We retrospectively screened 243 patients (200 male and 43 female), with a median age of 63 years (range 35–82 years), who underwent hepatic resection for HCC between January 1997 and December 2006 at our institution. After surgery, patients were followed up by monitoring dynamic computed tomography and/or magnetic resonance imaging and tumor markers every 3 months on an outpatient basis. Combined examination of tumor markers and imaging studies was used for diagnosis of recurrence of HCC. We reviewed the medical records of all patients for clinical information, including sex, age, markers of hepatitis B virus and hepatitis C virus (HCV), serum albumin, serum a-fetoprotein (AFP), and protein induced by vitamin K absence or antagonist(PIVKA)II. We reviewed tumor size, tumor number, vascular invasion, lymph node metastasis, and the pathological findings of background liver. We used the criteria of the Liver Cancer Study Group of Japan (6th edition) to determine tumor-node-metastasis (TNM) stage of HCC (16). Informed consent of patients between 1997 and 2000 was obtained in the form of opt-out on the web site of Hokkaido University Hospital and patients between 2001 and 2006 signed the written informed consent. This research was approved by the Institutional Review Board of our institution (017–0237) and performed in compliance with the Declaration of Helsinki.

Four-micrometer-thick sections of formalin-fixed and paraffin-embedded specimens were used for immunohistochemical staining. They were deparaffinized using xylene and ethanol, and antigen retrieval was performed using Target Retrieval Solution (pH 9.0; 415211; Nichirei Biosciences Inc., Tokyo, Japan), heated for 30 min at 95°C. The samples were incubated with Block Ace (UKB80; KAC Co., Ltd.) for 5 min to block nonspecific antibody reactions and incubated overnight at 4°C with anti-Wnt5a antibody (LS-C47384, diluted 1:2,000; LifeSpan BioSciences Inc.). The samples were incubated in Histofine Simple Stain MAX PO (MULTI; 724152; Nichirei Biosciences) for 30 min at room temperature. Immunohistochemical staining was visualized using 3,3′ diaminobenzidine, and sections were counterstained with hematoxylin. Immunoreactivity was evaluated according to the distribution of positive cells. Wnt5a-positive cells were defined according to the immunoreactivity on the cell membrane, regardless of cytoplasmic immunoreactivity. The immunohistochemical staining pattern of Wnt5a in HCC was heterogeneous and Wnt5a positivity was recorded if the proportion of Wnt5a-positive cells was >50% (Fig. 1). Two authors who were blinded to the clinical and pathological parameters evaluated the results of immunohistochemical staining.

Human liver cancer cell lines HLE, HLF, HepG2 and Huh7 were obtained from the Japanese Collection of Research Bioresources Cell Bank (Osaka, Japan). They were authenticated by short tandem repeat profiling. They were cultured in Dulbecco's modified Eagle's medium containing 10% fetal bovine serum (FBS). The medium was replaced every second day and all cell cultures were incubated at 37°C in 5% CO₂.

To establish Wnt5a knockdown cells, we used a lentiviral short hairpin RNA (shRNA) targeting Wnt5a (TL320572; OriGene). Lenti-X HTX Packaging System (631251; Takara Bio Inc.) was used for transfection into cell lines. To create overexpressing cells, we obtained the full-length cDNA encoding human Wnt5a from cDNA of Huh7 cells by reverse transcription polymerase chain reaction. The Wnt5a primers were: Forward, 5′-CAGTGTGGTGGAATTGCCACCATGAAGAAGTCC-3′ and reverse, 5′-GATATCTGCAGAATTCTACTTGCACACAAACTGG-3′. The cDNA was cloned into pcDNA3.1 (V79020; Thermo Fisher Scientific, Tokyo, Japan). HLF cells were seeded in cell culture dishes, and after reaching 70% confluence, they were transfected with pcDNA3.1-Wnt5a overexpression vector. Lipofectamine 2000 (Invitrogen; Life Technologies) was used for cell transfection. The medium was replaced after 48 h with G418-containing medium, and stably transfected cells were selected using antibiotic resistance preferentially.

