Transposon mutagenesis identifies genes and cellular processes driving epithelial-mesenchymal transition in hepatocellular carcinoma Takahiro Kodamaa, Justin Y. Newberga, Michiko Kodamaa, Roberto Rangela, Kosuke Yoshiharab, Jean C. Tienc, Pamela H. Parsonsd, Hao Wud, Milton J. Finegoldd, Neal G. Copelanda, and Nancy A. Jenkinsa,1 a Cancer Research Program, Houston Methodist Research Institute, Houston, TX 77030; bDepartment of Obstetrics and Gynecology, Niigata University Graduate School of Medical and Dental Sciences, Niigata 9518510, Japan; cMichigan Center for Translational Pathology, Department of Pathology, University of Michigan, Ann Arbor, MI 48109; and dDepartment of Pathology and Immunology, Baylor College of Mecidine and Texas Children’s Hospital, Houston, TX 77030

Contributed by Nancy A. Jenkins, April 30, 2016 (sent for review March 23, 2016; reviewed by Zoltán Ivics and Roland Rad)

Epithelial-mesenchymal transition (EMT) is thought to contribute to metastasis and chemoresistance in patients with hepatocellular carcinoma (HCC), leading to their poor prognosis. The genes driving EMT in HCC are not yet fully understood, however. Here, we show that mobilization of Sleeping Beauty (SB) transposons in immortalized mouse hepatoblasts induces mesenchymal liver tumors on transplantation to nude mice. These tumors show significant down-regulation of epithelial markers, along with up-regulation of mesenchymal markers and EMT-related transcription factors (EMTTFs). Sequencing of transposon insertion sites from tumors identified 233 candidate cancer genes (CCGs) that were enriched for genes and cellular processes driving EMT. Subsequent trunk driver analysis identified 23 CCGs that are predicted to function early in tumorigenesis and whose mutation or alteration in patients with HCC is correlated with poor patient survival. Validation of the top trunk drivers identified in the screen, including MET (MET proto-oncogene, receptor tyrosine kinase), GRB2-associated binding protein 1 (GAB1), HECT, UBA, and WWE domain containing 1 (HUWE1), lysine-specific demethylase 6A (KDM6A), and protein-tyrosine phosphatase, nonreceptor-type 12 (PTPN12), showed that deregulation of these genes activates an EMT program in human HCC cells that enhances tumor cell migration. Finally, deregulation of these genes in human HCC was found to confer sorafenib resistance through apoptotic tolerance and reduced proliferation, consistent with recent studies showing that EMT contributes to the chemoresistance of tumor cells. Our unique cell-based transposon mutagenesis screen appears to be an excellent resource for discovering genes involved in EMT in human HCC and potentially for identifying new drug targets.

metastasis (10, 11). Two recent publications have challenged this dogma for breast and pancreatic cancers and have suggested instead that EMT confers chemoresistance rather than metastatic potential to tumor cells (12, 13). In the case of HCC, growing evidence suggests that EMT contributes to metastasis and poorer patient prognosis (14–17). EMT is also positively correlated with resistance to sorafenib, cisplatin, and doxorubicin (18–20). Interestingly, sorafenib has been shown to have an inhibitory effect on the migration of HCC cells through inactivation of the EMT program (21), which is one of the potential mechanisms of its antineoplastic action in HCC. Collectively, the foregoing studies show that EMT has a significant negative impact on survival in patients with HCC through the induction of metastasis and drug resistance. To understand the mechanism(s) regulating EMT in HCC, it will be important to discover all EMT genes to identify therapeutic targets and improve the high mortality of patients with HCC. Insertional mutagenesis is a powerful in vivo screening tool for cancer gene discovery. Our laboratory has successfully modeled more than 20 different types of human cancers in mice using Sleeping Beauty (SB) transposon mutagenesis and has identified hundreds of candidate cancer genes (CCGs), which show striking overlaps with genes mutated or deregulated in human cancers (22– 24). We also have developed in vitro cell-based SB mutagenesis screens that have made it possible to identify genes involved in the malignant transformation of neural stem cells into glioma-initiating

| EMT | HCC | Met | ubiquitin-mediated proteolysis

Epithelial-mesenchymal transition (EMT) contributes to metastasis and chemoresistance in patients with hepatocellular carcinoma (HCC), but the genes driving EMT are poorly understood. Here, we describe a transposon mutagenesis screen that made it possible to identify 233 candidate cancer genes (CCGs) that are enriched for genes driving EMT in HCC. Twenty-three CCGs are predicted to function early in tumorigenesis, and alterations in these genes are associated with poor HCC patient survival. Validation studies showed that deregulation of the most highly mutated CCGs activates an EMT program that enhances HCC cell migration and also confers sorafenib resistance. Thus, transposon mutagenesis appears to provide an excellent resource for identifying genes regulating EMT in human HCC and for potentially identifying new drug targets for HCC.

transposon

Significance

H

CC is the sixth most common cancer and the third-leading cause of cancer-related deaths worldwide (1). In the United States, the incidence of HCC is increasing, and overall 5-y survival is now 5 mm

cells (25). This versatility and high relevance to human cancers makes SB an invaluable tool for cancer gene identification in mice. Here, we report that mobilization of SB transposons in immortalized mouse hepatoblasts induces mesenchymal liver tumors on transplantation into nude mice. Cloning and sequencing of the transposon insertions sites from these tumors enabled us to identify 233 CCGs that are enriched for genes driving EMT in human HCC. Twenty-three of these CCGs are predicted to function early in tumorigenesis, and alterations in these genes are associated with poor survival in patients with HCC. Validation studies showed that deregulation of the most highly mutated CCGs activates an EMT program that enhances HCC cell migration and also confers sorafenib resistance. Thus, this unique cell-based transposon mutagenesis screen appears to be an excellent resource for discovering genes involved in EMT in human HCC and for potentially identifying new drug targets for HCC.

Fig. 1. Sleeping Beauty transposition in immortalized hepatoblasts results in the formation of mesenchymal tumors following transplantation. (A) Excision PCR analysis of tail DNA from a T2Onc2/+; Rosa26-lsl-SB11/+ mouse (Tail/SB) and three immortalized hepatoblast cell lines derived from Alb-Cre/+; T2Onc2/+;Rosa26-lsl-SB11/+ mice (IHBC/SB). The 2,200-bp band in tail DNA denotes unmobilized transposons, whereas the 225-bp band in the three hepatoblast lines represents mobilized transposons. (B–D) Afp (B), Dlk1 (C), and Epcam (D) mRNA expression in three IHBC/SB lines and one C57BL6/J adult mouse liver was quantitated by qPCR. (E–J) To induce hepatocyte differentiation, an IHBC/SB line was cultured with 20 ng/mL HGF, 20 ng/mL oncostatin M, and 10−6 M dexamethazone for 14 d. mRNA expression levels of mature hepatocyte markers Alb (E), Apob (F), Aldob (G), and Glul (H), and two hepatoblast markers, Dlk1 (I) and Epcam (J), were then quantitated by qPCR. (K) The rate of tumor formation after s.c. injection of seven IHBC/SB and six IHBC/WT lines into the flanks of nude mice. Each line was injected into four flanks each. *P < 0.05, log-rank test. (L–Q) Histological analysis of tumors derived from IHBC/SB lines: H&E (L), Ki-67 (M), Vimentin (N), Pan-CK (O), Epcam (P), and SB transposase (Q). (Original magnification, 200×; scale bar: 100 μm.)

