Oncostatin M (OSM), SDF-1α and CXCL11 were purchased from Peprotech (Rocky Hill, NJ, USA). U0126 and dimethyl sulfoxide (DMSO) were purchased from Sigma-Aldrich (St. Louis, MO, USA).

Monoclonal antibodies used in the flow cytometric analysis were mouse anti-human monoclonal antibodies (mAbs) CD44-PE, CD133-APC, CD90-PE, CD24-FITC, the IgG-PE isotype, the IgG-APC isotype and the IgG-FITC isotype (all purchased from Miltenyi Biotec, Bergisch Gladbach, Germany). Antibodies used for immunofluorescence, immunoblotting and immunohistochemistry were as follows: mouse anti-human monoclonal CD90 (Abcam, Cambridge, MA, USA), rabbit anti-human polyclonal CD133 (Abnova, Walnut, CA, USA), mouse anti-human monoclonal CD44, rabbit anti-human albumin (ALB), phospho-p44/42 MAPK (Erk1/2), and rabbit anti-human HNF4α mAb (Cell Signaling Technologies, Danvers, MA, USA), rabbit anti-human polyclonal CD24, rabbit anti-human β-actin mAb (Epitomics, Burlingame, CA, USA), rabbit anti-human polyclonal transferrin (TF) (Proteintech Group Inc., Chicago, IL, USA), rabbit anti-human AFP mAb, rabbit anti-human C/EBPα and C/EBPβ mAb, Erk2/p42 MAPK antibody (Epitomics), rabbit anti-human CXCR7 IgG (Novus Biologicals, Littleton, CO, USA), and horseradish peroxidase-conjugated goat anti-rabbit IgG F(ab’)2 antibodies (Jackson ImmunoResearch, West Grove, PA, USA).

Human HCC cell lines with elevated lung metastatic potential (MHCC97L, MHCC97H, and HCCLM3) were established at the Liver Cancer Institute of Fudan University. The human HCC cell lines with low metastatic potential were SMMC-7721 (established at the Second Military Medical University), Huh7 and HepG2 (obtained from the American Type Culture Collection). These cell lines were cultured in high-glucose DMEM (Gibco-BRL, Grand Island, NY, USA) supplemented with 10% fetal bovine serum (Hyclone, Logan, UT, USA).

To induce differentiation, 10 to 40 ng/ml OSM and 1 or 5 μM dexamethasone (Dex) were used. Chemical differentiation inducer DMSO was also used to help establish a model of differentiation induction. For the CXCR7 stimulation assay, 100 ng/ml recombinant human SDF-1α and 200 ng/ml CXCL11 were used. For the inhibition assay, cells were serum-starved for at least 8 h before the MEK1/2 inhibitor U0126 (10 μM) was added.

The expression levels of cancer stem cell (CSC)-related markers were determined by flow cytometry. Briefly, tumor cells were grown to 80% confluency. After trypsin digestion, the cells were re-suspended in medium at a concentration of 1×10⁶ cells/ml and incubated with antibodies against CD44, CD133, CD90 and CD24 (diluted 1:11) at 4°C for 15 min. After washing 3 times with PBS, the cells were analyzed using a FACS flow cytometer (BD Biosciences, Franklin Lakes, NJ, USA).

Cell surface expression of CSC-related markers and CXCR7 was also determined by immunofluorescence analysis. Cells were grown on glass coverslips to 40–50% confluency and fixed, permeabilized, blocked, and incubated with primary monoclonal antibodies overnight at 4°C. Slides were washed and incubated with an anti-mouse or an anti-rabbit Cy3-conjugated secondary antibody (Jackson ImmunoResearch). Cells were counterstained with 4–6-diamidino-2-phenylindole to visualize cell nuclei and for inspection by fluorescence microscopy (Olympus).

Western blotting was performed according to the wet transfer protocol using the Bio-Rad Transfer Cell System (Bio-Rad, Ontario, Canada). Analyses of protein expression were performed according to the manufacturer’s instructions. Images were examined using Image Lab Software^® (Bio-Rad).

Small interfering RNAs (siRNAs; GenePharma Corp., Shanghai, China) demonstrated to be effective for the knockdown of CXCR7 expression (15) were used. siRNA transfection of the cells was performed using the Lipofectamine 2000 protocol (18). A negative control siRNA, 5′-UUC UCC GAA CGU GUC ACG UTT-3′, was also used. A FAM-labeled negative control siRNA was used to monitor transfection efficiency.

The Oligo GEArray Human Chemokine and Chemokine Receptors Chip (SuperArray Bioscience, Frederick, MD, USA) was used to compare profiles of the HCCLM3, MHCC97-L, and SMMC-7721 cells according to the manufacturer’s protocol. Total-RNA was extracted, isolated, and purified according to the TRIzol reagent (Invitrogen Corp., Carlsbad, CA, USA) protocol. GEArray^® Analyzer Software (SuperArray Bioscience) was used for data analysis. The microarray analysis was performed twice with similar results.

