High expression FUT1 and B3GALT5 is an independent predictor of postoperative recurrence and survival in hepatocellular carcinoma

Cancer may arise from dedifferentiation of mature cells or maturation-arrested stem cells. Previously we reported that definitive endoderm from which liver was derived, expressed Globo H, SSEA-3 and SSEA-4. In this study, we examined the expression of their biosynthetic enzymes, FUT1, FUT2, B3GALT5 and ST3GAL2, in 135 hepatocellular carcinoma (HCC) tissues by qRT-PCR. High expression of either FUT1 or B3GALT5 was significantly associated with advanced stages and poor outcome. Kaplan Meier survival analysis showed significantly shorter relapse-free survival (RFS) for those with high expression of either FUT1 or B3GALT5 (P = 0.024 and 0.001, respectively) and shorter overall survival (OS) for those with high expression of B3GALT5 (P = 0.017). Combination of FUT1 and B3GALT5 revealed that high expression of both genes had poorer RFS and OS than the others (P < 0.001). Moreover, multivariable Cox regression analysis identified the combination of B3GALT5 and FUT1 as an independent predictor for RFS (HR: 2.370, 95% CI: 1.505–3.731, P < 0.001) and OS (HR: 2.153, 95% CI: 1.188–3.902, P = 0.012) in HCC. In addition, the presence of Globo H, SSEA-3 and SSEA-4 in some HCC tissues and their absence in normal liver was established by immunohistochemistry staining and mass spectrometric analysis.

Liver cancer is the sixth most common cancer and the third leading cause of cancer-related deaths worldwide. The majority of primary liver cancers are hepatocellular carcinoma (HCC), which arises from hepatocytes. Hepatitis B virus (HBV) infection and aflatoxin B exposure are the two critical pathogenic factors leading to HCC in Eastern Asia and sub-Saharan Africa. In contrast, hepatitis C virus (HCV) infection is the major risk factor in Northern America, Europe and Japan 1-3 . Patients with advanced unresectable HCC had dismal outcomes with estimated survival rates of 17.5% at 1 year and 7.3% at 2 years 4 . Although patients with early-stage HCC are amenable to treatment by surgical resection, nonsurgical ablation procedures or liver transplantation, the overall survival remains poor because of high recurrence rate. Patients who received surgical resection had approximately 70% recurrence rate at 5 years 5 . Thus, identification of biomarkers for predicting relapse of patients with HCC is important. So far, vascular invasion, large tumor size and tumor multiplicity have been considered as poor prognostic factors for HCC recurrence after surgical treatments 6 . With the use of a variety of molecular profiling, including oligonucleotide microarray, quantitative reverse-transcription polymerase chain reaction (qRT-PCR), mass spectrometry, several biomarker candidates associated with HCC recurrence have been reported 7 . However, none of the candidate biomarkers have been put into clinical practice.
To date, no study has focused on the prognostic values of glycan structures or genes involved in their biosynthesis, even though aberrant glycosylation on glycoproteins or glycolipids is a common feature of cancer cells.

Results
Expression of FUT1, FUT2, B3GALT5 and ST3GAL2 in HCC tissues. Since cancer cells may arise from maturation arrest of immature stem cells or from dedifferentiation of mature cells 29 , it is likely that Globo H and its related biosynthetic enzymes expressed on endodermal cells might also be expressed on liver cancer cells. We examined the mRNA expression levels of the FUT1, FUT2, B3GALT5 and ST3GAL2 in tumor specimens of 135 HCC patients by qRT-PCR. Their clinical characteristics and demographic information were summarized in Table 1. The mean age was 57.7 ± 13.1 (range: 19-87) with 82.2% being male. All except 5 of 135 patients had chronic hepatitis infection with a majority (70.4%) infected by HBV. Six patients were infected by both HBV and HCV and remaining were infected by HCV. Liver cirrhosis was found in 59 cases (43.7%). Tumor sizes ranged from 0.8 to 16 cm with a mean size of 5.4 ± 3.9 cm, but less than 5 cm in 65.2%. Furthermore, 74 patients (54.8%) had histological grade 1-2 tumors and 66 patients (48.9%) showed evidence of vascular invasion in their tumors. According to the tumor-node-metastasis (TNM) system, 102 patients (75.6%) presented with TNM I + II stages. Thirteen patients (9.6%) had tumor metastasis at diagnosis. The median follow-up time was 58.6 months (range, 1.4 to 126.24 months). At the time of this report, 47 patients (34.8%) died and the 1 y, 3 y and 5 y overall survival rate was 87.2%, 73.7% and 66.4%, respectively. All deaths were related to HCC except three patients who died of other diseases. The results of qRT-PCR analyses of the 4 globoside biosynthetic genes in these samples were shown as −ΔCT, after subtracting the average of 2 reference genes, GAPDH and GUSB (Fig. 1A). The mean expression level was −6.05 ± 1.71 (−10.34 to −1.70) for FUT1, −7.02 ± 3.74 (−16.67 to 0.92) for FUT2, −7.48 ± 2.85 (−13.53 to 0.25) for B3GALT5 and −3.28 ± 1.02 (−5.83 to −0.53) for ST3GAL2. The expression levels of these 4 genes varied greatly among patients (ANOVA, P < 0.001) and ST3GAL2 expression was significantly higher than the other three genes.