PNAS | Published online May 31, 2016 | E3385

chymal tumors (Fig. 1 L–N). However, immunohistochemical analysis also showed that these tumors also expressed the epithelial markers pan-cytokeratin and Epcam, as well as SB transposase (Fig. 1 O–Q). These findings support the hypothesis that these tumors originate from transposase-active hepatoblasts, but acquire mesenchymal characteristics during their growth in vivo. To confirm these

diameter) were obtained from five of the seven IHBC/SB lines between 47 and 80 d after transplantation (Fig. 1K). Thus, transposon mutagenesis appears to be required for the malignant transformation of IHBC cells. Histological analysis showed that tumors were composed of highly proliferative, spindle-shaped cells with vimentin (Vim) immunopositivity, typical of mesen-

A

Ahnak Akt1 Bmp1 Bmp7 Cald1 Camk2n1 Cav2 Cdh1 Cdh2 Col1a2 Col3a1 Col5a2 Ctnnb1 Dsc2 Dsp Egfr Erbb3 Esr1 F11r Fgfbp1 Fn1 Foxc2 Fzd7 Gng11 Gsc Gsk3b Igfbp4 Il1rn Ilk Itga5 Itgav Itgb1 Jag1 Krt14 Krt19 Krt7 Mitf Mmp2 Mmp3 Mmp9 Msn Mst1r Mtap1b Nodal Notch1 Nudt13 Ocln Pdgfrb Plek2 Pppde2 Ptk2 Ptp4a1 Rac1 Rgs2 Serpine1 Sip1 Smad2 Snai1 Snai2 Snai3 Sox10 Sparc Spp1 Stat3 Steap1 Tcf4 Tcf7l1 Tfpi2 Tgfb1 Tgfb2 Tgfb3 Timp1 Tmeff1 Tmem132a Tspan13 Twist1 Vcan Vim Vps13a Wnt11 Wnt5a Wnt5b Zeb1 Zeb2

IHBC/SB-3 IHBC/SB-1 IHBC/SB-2 IHBC/WT AML12 IHBC/SB-4 IHBC/SB-5 IHBC/SB-2 tumor IHBC/SB-1 tumor IHBC/SB-3 tumor IHBC/SB-4 tumor

*

*

*

*

*

I *

H

*

*

*

*

G *

*

*

F *

E

D *

C

B

Fig. 2. An EMT program is active in IHBC/SB-derived tumors. (A) Heat map and hierarchical clustering depicting the expression of 84 EMT-related genes in five IHBC/SB lines, one IHBC/WT line, the mouse immortalized hepatocyte cell line (AML12), and four IHBC/SB-derived tumors. (B–I) mRNA expression levels of EMT markers and EMT-TFs quantitated by qPCR in five IHBC/SB lines, 45 IHBC/SB-derived tumors, and five well-differentiated HCCs derived from SB mice: Col1a2 (B), Vim (C), Fn1 (D), Cdh1 (E), Snai1 (F), Twist1 (G), Zeb1 (H), and Zeb2 (I). *P < 0.05, Kruskal–Wallis test.

E3386 | www.pnas.org/cgi/doi/10.1073/pnas.1606876113

Kodama et al.

Identification of CCGs Driving the Development of Mesenchymal Tumors. To identify CCGs driving mesenchymal tumor develop-

ment, we PCR-amplified and sequenced the transposon insertion sites from seven IHBC/SB lines and all 52 IHBC/SB-derived tumors. To obtain the maximum number of sequencing reads and avoid PCR bias (29), we used Illumina sequencing and acoustic shearing of tumor DNA (Materials and Methods). Using this modified method, we identified 871,231 and 6,541,525 mapped reads, corresponding to 50,215 and 118,530 nonredundant transposon insertion sites from IHBC/SB lines and IHBC/SBderived tumors, respectively (Datasets S1 and S2). Using the gene-centric common integration site (CIS) calling method (gCIS) (30), we identified 61 CCGs from IHBC/SB lines and 233 CCGs from IHBC/SB-derived tumors (P < 0.05, χ2 test followed by Bonferroni correction) (Tables S1 and S2). CISs are genomic regions that contain more transposon insertions than predicted by chance and thus are likely to mark the location of CCGs (22). Only one CCG overlapped between these two datasets (Fig. 3A), indicating that tumors have acquired new insertional mutations important for their development during growth in vivo. To determine whether CCGs identified in tumors are significantly enriched for human cancer genes, we compared our list of CCGs to the list of 570 known human cancer genes in the Cancer Gene Census database (31), and found that 28 GCCs are human cancer genes (Fig. 3B), which is a highly significant enrichment (P < 0.0001). These results provide additional evidence that insertional mutations in these CCGs are involved in driving the development of mesenchymal tumors. Previous studies have suggested that most CCGs identified by SB mutagenesis function during late stages of tumor progression Kodama et al.

B

7 IHBC/SB lines 1

Cancer Gene Census

IHBC/SB-tumor CISs

52 IHBC/SBderived tumors

60

28

199

542

232 p < 0.0001

C 3000

2500

IHBC/SB Tumors Have an Activated EMT Program. The mesenchymal

Gab1

Ptpn12 Fto

Average read counts

nature of these tumors prompted us to ask whether an EMT program had been activated in tumors. We did this by assaying the mRNA expression of 84 EMT-related genes in five IMBC/SB lines and one IMBC/WT line, four IMBC/SB tumors, and an immortalized mouse hepatocyte cell line (AML12). Hierarchical clustering showed distinct expression differences in the IHBC/SB line-derived and IHBC/SB-derived tumors (Fig. 2A), whereas similar expression patterns were observed in the IHBC/SB lines, IHBC/WT lines and AML12 cells. We also individually checked the expression levels of EMT-related genes in the IHBC/SB lines, IHBC/SB-derived tumors, and HCC tumors obtained from SB transposon expressing mice (SB/HCCs) by quantitative PCR (qPCR). SB/HCC tumors are well-differentiated HCCs and thus maintain their epithelial characteristics. Thus, they are different from poorly differentiated HCCs, which often exhibit mesenchymal properties (28). Mesenchymal markers Col1a2, Vim, and fibronectin 1 (Fn1) were markedly elevated in IHBC/SB-derived tumors but not in IHBC/SB lines and well-differentiated SB/ HCC tumors (Fig. 2 B–D). We also observed a significant decrease in expression of epithelial markers Cdh1 (cadherin 1, type 1, E-cadherin) in IHBC/SB-derived tumors (Fig. 2E). Finally, we found significant up-regulation of EMT-TFs, snail family zinc finger 1 (Snai1), twist family BHLH transcription factor 1 (Twist1), zinc finger E-box binding homeobox 1 (Zeb1), and Zeb2 (6) in IHBC/SB-derived tumors (Fig. 2 F–I). These data indicate that an EMT program has been activated in IHBC/SB-derived tumors.

PNAS PLUS

A

2000

1500

Mllt10

Huwe1 Lpp

Kdm6a

1000 Stag2 Epc2 Met Rasa1 Epb4.1l2 Nf1 Chd9 Psen1 Dnajc1 Ttc3 Fam120a Trio

Ltbp1 Arhgef28

500

Pbx1 Nedd4l 0 1

10

100

1000

- log P

Fig. 3. CCGs and trunk drivers identified in IHBC/WT lines and IHBC/SBderived tumors. (A) Venn diagram showing the overlap between 61 CCGs identified from seven IHBC/SB lines and 233 CCGs identified from 52 IHBC/SBderived tumors. (B) Venn diagram showing the overlap of 227 CCGs with human orthologs identified from 52 IHBC/SB-derived tumors and 570 human cancer genes listed in the Cancer Gene Census database. P values were calculated using Fisher’s exact test. (C) Transposon insertion sites identified in 52 IHBC/SB-derived tumors that were represented by >100 sequence read counts were selected for gCIS analysis, which identified 23 trunk drivers that were mutated in six or more tumors. Shown are the 23 trunk drivers, including their average sequence read counts and –log P values after Bonferroni correction of the gCIS results. The seven trunk drivers listed in the Cancer Gene Census are shown in red.

