Nine susceptibility loci for hepatitis B virus-related hepatocellular carcinoma identified by a pilot two-stage genome-wide association study.

Previous studies have indicated that complex interactions among viral, environmental and genetic factors lead to hepatocellular carcinoma (HCC). To identify susceptibility alleles for hepatitis B virus (HBV)-related HCC, the present study conducted a pilot two-phase genome-wide association study (GWAS) in 660 Han Chinese individuals. In phase 1, a total of 500,447 single-nucleotide polymorphisms (SNPs) were genotyped in 50 HCC cases and 50 controls using Affymetrix GeneChip 500k Array Set. In phase 2, 1,152 SNPs were selected from phase 1 and genotyped in 282 cases and 278 controls using the Illumina GoldenGate platform. The prior probability of HCC in control subjects was assigned at 0.01, and false-positive report probability (FPRP) was utilized to evaluate the statistical significance. In phase 1, one SNP (rs2212522) showed a significant association with HCC (Pallele=5.23×10-8; ORallele=4.96; 95% CI, 2.72-9.03). In phase 2, among 27 SNPs with unadjusted Pallele<0.05, 9 SNPs were associated with HCC based on FPRP criteria (FPRP <0.20). The strongest statistical evidence for an association signal was with rs2120243 (combined ORallele=1.76; 95% CI, 1.39-2.22; P=2.00×10-6), which maps within the fourth intron of VEPH1. The second strongest statistical evidence for an association was identified for rs1350171 (combined ORallele=1.66; 95% CI, 1.33-2.07; P=6.48×10-6), which maps to the region downstream of the FZD4 gene. The other potential susceptibility genes included PCDH9, PRMT6, LHX1, KIF2B and L3MBTL4. In conclusion, this pilot two-phase GWAS provides the evidence for the existence of common susceptibility loci for HCC. These genes involved various signaling pathways, including those associated with transforming growth factor β, insulin/phosphoinositide 3 kinase, Wnt and epidermal growth factor receptor. These associations must be replicated and validated in larger studies.

Introduction

Hepatocellular carcinoma (HCC) is the fifth most common cancer worldwide and the third leading cause of cancer-related mortality (1). In recent years, the incidence of HCC in western countries has increased markedly (2). HCC has a number of notable epidemiological features, including marked variations between geographical regions, racial and ethnic groups, and genders. Previous studies revealed that men have a higher prevalence of HCC than women; the ratio of affected men to affected women varies between 2:1 and 4:1 (3). In high-risk Chinese populations, the male:female ratio of HCC patients is ~2.65:1 (2). HCC frequently occurs within an established background of chronic liver disease and cirrhosis. Major causes of cirrhosis in patients with HCC include hepatitis B virus (HBV) and hepatitis C virus (HCV), alcoholic liver disease and, possibly, non-alcoholic steatohepatitis (3,4). China has a high incidence of HCC, and the newly diagnosed HCC patients in China account for ~55% of the newly diagnosed HCC patients globally each year (5). HBV infection is one major risk factor for HCC occurrence; ≥75% cases of HCC are associated with HBV infection in China (6). However, not all individuals in HBV-infected populations develop HCC over their lifetime. This phenomenon indicates that an individual's genetic background is important in HBV-related hepatocarcinogenesis (3).

To date, the predominant strategy employed to investigate HCC-associated genes has been the candidate gene-based case-control association study. The candidate genes have included GSTT1, GSTM1, UGT1A7, CYPs, NAT2, HFE, MTHFR, TGFβ, TNFα and MnSOD (7–11). By reviewing these studies, we observed numerous inconsistent results, and the associations reported were based on strong evidence. As the molecular mechanisms of most complex diseases, including HCC, are still unknown, the candidate strategy has inherent limitations; in fact, it is a process of validation of the equivocal hypothesis and the conclusions of such studies should also be drawn cautiously. Based on the success of the human genome project, the HapMap Project, and the availability of high throughput gene chips, the whole genome-wide association study (GWAS) has become the most powerful approach to search for and map the susceptibility genes of complex diseases (12,13). Since 2005, GWAS has achieved great success in identifying susceptibility genes in complex diseases, including diabetes (14), cancer (15,16) and systemic lupus erythematosus (17). Recently, a GWAS conducted in China reported 1p36.22 as a novel susceptibility locus for HCC (18).