Correlation of clinicopathological parameters with the expression levels of FUT1, FUT2, B3GALT5 and ST3GAL2. To explore the clinical relevance of the expression levels of these 4 globoside glycosyltransferases, we used the area under curve (AUC) of receiver operating characteristic (ROC) curve to evaluate their predictive values for disease recurrence. Among 135 patients, 4 patients who had persistent disease were excluded from relapse analyses, and 83 of 131 patients (63.4%) had disease recurrence. Since the date of relapse was not available in 2 patients and one patient in remission died of unrelated cause, the ensuing RFS analyses were conducted in 128 patients. As shown in Fig. 1B, the AUCs for FUT1, FUT2, B3GALT5 and ST3GAL2 were 0.591, 0.512, 0.585 and 0.557, respectively. Based on these findings, we further analyzed the prognostic significance of their expression levels in HCC, using Youden index to determine the optimal cutoff values defining high and low expression groups. The best cutoff value for predicting recurrence of HCC was −6.433 for FUT1, −7.633 for FUT2, −7.724 for B3GALT5 and −3.652 for ST3GAL2. The association of expression of each of the four genes with clinical pathological parameters in 135 patients with HCC was analyzed ( Table 2). The expression levels of FUT1 were significantly correlated with TNM stage (P = 0.002), tumor number (P = 0.001), vascular invasion (P = 0.039), metastasis (P = 0.011) and relapse (P = 0.008). Patients with High expression of FUT1 were at higher risk than those with low expression of FUT1 for having TNM III + IV stages (OR: 4. 16 (Fig. 2). Patients with high expression of ST3GAL2 tend to have shorter RFS, albeit the difference did not reach statistical significance (P = 0.063). Kaplan-Meier analyses of overall survival (OS) showed that patients with high expression of B3GALT5 had significantly poorer OS than those with low expression (P = 0.017), whereas the levels of FUT1 and ST3GAL2 did not significantly correlate with OS ( Fig. 3).
To further evaluate the joint effects of FUT1 and B3GALT5, patients were stratified into three groups: high for both, low for both and the others, for Kaplan-Meier survival analysis. The RFS and OS were not statistically different between the low for both group and others (P = 0.303 and 0.195, respectively) (supplementary Fig. 1s). Thus, these two groups were combined into one for comparison with the high for both group. As shown in Fig. 2D, patients with high expression of FUT1 and B3GALT5 combined had significantly lower rate of RFS than the remaining group (P < 0.001), at 1 year (46.0% vs. 79.2%), 3 years (23.7% vs. 60.2%), and 5 years (21.3% vs. 52.9%). In addition, these patients also had significantly shorter OS than the remaining group (P < 0.001) at 1 year (78.4% vs. 92.7%), 3 years (56.4% vs. 84.8%) and 5 years (47.8 vs. 78.6%) (Fig. 3D).