Cells were cultured to reach 80% confluence and harvested using lysis buffer on ice. A total of 10 µg of each lysate was run on sodium dodecyl sulfate-polyacrylamide gel electrophoresis and transferred to polyvinylidene difluoride membranes. After blocking with 3% bovine serum albumin, membranes were immunoblotted using primary antibodies against Wnt5a (ab174963, diluted 1:500; Abcam); glyceraldehyde 3-phosphate dehydrogenase (GAPDH; #3683, diluted 1:1,000; Cell Signaling Technology); E-cadherin (#3195, diluted 1:1,000; Cell Signaling Technology); ZO-1 (#8193, diluted 1:1,000; Cell Signaling Technology); N-cadherin (ab76011, diluted 1:5,000; Abcam); and vimentin (#5741, diluted 1:1,000; Cell Signaling Technology). The blots were then reacted with secondary anti-rabbit antibodies (#7074, diluted 1:5,000; Cell Signaling Technology), followed by detection of the proteins with SuperSignal West Dura Extended Duration Substrate (34076; Thermo Fisher Scientific).

Cells were seeded at 5,000/well in 96-well culture plates. We determined the number of viable cells by a colorimetric method using CellTiter 96 (G5430; Promega Corporation). The measurement was performed on 0, 1, 2 and 3 days after seeding.

Cell invasion assays were performed in Matrigel invasion chambers (24 wells, 8 µm; 354480; Corning). Cells (2.5×10⁴) in FBS-free medium were seeded in the upper chambers and medium containing 10% FBS was added to the lower chambers. Cells above the membrane were wiped off using a cotton swab after 24 h. Membranes were stained and average values were obtained by counting five fields per membrane under a microscope (×10).

Purification of total RNA from HCC cell lines was performed using RNeasy Mini Kit (74106; Qiagen). Comprehensive analysis of gene expression was performed using SurePrint G3 Human 8×60 K version 3.0 (Agilent) at Hokkaido System Science Co., Ltd. The fold changes in normalized signal values were used for comparison of gene expression levels.

Statistical analysis was performed using EZR version 1.35 (Saitama Medical Center, Jichi Medical University, Saitama, Japan) (17). Fisher's exact test was used to examine the correlation between expression of Wnt5a and clinical and pathological variables. The Kaplan-Meier method was used to plot OS and relapse-free survival (RFS) curves and they were compared using the log-rank test. The Cox proportional hazards regression model was used to perform multivariate analyses. Values in vitro were examined using Student's t-test. P<0.05 was considered statistically significant.

Clinical and pathological characteristics are summarized in Table I. Wnt5a expression was positive in 63 patients (25.9%) and negative in 180 (74.1%).

Correlations between clinical and pathological characteristics and Wnt5a expression are shown in Table II. Wnt5a negativity was significantly associated with HCV (P=0.011), poor tumor differentiation (P=0.001) and positive vascular invasion (P=0.046).

The 5-year OS rate in the Wnt5a-positive and Wnt5a-negative groups was 79.7 and 66.6%, respectively. OS in the Wnt5a-negative group was significantly poorer than in the Wnt5a-positive group (P=0.018). The 5-year RFS rate in the Wnt5a-positive and Wnt5a-negative groups was 43.7 and 30.7%, respectively. RFS in the Wnt5a-negative group tended to be poorer than in the Wnt5a-positive group, but there was no significant difference between the groups (Fig. 2A and B).

Univariate analysis showed that the following were predictive factors for lower OS in HCC patients: Albumin level [hazard ratio (HR) 1.826, 95% confidence interval (CI) 1.204–2.770, P=0.005]; AFP level (HR 1.668, 95% CI 1.090–2.553, P=0.018); PIVKAII level (HR 1.570, 95% CI 1.002–2.459, P=0.049); tumor number (HR 1.824, 95% CI 1.134–2.932, P=0.013); tumor size (HR 2.110, 95% CI 1.384–3.218, P=0.001); vascular invasion (HR 2.630, 95% CI 1.645–4.205, P<0.001); cirrhosis (HR 1.721, 95% CI 1.130–2.623, P=0.012); and Wnt5a negativity (HR 1.993, 95% CI 1.109–3.368, P=0.020). Multivariate analysis showed that the following were predictive factors for lower OS: Albumin level (HR 1.571, 95% CI 1.004–2.458, P=0.048); tumor number (HR 1.965, 95% CI 1.292–3.239, P=0.008); tumor size (HR 1.829, 95% CI 1.100–3.041, P=0.020); vascular invasion (HR 2.256, 95% CI 1.317–3.865, P=0.003); cirrhosis (HR 1.675, 95% CI 1.075–2.610, P=0.023); and Wnt5a negativity (HR 1.939, 95% CI 1.076–3.497, P=0.028; Table III).

