure.

MYC, TP53, and KRAS corporately transform PHHs into HCC in vivo

To validate whether the combination of MYC, TP53^R249S, and KRAS^G12D can transform PHHs into iHCC, we transduced PHHs with a mixture of lentiviruses overexpressing MYC, TP53^R249S, and KRAS^G12D (MTK) and intrasplenically infused them into NSIF mice. Six months later, 63.6% (14 out of 22 mice) of the recipient mice developed iHCC. Tumorigenic rate of MTK‐transduced PHHs is significantly higher than that of 16.6% (2 out of 12 mice) MT‐transduced PHHs (Fig 4A and Table 1). While tumors were not detected in the 9 NSIF mice that were transplanted with PHHs transduced only with EGFP (Table 1 and Appendix Table S2). In addition, colonies derived from single cells of different tumors, such as iHCC1‐1 and iHCC2‐1 contained all MYC, TP53^R249S , and KRAS^G12D (Fig 4B), suggesting that individual iHCC cells expressed all the three OCs. The H&E‐stained sections of orthotropic iHCC samples presented heterogeneous morphologies ranging from solid/compact to pseudoglandular rosette structures of HCC, in contrast to the homogeneous structure of normal human hepatocytes (Fig EV2). Tumors derived from MTK‐transduced PHHs expressed HLA, ALB, and HCC markers, including AFP and GPC3 (Figs 4C and EV3A). Notably, iHCC samples (iHCC1‐1 and iHCC3‐1) expressed cytokeratin‐19 (CK19) and EpCAM, which are expressed in hepatic progenitors (Rao et al, ²⁰⁰⁸) and are considered as poor prognostic markers (Chan et al, ²⁰¹⁴; Govaere et al, ²⁰¹⁴; Llovet et al, ²⁰¹⁵) in HCC patients (Figs 4C and EV3A). To investigate whether wild‐type TP53 affects tumorigenesis in PHHs that overexpress MYC, TP53^R249S, and KRAS^G12D, we inactivated TP53 by CRISPR/Cas9 in MTK‐transduced PHHs and orthotopically transplanted these genetically modified PHHs into NSIF mice for tumor watch (Fig EV3B and C). Tumors were detected in three out of four recipient mice (Fig EV3D). These results demonstrate that overexpression of TP53^R249S, MYC, and KRAS^G12D induced PHHs with WT p53 inactivation to transform into iHCC in vivo.

Figure 4. Generation of genetically defined in situ humanized liver cancers.

1. Representative images of in situ liver carcinomas derived from PHHs transduced with a combination of MYC, TP53^R249S, and KRAS^G12D (MTK, n = 22) or a combination of MYC and TP53^R249S (MT, n = 12) in NSIF mice. No tumors were detected in any of the mice injected with PHHs that were transduced with TP53^R249 and KRAS^G12D (TK, n = 9) or PHHs that were transduced with MYC and KRAS^G12D (MK, n = 9). Scale bar, 1 cm.
2. Genomic DNA of colonies that were derived from single cells of iHCC1‐1 and iHCC2‐1 tumor samples were extracted for PCR amplification of MYC, TP53^R249S, and KRAS^G12D .
3. Immunofluorescence analysis of the human cell marker HLA, the ductal/CC markers KRT19 and EpCAM, and the HCC markers AFP and GPC3 in a iHCC sample from NSIF mice transplanted with MTK‐transduced PHHs from the donor PHH1 (iHCC1‐1) and a clinical patient HCC sample (pHCC1). Nuclei were counterstained with DAPI. Scale bars, 20 μm.
4. Growth curves of PHHs after transduction with various combinations of MYC (M), TP53^R249S (T) and KRAS^G12D (K) in culturing medium supplemented with hHGF (10 ng/ml), hEGF (5 ng/ml), and WNT3a (5 ng/ml), P = 0.0038 for MTK‐ versus MT‐transduced PHH by two‐way ANOVA with Tukey's multiple comparison test; **P = 0.0038 for MTK‐ versus MT‐transduced PHH by two‐way ANOVA with Tukey's multiple comparison test; data are presented as the mean ± SD. N‐numbers refer to biological replicates (n = 3).
5. Cell cycle distribution analysis of mock‐PHHs, MTK‐transduced PHHs, and MT‐transduced PHHs on day 10. **P = 0.0034 for MT‐transduced PHHs versus mock‐PHHs and ***P = 0.0009 for MTK‐transduced PHHs versus mock‐PHHs by two‐way ANOVA with Tukey's multiple comparison test; data are presented as the mean ± SD. N‐numbers refer to biological replicates (n = 3).
6. Morphology of PHHs transduced with various combinations of MYC (M), TP53^R249S (T), and KRAS^G12D (K) at day 10. EGFP positivity indicates transduction of oncogenic candidates: scale bars, 100 μm.

TABLE 1.

Validation of the combination of MYC, TP53^R249S, and KRAS^G12D for transforming PHHs into HCC.

Hepatocytes a OC b	PHH1 (AKB)	PHH2 (XSM)	PHH3 (ANG)	In total d
MTK	6/10 c	4/6	4/6	14/22
MT	1/6	1/3	0/3	2/12
TK	0/3	0/3	0/3	0/9
MK	0/3	0/3	0/3	0/9
Mock	0/3	0/3	0/3	0/9

^a Hepatocytes: PHHs from three donors, including PHH1 (AKB, female, 39 years old), PHH2 (XSM, female, 59 years old), and PHH3 (ANG, male, 3 months old), were purchased from Bioreclamation IVT (Baltimore, MD, USA) and were used in the experiment.

^b OC: A cocktail of lentivirus containing different combinations of oncogenic candidates (M for MYC, T for TP53R249S, and K for KRASG12D) were transduced into PHHs from different donors. Mock: lentivirus containing EGFP only.

^c 6/10: The frequency of mice that developed iHCC in the group where PHHs from different donors (PHH1~PHH3) were transduced with indicated combination of oncogenic candidates. For example, 6 out of 10 mice, in which PHHs from the donor PHH1 were transduced with MTK and were transplanted, were observed bearing tumors.

^d In total: ***P < 0.001 for MTK‐ versus MT/Mock‐transduced PHH by one‐way ANOVA with Tukey's multiple comparison test; data are presented as the mean ± SD.

Figure EV2. Morphologic characteristics of iHCC.

