EPHB2 Activates β-Catenin to Enhance Cancer Stem Cell Properties and Drive Sorafenib Resistance in Hepatocellular Carcinoma

Hepatocellular carcinoma (HCC) ranks the sixth most commonly diagnosed cancer with a poor prognosis (1). The first-line treatments for HCC are liver transplantation and surgical resection. However, most HCCs are inoperable, as patients typically are diagnosed at advanced stages. HCC treatment recently entered a new era with the development of molecular-targeted therapies, as sorafenib improved survival in patients with advanced HCC (2). However, the survival benefit of sorafenib is modest, with median survival times 2.8 and 2.3 months longer than those of the placebo arm found in two large-scale trials (2, 3). Therefore, studies on the mechanism underlying sorafenib resistance are urgently needed. We and others have shown that sorafenib-resistant HCC cells are endowed with enhanced cancer stem cell (CSC) properties (4–6). These data strongly support the hypothesis that targeting the signaling pathways that mediate the functions of liver CSCs may be a strategy to reverse drug resistance in HCC.

To mimic this clinical situation, many researchers have employed in vitro establishment of sorafenib-resistant cells. However, the translational potential of this model is limited because the model is not representative of the true clinical situation. For this purpose, we employed in vivo models established with patient-derived tumor xenografts (PDTX) derived directly from two samples from patients with HCC and developed sorafenib resistance in vivo by administration of several rounds of sorafenib treatment. Using this strategy, we were able to enrich liver CSCs in these sorafenib-resistant PDTXs, as evidenced by an increase in self-renewal. Using RNA-sequencing analysis, we compared the expression profiles of sorafenib-resistant PDTXs and their corresponding mock controls. This analysis showed that EPH receptor B2 (EPHB2) showed remarkably higher expression in the liver CSC-enriched sorafenib-resistant PDTXs than in their mock counterparts, suggesting a role for EPHB2 in liver CSCs.

EPHB2 belongs to the Eph receptor family of receptor tyrosine kinase transmembrane glycoproteins and was first identified as a tumor suppressor because it has been shown to exhibit significantly downregulated expression in a variety of cancers, including colon (7) and head and neck cancers (8), mediated by chromosomal aberrations and promoter methylation. A number of studies have shown that EPHB2 is overexpressed and plays crucial roles in tumor growth, migration, and chemoresistance (9, 10). However, the clinical relevance and functional role of EPHB2 in HCC remain uncharacterized. In mouse and human HCCs, we found that EPHB2 became overexpressed in a stepwise manner from healthy liver tissue to cirrhotic liver tissue and then HCC tissue. High EPHB2 expression in patients was significantly correlated with poor patient survival. Interestingly, EPHB2 expression was markedly upregulated in our established sorafenib-resistant PDTXs. By overexpression and knockdown approaches, EPHB2 was found to be critically involved in the regulation of liver CSCs. Likewise, EPHB2^High HCC cells were shown to possess enhanced liver CSC properties. EPHB2 was found to regulate cancer stemness and drug resistance via a Wnt/β-catenin positive feedback loop. Targeting EPHB2 with rAAV-8-shEPHB2 suppressed HCC tumor growth and sensitized HCC cells to sorafenib in an immunocompetent mouse model. In summary, targeting the TCF1/EPHB2/β-catenin pathway may be a promising strategy for HCC treatment.

Lentiviral expression open reading frame clone, cytomegalovirus, EPHB2 (EX-E2415-Lv105) was purchased from GeneCopoeia. Δ-45-β-catenin and wild-type (WT)-β-catenin were obtained from Stephanie Ma (University of Hong Kong, Hong Kong, P.R. China).

Doxorubicin was purchased from EBEWE Pharma. Sorafenib was purchased from LC Laboratories.