(24). To identify the subset of CCGs that function early in tumor development, we selected transposon insertion sites represented by ≥100 sequencing reads, arguing that these insertions would be present in the largest number of tumor cells. We then performed gCIS analysis on these insertions and identified 23 CCGs that were mutated in six or more tumors. We refer to these CCGs as trunk drivers (Fig. 3C). Strikingly, Met (MET proto-oncogene, receptor tyrosine kinase) and its binding partner GRB2-associated binding protein 1 (Gab1) were the first and third most significantly mutated trunk drivers. The Met pathway has various important roles in tumor development, including increased proliferation, migration, invasion, and angiogenesis. It is also an inducer of EMT (32). Seven of the 23 trunk drivers are known human cancer genes, whereas five—Met, Lpp, Pbx1, lysine-specific demethylase 6A (Kdm6a), and Nf1—are reportedly involved in regulating EMT (32–37). Another trunk driver, Rasa1, is a Ras GTPase-activating protein that suppresses oncogenic Ras signaling, another EMT inducer (5). The trunk driver Ltbp1 controls the activity of TGF-β, which is also an EMT inducer (38). Although there could be some PCR bias in this trunk driver analysis, these analytical results are certainly in agreement with the hypothesis that transposon mutagenesis plays an important role in activating the EMT program and driving a mesenchymal tumor phenotype in tumors generated from immortalized hepatoblasts. Importantly, examination of the cancer genome atlas (TCGA), which contains the mutation and copy number data for 193 HCC patients, shows that deregulation of the 23 trunk drivers in human HCC is significantly correlated with poor patient survival (Fig. S3), suggesting the important role of these trunk drivers in human HCC. PNAS | Published online May 31, 2016 | E3387

GENETICS

results, we reinjected the five tumorigenic IHBC/SB lines into nude mice and generated a total of 52 tumors. Once again, all of these tumors were spindle-shaped mesenchymal tumors. To determine whether the subcutaneous microenvironment might have provided a selective pressure that led to the generation of mesenchymal tumors, we orthotopically injected IHBC/ SB cells into the livers of nude mice. All 10 tumors that developed in the liver were histologically identical to those that developed in the flank (Fig. S2). These results indicate that the subcutaneous microenvironment did not provide a selective pressure that resulted in the formation of mesenchymal tumors.

A

Met/Gab1 pathway (40/52, 76.9%) Gab1 61.5% Met

38.5%

B Gab1

C

5′

3′

Met 5′

3′

E

D

F *

*

SNU-423 SNU-449 SNU-475 p-Met Met p-Gab1 Gab1

I *

Oncogenic Activation of the Met/Gab1 Pathway Induces EMT and Enhances the Migration of Human HCC Cells. We next sought to

confirm that the Met/Gab1 pathway plays an important role in regulating EMT in human HCC. We focused on this pathway because 40 of 52 IHBC/SB-derived tumors (76.9%) contain insertions in either or both Met and Gab1 genes (Fig. 4A). Transposon insertions are located in the 5′ end of these genes, before exon 3 or 2 (Fig. 4 B and C), respectively, suggesting that they are functioning as oncogenes. Consistent with this, the expression of Met and Gab1 was significantly higher in tumors with insertions in Met or Gab1 compared with IHBC/SB lines (Fig. 4 D and E). To determine whether the Met/Gab1 pathway plays an important role in regulating EMT in human HCC cells, we treated three human HCC cell lines with hepatocyte growth factor (HGF), which is the ligand for Met (32). HGF strongly increased Met phosphorylation and slightly up-regulated Gab1 phosphorylation in human HCC cell lines (Fig. 4F). HGF treatment decreased the expression of epithelial markers, such as E-cadherin, in all three HCC cell lines (Fig. 4G). Finally, in vitro scratch assays (39) showed that HGF treatment significantly enhanced the migration of HCC cells in vitro (Fig. 4 H–J). These results, along with those reported by others (32), indicate that activation of the Met/ Gab1 pathway is an important driver of EMT in human HCC cells. CCGs Function in Multiple Pathways Important for EMT. We next used KEGG pathway analysis to determine whether any signaling pathways or cellular processes are enriched among the CCGs identified in tumors. Strikingly, multiple pathways and cellular processes known to be involved in EMT, including Wnt signaling, adherens junctions, focal adhesion, Notch signaling, MAPK signaling, regulation of actin cytoskeleton, and TGFβ signaling (5), were enriched among the CCGs identified in tumors (Fig. 5A), further supporting the hypothesis that transposon mutaE3388 | www.pnas.org/cgi/doi/10.1073/pnas.1606876113

Vehicle HGF 1h HGF 3h

J *

H

*

G

Vehicle HGF 1h HGF 3h

Vehicle HGF 1h HGF 3h

β-actin

Fig. 4. Met/Gab1 pathway activation induces EMT and enhanced migration of human HCC cells. (A) Oncoprints showing the number of transposon insertions in Gab1 and Met in 52 IHBC/SB-derived tumors. Color density is determined by the sequence read counts for each insertion. The higher the read count, the darker the color. Each column represents a single tumor. (B and C) Insertion maps showing the location of transposon insertions in Gab1 (B) and Met (C) in 52 IHBC/SB-derived tumors. Each bar represents a single transposon insertion event. (D) mRNA expression levels of Gab1 determined by qPCR in five IHBC/SB lines and 20 IHBC/SB-derived tumors that had transposon insertions in Gab1. *P < 0.05, Mann– Whitney U test. (E) mRNA expression levels of Met determined by qPCR of five IHBC/SB lines and 13 IHBC/ SB-derived tumors that had transposon insertions in Met. *P < 0.05, Mann–Whitney U test. (F) Protein expression levels of phosphor-Met (Tyr1234), Met, phosphor-Gab1 (Tyr307), Gab1, and β-actin determined by Western blot analysis of three human HCC cell lines, including SNU-423, SNU-449, and SNU-475, after treatment with 10 ng/mL of recombinant human HGF or DMSO. (G) Expression levels of CDH1 determined by qPCR in SNU-423, SNU-449, and SNU-475 cells at 2 d after treatment with 10 ng/mL recombinant human HGF or DMSO. (H–J) The migration capacity of SNU-423 (H), SNU-449 (I), and SNU-475 (J) cells measured using a scratch assay at 2 d after treatment with 10 ng/mL recombinant human HGF or DMSO. n = 4. *P < 0.05, Student’s t test.

genesis plays an important role in driving EMT and inducing a mesenchymal phenotype in IHBC/SB-derived tumors. Ubiquitin-mediated proteolysis was the most enriched pathway in tumors with 26 CCGs functioning in this process. Whereas ubiquitin-mediated proteolysis is not commonly associated with EMT, all IHBC/SB-derived tumors had one or more mutations in a CCG that functions in this process (Fig. 5B). Interestingly, these 26 CCGs also appear to be clinically relevant in human HCC. TCGA data show that deregulation of these CCGs in human HCC was significantly correlated with poor patient survival (Fig. S4). Ubiquitin-Mediated Proteolysis Is Linked to EMT in Human HCC Cells.