Representative H&E‐stained sections of iHCC tissues (iHCC1‐5) from tumor‐bearing mice transplanted with PHHs from 5 different donors (PHH1~5) transduced with the combination of MYC, TP53^R249S , and KRAS^G12D overexpression lentiviruses, normal human liver tissue (Human liver), and primary HCC tissues from three different patients (pHCC1~3). Tumor cells are pointed with black arrows. Scale bar, 20 μm.

Figure EV3. iHCC samples express HCC markers.

1. Representative IHC images of four iHCC samples (iHCC1‐1, iHCC2‐1, iHCC3‐1, and iHCC4‐1) from NSIF mice transplanted with MTK‐transduced PHHs from four different PHH donors (PHH1~4) and an HCC tissue from a patient (pHCC2). Anti‐HLA‐ABC, anti‐ALB, anti‐AFP, anti‐GPC3 and anti‐KRT19 antibodies were used. Scale bars, 20 μm.
2. DNA sequencing confirmed mutations of TP53 in the genomic DNA of sgTP53 transduced PHH cells.
3. Expression of p53 was detected in the PHH cells transduced with sgMock or sgTP53 by Western blotting.
4. Representative images of in situ liver carcinomas derived from PHHs transduced with a combination of MTK with or without deletion of TP53 in NSIF mice. 3 out of 4 mice harbored tumor in both MTK+ sgMock and MTK+ sgTP53 groups after 11 weeks (n = 4 per group). N‐numbers refer to biological replicates.
5. Representative IHC images Ki67 staining of liver sections from NSIF mice transplanted with PHHs that were transduced with the combinations of MYC/TP53^R249S/KRAS^G12D (MTK) or MYC/TP53^R249S (MT) lentiviruses. Scale bars, 20 μm.

Source data are available online for this figure.

We next sought to determine the minimal oncogene profile required for the HCC transformation by transducing PHHs with two combinations of lentiviruses expressing MYC, TP53^R249S, KRAS^G12D, and transplanting these PHHs into NSIF mice. Tumors were detected in only 16.6% (2 out of 12 mice) of recipient mice transplanted with PHHs overexpressing MYC and TP53^R249S (MT group). Still, they were absent in both the TP53^R249S and KRAS^G12D (TK) combinational group and the MYC and KRAS^G12D (MK) combinational group (Fig 4A, and Table 1, and Appendix Table S3). Moreover, MTK‐transduced tumor samples contained more Ki67⁺ cells than MT‐transduced iHCC samples (Fig EV3E), suggesting that RAS signaling promotes the proliferation of tumor cells in iHCC. In culture, 1 million PHHs were transduced with different combinations of oncogenic lentiviruses leading on day 2, to 0.8 ± 0.1 million transduced PHHs (Fig 4D). MTK‐ and MT‐transduced PHHs started to thrive on day 6, while MYC‐, TP53^R249S‐, KRAS^G12D‐, MK‐, and TK‐transduced PHHs did not expand in culture (Fig 4D). Consistent with the in vivo results, PHHs transduced with the MTK oncogene combination proliferated faster than MT‐transduced cells (Fig 4D). Cell cycle analysis results showed that the MTK‐ and MT‐transduced PHH populations had significantly more proliferative cells (S‐phase) than the mock‐transduced PHH population (Fig 4E). Furthermore, MT‐ and MTK‐transduced PHHs showed a squamous morphology compared to the flattened, senescent morphology seen in cultures of other transduced PHHs (Fig 4F). Taken together, these results suggest that MYC and TP53^R249S are indispensable for inducing PHHs to transform into HCC cells and that RAS signaling further promotes this transformation. Thus, we established a humanized mouse model of iHCC via cotransfection of MYC, TP53^R249S, and KRAS^G12D for the following experiments.

iHCC recapitulates the transcriptional profiles and hallmarks of HCC

We next compared the global gene expression profiles of iHCC to those of corresponding parental PHHs using RNA‐sequencing analysis. Principal component analysis (PCA) showed that iHCC from various donors were clustered together and were separated from PHHs of their corresponding donors (Fig 5A). The analysis of relative transcript abundance of 17,667 genes revealed that hepatic transcription factors, including FOXA2, FOXA3, HNF4A, and ATF5, were highly expressed in PHHs, while tumor‐associated genes (MYC, TP53, and KRAS) were upregulated in iHCC cells (Fig 5B and Dataset EV1). Moreover, gene ontology (GO) analysis confirmed that the differentially expressed genes (DEGs) enriched in iHCC were associated mainly with the cell cycle, chromosome segregation, telomere maintenance, and cellular responses to DNA damage. In contrast, DEGs enriched in PHHs were associated with liver metabolic functions (Fig 5C). The heatmap shows that iHCC cells expressed both HCC/hepatocytes markers (Trerotoli et al, ²⁰⁰⁹) (AFP, GPC3, MYC, and ALB) and hepatic progenitor markers (EPCAM, KRT18, KRT19) (Fig 5D). Interestingly, genes related to epithelial‐to‐mesenchymal transition (EMT), including TWIST1, VIM, and SNAI1, were upregulated in iHCC cells compared to the corresponding PHHs (Fig 5D), consistent with the mesenchymal morphology of iHCC cells and epithelial morphology of PHHs (Fig 4F). Karyotype analysis revealed that compared to PHHs, iHCC cells had numerical aberrations, including loss and trisomy of chromosomes (Appendix Fig S2). Notably, copy number gains of chromosome 1, commonly identified in HCC samples from patients(Kusano et al, ¹⁹⁹⁹), were also detected in iHCC2‐1 and iHCC3‐1 samples (Appendix Fig S2). These results indicate that iHCC exhibit upregulation of oncogenes, downregulation of hepatic transcription factors, and genomic instability.

Figure 5. Genetically defined iHCCs recapitulate the transcriptional profiles of human HCC.