High expression of FUT1 and B3GALT5 combined is an independent prognostic factor for HCC relapse and overall survival. To evaluate the potential value of the expression levels of FUT1, B3GALT5 and ST3GAL2 for predicting HCC recurrence and OS, univariate Cox proportional hazard regression analyses were conducted. The results indicated that RFS correlated with the patient older than 55-year (HR: 1.960, 95% CI: 1.233-3.126, P = 0.005), tumor size greater than 5 cm (HR: 3.113, 95% CI: 1.975-4.906, P < 0.001), presence of metastasis (HR: 3.632, 95% CI: 1.891-6.973, P < 0.001), and presence of vascular invasion (HR: 3.317, 95% CI: 2.098-5.244, P < 0.001). In addition, RFS duration of HCC patients was significantly associated with high expression of FUT1 (HR: 1.698, 95% CI: 1.067-2.701, P = 0.025) and B3GALT5 (HR: 2.082, 95% CI: 1.313-3.302, P = 0.002). Similarly, the above-mentioned factors except FUT1 also significantly correlated with OS (Table 3). Next, to identify the independent variables associated with poor RFS and OS, we selected those important covariates which showed statistical significance in the univariate analysis for multivariable Cox regression analysis in a stepwise manner. As shown in Table 3 in addition to age ≥55, tumor ≥5 cm and vascular invasion, the expression levels FUT1 and B3GALT5 combined were independent risk factors for both RFS (HR: 2.370, 95% CI:   Next, we performed IHC to examine the expression of SSEA-3, -4 and Globo H in normal liver tissues obtained from two trauma patients. None of these three globosides were detected by IHC staining of normal liver, although weak non-specific granular signals were observed for SSEA-3 staining (Fig. 5A). This was further confirmed by MALDI-TOF MS analyses of the GSLs extracted from the same tissues, showing that the majority of GSLs in the neutral fraction were lactosylceramide (LacCer), globotriaosylceramide (Gb3) and globoside (Gb4)/ neolactotetraosylceramide (nLc4) (Fig. 5B). No signal for Globo H or SSEA-3 was detected. In the acidic fraction,

Discussion
In this study, we provide the first evidence that expression level of FUT1, B3GALT5 and ST3GAL2, the enzymes catalyzing the synthesis of three globosides, Globo H, SSEA-3 and SSEA-4, was associated with adverse clinical features. Greater expression of either FUT1 or B3GALT5 in HCC tissue was associated with poor prognosis. More importantly, high expression of these two genes combined emerged as an independent predictor for RFS and OS in HCC.
Overexpression of FUT1 had been reported in colon cancer 32 and shown to confer resistance to serum starvation-induced cell death 33 . Since serum starvation elicits autophagy, the protection conferred by FUT1 may be explained by our recent findings that FUT1 impedes perinuclear localization of lysosomes via fucosylation of LAMP1 and LAMP2, thereby preventing autophagic death 34 37 . Also, data mining of breast cancer by Milde-Langosch et al. showed that high expression of FUT1 was a poor prognostic factor for breast cancer and associated with distant metastasis 38 . These studies were in line with our findings that high expression of FUT1 in HCC was associated with vascular invasion and multiple tumor nodules. On the other hand, Mathieu et al. reported that introducing FUT1 into HepG2 cells inhibited angiogenesis and tumor growth 39 . However, the HepG2 cell is a hepatoblastoma 40 , which is an embryonal malignancy of hepatocellular origin and has different clinical features distinctive from HCC. Such differences may account for the differential roles of FUT1 in hepatoblastoma HepG2 cell and HCC.
Besides FUT1, we also showed high expression of B3GALT5 in HCC tissues to be associated with advanced TNM stage, metastasis, vascular invasion and tumor recurrence. Among the 6 members of beta 1-3 galactosyltransferase family, B3GALT5 has broadest range of acceptors. In addition to transferring a galactose to the terminal GalNAc of GB4 to generate SSEA-3, B3GALT5 is also responsible for the synthesis of type 1 chain (Galβ1-3GlcNAc), which is the precursor of Lewis A and sialyl Lewis A (SLe a ), the latter is also known as CA19-9 antigen. Along this line, the expression of B3GALT5 was shown to be correlated with CA19-9 in pancreatic cell lines and cancer tissues 41,42 . SLe a is a ligand for selectins and mediates adhesion of cancer cell to endothelial cells and promotes tumor progression 43 . Similarly Chung et al. 44 reported that overexpression of HBV X protein in Chang cell enhanced SLe a expression and mediated cell adhesion to endothelial cell and metastasis through Moreover, we found that higher expression of ST3GALT2 was associated with the number of HCC tumor nodules and high histological grades. The molecular mechanisms by which ST3GAL2 directly or indirectly contributes to tumor progression are not clear. SSEA-4 is a well-known product generated by ST3GAL2. Although the biological function of SSEA-4 is still unclear, it has been identified as a marker for human embryonic stem cells as well as cancer stem cells in breast cancer 45 . In addition, Yang et al. reported that knockdown of ST3GAL2 significantly reduced the phosphorylation of ERK1/2 and EGFR in the osteoblasts 46 . Since EGFR pathway is known to contribute to the inflammation in the liver and hepatocarcinogenesis 47 , ST3GAL2 might promote tumorigenesis through EGFR pathway. Saito et al. reported higher expression of ST3GAL2 in renal cell carcinoma tissues than in non-tumor kidney, suggesting that ST3GAL2 might be involved in the malignant processes 12 . In addition, Aloia et al. identified a subpopulation of breast cancer cells carrying high levels of SSEA-4 to be resistant to chemotherapy drugs and showed ST3GAL2 to be a marker of poor outcome in breast cancer and ovarian cancer patients undergoing chemotherapy 48 . These reports are in line with our finding that high expression of ST3GAL2 is associated with HCC progression.