MHCC-97L (Liver Cancer Institute, Fudan University, Shanghai, P.R. China), Huh7 and PLC/PRF/5 (Japan Cancer Research Bank), were maintained in DMEM containing high glucose (Gibco BRL) with 10% heat-inactivated FBS (Gibco BRL), 100 mg/mL penicillin G, and 50 μg/mL streptomycin (Gibco BRL) at 37°C in a humidified atmosphere containing 5% CO₂. These three HCC cell lines were established from male origin. All cell lines used in this study were obtained between 2013 and 2016, regularly authenticated by morphologic observation and AuthentiFiler STR (Invitrogen) and tested for the absence of Mycoplasma contamination (MycoAlert, Lonza). Experiments were performed within 20 passages after cell thawing.

Paired patient HCC and adjacent noncancerous liver tissue specimens were collected at the time of surgical resection at Queen Mary Hospital, Hong Kong, P.R. China, from 1991 to 2013. These samples were obtained from patients with their written consent. The use of human clinical specimens was approved by the Institutional Review Board of the University of Hong Kong/Hospital Authority Hong Kong West Cluster.

For suppression of EPHB2 in HCC cells, short hairpin (shRNA) sequences against human EPHB2 were cloned to pLKO.1 vector (Addgene). Transduced cells were selected with 2 μg/mL puromycin. Supplementary Table S1 lists the sequences of the EPHB2 and NTC used. For overexpression, stable HCC cells ectopically overexpressing EPHB2 were selected with 2 μg/mL puromycin.

Total RNA was isolated using TRIzol reagent according to the manufacturer's protocol (Invitrogen). cDNA was synthesized using PrimeScript RT Reagent Kit (Takara) according to the manufacturer's instructions and then subjected to qPCR with BrightGreen 2× qPCR Master Mix (Applied Biological Materials) using QuantStudio 7 Flex Read Time PCR System (Applied Biosystems) with primers specific to the sequences of genes of interest, which are provided in Supplementary Table S2. Relative expression differences were calculated using 2^−ΔΔC_T method with reference to HPRT/GAPDH.

The cells were processed for chromatin immunoprecipitation (ChIP) assays according to Upstate CHIP protocol. Briefly, cells were cross-linked for 10 minutes with 1% formaldehyde and lysed. Lysate pellets were resuspended and enzymatic digested with micrococcal nuclease for 5 minutes. Protein–DNA complexes were immunoprecipitated using TCF1 antibody (Cell Signaling Technology) or Normal Rabbit IgG (Invitrogen) bound to protein A agarose, eluted, and digested with proteinase K. For qPCR analysis of the ChIP DNA samples before amplicon generation, QIAquick-purified immunoprecipitates were dissolved in 30 μL of water. Standard PCR reactions using 2 μL of the immunoprecipitated DNA were performed with a SYBR Green PCR Kit (Applied Biosystems). The EPHB2 promoter region (CTGTTAATGATG) in the TCF1–precipitated chromatin was amplified by qPCR primers (forward primer sequence CACTCCACAGAGTTCCGAGC and reverse primer sequence CCTGCTCTGACAGCCTTCAT. The amplicon is at −1,014 bp downstream of transcription start site (TSS). Calculation of TCF1 occupancy on the EPHB2 promoter was performed according to the ChIP-qPCR primer assay data analysis template from Sigma-Aldrich.

Total RNA of PLC/PRF/5 of nontarget control (NTC) and EPHB2 knockdown (shEPHB2-2) groups and sorafenib-resistant PDTXs (PDTX#1 and #5) and their corresponding mock controls were extracted using TRIzol Reagent (Life Technologies) according to manufacturer's protocol. The quality of total RNA was checked by Agilent 2100 bioanalyzer (Agilent Technologies Inc.) to have OD260/280 ratio of between 1.8 and 2.0 and RNA integrity number value higher than 8.0. The RNA samples that met the quality assessment were then subjected to Illumina Solexa sequencing using Hiseq 1500 sequencer (Illumina) for performing HiSeq sequencing run (pair-end sequencing of 101 bp). Each sample had an average throughput of 10.8 Gb and a total throughput of 21.5 Gb. An average of 94% of the bases achieved a quality score of Q30 where Q30 denotes the accuracy of a base call to be 99.9%. Expression estimation and tests for differential expression were processed by Cufflinks v2.1.1. All data were expressed as fragments per kilobase of exon per million fragments mapped values and fold changes in transcript levels relative to the NTC group from PLC/PRF/5. RNA-sequencing data are available publicly at GSE168783 and GSE168839.