1. PHHs from five donors (PHH1 (AKB, female, 39 years old), PHH2 (XSM, female, 59 years old), PHH3 (ANG, male, 3 months old), PHH4 (HVN, male, 33 years old), and PHH5 (QBU, male, 50 years old)) and their corresponding iHCC cells were used for RNA‐seq analysis. Four iHCC samples (iHCC1‐1~iHCC1‐4) from the donor PHH1, three iHCC samples (iHCC2‐1~iHCC2‐3) from the donor PHH2, three iHCC samples (iHCC3‐1~iHCC3‐3) from the donor PHH3, one iHCC sample (iHCC4‐1) from the donor PHH4, and an iHCC sample (iHCC5‐1) from the donor PHH5 were used. The PCA plot shows samples plotted in two dimensions using their projections onto the first two principal components (PC1 and PC2). Each data point represents one sample (blue, PHHs; red, iHCC). PC1 was correlated with the type of sample (iHCCs versus PHHs).
2. Volcano plot of RNA‐seq data showing genes with differential expression (1,591 upregulated and 2,243 downregulated genes, fold change ≥ 2 and P value ≤ 0.05) in iHCC cells compared to PHHs. TP53, KRAS, MYC, MUC1, and AFP were highly expressed in iHCC cells, whereas hepatocyte signature markers, including FOXA2, FOXA3, ATF5, HNF4A, and ALB were highly expressed in PHHs.
3. Gene ontology (GO) analysis of the pathways enriched with DEGs (fold change ≥ 2 and P value ≤ 0.05) in iHCC cells and PHHs. Genes associated with telomere maintenance and the cell cycle were highly enriched in iHCC cells (top), while genes associated with biomacromolecule metabolisms were highly enriched in PHHs (bottom).
4. Heat map analysis of the log₂(RPKM) values of selected genes found to be highly expressed or downregulated in PHHs and iHCC cells. These genes were classified into three categories based on their functions.
5. UMAP visualization of iHCC and clinical HCC samples from the TCGA‐LIHC database based on their gene expression profiles.
6. UMAP visualization of iHCC and three subtypes of clinical HCC samples from the TCGA‐LIHC database (iClust 1–3, TCGA‐LIHC) based on their gene expression profiles.
7. Unsupervised clustering analysis of iHCC and clinical HCC samples in relative to normal PHHs. The distribution of molecular attributes (TP53 and KRAS mutation, MYC amplification) of iHCC, clinical HCC samples, and PHHs were displayed.

We then investigated whether iHCC recapitulate molecular profiles of HCC clinical samples. We compared the global transcription profiles of iHCC and PHHs based on our RNA‐seq analysis to those of HCC samples containing normal and mutated TP53 and amplified MYC and normal hepatocytes, which were available in TCGA database (TCGA‐LIHC, 374 HCC and 50 normal samples). Uniform Manifold Approximation and Projection (UMAP) analysis showed that iHCCs and HCC samples shared similar gene expression profiles and were different from those of PHHs from our samples and from the database (Fig 5E). In addition, HCC samples from the database were further classified into three major subtypes (iClust1‐3). iClust1 is considered more aggressive with robust proliferation and less differentiation than iClust2 and iClust3 (Cancer Genome Atlas Research Network. Electronic address wbe & Cancer Genome Atlas Research N, 2017). UMAP and dendrogram analysis show that iHCC samples resembled the iClust1 subtype in respect of transcriptional profiles (Fig 5F) and TP53/MYC/KRAS mutations (Fig 5G). In addition, GO analysis confirmed that the differentially expressed genes (DEGs) enriched in iHCCs and iClust 1 samples were associated mainly with cellular responses to DNA damage, telomere activity, and cell cycle, compared to that of iClust 2 and iClust 3 HCC (Fig EV4A). Though KRAS mutations have not been reported in HCC tumors that harbor p53 mutations or MYC amplification in the TCGA database, we have identified that 126 tumors with TP53 mutations, 140 tumors with mutations in the RTK/RAS/PI3K signaling pathways (KRAS, MET, PIK3CA, PTEN, NF1, and NRAS), and 197 tumors with the MYC/WNT pathways (CTNNB1, AXIN1, APC, and MYC) from 348 HCC tumors obtained from the TCGA database (Fig EV4B and Dataset EV2). Therefore, while KRAS mutations are not frequent, we believe that other modes of RTK activation can substitute for these mutations. In addition, 17 HCC samples contain MYC amplification and TP53 mutations, and 6 HCC samples harbor TP53 mutations and RAS expression upregulation. Furthermore, 35 HCC samples carry mutations in TP53, the RTK/RAS/PI3K and WNT/MYC pathways (Dataset EV2). Taken together, these analyses suggest this iHCC model represents an iClust1 subtype of HCC samples with similar transcription profiles and mutation profiles in TP53, RTK/RAS/PI3K and WNT/MYC pathways.

Figure EV4. Mutational signatures in liver hepatocellular carcinoma(TCGA‐LIHC).

1. Gene ontology (GO) analysis of the pathways enriched with DEGs (fold change ≥ 2 and P value≤ 0.05) in iHCC cells and clinical patient HCC (iClust 1, iClust 2 and iClust 3).
2. Alteration signatures of TP53 mutations, and mutations in the RTK/RAS/PI3K and WNT/MYC pathways in clinical HCC samples obtained from the TCGA‐LIHC database.

The iHCC model enables the identification of therapeutic targets for HCC

As iHCC derived from NSIF mice have not been edited by adaptive immunity, we used them to identify biomarkers of early HCC as potential immunotherapeutic targets. Based on RNA‐seq analysis of iHCC cells and PHHs (Fig 5B), we determined that 55 genes encoding cell surface markers were upregulated in iHCC cells compared to PHHs (Dataset EV3). We then selected the top 20 genes that were highly expressed in iHCC cells (Fig 6A). With GEPIA analysis, four out of the top 20 genes (SLC34A2, FBN2, FOLR1, and SLC39A10) were associated with poor prognosis as evidenced by their prognostic value in cohorts of primary HCC patients according to the TCGA datasets (Menyhart et al, ²⁰¹⁸) (TCGA‐LIHC) (Fig EV5A and Dataset EV3). We then focused on FAP and MUC1, as they are associated with HCC progression (Zou et al, ²⁰¹⁸; Kasprzak & Adamek, ²⁰¹⁹) and have been used as targets for CAR‐T cell therapy for cardiac fibrosis (Aghajanian et al, ²⁰¹⁹), breast cancer, and lung cancer (Wei et al, ²⁰¹⁷; Zhou et al, ²⁰¹⁹). FAP was highly expressed in 4 of the 5 donor‐derived iHCC, and MUC1 expression was detected in 3 of the 5 donor‐derived iHCC but not in normal livers (Fig EV5B and C). We then generated anti‐FAP and anti‐MUC1 CAR‐T (CARFAP‐T and CARMUC1‐T) cells by transducing human T cells with third‐generation CAR molecules incorporating CD28‐ and TLR2‐derived costimulatory domains (Lai et al, ²⁰¹⁸) (Fig EV5D and E). CARFAP‐T and CARMUC1‐T cells efficiently lysed iHCC2‐1 cells with high FAP and MUC1 expression in vitro, respectively (Fig 6B). To evaluate the antitumor effects of CARFAP‐T and CARMUC1‐T cells in vivo, we subcutaneously transplanted iHCC tumor tissues into NSI mice and intravenously transferred 5 × 10⁶ CAR‐T cells or mock T cells on day 10 post‐tumor transplantation. Tumors were efficiently suppressed in xenografts that were injected with CARFAP‐T and CARMUC1‐T cells but not in the mock T or PBS groups (Fig 6C).