At the time of our investigation, 171 of the 243 patients (70.4%) had experienced recurrence. This included 133 of the 180 Wnt5a-negative patients (73.9%) and 38 of the 63 Wnt5a-positive patients (60.3%). The recurrence patterns and their management are summarized for each group (Table SI). The recurrence location did not differ significantly between the groups. There were also no significant differences in tumor number and diameter. The management strategies for recurrence varied, including transcatheter arterial chemoembolization, radiofrequency ablation, microwave coagulation therapy, percutaneous ethanol injection therapy, liver resection, metastatic resection, chemotherapy and radiotherapy. There was no significant difference in recurrence management between the Wnt5a-positive and Wnt5a-negative groups.

Prognosis after recurrence is shown in Fig. 2C and D. The 5-year OS after treatment of recurrence was 67.9% in the Wnt5a-positive group and 43.5% in the Wnt5a-negative group. OS after recurrence in the Wnt5a-negative group was significantly poorer than in the Wnt5a-positive group (P=0.026). Median RFS after treatment of recurrence was 10 months in the Wnt5a-positive group and 8 months in the Wnt5a-negative group. There was no significant difference in RFS after treatment of recurrence (P=0.576).

Analysis using clinical samples revealed that Wnt5a expression was related to tumor differentiation, vascular invasion, and prognosis. Therefore, we used different cell lines to clarify the mechanism. We evaluated expression of Wnt5a in liver cancer cell lines using western blotting. We chose HLE and HLF as poorly differentiated cell lines, Huh7 as a well-differentiated cell line, and HepG2 as a hepatoma/liver cancer cell line. The intensity ratio of expression to GAPDH was 20.2% in HLE cells, 23.5% in HLF cells, 40.2% in HepG2 cells and 85.5% in Huh7 cells (Fig. S1). Wnt5a expression was lower in HLE and HLF cells than in HepG2 and Huh7 cells.

Huh7 cells had high expression of Wnt5a, and we examined changes in proliferation and invasiveness of Huh7 cells by knockdown of Wnt5a. The cell proliferation assay showed that knockdown of Wnt5a significantly increased proliferation of Huh7 cells (P=0.043; Fig. 3A). The invasion assay showed that knockdown of Wnt5a significantly increased invasiveness of Huh7 cells (P=0.009; Fig. 3B).

We examined changes in proliferation and invasiveness caused by overexpression of Wnt5a on HLF cells, which have low expression of Wnt5a. Overexpression of Wnt5a did not change proliferation of HLF cells (Fig. 3C) in the cell proliferation assay. Overexpression of Wnt5a significantly decreased invasiveness of HLF cells in the cell invasion assay (P=0.003; Fig. 3D).

To elucidate the molecular mechanism involved in changes in invasiveness induced by Wnt5a, changes in EMT-related molecules were examined using western blotting (Fig. 4). We investigated changes in the expression of E-cadherin and ZO-1 as epithelial markers and N-cadherin and vimentin as mesenchymal markers. Knockdown of Wnt5a in Huh7 cells with high expression of Wnt5a decreased E-cadherin expression, and there was no change in ZO-1, N-cadherin, or vimentin expression. Overexpression of Wnt5a in HLF cells with low expression of Wnt5a decreased vimentin expression, and there was no change in E-cadherin, ZO-1 or N-cadherin expression.

To clarify the relationship between Wnt5a and EMT, we investigated changes in EMT-related genes with Wnt5a overexpression in HLF cells using DNA microarray analysis. The changes in SNAIL, ZEB and TWIST families as EMT transcriptional repressors are shown in Fig. S2. Expression of SNAI2 and SNAI3 was decreased by overexpression of Wnt5a. The changes in the matrix metalloproteinase (MMP) family are shown in Fig. S3. Expression of MMP2, MMP7, MMP9, MMP10 and MMP23B was decreased by overexpression of Wnt5a.