Six to eight weeks old male WT C57BL/6 mice were used and the procedure was performed as described previously (11). In brief, 7.5 μg of plasmids encoding human AKT1 (myristylated AKT1 or myr-AKT1) and human neuroblastoma Ras viral oncogene homolog (N-RasV12) along with sleeping beauty (SB) transposase in a ratio of 25:1 were diluted in 2 mL saline (0.9% NaCl), filtered through 0.22 μm filter and injected into the lateral tail vein of C57BL/6 mice in 5–7 seconds. The constructs used in this study showed long-term expression of genes via hydrodynamic injection. To examine the role of EPHB2 in liver carcinogenesis using this model, eSpCas9-2A-GFP [PX458 against mouse EPHB2 (CGGCCCTGTAGTAGCCGTTG, Genscript) or eSpCas9-2A-GFP against NTC (Genscript)] will also be injected together with the AKT1, N-RasV12, and SB transposase. The study protocol was approved by and performed in accordance with the Committee of the Use of Live Animals in Teaching and Research at the University of Hong Kong and the Hong Kong Polytechnic University (Hong Kong, P.R. China).

Six to 8 weeks old male WT C57BL/6J mice were used and the procedure was performed as described previously (11). In brief, 7.5 μg of plasmids encoding human AKT1 (myristoylated AKT1 or myr-AKT1) and human neuroblastoma Ras viral oncogene homolog (N-RasV12) along with sleeping beauty transposase in a ratio of 25:1 were diluted in 2 mL saline (0.9% NaCl), filtered through 0.22 μm filter and injected into the lateral tail vein of C57BL/6J mice in 5–7 seconds. The constructs used in this study showed long-term expression of genes via hydrodynamic injection. Upon injection of plasmids for 2 weeks, the mice were injected with rAAV8-shEPHB2/rAAV8-NTC at the dosage of 1 × 10¹² vg/mouse intravenously (Vigene Biosciences). The sequence of shRNA against mouse EPHB2 is 5′-ACGAGAACATGAACACTAT-3′. After 1 week, we started to treat mice with sorafenib at 30 mg/kg for 21 days. At this point, the mice were separated into the following four subgroups: (i) DMSO with rAAV8-NTC; (ii) DMSO and rAAV8-shEPHB2; (iii) rAAV8-NTC and sorafenib (30 mg/kg); and (iv) rAAV8-shEPHB2 and sorafenib. The mice were sacrificed at day 21 after sorafenib treatment. The study protocol was approved by and performed in accordance with the Committee of the Use of Live Animals in Teaching and Research at the University of Hong Kong and the Hong Kong Polytechnic University (Hong Kong, P.R. China).

Statistical significance of qPCR, sphere formation assay, flow cytometry analysis, expression of liver CSC markers were determined by Student t test using Microsoft Office Excel software (Microsoft Corporation). The displayed results showed the means and the SDs, and those with P values less than 0.05 were considered statistically significant (*P < 0.05; ^**, P < 0.01; ^***, P < 0.001; ^****, P < 0.0001). Kaplan–Meier survival analysis was used to analyze disease-free survival and a log-rank test was used to determine the statistical significance; these analyses were carried out using SPSS 20 software.

The methods have been described in Supplementary Materials and Methods.