Figure 6. Genetically defined iHCC as a platform for validation of potential therapeutic targets.

1. Heat map shows the top 20 genes encoding cell surface markers found to be highly expressed in iHCC samples compared to the corresponding PHHs. Within these 20 genes, seven genes highlighted in red were selected because their expression was correlated with poor prognosis of HCC patients based on TCGA‐HCC (TCGA‐LIHC) analysis.
2. Representative cytotoxicity CARFAP‐T and CARMUC1‐T cells assessed by coincubation with iHCC2‐1 cells. P = 0.0038 (CARFAP‐T versus mock T), and P < 0.0478 (CARMUC1‐T versus mock‐T). P values were determined by paired t‐test, **P ≤ 0.05. Cells were seeded at E: T ratios ranging from 4:1 to 1:4; data are presented as the mean ± SD. N‐numbers refer to biological replicates (n = 3).
3. Growth of iHCC in xenografts (Left) and tumor weights of iHCC dissected from xenografts (Right) (n = 5), the xenografts were treated with mock, CARFAP‐, or CARMUC1‐T cells when tumors reach 50 mm³ volume. ***P ≤ 0.001 (tumor volume of CARFAP‐/CARMUC1‐T versus mock T) with two‐way ANOVA with Tukey's multiple comparison test. **P ≤ 0.01 (tumor weight of CARFAP‐/CARMUC1‐T versus mock T) with one‐way ANOVA with multiple comparison test. Data are presented as the mean ± SD. N‐numbers refer to biological replicates.
4. Clonogenic assay of iHCC1‐1 cells treated with a MEK inhibitor (trametinib) and/or a CDK4/6 inhibitor (palbociclib); Representative images were shown based on three biological replicates.
5. Luciferase and EGFP marked iHCC1‐1 cells (1 × 10⁶) were intrasplenically injected into NSI mice on day 0. iHCC‐bearing mice were treated with DMSO (mock group, n = 4), or trametinib (3 mg/kg body weight) plus palbociclib (150 mg/kg body weight, n = 5) for 2 weeks (Daily oral gavage of drugs from Day 3 to Day 17). Top, on days 0, 7, 14, and 21, bioluminescence imaging was conducted. Bottom, quantification of the total flux analyzed by in vivo bioluminescence imaging of luciferase activity. P = 0.0057 on day 21 by two‐way ANOVA with Sidak's multiple comparison test, **P ≤ 0.01. Data are presented as the mean ± SD. N‐numbers refer to biological replicates.
6. Clonogenic assays (Top) of iHCC1‐1 cells treated with XL413. The data are the representative of three biological replications. (Bottom) Growth curves of iHCC in xenografts (n = 5) treated with PBS or XL413 at 50 mm³ tumor volume. P = 0.0486 (XL413 versus PBS group) by two‐way ANOVA with Sidak's multiple comparison test at Day 24, *P ≤ 0.05. Data are presented as the mean ± SD. N‐numbers refer to biological replicates.

Figure EV5. Cell surface markers associated with poor prognosis and the design of CAR molecules.

• A
  Kaplan‐Meier analysis of the TCGA‐HCC (TCGA‐LIHC) cohorts based on the expression levels of SLC34A2, FBN2, FOLR1, and SLC39A10 in the cohort samples (n = 364, high expression in red, low expression in black, the number of patients were indicated in figures, Statistical significance was determined using a log‐rank test).
• B, C
  Representative IHC staining of FAP and MUC1 in a normal liver (PHH), primary HCC (pHCC) and five MTK‐transduced iHCC tissues (iHCC1‐1, iHCC2‐1, iHCC3‐1, iHCC4‐1, and iHCC5‐1) that were derived from five different donors (PHH1~5). Scale bars, 20 μm.
• D
  Design of the pWPXLD‐CAR vector. pWPXLD‐CAR contains an anti‐FAP/MUC1 single‐chain fragment variation (scFv), CD28 transmembrane and endodomain, TLR2 endodomain, CD3ζ domain, and EGFP.
• E
  The transduction efficiency of pWPXLD‐CAR is shown. Data are presented as the mean ± SD. N‐numbers refer to biological replicates.

Recent studies have demonstrated that the combination of the MEK inhibitor trametinib (Tra) and the CDK4/6 inhibitor palbociclib (Pal) leads to efficient tumor control through the induction of retinoblastoma (RB) protein‐mediated cellular senescence in KRAS‐ and TP53‐ mutant cancers (Ruscetti et al, ²⁰¹⁸), revealing that the induction of senescence may constitute a strategy for the treatment of TP53 mutant liver cancers (Wang et al, ²⁰¹⁹) or pancreatic cancers(Ruscetti et al, ²⁰²⁰). We thus evaluated the efficacy of the combination treatment of trametinib and palbociclib against iHCC with KRAS and TP53 mutations. Palbociclib alone did not affect the growth of iHCC cells in vitro, whereas trametinib repressed iHCC cell expansion (Fig 6D). Of note, the combination of trametinib and palbociclib was more effective than trametinib alone for inducing senescence, as shown by the staining of SA‐β‐gal (Ruscetti et al, ²⁰²⁰), in iHCC cells (Appendix Fig S3A). In addition, the combination treatment with trametinib and palbociclib suppressed the growth of iHCC in the xenograft model (Fig 6E). We further evaluated the efficacy of XL413, a CDC7 kinase inhibitor that specifically induces senescence in HCC samples (Wang et al, ²⁰¹⁹) in the iHCC model. Of interest, XL413 inhibited the growth of iHCC in culture and in xenografts (Fig 6F) and induced senescence in iHCC (Appendix Fig S3B). Taken together, these results indicate that iHCC is a model for the identification of therapeutic targets and evaluation of anticancer drug efficacy for HCC.