Ethical approval was obtained from the Zhongshan Hospital Research Ethics Committee, and informed consent was obtained from each patient. Data from 112 patients were retrieved from the prospectively designed database. These patients underwent hepatectomy by the same surgical team from January 2000 to May 2004. The hepatectomy for HCC was carried out as described previously (19). All patients were classified as Child-Pugh A, and all tumors were identified as HCC by histological analysis.

Regular follow-up procedures in our clinic include the following: AFP assay and liver ultrasonography every 3 months during the first year and every 6 months thereafter; and magnetic resonance imaging or computed tomography scanning after 1 month and every 6 months thereafter. Chest computed tomography scanning is regularly used to identify lung metastasis. Lung metastasis was confirmed by biopsy through endoscopy or histology after partial pulmonary resection. Until May 2009, lung metastasis was found in 56 patients. Ten patients with resectable lung metastases received partial resection of the lung.

Hematoxylin and eosin-stained slides were screened for optimal tumor content and tissue adjacent to the tumor (TAT) with a distance of 2 cm. The TMA was constructed in accordance with standard procedures (20) based on 112 tumor tissues and 46 TATs. Two cores were taken from each formalin-fixed, paraffin-embedded HCC sample by using punch cores that measured 1.0 mm in diameter from the center of the tumor foci and TAT. A two-step method of IHC including heat-induced antigen-retrieval procedure was performed as previously described (20). Detection without the primary antibody was considered as the negative control.

CXCR7 expression was defined by staining intensity and the percentage of positive tumor cells as described previously (21). Two pathologists observed the results independently. We simplified the assay results of CXCR7 into CXCR7^Low (weak staining) and CXCR7^High (strong staining). Similarly, HNF4α expression was categorized as HNF4α^Low or HNF4α^High, except that HNF4α^Low included the negatively stained population due to the generally weak staining.

When two groups of cells or tissues were compared, analysis was performed with the Student’s t-test. The Pearson χ² test was used to compare qualitative variables in the clinical pathology analysis. When expected sample values were <5, Fisher’s exact test was used. Spearman’s rank test was used to detect the correlation between variables. Overall survival (OS), time to progression (TTP), and time to extrahepatic metastasis were observed in the survival analysis. OS was calculated from the date of hepatectomy to the date of death regardless of cause. TTP was calculated from the date of hepatic resection to the date of recurrence. Time to lung metastasis was calculated from the date of hepatectomy to the date of lung metastasis with definite clinical diagnosis. The patients lost in the follow-up and the patients who had not achieved the desired results at the end of this study were recognized as censored cases. The Kaplan-Meier method was used to describe the survival curves, and the log-rank test was used to compare survival distributions between groups. The Breslow test was also used when survival curves indicated greater differences during the early follow-up period. All P-values were obtained using two-tailed tests and the statistical significance was set at 0.05. Statistical analyses were carried out by SPSS 18.0 Software (SPSS Inc., Chicago, IL, USA).

Due to the highly heterogenetic character of HCC, we first detected HCC-related stem cell biomarkers in four HCC cell lines (HepG2, SMMC-7721, Huh7 and HCCLM3) with different characteristics and backgrounds. Flow cytometric analysis indicated that Huh7 cells had the highest ratios of CD133 (72.6%) and CD24 (98.8%), which was confirmed by western blot and immunofluorescence analyses (Fig. 1). HCCLM3 cells had the highest CD90-positive ratio (3.2%) and a CD44-positive ratio of >90%. Since these CSC-related markers have been linked to HCC progression and dysregulated differentiation potential (22–24), Huh7 and HCCLM3 cells were selected for further analysis.

The differentiation induction model for HCC was established using potential differentiation inducers. DMSO was used as a chemical inducer of differentiation. OSM, which has the ability to maintain the maturation and differentiation capacity of hepatocytes (25), was also used. Both in the Huh7 and in the HCCLM3 cells, DMSO showed a strong ability to induce HCC differentiation into normal hepatocytes, based on hepatocyte differentiation markers or elevated expression of plasma proteins ALB and TF, and the typical morphological features of differentiated hepatocytes (Fig. 2A and B). The typical morphology of differentiation was maintained from day 2 until late stages of the induction process. Meanwhile, results from HCCLM3 cells revealed that OSM had the ability to induce elevated expression of ALB and TF (Fig. 2C). However, no typical morphological features were noted, which indicated the main role of OSM in maintaining the maturation of cells. Dex was unable to induce differentiation when used alone (data not shown). In addition, AFP was also increased early during induction, which was consistent with previously reported results (26). Next, AFP levels decreased gradually accompanied by the elevated ALB and TF, and the occurrence of hepatocyte-like morphology.