To determine whether ubiquitin-mediated proteolysis is linked to EMT in human HCC, we focused on the HECT, UBA, and WWE domain containing 1 (Huwe1) gene, the most frequently mutated gene linked to ubiquitin-mediated proteolysis in tumors (Fig. 5B). Transposon insertions in Huwe1 are scattered throughout the gene (Fig. 5C), and its expression is significantly down-regulated in tumors with insertions in Huwe1 (Fig. 5D), suggesting that Huwe1 is a tumor suppressor gene. Therefore, we investigated whether HUWE1 knockdown effects EMT in human HCC cells. Knockdown of HUWE1 using two different siRNAs increased the expression of mesenchymal markers CDH2 (cadherin 2, type 1, Ncadherin), FN, and VIM (Fig. 5E) and the EMT-TFs SNAI1 and ZEB1 (Fig. 5E). HUWE1 knockdown also enhanced the migration of cells from three human HCC cell lines (Fig. 5 F–H), suggesting that HUWE1 is an EMT suppressor in human HCC. Importantly, we also found a significant negative correlation between the expression of HUWE1 and poor prognosis in patients with HCC, indicating that HUWE1 also has clinical relevance in human HCC (Fig. 5I). Protein-Tyrosine Phosphatase, Nonreceptor-Type 12 and Kdm6a Also Regulate EMT in Human HCC Cells. We also asked whether two

other SB-identified trunk drivers, protein-tyrosine phosphatase, Kodama et al.

PNAS PLUS

A

Ubiquitin mediated proteolysis Wnt signaling pathway Pathway in cancer Adherens junction Endocytosis Focal adhesion Axon guidance Notch signaling pathway Chronic myeloid leukemia MAPK signaling pathway Regulation of actin cytoskeleton TGF-beta signaling pathway 0

2

4

6

8

- log P

D

71.2% 36.5% 32.7% 26.9% 26.9% 25.0% 23.1% 23.1% 23.1% 23.1% 21.2% 21.2% 19.2% 19.2% 19.2% 19.2% 17.3% 15.4% 11.5% 11.5% 9.6% 9.6% 9.6% 7.7% 5.8% 5.8%

E GENETICS

Huwe1 Fbxw7 Rfwd2 Cblb Trip12 Ube3c Anapc1 Ube3a Ube2l3 Cul1 Itch Cbl Cul3 Fbxw11 Wwp2 Ube2d3 Ube2k Cul5 Rchy1 Cul4a Uba2 Birc2 Ppil2 Ube2g2 Prpf19 Ube2b

12

*

B

10

C Huwe1 5′

3′

G

HUWE1 1.0

* *

*

I

H Ratio of survived patients

F

high low

0.8 0.6 0.4 0.2 logrank p = 0.03

0.0 0

2000

4000

(Days)

Fig. 5. Ubiquitin-mediated proteolysis plays a role in regulating EMT in human HCC cells. (A) KEGG pathway analysis of the CCGs identified from 52 IHBC/SBderived tumors. P values for enrichment are represented on a –log scale. Pathways known to be involved in EMT are shown in red. (B) Oncoprints showing the number of IHBC/SB-derived tumors with transposon insertions in 26 CCGs that function in ubiquitin-mediated proteolysis. Color density is determined by the sequence read counts for each insertion. The higher the read count, the darker the color. Each column represents a single tumor. (C) Insertion maps showing the location of transposon insertions in Huwe1 in IHBC/SB-derived tumors. Each bar represents a single transposon insertion event. (D) mRNA expression levels of Huwe1 quantitated by qPCR in five IHBC/SB lines and 21 IHBC/SB-derived tumors that had transposon insertion in Huwe1. *P < 0.05, Mann–Whitney U test. (E) mRNA expression levels of CDH2, FN1, VIM, SNAI1, ZEB1, and HUWE1 quantitated by qPCR in SNU-387 cells at 3 d after transfection with negative control (NC) or HUWE1 siRNA. (F–H) Migration capacity of SNU-387 (F), SNU-423 (G), and SNU-475 (H) cells measured using a scratch assay at 3 d after transfection with negative control (NC) or HUWE1 siRNA. n = 4. P < 0.05 vs. all, one-way ANOVA. (I) Kaplan–Meier plot of patient survival based on mRNA abundance of HUWE1 in human HCC tumor samples (GSE10141). Statistical analysis was performed using the log-rank test.

Kodama et al.

PNAS | Published online May 31, 2016 | E3389

A B

Ptpn12 5′

3′

Kdm6a 5′

3′

D *

*

C

E

F

G

H

I

*

J *

*

nonreceptor-type 12 (Ptpn12) and Kdm6a, regulate EMT in human HCC cells. We chose these genes because they had the highest average read counts of all trunk drivers identified in tumors (Fig. 3C). Similar to Huwe1, transposon insertions in both genes were located throughout the coding region (Fig. 6 A and B), and their expression was significantly down-regulated in tumors with SB insertions (Fig. 6 C and D), suggesting they are tumor suppressor genes as well. Knockdown of these genes using two independent siRNAs in human HCC cells once again led to the up-regulation of mesenchymal markers CDH2, FN, and VIM and the EMT-TFs TWIST1 and ZEB1 (Fig. 6 E and F). Down-regulation of these genes also significantly increased the migration of cells from two human HCC cell lines, SNU-387 and SNU-475 (Fig. 6 G–J), further confirming that these genes are suppressors of EMT in human HCC cells. Combined with our previous studies, all five of the top trunk drivers identified in IHBC/SB-derived tumors are confirmed drivers of EMT in human HCC cells. SB-Identified EMT Regulators Promote Sorafenib Resistance in Human HCC Cells. Two recent reports have shown that EMT promotes

chemoresistance of tumor cells (12, 13). To determine whether EMT contributes to chemoresistance in human HCC cells, we examined the ability of three SB-identified EMT suppressors— Kdm6a, Huwe1, and Ptpn12—to modify the chemosensitivity of HCC cells. We knocked down the expression of each of these genes in human HCC cells using siRNA and then treated the cells with sorafenib, the sole potentially effective antineoplastic agent against HCC identified to date (4). Intriguingly, on sorafenib treatment, the viability of HCC cells containing siRNAs targeting these genes was significantly higher than that of cells receiving negative control siRNA (Fig. 7 A–C). Thus, EMT also seems to confer chemoresistance to human HCC cells. E3390 | www.pnas.org/cgi/doi/10.1073/pnas.1606876113

*

Fig. 6. SB-identified trunk drivers Ptpn12 and Kdm6a regulate EMT in human HCC cells. (A and B) Insertion maps showing the location of transposon insertions in Ptpn12 (A) and Kdm6a (B) in IHBC/SB-derived tumors. Each bar represents a single transposon insertion event. (C) mRNA expression levels of Ptpn12 quantitated by qPCR in five IHBC/SB lines and seven IHBC/SB-derived tumors that had transposon insertions in Ptpn12. *P < 0.05, Mann–Whitney U test. (D) mRNA expression levels of Kdm6a quantitated by qPCR in five IHBC/SB lines and nine IHBC/SB-derived tumors that had transposon insertion in Kdm6a. *P < 0.05, Mann–Whitney U test. (E and F) mRNA expression levels of CDH2, FN1, VIM, TWIST1, ZEB, and PTPN12 (E) or KDM6A (F) quantitated by qPCR in human SNU-387 HCC cells at 3 d after transfection with negative control (NC) or PTPN12 siRNA (E) or KDM6A siRNA (F). (G and H) Migration capacity of human HCC cells in SNU-387 (G) and SNU475 (H) cells at 3 d after transfection with negative control (NC) or PTPN12 siRNA, assessed using a scratch assay. *P < 0.05 vs. all, one-way ANOVA. (I and J) Migration capacity of human HCC cells in SNU-387 (I) and SNU-475 (J) cells at 3 d after transfection with negative control (NC) or KDM6A siRNA, assessed using a scratch assay. n = 4. *P < 0.05 vs. all, one-way ANOVA.