Discussion

Genetically engineered mouse models are powerful tools to study PLC. Ectopic expression of Myc or Kras^G12D with p53 depletion induces mouse hepatocytes to transform into HCC or ICC (Klocke et al, ²⁰⁰¹; O'Dell et al, ²⁰¹²). However, mouse and human hepatocytes are different in numerous aspects (Odom et al, ²⁰⁰⁷). Therefore, humanized PLC models are in need for the identification of therapeutic targets and evaluation of antitumor drug efficacy. Though human liver cancer organoids have been established recently (Sun et al, ²⁰¹⁹), the organoids are derived from human‐induced hepatocytes (hiHeps) that are reprogrammed from human fibroblasts rather than PHHs (Huang et al, ²⁰¹⁴). Furthermore, organoid models lack a tumor microenvironment. Here, we first demonstrated that human PHHs can be transformed into iHCC cells in situ upon overexpression of MYC, TP53^R249S, and KRAS^G12D, providing us a model for investigating the initiation and progression of human HCC. iHCC mimics the histological architecture, tumor serum marker profile, and gene expression characteristics of human HCC. However, Ki‐67, a marker of proliferation (Sales‐Gil et al, ²⁰²¹), was highly expressed in iHCC, which is not common in primary human HCC, possibly indicating the aggressiveness of iHCC. To our knowledge, it is the first in vivo model of orthotopic transformation of human PHHs into HCC cells. Therefore, humanized HCC models are better than murine HCC models for the identification of therapeutic targets and evaluation of anti‐tumor drug efficacy. Of course, iHCC models have some disadvantages compared to murine spontaneous HCC models, such as the lack of an immune system.

Based on previous reports (Schulze et al, ²⁰¹⁵; Bailey et al, ²⁰¹⁸; Llovet et al, ²⁰¹⁸), we selected ten candidates for overexpression in PHHs and identified enrichment of MYC, TP53^R249S, and KRAS^G12D in tumors harvested from the livers of xenografted mice. Since individual oncogenes such as MYC and RAS fails to induce the transformation of PHHs (Sun et al, ²⁰¹⁹), we did not transduce individual oncogenes into PHHs in this study. However, it is worthy of performing a genetic screen for various combinations of these oncogenes that can transform PHHs into iHCC in the future. TP53^R249S, frequently detected in human HCC related to hepatitis B infection and aflatoxin B1, gains its function via MYC activation upon CDK4 phosphorylation at Serine 249 and consequent PIN1 binding (Hussain et al, ²⁰⁰⁷; Liao et al, ²⁰¹⁷). Though the loss of heterozygosity of TP53 in HCC patient samples was not uncommon (Hsu et al, ¹⁹⁹⁴; Zhu et al, ²⁰⁰⁴), mutant TP53 was able to inhibit the activity of WT TP53 (Kern et al, ¹⁹⁹²). The mechanistic basis for the cooperation among these oncoproteins has been elucidated. Mutant TP53 enhances KRAS^G12D activity through modulating RNA splicing in pancreatic ductal adenocarcinoma (Escobar‐Hoyos et al, ²⁰²⁰). Mutant KRAS and endogenous wild‐type p53 were shown to induce cellular senescence that limits liver cancer development in pre‐malignant hepatocytes, depending on an intact CD4⁺ T cell‐mediated adaptive immune response (Kang et al, ²⁰¹¹; Mehta et al, ²⁰²¹), which is absent in our iHCC model. Although the combination of MYC, TP53^R249S, and KRAS^G12D was uncommon in human HCC, iHCC cells derived from MTK‐transduced PHHs can be used to investigate further crosstalk among MYC, TP53^R249S, and KRAS^G12D in human HCC. In addition, new mechanisms underlying how TP53^R249S, MYC, and KRAS^G12D cooperatively transform primary human hepatocytes (PHHs) into HCC need to be addressed in future studies.

Whole‐genome sequencing has identified numerous genetic mutations and greatly enforced our understanding of human HCC (Cancer Genome Atlas Research Network. Electronic address wbe & Cancer Genome Atlas Research N, 2017). However, most of these mutations have not been characterized as to whether they can drive the initiation of HCC. We identified MYC, TP53^R249S, and KRAS^G12D out of the ten oncogenes with the capacity of inducing human PHHs into HCC in vivo. Due to robust lentiviral transduction efficiency in human PHHs, our iHCC humanized mouse model constitutes a platform for performing high‐throughput genetic screening to distinguish the driver mutations detected in patient samples by whole‐genome sequencing that contribute to the transformation of PHHs to HCC cells in the numerous genetic lesions. Nevertheless, reconstitution protocol needs to be further optimized for improving in vivo repopulation of PHHs (Azuma et al, ²⁰⁰⁷). In addition to gain‐of‐function, we can also perform loss‐of‐function genetic screening using the CRISPR/Cas9 system to identify new tumor suppressor genes in HCC using our model in future research (Michels et al, ²⁰²⁰).

Most patients with HCC present with advanced disease, and treatment options are limited (Llovet et al, ²⁰⁰³). CAR‐T/NK cell therapy is a promising immunotherapeutic strategy for treating multiple refractory blood malignancies. Nevertheless, further advances are required for treating solid tumors with CAR‐T/NK cells (Schnalzger et al, ²⁰¹⁹). One challenge is to identify safe and effective tumor antigens in solid tumors. There are four antigens as CAR‐T targets for treating HCC with limited success: glypican‐3 (GPC3)(Jiang et al, ²⁰¹⁶; Shi et al, ²⁰²⁰) (NCT03198546, NCT02395250, NCT03146234), α‐fetoprotein (AFP) (Liu et al, ²⁰¹⁷) (NCT 03349255), CD147 (Tseng et al, ²⁰²⁰), and CD133 (Dai et al, ²⁰²⁰) (NCT02541370). It is thus urgent to detect biomarkers of early‐stage HCC. We found that iHCC in our humanized mice expressed MUC1 and FAP, positively correlated with poor survival prognosis (Zou et al, ²⁰¹⁸; Kasprzak & Adamek, ²⁰¹⁹). MUC1 and FAP have been identified as CAR‐T targets for treating non‐small cell lung cancer (Wei et al, ²⁰¹⁷) (NCT03198052) and cardiac fibrosis respectively, but not for HCC treatment. Therefore, our model may serve as a tool for identification of biomarkers and targets of early‐stage HCC.