In an attempt to identify critical kinases involved in sorafenib resistance, we developed two sorafenib-resistant HCC cell lines from two PDTXs (PDTX#1 and PDTX#5; ref. 4), which were enriched for an increase in self-renewal and tumorigenicity (Fig. 1A–C). We have analyzed four tumor samples derived from these two sorafenib-resistant HCC cell lines and their corresponding mock counterparts by RNA-sequencing analysis. Upon analysis, we found 67 genes to be commonly upregulated in the two sorafenib-resistant PDTXs, among which, we found EPHB2 to be the most significantly upregulated kinase (Fig. 1D). Quantitative PCR analysis confirmed this observation, as elevated expression of EPHB2 was indeed found in the sorafenib-resistant PDTXs compared with their corresponding mock controls (Fig. 1E). In addition to these two PDTXs, we also observed similar upregulation of EPHB2 expression in sorafenib-resistant Huh7, PLC/PRF/5, and MHCC-97L cells (Supplementary Fig. S1; ref. 11). In an analysis of dataset (GSE25097), EPHB2 expression was significantly upregulated in tumor specimens, with stepwise upregulation from normal to cirrhotic to tumor tissues, suggesting an oncogenic role for EPHB2 in liver carcinogenesis (Fig. 1F). Similarly, we found stepwise increases in the mRNA expression of EPHB2 from normal to fibrotic to HCC tissues in the STAM nonalcoholic steatohepatitis (NASH) HCC mouse model (Fig. 1G). Interestingly, along with EPHB2 overexpression, a similar stepwise increase in the expression of liver CSC markers, including CD133, CD90, and CD24, was previously reported upon fibrosis and subsequent HCC development (12). Next, we examined the prognostic significance of EPHB2 expression in patients with HCC. The cut-off value of T/N > 6 was used to divide patients with HCC according to their EPHB2 expression. Patients with high EPHB2 expression were significantly correlated with poor disease-free survival (P = 0.041; Fig. 1H). Collectively, these data suggest that EPHB2 overexpression may be associated with liver carcinogenesis and sorafenib resistance via regulation of liver CSCs.

First, we performed endogenous knockout of EPHB2 in an immunocompetent mouse model to investigate the tumor-promoting role of EPHB2 in HCC. Using the hydrodynamic tail vein injection (HTVI) delivery approach, we induced HCC tumors in C57BL/6 mice using a combination of activated forms of myr-AKT and N-RasV12 proto-oncogenes and SB transposase (Fig. 2A; ref. 13). A stepwise increase in EPHB2 mRNA expression was noted in livers harvested from mice injected with the empty vector control alone to livers with early (1 to 2 weeks postinjection) to advanced HCC (2 to 4 weeks postinjection; Fig. 2B). Using the same model, we examined the role of EPHB2 in tumor formation by hydrodynamic injection of sgEPHB2 together with myr-AKT, N-RasV12, and SB transposase (Fig. 2C). Upon suppression of EPHB2 expression, the number of tumors and survival of mice were compared between the sgEPHB2 and sgNTC groups, and endogenous EPHB2 knockout significantly attenuated the size and number of tumor nodules compared with sgNTC treatment (Fig. 2D and E). In addition, EPHB2 knockout mice tended to show longer survival (P = 0.0888) than mice injected with sgNTC controls (Fig. 2F). HTVI of shEPHB2 is efficiently transducted to the livers of mice, as evidenced by the intense GFP signal and suppression of EPHB2 by qPCR analysis (Fig. 2G and H). The above findings support a key role for EPHB2 in promoting HCC growth.