Finally, to study the potential mechanism of EMT-induced sorafenib resistance, we assessed the proliferation and apoptosis of HCC cells on knockdown of these three EMT suppressors. Knockdown of these genes significantly reduced HCC cell proliferation (Fig. 7 D–F), and HCC cells with siRNAs targeting these genes significantly attenuated the apoptosis induced by sorafenib treatment (Fig. 7 G–I). These findings suggest that EMT confers sorafenib resistance to human HCC cells through reduced cell proliferation and increased apoptotic tolerance. Discussion In this study, we have shown that mobilization of SB transposons in immortalized mouse hepatoblasts induces mesenchymal liver tumors following transplantation into nude mice. The tumors have reduced gene expression of epithelial markers and increased expression of mesenchymal markers and EMT-TFs such as Snail, Twist1, and Zeb1. Sequencing of transposon insertion sites from 52 tumors identified 233 CCGs that are enriched for genes and cellular processes driving EMT, whereas trunk driver analysis identified 23 CCGs that appear to function early in tumorigenesis and whose mutation or alteration in patients with HCC correlates with poor survival of these patients. Validation studies have shown that deregulation of the most highly mutated CCGs in human HCC activates an EMT program that enhances cell migration and also confers sorafenib resistance to HCC cells. Thus, transposon mutagenesis appears to provide an excellent resource for identifying genes driving EMT in human HCC and for potentially identifying new drug targets for HCC. Strikingly, Met and its binding partner Gab1 were the first and third most significantly mutated trunk drivers identified in tumors. Activating insertional mutations in Met and/or Gab1 were identified in >75% of tumors and showed very high sequencing read Kodama et al.

Kodama et al.

B

C *

*

D

E

F

H *

*

*

*

G

PNAS PLUS

A

*

I *

*

Fig. 7. siRNA knockdown of KDM6A, HUWE1, and PTPN12 induces sorafenib resistance in human HCC cells. (A–C ) Cell viability following sorafenib treatment of human HCC cells measured using the WST-1 assay. SNU-387 human HCC cells were transfected with negative control (NC), KDM6A (A), HUWE1 (B), or PTPN12 (C) siRNA. At 2 d after transfection, the cells were treated with 4 μM sorafenib or vehicle for another 3 d. Relative cell viability was calculated as the ratio of the WST-1 value between sorafenib-treated and vehicletreated cells. n = 4. *P < 0.05 vs. all, one-way ANOVA. (D–F) Cell proliferation of human HCC cells assessed using the WST-1 assay. SNU-387 human HCC cells were transfected with negative control (NC), KDM6A (D), HUWE1 (E), or PTPN12 (F) siRNA. At 5 d after transfection, the WST-1 assay was performed. n = 4. *P < 0.05 vs. all, one-way ANOVA. (G–I) Measurement of apoptosis using the caspase-3/7 assay following sorafenib treatment of human HCC cells. SNU-387 human HCC cells were transfected with negative control (NC), KDM6A (G), HUWE1 (H), or PTPN12 (I) siRNA. At 2 d after transfection, the cells were treated with 4 μM sorafenib or vehicle for another 2 d. Relative caspase-3/7 activity was calculated as the ratio of the caspase-3/7 activity value between sorafenib-treated and vehicle-treated cells. n = 4. *P < 0.05 vs. all, one-way ANOVA.

Kdm6a is an H3K27me3 demethylase that can reverse histone lysine methylation induced by Ezh2, a member of the polycomb repressive complex 2 (58). Loss-of-function mutations of KDM6A are frequently observed in various human cancers, including bladder cancers, medulloblastomas, and renal cancers (59), suggesting that KDM6A functions as a tumor suppressor. The role of Kdm6a in cancer progression remains controversial, however. Choi et al. (36) reported that Kdm6a represses EMT genes and inhibits EMTinduced cancer stemness properties in breast cancer. Similarly, van den Beucken et al. (37) showed that hypoxia inhibits KDM6A and induces the hypermethylation of DICER, leading to the up-regulation of ZEB1 expression and promotion of EMT and cancer stemness. In contrast, Thieme et al. (58) showed that Kdm6a positively regulates the migration of hematopoietic stem cells. In addition, Kim et al. (60) reported that knockdown of KDM6A decreased the migration and invasion capacity of breast cancer cell lines. In the present study, Kdm6a was inactivated in IHBC/SBderived tumors by transposon insertions. Knockdown of Kdm6a accelerated the migration of cells from two human HCC cell lines and increased ZEB1 expression. Because histone lysine demethylase can potentially regulate the activation of hundreds of genes, its function may vary and may depend on cell context, which could possibly generate contradictory findings. Why IHBC/SB lines preferentially generate mesenchymal tumors following transplantation into nude mice remains unclear. PNAS | Published online May 31, 2016 | E3391

GENETICS

counts, suggesting that activation of this signaling pathway is a major driver of these tumors. This pathway has been extensively studied in various cancers and shown to be a potent oncogenic pathway involved in cancer formation, progression, and metastasis (32, 40, 41). This pathway is activated in human HCC mainly through copy number alterations (42), and its overexpression has been correlated with an increased incidence of intrahepatic metastases and poor prognosis in patients with HCC (43). In HCC, HGF stimulation has been shown to activate an EMT program and to enhance the migration capacity of PLC/PRF/5 and HepG2 human liver cancer cells (21). In the present study, we have shown that activation of the Met/Gab1 pathway by HGF treatment decreases the expression of epithelial markers and increases the migration capacity of SNU-423, SNU-449, and SNU-475 human liver cancer cells, confirming that this pathway is an important activator of EMT in HCC. These data suggest that the Met pathway could be a promising therapeutic target in HCC. Indeed, tivantinib, a selective Met inhibitor, has recently been shown to significantly improve the overall survival of Met-positive HCC patients who experienced failure of or intolerance to previous systemic therapy (44). Phase III trials of several MET inhibitors in patients with HCC are currently underway, but the results of these trials have not yet been reported. Pathway analysis revealed that 26 CCGs identified in mesenchymal tumors function in ubiquitin-mediated proteolysis. Huwe1 was the most frequently mutated CCG and was inactivated by SB insertions in more than 70% of tumors. Huwe1 is a proven tumor suppressor gene that acts via the degradation of c-Myc/Miz1 in skin cancer (45) or Mcl-1 in HCC (46); however, a role of Huwe1 in EMT has not yet been described. In this study, we found that siRNA-mediated knockdown of Huwe1 led to increased expression of a variety of mesenchymal markers as well as EMT-TFs SNAI1 and ZEB1, leading to enhanced migration and sorafenib resistance of multiple HCC cell lines, demonstrating a new role for Huwe1 as an EMT suppressor. We also showed that down-regulation of Huwe1 is correlated with poor prognosis in patients with HCC, suggesting that inactivation of HUWE1 may contribute to the progression of HCC through activation of EMT. The second most frequently mutated CCG that functions in ubiquitin-mediated proteolysis was Fbxw7, which was mutated in 35% of tumors. Fbxw7 is highly mutated in various human cancers and functions as a tumor suppressor through degradation of several growth promoters, including c-Myc, m-TOR, cyclin E, and c-Jun (47). FBXW7 is also a known EMT suppressor and contributes to chemosensitivity in human HCC (48). In IHBC/SB-derived tumors, insertions in Fbxw7 are scattered throughout the coding regions, where they are predicted to inactivate Fbxw7 expression and potentially activate EMT in IHBC/SB-derived tumors. We also have shown that two other SB-identified trunk drivers, Ptpn12 and Kdm6a, regulate EMT in human HCC cells. PTPN12 tyrosine phosphatase was originally identified in a genetic screen for tumor suppressors in triple-negative breast cancers, where it inhibits multiple oncogenic tyrosine kinases, including HER2 and EGFR (49). In addition, studies of Ptpn12-deficient mice have shown that Ptpn12 regulates the migration of endothelial cells, dendritic cells, and macrophages, probably through dephosphorylation of Prk2, FAK, Cas, and paxillin (50–52). These antimigratory functions of PTPN12 also have been observed in several types of human cancers, including breast, ovary, and prostate cancers (53–55). Consistent with this, Ptpn12 was inactivated in our IHBC/SB-derived tumors by transposon insertion. We also showed that knockdown of PTPN12 significantly increased the migration capacity and expression of mesenchymal markers and EMT-TFs in multiple human HCC cell lines, indicating that PTPN12 is a universal EMT suppressor in cancer cells as well as in normal tissues. Decreased expression of PTPN12 is also negatively correlated with tumor recurrence and survival of HCC patients (56), further supporting the importance of this tumor suppressor in HCC. Interestingly, Ptpn12-deficient mouse embryos exhibited a failure of liver development, leading to lethality (57), suggesting that Ptpn12 may play a fundamental role in liver development as well.