Materials and Methods

Generation of NSIF mice

All experimental protocols were performed in accordance with the institutional guidelines of the China Council on Animal Care and approved by the Ethics Committee of Animal Experiments at Guangzhou Institutes of Biomedicine and Health, Chinese Academy of Sciences (GIBH). All mice were bred and maintained in specific pathogen‐free (SPF)‐grade cages and provided autoclaved food and water. NSI and ICR mouse strains were used as embryos and foster mothers, respectively. Female NSI mice (8–10 weeks of age) were superovulated by an intraperitoneal injection of 5 IU of PMSG (Sigma) and 5 IU of hCG (C1063, Sigma) at 48‐h intervals. The superovulated female mice were mated with male NSI mice, and the fertilized embryos were collected from the oviducts. TALEN mRNAs were injected into the cytoplasm of fertilized eggs displaying recognizable pronuclei in M2 medium using a piezo‐driven micromanipulator. A 25 ng/μl sample of Fah‐TALEN mRNAs was injected. After a 24‐h incubation at 37°C, the surviving embryos were selected and transferred into the oviducts of pseudopregnant foster mothers. Editing of the Fah gene was analyzed in the resulting offspring.

Recombinant vectors and viral packaging

The cDNA of each oncogene (MYC, TP53^R249S, KRAS^G12D, NRAS^G12D, CTNNB1^S45F, BRAF^V600E, AXIN1^G652S, IL6, PIK3CA^E542K , and CSF1R^Y969C ) or third‐generation anti‐FAP and anti‐MUC1 CAR vectors incorporating the CD28 and TLR2 costimulatory endodomains were cloned into the second‐generation lentiviral vector pWPXLD. sgRNA targeting TP53 was cloned into LentiCRISPR v2 (addgene 52961) vector (Appendix Table S4). Lentiviral particles were produced in HEK‐293T cells following polyethyleneimine (PEI) (49553‐93‐7, Polysciences, Inc., USA)‐mediated transfection with the pWPXLD‐based transfer plasmid and the packaging plasmids psPAX2 and pMD2.G. Lentivirus‐containing supernatant was harvested at 48 and 72 h post‐transfection and filtered through a 0.45 µm filter.

Cells and culture conditions

PHH were cultured in DMEM/F‐12 medium with 10% heat‐inactivated FBS (Gibco, Grand Island, NY, USA), 10 mM HEPES, 2 mM glutamine (Gibco, Grand Island, NY, USA), and 1% penicillin/streptomycin (Gibco, Grand Island, NY, USA) supplemented with hHGF (10 ng/ml, Cat#100‐39, Peprotech, NJ, USA), hEGF (5 ng/ml, Cat#AF‐100‐15, Peprotech), and WNT3a (5 ng/ml, Cat#315‐20). iHCC were cultured in DMEM/F‐12 (Gibco, Grand Island, NY, USA), supplemented with 10% heat‐inactivated FBS (Gibco, Grand Island, NY, USA), 10 mM HEPES, 2 mM glutamine (Gibco, Grand Island, NY, USA) and 1% penicillin/streptomycin (Gibco, Grand Island, NY, USA). All cells were cultured at 37°C in an atmosphere of 5% carbon dioxide.

Transduction of hepatocytes and tracking of hepatocyte engraftment

Primary human hepatocytes from five different donors (PHH1 (AKB, female, 39 years old), PHH2 (XSM, female, 59 years old), PHH3 (ANG, male, 3 months old), PHH4 (HVN, male, 33 years old) and PHH5 (QBU, male, 50 years old)) were purchased from Bioreclamation IVT (Baltimore, MD, USA). These cells were transduced with a lentivirus which expresses an oncogene and EGFP or luciferase as indicated in the experiment. Twenty‐4 h after transduction, hepatocytes were cultured in DMEM/F‐12 medium supplemented with hHGF (10 ng/ml, Cat#100‐39, Peprotech), hEGF (5 ng/ml, Cat#AF‐100‐15, Peprotech), and WNT3a (5 ng/ml, Cat#315‐20, Peprotech). The transduction efficiency was determined by fluorescence‐activated cell sorting (FACS). EGFP‐ and luciferase‐positive hepatocytes (0.5 × 10⁶ cells/mouse) were transplanted into the spleens of NSIF mice. Hepatocyte engraftment was evaluated using a cooled charge‐coupled device (CCD) camera system (IVIS 100 Series Imaging System, Xenogen, Alameda, CA, USA). Mice were intraperitoneally injected with 75 mg/kg D‐luciferin firefly potassium salt and imaged 5 min later with auto‐exposure time. The total and average emissions were quantified using Living Image software (Xenogen, Alameda, CA, USA). For oncogene screening and validation, viral titers in the mixture were adjusted to the same multiplicity of infection (MOI = 5.0) to obtain a transduction rate of > 90% on day 2 after infection and 0.5 million GFP‐positive transduced hepatocytes were transplanted into NSIF mice via intrasplenic injection. For the MEK inhibitor (Trametinib, T2125, Targetmol) and a CDK4/6 inhibitor (Palbociclib, T1785, Targetmol) treatment assay, MYC, TP53^R249S, and KRAS^G12D cells that were further transduced with the EGFP and luciferase lentiviral expression vectors (2 × 10⁶ cells per mouse) were injected intrasplenically into 6‐ to 8‐week‐old immunodeficient NSI mice. Mice were randomly assigned to treatment with vehicle or trametinib (3 mg/kg body weight, daily gavage)/palbociclib (150 mg/kg body weight, daily gavage) 5 days per week for 2 weeks. On days 0, 7, 14, and 21, bioluminescence imaging was conducted. For XL413 (S7547, Selleck) treatment, iHCC cells were subcutaneously injected into NSI mice. Xenografted mice were treated with PBS or XL413 (50mg/kg for 2 weeks) at 50 mm³ tumor volume. The tumor volume was calculated as (length × width²)/2.