Based on the model of HCC differentiation in vitro, we further observed CXCR7 levels following induced differentiation by 1% DMSO and 40 ng/ml OSM. After DMSO treatment, CXCR7 expression was increased early and was decreased during the late stages of differentiation, which showed an inverse expression trend to HNF4α, which has a critical role in liver-specific gene expression. As shown in Fig. 3A, during early stages of induced differentiation, CXCR7 was elevated and the HNF4α level was decreased in both Huh7 and HCCLM3 cells. Whereas during the late stage, CXCR7 levels were decreased and HNF4α levels were elevated. In addition, CXCR7 levels were elevated while the opposite trend was observed for HNF4α during the early stage of differentiation in response to 40 ng/ml OSM treatment (Fig. 3B).

Furthermore, IHC analysis of the TMA was performed to confirm the relationship between CXCR7 and HNF4α. Immunopositivity for CXCR7 was mainly observed at the membrane or in the cytoplasm of the HCC cells, whereas for HNF4α it was expressed mainly in the nucleus. The expression intensity and localization of CXCR7 and HNF4α are shown in Fig. 4A and B, respectively. Immunostaining of normal hepatocytes indicated the specificity and selectivity of anti-HNF4α. Stratum analysis based on cell differentiation (Edmondson grade 1/2; n=57) indicated that CXCR7 was negatively correlated with HNF4α expression (Spearman’s rho=−0.296; P=0.025). Among the tumors with low HNF4α expression (n=48), 62.5% were CXCR7^High, whereas only 22.2% of the tumors with high HNF4α expression (n=9) were CXCR7^High. Furthermore, immunostaining analyses indicated that the distribution and expression levels of CXCR7 were inversely associated with those of HNF4α on a relatively well-differentiated background (Fig. 4C). In addition, CXCR7^HighHNF4α^Low tumors tended to be larger (Fisher’s exact test, P=0.002), suggesting a relationship between differentiation and proliferation of HCC (Table I).

There was a statistically significant difference in OS between the CXCR7^High and CXCR7^Low groups on the HNF4α^Low background (log-rank test; P=0.032). In particular, there was a strong difference during the early stages of follow-up (Breslow test; P=0.007; Fig. 5A). The median OS of CXCR7^High patients was much shorter than that of CXCR7^Low patients (36 vs. 81.2 months). Similarly, on the HNF4α^Low background, the TTP of the CXCR7^High patients was significantly shorter than that for the CXCR7^Low patients (log-rank test; P=0.038), particularly during the early stages of follow-up (Breslow test; P=0.008; Fig. 5B). In addition, the time that relapsed before observing lung metastasis was shorter for CXCR7^High patients than for CXCR7^Low patients (log-rank test; P=0.007) (Fig. 5C). In addition, in patients with high HNF4α expression in tumors, there was no significant difference between CXCR7^High patients and CXCR7^Low patients (data not shown).

Since there was a greater correlation between CXCR7 and HNF4α in relatively well-differentiated HCC, the prognostic value of CXCR7 was further evaluated in patients with Edmondson grade 1/2 HCC (n=48) (Fig. 5D and E). We found that CXCR7 expression had a strong prognostic value in the time-to-metastasis analysis (log-rank test; P=0.004; Fig. 5F).

Since CXCR7 is an atypical G-protein-coupled receptor that has been reported to mediate extracellular regulated protein kinase (ERK) signaling, we explored the possibility of a CXCR7-ERK-HNF4α pathway. Gene chip analysis indicated that HCCLM3 cells highly expressed CXCR7. CXCR4 and CXCR3, however, were nearly undetectable (Fig. 6). Therefore, we utilized HCCLM3 cells as the CXCR7⁺CXCR4⁻CXCR3⁻ model to observe CXCR7 signaling stimulated by SDF-1α and CXCL11. Phosphorylation of MAPKs ERK1/2 was decreased at 1, 8 or 16 h, and was accompanied by increased HNF4α expression (Fig. 7A), which indicated suppression of the ERK pathway by HNF4α. Then, following stimulation with SDF-1α or CXCL11, elevated phosphorylation of ERK1/2 was observed (Fig. 7B). Meanwhile, western blotting indicated elevated HNF4α levels after exposure to U0126 (Fig. 7C), which indicated that the use of U0126 abrogated the inhibitory effect of CXCR7 ligands on HNF4α expression. Furthermore, after downregulating CXCR7 in the HCCLM3 cells, HNF4α was increased in all positive siRNA groups (Fig. 7D). However, another hepatic-enriched nuclear factor C/EBPα was not obviously affected. Another cell line, MHCC97H, with the same genetic background as HCCLM3 and also expressing high levels of CXCR7 and extremely low levels of CXCR4 and CXCR3 was used to confirm these results. The results with this cell line were similar to those using the HCCLM3 cell line.