To further examine whether EPHB2 functionally drives self-renewal, tumor formation, and drug resistance, we performed EPHB2-knockdown and EPHB2-overexpression experiments using lentiviral-based overexpression and knockdown approaches. Western blot analysis showed successful establishment of knockdown clones derived from PLC/PRF/5 and MHCC-97L cells and overexpression clones derived from Huh7 cells (Fig. 3A). In a sphere formation assay performed with PLC/PRF/5 or MHCC-97L cells, knockdown of EPHB2 expression significantly reduced the number and size of primary spheres and their second passages. Accordingly, EPHB2 overexpression was shown to enhance the sphere-forming ability of Huh7 cells (Fig. 3B). Similar findings were observed in limiting dilution assay (Supplementary Fig. S2). In line with this finding, knockdown of EPHB2 expression reduced the size and number of tumors formed by both PLC/PRF/5 cells and MHCC-97L cells (Fig. 3C; Supplementary Table S3). Conversely, EPHB2 overexpression in Huh7 cells significantly enhanced tumorigenicity, with an increase in the estimated CSC frequency (Fig. 3C; Supplementary Table S3). We consistently found that shEPHB2 HCC cells exhibited decreases in the percentages of cells expressing CD90 and CD47, while EPHB2-overexpressing cells showed increased expression of these markers (Fig. 3D). In addition, OCT4, one of the core pluripotent genes, was consistently changed upon alteration of EPHB2 expression (Supplementary Fig. S3). Next, we found that knockdown of EPHB2 expression enhanced the number of apoptotic cells in response to both doxorubicin treatment and sorafenib treatments (Fig. 3E). Conversely, EPHB2-overexpressing Huh7 cells exhibited reduced cell death when treated with doxorubicin or sorafenib (Fig. 3E). To further confirm the role of EPHB2 knockdown in chemosensitization, we employed lentiviral-based doxycycline-inducible knockdown of EPHB2 expression in PLC/PRF/5 cells. In this experiment, we added doxycycline at the same time as sorafenib administration. Consistently, inducible suppression of EPHB2 led to sensitization of the PLC/PRF/5 cells to sorafenib treatment (Supplementary Fig. S4A and S4B).

Next, we examined whether EPHB2^High HCC populations possess enhanced liver CSC properties by separating MHCC-97L cells into EPHB2^High and EPHB2^Low populations by a cell sorting approach. Upon analysis, purities ranged from 99% to 100% for EPHB2^Low cells and 85% to 95% for EPHB2^High cells (Fig. 4A). Western blot analysis further confirmed successful separation of MHCC-97L cells into EPHB2^High and EPHB2^Low populations (Fig. 4B). By sphere formation assay, EPHB2^High cells generated significantly larger spheres and more spheres in the first and second passages (Fig. 4C), and this finding supported the greater self-renewal capability of EPHB2^High HCC cells. Furthermore, EPHB2^High cells showed higher tumorigenic ability, when compared with EPHB2^Low cells purified from MHCC-97L cells. A significant difference in tumor incidence was observed between the EPHB2^High and EPHB2^Low cells (Fig. 4D). As few as 500 EPHB2^High cells were sufficient for consistent tumor development in NOD/SCID mice (Fig. 4D). In parallel with this finding, EPHB2^High cells exhibited higher expression of CD90 and CD47 (Fig. 4E). Finally, EPHB2^High cells were more refractory to the effects of sorafenib and doxorubicin than EPHB2^Low cells by Annexin V staining (Fig. 4F).