One possibility is that IHBC/SB lines acquire mesenchymal properties during growth in culture, and the tumors simply reflect this preselection. Our qPCR EMT array results suggest that this is not the case, however, and instead indicate that the mesenchymal properties of tumors are acquired during growth in vivo. Consistent with this, there was almost no overlap between the CCGs detected in IHBC/SB lines and those detected in IHBC/SB-derived tumors, indicating that these EMT-related cancer genes were selectively mutated during growth in vivo. Another possible explanation is that the subcutaneous microenvironment provided selective pressure for the generation of mesenchymal tumors. This also does not appear to be the case, however, given that IHBC/SB-derived tumors generated after orthotopic injection into the liver were identical to the tumors generated after s.c. injection. EMT is linked to various fundamental processes in cancer development and progression, including migration, invasion, cancer stemness, and DNA repair (5, 6, 8, 61), and it is possible that transposon-induced activation of an EMT program was selected during growth of IHBC/SB cells in vivo because it gave the cells a strong survival advantage. Although these tumors originated from epithelial cells and acquired many mesenchymal properties, some of our findings were not consistent with typical EMT. For instance, whereas the expressions of many known EMT markers and EMT regulators were up-regulated in tumors, including extracellular matrix proteins, TGF-β family proteins, and EMT-TFs, the expression levels of β-catenin and N-cadherin were low in these tumors. Moreover, these tumors still maintained immunopositivity for epithelial markers as well as mesenchymal markers. Therefore, the phenotype that these tumors acquired was not complete EMT in a strict sense, which requires a complete transition, but rather incomplete EMT caused by epithelial plasticity of cancer cells, which is often a transient and reversible process. This important feature of cancer cells may contribute to metastasis, cancer stemness, and drug resistance; thus, our transposon screen provides important global insights into the genes regulating this process. In summary, our unique cell-based SB transposon screen provides, to our knowledge, the first high-throughput approach to identifying and validating genes driving EMT in human HCC. CCGs identified in this screen provide an important new resource for further studying the genes and cellular processes driving EMT in human HCC, as well as new potential therapeutic targets. Materials and Methods Mice. Rosa26-loxP-STOP-loxP-SB11 transposase knock-in mice [Gt(ROSA) 26Sortm2(sb11)Njen] (62); T2/Onc2 (6070) transposon transgenic mice [TgTn (sb-T2/Onc2)6070Njen] (62), which have 214 copies of the T2/Onc2 transposon inserted into a single site on chromosome 4; B6.Cg-Tg(Alb-cre)21Mgn/J transgenic mice (Jackson Laboratory strain 003574); and C.129S4-Ptentm1Hwu/J mice (Jackson Laboratory strain 004597) were used in this study. Mice were housed in a specific pathogen-free facility with a 12-h-light/12-h-dark cycle. The Institutional Animal Care and Use Committee of Houston Methodist Research Institute approved all mouse procedures. Mice heterozygous for the Alb-Cre transgene were crossed to mice homozygous for Rosa26-lsl-SB11 and T2/Onc2 (6070) to generate Alb-Cre/+;T2Onc2/+;Rosa26-lsl-SB11/+ embryos used for hepatoblast isolation. For qPCR analysis, well-differentiated HCCs were collected from Alb-Cre/+;T2Onc2/+;Rosa26-lsl-SB11/+;Pten flox/flox mice and confirmed histologically by a pathologist. Isolation and Immortalization of Embryonic Hepatoblasts. Immortalized hepatoblasts were obtained using the method reported by Strick-Marchand H et al. (26). In brief, Alb-Cre/+;T2Onc2/+;Rosa26-lsl-SB11/+ embryos were dissected at E13.5, and dissociated cells from embryonic livers were plated into type I collagen-treated dishes. Cells were cultured in vitro with DMEM/F12 medium containing 10% FCS, penicillin-streptomycin, 50 ng/mL EGF (Gibco), 30 ng/mL IGF-II (Peprotech), and 10 μg/mL insulin (Thermo Fisher Scientific). Spontaneously immortalized hepatoblast colonies were inoculated and further cultured to obtain immortalized IHBC/SB lines. Soon after the immortalized lines were established, they were cryopreserved for transplantation. Transposons in these lines are continuously transposing during in vitro culture as well as after transplantation, because of constitutive activity of the SB11 transposase. Embryos from C57BL6/J mice were used to establish IHBC/WT lines as a control.