Distinct clinical characteristics of HBV infection have been reported in different geographical regions of the world, and increasing evidence indicates an association with the genetic diversity of infected patients (19,20). However, such data are largely lacking in Qidong, China, where chronic HBV infection is highly endemic (21). In the present study, a pilot two-stage GWAS was conducted to search for susceptibility loci for HCC. In order to maximize the statistical power to identify associations, homogeneity of the samples was ensured by selecting participants from the male Chinese Han population with HBV surface antigen (HBsAg) seropositivity, and a high proportion of HCC cases with a family history of HCC was included.

Materials and methods

Participants

The present analysis used data and stored samples from a prospective cohort in Qidong, Jiangsu, China. The enrollment of the study cohort has been previously described (22–24). Briefly, male HCC patients with HBsAg seropositivity were recruited from inpatients at the Qidong Liver Cancer Institute (Qidong, China), between August 1, 2006 and July 31, 2008. HCC was diagnosed by histopathological biopsy, or by elevated α-fetoprotein (AFP) levels and distinct changes on imaging (ultrasonography, computed tomography and magnetic resonance imaging), according to the 2004 Barcelona guidelines (25). Blood samples were collected at the time of HCC diagnosis. All male control subjects were selected from a cohort consisting of HBsAg carriers, established by Qidong Liver Cancer Institute in 1996. The absence of HCC in the controls was verified by assessing AFP levels and ultrasonography when the blood samples were collected in 2006. All samples were HCV seronegative. According to the tenets of the Declaration of Helsinki (26), the study was approved by the ethics committees of Qidong County Liver Cancer Institute, (Qidong, China) and Fudan University (Shanghai, China). All participants provided written informed consent, including consent for genetic studies. Data do not contain any information that may lead to the identification of the patients.

Genotyping

DNA was extracted from peripheral blood mononuclear cells using SHENamp Blood DNA Kit according to the manufacturer's instructions (Shengyou Biotechnology Co., Ltd., Hangzhou, China). In phase 1, a total of 500,447 single-nucleotide polymorphisms (SNPs) were genotyped using an Affymetrix GeneChip Mapping 500k Array Set (Affymetrix, Santa Clara, CA, USA) in 50 cases and 50 controls. In phase 2, a total of 1,152 SNPs, selected from phase 1, were genotyped in 282 cases and 278 controls using an Illumina GoldenGate genotyping assay (Illumina, San Diego, CA, USA) at Shanghai Biochip Co., Ltd. (Shanghai, China). The quality control, genotyping and data analyses were performed according to the protocols of the respective manufacturers. For phases 1 and 2, a DNA sample was deemed to have failed if it generated genotypes at <93% of loci. A SNP was deemed to have failed if <90% of DNA samples generated a genotype at the locus. In phase 2, 8 duplicate samples were genotyped to ensure quality of genotyping. For all 1,152 SNPs, >99.8% concordant results were obtained. The genotyping results of 11 SNPs were also validated in 41 samples using the iPLEX Gold assay on the MassARRAY® platform (Sequenom, Inc., San Diego, CA, USA) and >98.4% concordant results were obtained.

Statistical analysis

The statistical analysis of the GeneChip data was performed using Genotyping Console Software, version 4.0 (Affymetrix, Santa Clara, CA, USA) and BeadStudio Genotyping Module, version 3.2 (Illumina, San Diego, CA, USA). Combined analysis was conducted using Stata software version 10.0 (StataCorp, College Station, TX, USA). For each SNP, allele P-value, Cochran-Armitage trend P-value, odds ratio (ORs) and 95% confidence interval (CI) were calculated. The prior probability of HCC in control subjects was assigned at 0.01, and false-positive report probability (FPRP) was used to assess the reliability of the associations (27). FPRP<0.20 was set as the significance threshold for the associations.