To determine the major downstream effector of EPHB2, we employed RNA-sequencing analysis to compare the gene expression profiles of shEPHB2-2 and NTC-transfected cells derived from the PLC/PRF/5 cell line. By gene set enrichment analysis (GSEA), we found several pathways to be downregulated, with over 60% of genes related to the Wnt/β-catenin signaling pathway being significantly altered (Fig. 5A–C). To confirm this finding, we examined the effect of EPHB2 alternations on the transactivation activity of β-catenin. A TOP/FOP reporter assay showed that the transactivation activity of β-catenin was suppressed when EPHB2 was repressed, while this activity was elevated in EPHB2-overexpressing cells (Fig. 5D). Consistent with the results for the transactivation activity of β-catenin, the expression of known downstream targets of the Wnt/β-catenin pathway, including cyclin D1 and CMYC, showed similar trends in response to alterations in EPHB2 expression (Fig. 5D). Consistently, higher β-catenin transactivation activity was observed in EPHB2^High Huh7 cells than in EPHB2^Low cells (Supplementary Fig. S5A and S5B). By immunofluorescence (IF) staining, increased cytoplasmic and nuclear β-catenin protein levels were observed in EPHB2 OE cells, and the opposite effect was observed upon EPHB2 suppression (Fig. 5E). Next, we examined the mediator linking EPHB2 and the Wnt/β-catenin signaling pathway by identifying the direct substrate of EPHB2. By immunoprecipitation analysis, we found that EPHB2 physically interacted with SRC (Fig. 5F). Upon interaction, EPHB2 was able to phosphorylate SRC, as evidenced by the corresponding changes in phosphorylation at Y416 upon suppression or overexpression of EPHB2, which subsequently led to changes in the AKT/GSK3β/β-catenin signaling pathway (Fig. 5G). In a The Cancer Genome Atlas (TCGA) cohort analysis, we found that the Wnt/β-catenin signaling pathway was the only pathway that was significantly enriched in patients with HCC with high EPHB2 expression, which further confirm Wnt/β-catenin as the major effector of EPHB2 signaling in our in vitro finding (Fig. 5H).

EPHB2 is known to be a Wnt target gene in colorectal cancer (14); however, the mechanism remains unclear. To examine whether Wnt/β-catenin alterations affect EPHB2 expression, we analyzed the impact of introducing constitutively active β‐catenin (Δ45β‐cat) into Huh7 cells. Constitutively active β‐catenin is mutated at serine 45 and, as a result, cannot be phosphorylated or undergo ubiquitination and degradation (15). WT β‐catenin was also overexpressed as a control. Upon activation of β‐catenin transactivation activity (Fig. 6A and B), we found that both the mRNA and protein levels of EPHB2 were upregulated (Fig. 6C). To further confirm this finding, we administered Wnt3a-supplemented medium to activate the Wnt/β-catenin pathway, and EPHB2 mRNA and protein levels were consistently upregulated (Fig. 6D). Echoing the above in vitro observations, we found that the EPHB2 mRNA level was consistently upregulated in an HCC mouse model (Fah^−/−) with hydrodynamic injection of constitutively active β‐catenin-S33Y, together with shp53 in Fah/SB11 mice (Fig. 6E; ref. 16). On the basis of these in vitro and in vivo observations, we postulated that Wnt/β-catenin transcriptionally activates EPHB2 expression. In silico analysis (PROMO 3.0) identified one putative TCF1 binding site (CTGTTAATGATG) at the position of −1,014 bp to −1,003 bp upstream of the TSS of the EPHB2 gene (Fig. 6F). Upon suppression of TCF1 in EPHB2-knockdown HCC cells, we found that there was a significant decrease in the fold enrichment of TCF1 binding relative to that in NTC cells (Fig. 6G and H), suggesting that TCF1 physically binds to and activates the promoter of EPHB2, which forms a Wnt/β-catenin positive feedback loop. Intriguingly, high EPHB2 expression significantly co‐occurred with β‐catenin (CTNNB1) driver mutations (Fig. 6I), which further confirms the role of Wnt/β-catenin in driving EPHB2 expression in HCC clinical samples.