E3392 | www.pnas.org/cgi/doi/10.1073/pnas.1606876113

In Vitro Hepatocyte Differentiation of Hepatoblasts. To induce hepatocyte differentiation in vitro, IHBC/SB lines were cultured for 14 d in medium supplemented with 20 ng/mL recombinant mouse HGF (Peprotech), 20 ng/mL recombinant mouse oncostatin M (R&D Systems), 10−6 M dexamethazone (Sigma-Aldrich), and insulin-transferrin-selenium (Thermo Fisher Scientific) (27). Allograft Tumor Formation Assay. Female athymic nude mice [Crl:Nu(NCr)Foxn1nu] were purchased from Charles River Laboratories. A total of 1.0 × 106 IHBC/SB or IHBC/WT cells were injected s.c. into the flanks of nude mice within 2 wk after their resuscitation from cryopreservation. The mice were monitored for tumor formation until 90 d of age. These experiments were repeated several times. Tumors >5 mm in diameter were collected for sequence and biochemical analysis. Sequencing of Transposon Insertion Sites. Transposon insertion sites were sequenced using splinkerette PCR to produce barcoded PCR products, which were then pooled and sequenced (22–24). To obtain the maximum amount of sequencing reads and avoid PCR bias (29), we used Illumina sequencing and acoustic shearing of genomic DNA (gDNA). gDNA (3 μg per sample) was acoustically sheared to 300-bp fragments using a Covaris S220 Focusedultrasonicator and end-repaired using a Fast DNA End Repair Kit (Thermo Fisher Scientific). Adapters (linker+: 5′-GTA ATA CGA CTC ACT ATA GGG CTC CGC TTA AGG GAC-3′; linker−: 5′-phos-GTC CCT TAA GCG GAG-C3spacer-3′) were then annealed and ligated to repaired DNA. Purified DNA was PCRamplified using linker (5′-GTA ATA CGA CTC ACT ATA GGG C-3′), IRR (5′-GGA TTA AAT GTC AGG AAT TGT GAA AA-3′), and IRL (5′-AAA TTT GTG GAG TAG TTG AAA AAC GA-3′) primers. Nested secondary PCR was performed using indexed SB-2ndR primer [5′-CAA GCA GAA GAC GGC ATA CGA GAT (8-bp index) GTG ACT GGA GTT CAG ACG TGT GCT CTT CCG ATC TTA GGG CTC CGC TTA AGG GAC-3′], SB-2ndRF primer [AAT GAT ACG GCG ACC ACC GAG ATC TAC ACT CTT TCC CTA CAC GAC GCT CTT CCG ATC T (1- to 4-bp stagger) (8-bp index) GGC TAA GGT GTA TGT AAA CTT CCG ACT TCA ACT G-3′], and SB-2ndLF primer [AAT GAT ACG GCG ACC ACC GAG ATC TAC ACT CTT TCC CTA CAC GAC GCT CTT CCG ATC T (1- to 4-bp stagger) (8-bp index) AAC TTA AGT GTA TGT AAA CTT CCG ACT TCA ACT G-3′]. All oligos were purchased from Integrated DNA Technologies with standard desalting. DNA fragments of 300–400 bp from amplicon libraries were excised from the agarose gel and quantified using a Qubit fluorometer (Thermo Fisher Scientific), an Agilent 2100 bioanalyzer, and real-time qPCR. Excised libraries were then pooled, loaded onto two flow cell lanes, and sequenced using an Illumina Hi Seq2500 sequencer at Eurofins Genomics. Sequencing reads were converted to FASTQ format and demultiplexed using bcl2fastq-1.8.4 (Illumina) at Eurofins Genomics. To reduce bleed-through across samples in the same flow cell, a custom dual-indexing routine was incorporated. This checked that a dual index sequence near the beginning of a read matched the barcode associated with the read, and then stripped away the index and the preceding sequence. After these preprocessing steps, reads to the mouse genome were mapped using a modified form of our SBCapSeq pipeline. The likelihood of local hopping of the SB transposon is increased on the chromosome where the transposon concatamer is located (62); thus, all insertions on chromosome 4, the site of the T2/Onc2 donor concatamer, were removed from the dataset before subsequent analysis, unless indicated otherwise. Sfi1 is listed as a singlecopy gene in the reference mouse genome; however, it is estimated that the mouse genome has 20–30 copies of Sfi1 (63). Insertions in multiple different Sfi1 loci are erroneously annotated to a single Sfi1 gene located on chromosome 11; thus, we removed Sfi1 from our lists of CIS genes, because this is a known artifact. We reported the resulting TA sites and their read depths in BED format. A total of 871,231 and 6,541,525 mapped reads, corresponding to 50,215 and 118,530 nonredundant insertion sites, were identified from seven IHBC/SB lines and 52 IHBC/SB-derived tumors, respectively (Datasets S1 and S2). CCGs were identified using gCIS (30), which looks for a higher density of transposon insertions within the coding regions of all RefSeq genes than would be predicted by chance. gCIS analysis was performed using insertions with >10 sequence read counts. Genes with a P value 100 sequence read counts. Genes with a P value < 0.05, as calculated by the χ2 test followed by Bonferroni correction, and detected in more than six tumors were defined as candidate trunk drivers. For pathway analysis, gCIS analysis was performed using insertions with >10 sequence read counts, and genes with a P value < 0.05 calculated by the χ2 test followed by false discovery rate correction were used. For all other information on Materials and Methods, please see SI Materials and Methods. A list of siRNA oligonucleotides and TaqMan probes is also provided in Table S3.

Kodama et al.

1. Forner A, Llovet JM, Bruix J (2012) Hepatocellular carcinoma. Lancet 379(9822): 1245–1255. 2. Mittal S, El-Serag HB (2013) Epidemiology of hepatocellular carcinoma: Consider the population. J Clin Gastroenterol 47(Suppl):S2–S6. 3. National Comprehensive Cancer Network (2016) NCCN Clinical Practice Guidelines in Oncology: Hepatobiliary Cancers (National Comprehensive Cancer Network, Fort Washington, PA). 4. Llovet JM, et al.; SHARP Investigators Study Group (2008) Sorafenib in advanced hepatocellular carcinoma. N Engl J Med 359(4):378–390. 5. Lamouille S, Xu J, Derynck R (2014) Molecular mechanisms of epithelial-mesenchymal transition. Nat Rev Mol Cell Biol 15(3):178–196. 6. De Craene B, Berx G (2013) Regulatory networks defining EMT during cancer initiation and progression. Nat Rev Cancer 13(2):97–110. 7. Kalluri R, Weinberg RA (2009) The basics of epithelial-mesenchymal transition. J Clin Invest 119(6):1420–1428. 8. Thiery JP, Acloque H, Huang RY, Nieto MA (2009) Epithelial-mesenchymal transitions in development and disease. Cell 139(5):871–890. 9. Tsai JH, Donaher JL, Murphy DA, Chau S, Yang J (2012) Spatiotemporal regulation of epithelial-mesenchymal transition is essential for squamous cell carcinoma metastasis. Cancer Cell 22(6):725–736. 10. Mani V, et al. (2008) Serial in vivo positive contrast MRI of iron oxide-labeled embryonic stem cell-derived cardiac precursor cells in a mouse model of myocardial infarction. Magn Reson Med 60(1):73–81. 11. Sampieri K, Fodde R (2012) Cancer stem cells and metastasis. Semin Cancer Biol 22(3): 187–193. 12. Fischer KR, et al. (2015) Epithelial-to-mesenchymal transition is not required for lung metastasis but contributes to chemoresistance. Nature 527(7579):472–476. 13. Zheng X, et al. (2015) Epithelial-to-mesenchymal transition is dispensable for metastasis but induces chemoresistance in pancreatic cancer. Nature 527(7579):525–530. 14. Yang MH, et al. (2009) Comprehensive analysis of the independent effect of twist and snail in promoting metastasis of hepatocellular carcinoma. Hepatology 50(5): 1464–1474. 15. Jou J, Diehl AM (2010) Epithelial-mesenchymal transitions and hepatocarcinogenesis. J Clin Invest 120(4):1031–1034. 16. Xia L, et al. (2013) Overexpression of forkhead box C1 promotes tumor metastasis and indicates poor prognosis in hepatocellular carcinoma. Hepatology 57(2):610–624. 17. Ding W, et al. (2010) Epithelial-to-mesenchymal transition of murine liver tumor cells promotes invasion. Hepatology 52(3):945–953. 18. Zhai B, Sun XY (2013) Mechanisms of resistance to sorafenib and the corresponding strategies in hepatocellular carcinoma. World J Hepatol 5(7):345–352. 19. Chow AK, et al. (2013) The enhanced metastatic potential of hepatocellular carcinoma (HCC) cells with sorafenib resistance. PLoS One 8(11):e78675. 20. Chen X, et al. (2011) Epithelial mesenchymal transition and hedgehog signaling activation are associated with chemoresistance and invasion of hepatoma subpopulations. J Hepatol 55(4):838–845. 21. Nagai T, et al. (2011) Sorafenib inhibits the hepatocyte growth factor-mediated epithelial mesenchymal transition in hepatocellular carcinoma. Mol Cancer Ther 10(1): 169–177. 22. Bard-Chapeau EA, et al. (2014) Transposon mutagenesis identifies genes driving hepatocellular carcinoma in a chronic hepatitis B mouse model. Nat Genet 46(1):24–32. 23. Mann MB, et al. (2015) Transposon mutagenesis identifies genetic drivers of Braf (V600E) melanoma. Nat Genet 47(5):486–495. 24. Takeda H, et al. (2015) Transposon mutagenesis identifies genes and evolutionary forces driving gastrointestinal tract tumor progression. Nat Genet 47(2):142–150. 25. Koso H, et al. (2012) Transposon mutagenesis identifies genes that transform neural stem cells into glioma-initiating cells. Proc Natl Acad Sci USA 109(44):E2998–E3007. 26. Strick-Marchand H, Weiss MC (2002) Inducible differentiation and morphogenesis of bipotential liver cell lines from wild-type mouse embryos. Hepatology 36(4 Pt 1): 794–804. 27. Zhang W, et al. (2012) Efficient generation of functional hepatocyte-like cells from human fetal hepatic progenitor cells in vitro. J Cell Physiol 227(5):2051–2058. 28. van Zijl F, et al. (2009) Epithelial-mesenchymal transition in hepatocellular carcinoma. Future Oncol 5(8):1169–1179. 29. March HN, et al. (2011) Insertional mutagenesis identifies multiple networks of cooperating genes driving intestinal tumorigenesis. Nat Genet 43(12):1202–1209. 30. Brett BT, et al. (2011) Novel molecular and computational methods improve the accuracy of insertion site analysis in Sleeping Beauty-induced tumors. PLoS One 6(9): e24668. 31. Futreal PA, et al. (2004) A census of human cancer genes. Nat Rev Cancer 4(3): 177–183. 32. Gherardi E, Birchmeier W, Birchmeier C, Vande Woude G (2012) Targeting MET in cancer: Rationale and progress. Nat Rev Cancer 12(2):89–103. 33. Ngan E, Northey JJ, Brown CM, Ursini-Siegel J, Siegel PM (2013) A complex containing LPP and α-actinin mediates TGFβ-induced migration and invasion of ErbB2-expressing breast cancer cells. J Cell Sci 126(Pt 9):1981–1991.