Results

Two stage GWA study identified 9 susceptibility loci

In phase 1, the age distribution of 50 cases [mean age (± SD) at diagnosis, 52.3±7.9 years] and 50 controls (mean age at diagnosis, 54.88±8.68 years) was matched. Of the 50 cases, 11 had one or more first-degree relatives affected by HCC (mean age at diagnosis, 54.1±8.1 years). All 100 samples had a call rate of >96%. The call rate is the percentage of successful genotype calls per passing SNP. Exclusion criteria for SNPs included the following: i) overall frequency of SNP <90%; ii) minor allele frequency (MAF) <0.05; and iii) P-value of Hardy-Weinberg equilibrium (HWE) <0.001. Altogether, 279,757 eligible SNPs passed quality control and were included in further analyses. For each SNP, allele P-values, genotype P-values and Cochran-Armitage trend P-values were calculated. The minimum P-value was designated as minP. For autosome, min P = min (allele_P, genotype_P, P_trend); for sex chromosome, min P =min (allele_P, allele_exact_P). There were 355 SNPs on autosomes and 2 SNPs on the X chromosome with minP<1×10−3. A total of 26 SNPs were identified with minP<1×10−4 (Table I). One SNP (rs2212522) demonstrated a significant association with HCC (Pallele=5.23×10−8; ORallele=4.96; 95% CI, 2.72–9.03).

Table I.

Twenty-six SNPs with minP<1×10−4 in phase 1.

		Cases, n		Controls, n
dbSNP rsID[a]	Associated gene(s)	A	B	A	B	minP-value	ORallele
rs2212522	L3MBTL4	68	32	30	70	7.65×10−8	0.2016807
rs4713039	NO145	44	56	77	23	1.81×10−6	4.2608696
rs4539982	TBL1XR1	79	15	49	45	2.68×10−6	0.2067511
rs9877175	TBL1XR1	18	80	49	51	5.25×10−6	4.2701525
rs4277177	TMEM16F	51	45	62	36	8.52×10−6	1.5196078
rs8031646	ARRDC4	72	26	42	56	1.40×10−5	0.2708333
rs11057529	FAM101A	17	83	45	55	1.40×10−5	3.9946524
rs7069096	LYZL1	36	62	67	33	2.03×10−5	3.4966332
rs946351	NO145	51	49	22	78	2.05×10−5	0.2709904
rs12044483	FLJ32784, UBXD3	39	61	69	31	2.08×10−5	3.4813896
rs10926832	PLD5	32	66	61	39	2.16×10−5	3.2259615
rs1334125	PRMT6	34	66	64	36	2.19×10−5	3.4509804
rs12580388	TMEM132D	80	20	53	47	3.17×10−5	0.2819149
rs6910232	NO145	64	34	35	63	3.43×10−5	0.2951389
rs7854810	ENST00000380100	97	3	76	22	3.80×10−5	0.1068416
rs12034802	PRMT6	60	30	37	63	4.42×10−5	0.2936508
rs7870157	ENST00000387810	4	96	24	76	4.59×10−5	7.5789474
rs1883165	FLJ32784	65	31	35	57	4.60×10−5	0.2928475
rs2292723	TMEM132D	11	89	32	66	5.63×10−5	3.9228651
rs10735541	TLE4	32	68	13	87	5.93×10−5	0.3175287
rs1543940	LOC440337, FAM86A	13	79	36	54	7.63×10−5	4.0512821
rs9571852	NBEA	14	86	38	62	7.89×10−5	3.7649773
rs9949516	L3MBTL4	31	63	60	38	8.92×10−5	3.2088285
rs6828409	EREG, EPGN	75	25	47	51	9.16×10−5	0.3071895
rs7105477	AP2A2	82	8	68	32	9.56×10−5	0.2073171
rs4417097	PRMT6, AMY1C	39	53	69	29	9.76×10−5	3.2334218

Accessible at http://www.ncbi.nlm.nih.gov/SNP/. SNP, single-nucleotide polymorphism. A represents the total number of major allele in cases or controls (n); B represents the total number of minor allele in cases or controls (n).