Currently, EPHB2 kinase inhibitors are not commercially available. Therefore, we examined the therapeutic targeting of EPHB2 by intravenous injection of rAAV-8-EPHB2 alone or in combination with sorafenib in an NRAS/AKT-induced HCC mouse model (11). Two weeks after hydrodynamic injection of plasmids encoding the activated forms of NRAS and AKT, we intravenously injected the mice with either rAAV8-shEPHB2 or rAAV8-NTC. After 1 week, we started to treat the mice with sorafenib at 30 mg/kg for 21 days. At this point, the mice in a group of 5–6 were separated into the following four subgroups: (i) DMSO and rAAV8-NTC, (ii) DMSO and rAAV8-shEPHB2, (iii) rAAV8-NTC and sorafenib (30 mg/kg), and (iv) rAAV8-shEPHB2 and sorafenib (Fig. 7A). First, we found that rAAV8-NTC/rAAV8-shEPHB2 were efficiently delivered to the livers of mice in these four groups, as evidenced by the intense GFP signal (Fig. 7B). The efficacy of the combined drug treatment was evaluated by calculating the liver weight to body weight ratio was compared among the groups. We found that rAAV8-shEPHB2/sorafenib produced significantly maximal suppression of tumor growth, indicating that shEPHB2 treatment can synergize with sorafenib treatment and is effective against liver tumors in vivo (Fig. 7C and D), which is accompanied with increase in number of apoptotic cells by TUNEL (terminal deoxynucleotidyl transferase–mediated dUTP nick-end labeling) assay (Supplementary Fig. S6). IHC staining showed the efficient suppression of EPHB2 in the livers of rAAV8-shEPHB2 transduced groups (Fig. 7E).

Liver CSCs have been identified by several cell-surface antigens, including CD133 (17), CD90 (18), CD13 (19), CD24 (20), and epithelial cell adhesion molecule (EpCAM; ref. 21). However, there are no druggable kinase inhibitors that show specific targeting against the cell surface of liver CSCs. Identification of kinases in liver CSCs could offer a new paradigm for HCC treatment by developing specific kinase inhibitors to target these cells. In an attempt to identify specific cell-surface kinases crucial for the regulation of liver CSCs, we established sorafenib-resistant PDTXs with increased self-renewal. By RNA-sequencing analysis, we found that EPHB2 was one of the most commonly upregulated kinases in the two sorafenib-resistant PDTX models.

EPHB2, belongs to the Eph receptor family of receptor tyrosine kinase transmembrane glycoproteins. These receptors bind ligands called ephrins and are involved in diverse cellular processes, including motility, division, and differentiation (22). A distinguishing characteristic of Eph-ephrin signaling is that both the receptors and the ligands are competent to transduce a signaling cascade, resulting in bidirectional signaling (23). Two ligands of EPHB2, EFNB1, and EFNB2, have been reported (24). The role of EPHB2 in cancer remains controversial, and it is still unclear whether EPHB2 acts as a tumor suppressor gene or oncogene. We found that EPHB2 was overexpressed in a stepwise manner, increasing from healthy tissue to cirrhotic tissue and then to HCC tissue. This result is consistent with a previous report showing the role of EPHB2 in the development of cirrhosis in a CCL4-induced fibrosis model (25), which at least in part involves activation of stellate cells (26). In addition, EPHB2 overexpression was significantly correlated with poor patient survival.

EphB2 was first determined to play a critical role in regulating the homeostasis of stem cells in intestinal crypts (27). Subsequently, EPHB2 was found to regulate cell movement in intestinal crypts through a kinase-independent PI3K signaling mechanism (9). Recently, a role for EPHB2 in liver CSCs has been suggested. In colon cancer, EphB2-sorted intestinal stem cell-like tumor cells display a robust tumor-initiating capacity in immunodeficient mice and long-term self-renewal potential (28). Although these findings suggest a potential role for EPHB2 in regulating CSCs, the molecular mechanism by which EPHB2 regulates CSC properties is poorly understood. Through overexpression and knockdown approaches, we showed that EPHB2 regulates the self-renewal, tumorigenicity and drug resistance of HCC cells, and expression of liver CSC markers, including CD90 (18) and CD47 (29). By using a cell sorting approach, we consistently found that EPHB2^High HCC cells contained enhanced liver CSC populations. This result is consistent to previous report showing the role of EPHB2 in CSCs in colon cancer (28). Mechanistically, we found that Wnt/β-catenin signaling is the downstream signaling pathway in EPHB2-mediated liver CSC functions, as evidenced by the transactivation activity of β-catenin and its downstream targets. This result was supported by TCGA analysis showing enrichment of Wnt/β-catenin signaling in patients with high EPHB2 expression. By immunoprecipitation, we found that EPHB2 directly interacts with Src and induces phosphorylation at Y416. This result is consistent with a previous report demonstrating the interaction between these two proteins (30, 31). Through Src activation, EPHB2 drives activation of the AKT/GSK3β/β-catenin signaling cascade. EPHB2 is a known Wnt target gene in colorectal cancer (28), but the exact mechanism remains unclear. Through bioinformatic analysis, we found one putative TCF1 binding site upstream of the TSS in the EPHB2 promoter. Using a ChIP assay, we found a physical interaction between TCF1 and the EPHB2 promoter. Upon knockdown of EPHB2 expression, we found a consistent decrease in TCF1 expression, leading to a decrease in the site occupancy of TCF1 binding in knockdown cells relative to NTC cells. Consistent with this finding, HCC tumors with a β-catenin driver mutation exhibited high EPHB2 expression, further confirming the role of β-catenin/TCF1 as an upstream activator of EPHB2 signaling.