Transplantation of human hepatocytes

All animal experiments were performed in accordance with institutional regulations. Primary human liver cells (0.5–1 million) were transplanted into the spleens of NSIF mice after the withdrawal of NTBC‐containing drinking water (Azuma et al, ²⁰⁰⁷). The NTBC concentration was gradually decreased (3.5 mg/l, days 0 to 2; 1.75 mg/l, days 3 and 4; and 0.8 mg/l, days 5 and 6), and NTBC was then completely withdrawn 1 week after transplantation. NTBC was readministered for 5 days, 3–4 weeks later when a transplanted mouse had lost > 20% of its pretransplant weight. The animals' body weights were monitored twice weekly after transplantation. The surviving recipient mice were euthanized for collection of blood and liver samples.

Measurement of human ALB, AFP, APOA1, and glycogen

Small amounts of blood were collected weekly from the left saphenous vein with a heparinized blood capillary tube. The human ALB concentration was measured with a Human Albumin ELISA Quantitation Kit (E80‐129, Bethyl Laboratories). The AFP concentration was measured with an AFP kit (EHAFP, Invitrogen, Thermo Fisher) according to the corresponding manufacturer's protocol. To determine the levels of stored glycogen, 1 × 10⁶ cells were collected. Human albumin was measured using the Human albumin ELISA quantitation set. Glycogen storage was measured with a Glycogen assay kit (MAK016, Sigma‐Aldrich) and human APOA1 was measured using a APOA1 detection kit (SEKH00903, Solarbio) according to the manufacturers' instructions.

Quantitative real‐time PCR

mRNA was extracted from cells with TRIzol reagent (15596026, Thermo Fisher) and reverse transcribed into cDNA using the PrimeScriptTM RT reagent Kit (Takara, Kasatsu, Japan). All reactions were performed with TransStart Tip Green qPCR SuperMix (TransGene, Beijing, China) on a Bio‐Rad CFX96 real‐time PCR machine (Bio‐Rad, Hercules, CA), using the primers shown in Appendix Table S4. Delta CT calculations were relative to β‐actin and corrected for PCR efficiencies.

Histological and immunofluorescence assays

Tissue samples were fixed with 4% formalin, embedded in paraffin, sectioned at a thickness of 4 μm, and stained with H&E. Images were acquired with a microscope (Leica DMI6000B, Leica Microsystems, Wetzlar, Germany). Paraffin sections were used for immunofluorescence (IF) and immunohistochemical (IHC) staining. A primary antibody (1:100–1:2,000) was added to the tissue sections on slides and incubated overnight at 4°C. Alexa Fluor 647‐conjugated goat anti‐mouse (1:2,000) and Alexa Fluor 568/488‐conjugated goat anti‐rabbit (1:2,000) antibodies were added and incubated for 1 h at room temperature. DAPI was used to stain nuclei, and the prepared slides were analyzed using a Zeiss LSM800 microscope. Anti‐HLA Class I ABC antibody (ab70328), Anti‐Cytokeratin 19 antibody (ab76539), Anti‐c‐Myc antibody (ab32072), Anti‐p53 antibody (ab32389), Anti‐KRAS antibody (ab275876), Anti‐EpCAM antibody (ab223582), Anti‐AFP antibody (ab169552), Anti‐GPC3 antibody (ab95363), Anti‐MUC1 antibody (ab109185), and anti‐FAP antibody (ab207178) were purchased from Abcam.

RNA sequencing

mRNA from primary PHH or iHCC was prepared according to the TruSeqTM RNA Sample Preparation Guide, and sequencing was performed on BGISEQ‐500 (BGI, Wuhan, China). Sequencing reads were mapped to the human RefSeq‐RNA reference sequence using the FANSe2 algorithm. Reads mapped with tophat2 were associated with genes using the custom Perl scripts that allowed no more than 2 unmapped bases. Cufflinks (version 2.1.1) was used to identify reads consistent with the annotated genes. These genes were quantified using the reads per kilobase million (RPKB) method. For small genes (< 200 bps), a minimum of 10 mappable reads were required. Mappable reads were imported into the DEGseq software package to calculate up‐ or downregulation of genes.

Integrative clustering of iHCC and clinical HCC using iCluster

We used publicly available data provided by the TCGA Research Network (http://xena.ucsc.edu) to perform hierarchical clustering and GSEA analyses. The RNA sequencing data of TCGA‐HCC (374 HCC samples and 50 normal samples) were downloaded from TCGA. To understand the subgroups formed by integrating various molecular platforms of iHCC, we utilized iCluster, which formulates the problem of subgroup discovery as a joint multivariate regression of multiple data types with reference to a set of common latent variables that represent the underlying tumor subtypes (Mo et al, ²⁰¹³; Cancer Genome Atlas Research Network. Electronic address wbe & Cancer Genome Atlas Research N, ²⁰¹⁷).

DNA extraction and sequence analysis

DNA was extracted from samples using a HiPure Tissue DNA Mini Kit (D3121, Magen, China). In brief, samples of approximately 10 mg were digested in 250 µl of ATL buffer supplemented with proteinase K overnight at 55°C and were then treated with RNase for 30 min at 37°C. Genomic DNA was subjected to PCR amplification, and mutations were identified by direct sequencing. To avoid amplification of endogenous genes, we ensured that the PCR primer pairs flanked the introns in the 10 oncogenes (Appendix Table S4).

Western blot assays

Western blotting was performed following a routine protocol. The hiHep organoids were lysed in RIPA lysis buffer (supplemented with NaF (1:100), Na3VO4 (1:200), P8340 (1:100)) and an equal amount of total protein lysates were separated by 6–15% SDS–PAGE and transferred to a PVDF membrane (Millipore). The membrane was blocked with 5% (w/v) reagent‐grade non‐fat milk and incubated overnight with primary antibodies (MYC (5605, CST), KRAS^G12D (14429, CST,), p53(48818, CST)) at 4°C followed by secondary antibody incubation for 1 h. The protein bands were visualized using Luminata Western ECL substrate (Sigma).

Long‐term cell proliferation assays (colony formation assays)

Cells were seeded into 6‐well plates (1 × 10⁴ cells per well) and cultured with/without drugs as indicated for 10–14 days. The medium was changed twice weekly. Cells were fixed with 4% paraformaldehyde and stained with 0.1% crystal violet (in water) (Lin et al, ²⁰²⁰).