34. Risolino M, et al. (2014) Transcription factor PREP1 induces EMT and metastasis by controlling the TGF-β-SMAD3 pathway in non-small cell lung adenocarcinoma. Proc Natl Acad Sci USA 111(36):E3775–E3784. 35. Arima Y, et al. (2010) Decreased expression of neurofibromin contributes to epithelial-mesenchymal transition in neurofibromatosis type 1. Exp Dermatol 19(8): e136–e141. 36. Choi HJ, et al. (2015) UTX inhibits EMT-induced breast CSC properties by epigenetic repression of EMT genes in cooperation with LSD1 and HDAC1. EMBO Rep 16(10): 1288–1298. 37. van den Beucken T, et al. (2014) Hypoxia promotes stem cell phenotypes and poor prognosis through epigenetic regulation of DICER. Nat Commun 5:5203–5215. 38. Chandramouli A, Simundza J, Pinderhughes A, Cowin P (2011) Choreographing metastasis to the tune of LTBP. J Mammary Gland Biol Neoplasia 16(2):67–80. 39. Liang CC, Park AY, Guan JL (2007) In vitro scratch assay: A convenient and inexpensive method for analysis of cell migration in vitro. Nat Protoc 2(2):329–333. 40. Mazzone M, Comoglio PM (2006) The Met pathway: Master switch and drug target in cancer progression. FASEB J 20(10):1611–1621. 41. Cecchi F, Rabe DC, Bottaro DP (2012) Targeting the HGF/Met signaling pathway in cancer therapy. Expert Opin Ther Targets 16(6):553–572. 42. Giordano S, Columbano A (2014) Met as a therapeutic target in HCC: Facts and hopes. J Hepatol 60(2):442–452. 43. Ueki T, Fujimoto J, Suzuki T, Yamamoto H, Okamoto E (1997) Expression of hepatocyte growth factor and its receptor c-met proto-oncogene in hepatocellular carcinoma. Hepatology 25(4):862–866. 44. Qi XS, Guo XZ, Han GH, Li HY, Chen J (2015) MET inhibitors for treatment of advanced hepatocellular carcinoma: A review. World J Gastroenterol 21(18):5445–5453. 45. Inoue S, et al. (2013) Mule/Huwe1/Arf-BP1 suppresses Ras-driven tumorigenesis by preventing c-Myc/Miz1-mediated down-regulation of p21 and p15. Genes Dev 27(10): 1101–1114. 46. Gruber S, et al. (2013) Obesity promotes liver carcinogenesis via Mcl-1 stabilization independent of IL-6Rα signaling. Cell Reports 4(4):669–680. 47. Yumimoto K, et al. (2015) F-box protein FBXW7 inhibits cancer metastasis in a non– cell-autonomous manner. J Clin Invest 125(2):621–635. 48. Yu J, et al. (2014) FBW7 increases chemosensitivity in hepatocellular carcinoma cells through suppression of epithelial-mesenchymal transition. Hepatobiliary Pancreat Dis Int 13(2):184–191. 49. Sun T, et al. (2011) Activation of multiple proto-oncogenic tyrosine kinases in breast cancer via loss of the PTPN12 phosphatase. Cell 144(5):703–718. 50. Rhee I, Zhong MC, Reizis B, Cheong C, Veillette A (2014) Control of dendritic cell migration, T cell-dependent immunity, and autoimmunity by protein tyrosine phosphatase PTPN12 expressed in dendritic cells. Mol Cell Biol 34(5):888–899. 51. Souza CM, et al. (2012) The phosphatase PTP-PEST/PTPN12 regulates endothelial cell migration and adhesion, but not permeability, and controls vascular development and embryonic viability. J Biol Chem 287(51):43180–43190. 52. Rhee I, Davidson D, Souza CM, Vacher J, Veillette A (2013) Macrophage fusion is controlled by the cytoplasmic protein tyrosine phosphatase PTP-PEST/PTPN12. Mol Cell Biol 33(12):2458–2469. 53. Sahu SN, Nunez S, Bai G, Gupta A (2007) Interaction of Pyk2 and PTP-PEST with leupaxin in prostate cancer cells. Am J Physiol Cell Physiol 292(6):C2288–C2296. 54. Villa-Moruzzi E (2013) PTPN12 controls PTEN and the AKT signalling to FAK and HER2 in migrating ovarian cancer cells. Mol Cell Biochem 375(1-2):151–157. 55. Li J, et al. (2015) Loss of PTPN12 stimulates progression of ErbB2-dependent breast cancer by enhancing cell survival, migration, and epithelial-to-mesenchymal transition. Mol Cell Biol 35(23):4069–4082. 56. Luo RZ, et al. (2014) Decreased expression of PTPN12 correlates with tumor recurrence and poor survival of patients with hepatocellular carcinoma. PLoS One 9(1):e85592. 57. Sirois J, et al. (2006) Essential function of PTP-PEST during mouse embryonic vascularization, mesenchyme formation, neurogenesis and early liver development. Mech Dev 123(12):869–880. 58. Thieme S, et al. (2013) The histone demethylase UTX regulates stem cell migration and hematopoiesis. Blood 121(13):2462–2473. 59. Suvà ML, Riggi N, Bernstein BE (2013) Epigenetic reprogramming in cancer. Science 339(6127):1567–1570. 60. Kim JH, et al. (2014) UTX and MLL4 coordinately regulate transcriptional programs for cell proliferation and invasiveness in breast cancer cells. Cancer Res 74(6):1705–1717. 61. Zhang P, et al. (2014) ATM-mediated stabilization of ZEB1 promotes DNA damage response and radioresistance through CHK1. Nat Cell Biol 16(9):864–875. 62. Dupuy AJ, Akagi K, Largaespada DA, Copeland NG, Jenkins NA (2005) Mammalian mutagenesis using a highly mobile somatic Sleeping Beauty transposon system. Nature 436(7048):221–226. 63. Quinlan AR, et al. (2010) Genome-wide mapping and assembly of structural variant breakpoints in the mouse genome. Genome Res 20(5):623–635.

Kodama et al.

PNAS | Published online May 31, 2016 | E3393

PNAS PLUS

(CPRIT) and Public Health Service Grant DK56338, which funds the Texas Medical Center Digestive Diseases Center. N.G.C. and N.A.J. are CPRIT Scholars in Cancer Research.

GENETICS

ACKNOWLEDGMENTS. We thank E. Freiter and H. Lee for monitoring mice and providing technical assistance with the animals. This research was supported in part by the Cancer Prevention Research Institute of Texas