In this study, we provided the first evidence for SSEA-3 expression in HCC by immunohistochemistry and mass spectrometry. We also confirmed the presence of both SSEA-4 and Globo H in HCC tumor cells by IHC staining and mass spectrometric analysis of GSLs from HCC tumor tissues. Furthermore, none of these three globosides were found in normal liver tissues by either IHC or mass spectrometry. There are only a few studies addressing the biological functions of Globo H, SSEA3 and SSEA-4. Sorting of SSEA3 + cells from breast cancer stem cell subpopulation of MCF-7 and MDA-MB-231 cell lines further enhanced their tumor initiating capacity 49 . In addition, clustering of SSEA-4 by the monoclonal antibody promotes invasion ability of the breast cancer cell line through activation of cSrc and FAK 50 . Apart from these studies, the functions of SSEA-3 and SSEA-4 remain an enigma. Previously, our studies demonstrated that Globo H could promote angiogenesis through its interaction with Translin-Associated Factor X (TRAX), with consequent release and activation of phospholipase C β1 from TRAX to trigger Ca 2+ mobilization 51 . In addition, Globo H acted as an immune checkpoint to facilitate the escape of tumor cells from immune surveillance by suppressing the activation of B-cell and T-cell 52 . These findings provided strong rationales for Globo H-targeted immunotherapy of cancer. Indeed, clinical trials of Globo H vaccine have generated promising results 53 .
In the present study, we found that high expression of either FUT1 or B3GALT5 was a significant risk factor for HCC relapse and high expression of B3GALT5 correlated with OS, but neither alone was an independent predictor for HCC recurrence or OS. Notably, multivariable analyses showed that a combination of high expression of FUT1 and B3TALT5 is an independent risk factor for recurrence and OS, in addition to vascular invasion and tumor size greater than 5 cm. The latter two were consistent with reports in many studies 6 . We are currently  investigating the expression of Globo H, SSEA-3 and SSEA-4 in more HCC samples, in relation to these three glycosyltransferases and their clinical relevance. The risk factors and incidence of HCC vary greatly according to geographic regions and ethnicities. The patients in our study were Taiwanese with HCC which were mostly (96.3%) associated with viral hepatitis. Thus, it remains to be determined if our finding of high expression of combined FUT1 and B3GALT5 as an independent risk factor for postoperative recurrence and OS in patients with resectable HCC will apply to HCC patients from different ethnicities or different etiologies, such as alcoholic and nonalcoholic steatohepatitis. In conclusion, our finding of combined high expression of FUT1 and B3TALT5 as an independent risk factor for postoperative recurrence and OS of HCC represents a simple and valuable addition to the existing clinicopathological parameters for the prognostication of HCC. Surgical resection is the main stay of HCC treatment. However, high incidence rate of recurrence and lack of simple and reliable biomarkers for predicting postoperative recurrence is still a challenge. If confirmed in large cohort study, our findings provide an important armamentarium to fill the void. Furthermore, we have shown here that Globo H, SSEA-3 and SSEA-4 were expressed in HCC tissues but not in normal liver tissue. Thus, these GSLs and glycosyltransferase could be potential therapeutic targets for HCC.

Materials and Methods
Clinical specimens. Complementary DNA (cDNA) or total RNA extracted from surgically resected tumors, along with relevant clinical and pathological data from 135 patients with American Joint committee on Cancer HCC stage I to IV were obtained from Linkou Chang Gung Memorial Hospital and Taiwan Liver Cancer Network. Snap-Frozen HCC tissues used for GSL extraction were obtained from Tissue Bank of Tri-Service General Hospital, Scale bars indicate 60 μm (original magnification 40X) (B) Total GSLs extracted from case 1 were separated into neutral and acidic fractions, permethylated and analyzed by MALDI-TOF mass spectrometry. The major GSLs in the neutral fraction were LacCer, Gb3, Gb4/nLc4, and GM3, GD3 and GM1a/b were found in the acidic fraction. GSLs with the same glycan moiety but with different fatty acyl contents were bracketed.