SA‐β‐gal staining

SA‐β‐gal activity was assessed using a Senescence β‐Galactosidase Staining Kit (C0602, Beyotime, China) according to the manufacturer's instructions. iHCC cells were plated in a 6‐well plate (1 × 10⁵ cells per well) and treated with a MEK inhibitor (trametinib, T2125, Targetmol), a CDK4/6 inhibitor (palbociclib, T1785, Targetmol), and XL413 as indicated. SA‐β‐gal‐positive cells (senescent cells) were identified as blue‐stained cells under standard light microscopy.

Cell cycle analysis

PHHs transduced with the MYC/TP53^R249S/KRAS^G12D combination were harvested by trypsinization on day 10. Cells were washed two times with PBS, fixed at −20°C in 70% ethanol for 12 h and stained in 300 μl of propidium iodide (final concentration of 50 mg/ml) and 0.1% Triton X‐100 at 37°C for 15 min. The distribution of cells in different phases of the cell cycle was analyzed by flow cytometry (Agilent NovoCyte, USA).

Isolation, transduction, and expansion of primary human T lymphocytes

Peripheral blood mononuclear cells (PBMCs) were isolated from healthy adult donors using Lymphoprep (Stem Cell Technologies, Vancouver, Canada). T cells were negatively selected from PBMCs with a MACS Pan T Cell Isolation Kit (Miltenyi Biotec, Bergisch Gladbach, Germany) and activated using microbeads coated with anti‐human CD3, anti‐human CD2, and anti‐human CD28 antibodies (Miltenyi Biotec, Bergisch Gladbach, Germany) at a bead: cell ratio of 1:2 and a density of 2.5 × 10⁶ cells/ml for 2 days in RPMI‐1640 medium supplemented with 10% heat‐inactivated fetal bovine serum (FBS), 100 IU/ml recombinant human IL‐2, 10 mM HEPES, 2 mM glutamine and 1% penicillin/streptomycin. On day 2 post‐activation, T cells were transfected with CAR vector lentiviral supernatants in the presence of 8 μg/ml polybrene at an MOI of 2.0 (Sigma‐Aldrich, St. Louis, USA). Twelve hours after transfection, T cells were cultured in fresh medium containing IL‐2 (300 U/ml); subsequently, fresh medium was added every 3 days to maintain cell density within the range of 0.5–1 × 10⁶ cells/ml. Healthy PBMC donors provided informed consent to use their samples for research purposes, and all procedures were approved by the Research Ethics Board of GIBH.

Xenograft models and in vivo assessment

Animal experiments were performed in the Laboratory Animal Center of GIBH, and all animal procedures approved by the Animal Welfare Committee of GIBH. All protocols were approved by the relevant Institutional Animal Care and Use Committee (IACUC). NSI mice were maintained in specific pathogen‐free (SPF)‐grade cages and provided autoclaved food and water. Mice were randomized into experimental groups of ≥ 5. Direct injection of indicated tumor cells in 200 μl PBS was performed to establish subcutaneous (flank) tumors. At the indicated time for each experiment, 5 × 10⁶ transduced human T (GFP⁺ or CAR⁺) cells in 200 μl PBS were adoptively transferred to tumor‐bearing mice systemically by tail vein injection. Peripheral blood was obtained by retro‐orbital bleeding. Tumors were measured every 3 days with a caliper. Tumor volume was calculated using the following equation: (length × width²)/2.

Statistical analysis

Kaplan–Meier's analysis of the correlation between candidate genes expression and overall survival was acquired from the GEPIA analysis tool (http://gepia.cancer‐pku.cn/index.html), which was based on datasets from The Cancer Genome Atlas (TCGA). Data are presented as the mean ± SD/SEM values, and significance was evaluated by Student's t‐test or ANOVA as indicated. Differences with P < 0.05 were considered statistically significant.

Author contributions

Zhiwu Jiang: Conceptualization; Resources; Formal analysis; Funding acquisition; Validation; Investigation; Visualization; Methodology; Writing—original draft; Writing—review & editing. Lin Cheng: Conceptualization; Formal analysis; Investigation; Methodology. Zhiping Wu: Data curation; Investigation; Methodology. Linfu Zhou: Resources; Formal analysis; Methodology. Haitao Wang: Software; Formal analysis; Methodology. Qilan Hong: Software; Formal analysis; Methodology. Qiting Wu: Resources; Methodology. Youguo Long: Resources; Methodology. Yunlin Huang: Investigation; Methodology. Gaoqi Xu: Methodology. Yao Yao: Resources; Supervision; Funding acquisition; Validation; Methodology. Zhaoyang Tang: Resources; Validation; Investigation; Methodology. Zhenfeng Zhang: Resources; Formal analysis; Methodology; Project administration. Lili Yang: Resources; Validation; Methodology. Wei Luo: Resources; Methodology. Jie Yang: Resources; Investigation; Methodology. Likun Gong: Resources; Investigation; Methodology. Pentao Liu: Resources; Supervision; Validation; Investigation; Methodology. Xinwen Chen: Resources; Validation; Methodology. ShuZhong Cui: Resources; Validation; Investigation. Qi Zhang: Resources; Validation; Methodology. Yinxiong Li: Conceptualization; Resources; Validation; Investigation; Visualization; Methodology; Writing—original draft; Writing—review & editing. Peng Li: Conceptualization; Resources; Formal analysis; Supervision; Funding acquisition; Validation; Investigation; Visualization; Methodology; Writing—original draft; Project administration; Writing—review & editing.

In addition to the CRediT author contributions listed above, the contributions in detail are:

Conceptualization, ZJ and PLi; Methodology, ZJ, LC, ZW, LZ, HW, QH, GX, and PLi; Investigation, ZJ, LC, ZW and PLi; Validation, YH, GX, YY, and LG; Formal Analysis, ZJ, and LC; Resources, QW, LZ, YLo, ZT, ZZ, WL, JY, SC and QZ; Writing—Original Draft, ZJ and PLi; Writing—Review and Editing, YLi, PLiu, ZJ, LY, and PLi; Supervision, YLi, PLiu, and XC; Project Administration, YLi. PLi; Funding Acquisition, ZJ, YY, JY, YL, and PLi.

Disclosure and competing interests statement

The authors declare that they have no conflict of interest.

Supporting information

Appendix

Expanded View Figures PDF

Dataset EV1

Dataset EV2

Dataset EV3

Source Data for Expanded View

Source Data for Figure 1C

Source Data for Figure 3I

EMBO reports (2022) 23: e54275.

Data availability

RNA‐Seq data: Gene Expression Omnibus GSE168852 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE168852).