Finally, we examined the therapeutic potential of targeting EPHB2 alone or in combination with sorafenib in an immunocompetent mouse model by intravenous administration of rAAV-8-shEPHB2. rAAV-8 was previously shown to efficiently introduce transgenes in vivo with a specific tropism for the liver (32). When compared with those in the rAAV8-NTC group, mice injected with rAAV8-shEPHB2 showed suppression of tumor growth in terms of the liver/body ratio. rAAV8-shEPHB2 showed maximal suppression of HCC tumor growth compared with sorafenib or rAAV8-shEPHB2 alone. On the basis of these data, suppression of EPHB2 could sensitize immunocompetent HCC mouse model to sorafenib.

In conclusion, we have demonstrated that EPHB2 regulates cancer stemness and drug resistance via the TCF1/EPHB2/β-catenin positive feedback loop. Targeting EPHB2 alone or in combination with other targets may be a novel therapeutic strategy for the treatment of HCC.

I.O.L. Ng reports grants from Hong Kong Research Grants Council during the conduct of the study. No disclosures were reported by the other authors.

H.W. Leung: Formal analysis, investigation, methodology, writing–original draft. C.O.N. Leung: Conceptualization, formal analysis, investigation, methodology, writing–review and editing. E.Y. Lau: Conceptualization, formal analysis, methodology. K.P.S. Chung: Data curation, investigation, methodology. E.H. Mok: Data curation, formal analysis, methodology. M.M.L. Lei: Data curation, formal analysis, methodology. R.W.H. Leung: Validation, methodology. M. Tong: Data curation, formal analysis. V.W. Keng: Data curation, methodology. C. Ma: Conceptualization, resources, supervision. Q. Zhao: Resources, validation. I.O.L. Ng: Conceptualization, resources, funding acquisition. S. Ma: Conceptualization, formal analysis, investigation. T.K. Lee: Conceptualization, resources, formal analysis, supervision, funding acquisition, investigation, project administration, writing–review and editing.

We thank the University Research Facility in Life Sciences at the Hong Kong Polytechnic University for providing and maintaining the equipment and technical support needed for flow cytometry and imaging work as well as the Centralized Animal Facility at the Hong Kong Polytechnic University for supporting our animal studies. We thank Chi-Ho Lin (University of Hong Kong) for his help to deposit the RNA-sequencing data to Gene Expression Omnibus database. This study was supported by the RGC General Research Fund (15103817, to T.K. Lee), the Theme-based Research Scheme project (T12-704/16-R, to T.K. Lee, S. Ma, and I.O.N. Ng), Project of Strategic Importance (PolyU; FAST-5, to T.K. Lee, Q. Zhao, and C. Ma), and Research Impact Fund (R5050-18F, to T.K. Lee).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.