Taipei, Taiwan. Normal liver tissues and sections were obtained from trauma patients from the Tissue Bank of Linkou Chang Gung Memorial Hospital, Taoyuan, Taiwan. Informed consent was obtained from all subjects before their tissues were deposited. All methods were carried out in accordance with relevant guidelines and regulations and approved by Institutional Review Board of Chang Gung Medical Foundation, Institutional Review Board of Tri-Service General Hospital, and Biobank Ethics Committee of National Health Research Institutes.
Immunohistochemistry. The primary antibodies used included the following: mAb VK9 for Globo H, (hybridoma provided by Dr. Govindaswami Ragupathi, Memorial Sloan-Kettering Cancer Center, New York, NY), anti-SSEA-4 (MC-813-70, R&D Systems, Minneapolis, MN) and anti-SSEA-3 (MC-631, R&D Systems, Minneapolis, MN). IHC analysis was performed on formalin-fixed, paraffin-embedded tissue. Sections (3 μm) on coated slides were deparaffinized and rehydrated, then subjected to antigen retrieval by autoclave at 121 °C for 5 minutes in antigen Retrieval AR-10 solution (BioGenex, Fremont, CA) for SSEA-3 and Globo H or in Target Retrieval Solution (Dako, Glostrup, Denmark) for SSEA-4. Sections were treated with H 2 O 2 and then incubated with primary antibodies at 4 °C overnight, followed by staining with Super Sensitive Polymer-HRP Detection System (BioGenex, Fremont, CA) and developed by DAB substrate (BioGenex, Fremont, CA). Slides were counter-stained with Harris' hematoxylin and mounted. Sections were examined by pathologists and digital images were captured by Aperio Scope AT Turbo Slide Scanner (Leica Biosystems) at 40X magnification.
Quantitative reverse-transcription real-time polymerase chain reaction. Total RNA (1 μg) was converted to cDNA using high capacity cDNA reverse transcription kit (Applied Biosystems, CA, USA) according to the manufacturer's instructions. The expression levels of FUT1, FUT2, B3GALT5 and ST3GAL2 were determined by using TaqMan Gene Expression Master Mix on an Applied Biosystems 7500 fast real-time PCR system. Glyceraldehyde-3-phosphate dehydrogenase (GAPDH) and glucuronidase beta (GUSB) were used as endogenous controls. The TaqMan probes for the detection of FUT1 (Hs00355741_m1), FUT2 (Hs00382834_m1), B3GALT5 (Hs00707757_s1), ST3GAL2 (Hs00911835_m1), GAPDH (Hs99999905_m1) and GUSB (Hs99999908_m1) were purchased from Applied Biosystems. Ten nanograms of cDNA sample were used for the qRT-PCR reaction following the manufacturer's instructions. The fluorescent signals were analyzed by 7500 software v2.06.

Matrix-assisted laser desorption ionization-time of flight (MALD-TOF) analysis of permethylated GSLs.
GSLs were extracted from HCC tissue as described 54 . Briefly, frozen tissue (approximately 25 mg) was homogenized and extracted with methanol and chloroform and bath-sonicated for 30 min. After centrifugation, the supernatant was collected, and the pellet was extracted three times with the same solvent. The supernatants containing neutral and acidic GSLs were pooled. Neutral and acidic GSLs were further separated by anion-exchange chromatography. The purified GSLs were subjected to permethylation according to NaOH/ DMSO method 55 . MS analyses were carried out on an SCIEX 5800 MALDI-TOF/TOF mass spectrometer using 2,5-dihydroxybenzoic acid as matrix. MS spectra were acquired in positive-ion mode and accumulated in 4000 laser shots with random sampling to increase the spectral reproducibility.
Statistical Analysis. Data are presented as mean ± SD, count and percentages. The prognostic performance of genes was calculated by the receiver operating characteristic curve and the area under the ROC curve. The Youden index (sensitivity + specificity −1) was used to determine the optimal cut-off value for high versus low gene expression level. Survival curves were plotted by Kaplan-Meier method, with the log-rank test applied for comparison. The Cox proportional-hazards regression model was employed to evaluate the independent prognostic factors. All tests were two-sided and P-values < 0.05 were considered statistically significant. The statistical analyses were performed with Prism 5.0 (GraphPad Software, La Jolla, CA, USA) and MedCalc 14.8 (MedCalc Software, Ostend, Belgium) software.