In phase 2, 282 HCC cases and 278 controls were recruited from Qidong. Of the HCC cases, 61 (21%) had a family history of HCC in a first or second-degree relative. The age distributions were 51.4±10.0 and 53.7±10.6 years in cases and controls, respectively. The selected 1,152 SNPs included 4 categories: i) SNPs whose minP in phase 1 was <1×10−3 or SNPs well clustered in chromosome with minP<0.05 (n=398); ii) predicted deleterious non-synonymous SNPs (nsSNPs) (n=430) of the nearest genes; iii) SNPs possibly associated with copy number variation (CNV) (n=315); and iv) SNPs located at 8q24 which had been reported to be associated with other types of cancers (n=9: rs13254738, rs6983561, rs16901979, rs13281615, rs10505447, rs10808556, rs6983267, rs7000448 and rs1447295) (28–33).

All 560 samples had a call rate of >93%. Of the selected 1,152 SNPs, 598 passed the quality control following the exclusion procedure (MAF <0.05, HWE disequilibrium P<0.001 and call frequency <0.95). There were 35 SNPs with unadjusted Pallele<0.05. Of these, 8 SNPs whose minor allele number was ≤3 were further excluded, leaving 27 SNPs with unadjusted Pallele<0.05. The most significant signal in phase 1, rs2212522, was replicated in phase 2 with Pallele<0.05 and combined Pallele=7.91×10−5. Out of 27 SNPs, 20 were genotyped in phases 1 and 2, and 7 SNPs were genotyped only in phase 2. Combined analysis of the 20 SNPs revealed that there were 15 SNPs with combined Pallele<0.05. Thus, there were 22 SNPs with combined allele P-value or with phase 2 allele P-value of <0.05 (Table II). As all controls were men with HBsAg seropositivity, a prior probability of 0.01 for HCC occurrence was assigned and FPRP was calculated (34). The results revealed that 9 of the 22 SNPs were associated with HCC (FPRP<0.20) (Table III). CNV was validated by genotyping 315 high density SNPs using the Illumina GoldenGate platform and quantitative polymerase chain reaction (qPCR) analysis; this identified a number of CNVs in HCC, reported in other studies (35,36). No associations were identified between the 9 SNPs in 8q24 and HCC.

Table II.

Twenty-two SNPs with allele P<0.05 in phase 2 or combined allele P<0.05.

				Allele P-value
dbSNP rsID[a]	Associated gene	Chromosome	Position	Phase 1	Phase 2	Combined
rs2120243	VEPH1	3q24–25	158630262	1.37×10−4	3.23×10−4	2.00×10−6
rs1350171	FZD4	11q14.2	86322037	1.68×10−4	9.16×10−4	7.47×10−6
rs1048338	FZD4	11q14.2	86310465	1.57×10−4	1.44×10−3	1.11×10−5
rs2212522	L3MBTL4	18p11.31	5890773	5.23×10−8	4.81×10−2	7.91×10−5
rs7116140	FZD4	11q14.2	86313003	2.14×10−4	1.20×10−2	1.76×10−4
rs4480667	PCDH9	13q14.3	66825714	6.60×10−1	1.92×10−4	3.02×10−4
rs4417097	PRMT6	1p13.3	106601576	8.78×10−5	2.79×10−2	4.27×10−4
rs9893681	LHX1	17q12	32362128	4.47×10−4	2.82×10−2	9.31×10−4
rs4767254	TBX3	12q24.1	113682775	6.13×10−4	3.22×10−2	1.18×10−3
rs132024	PHF21B	22q13.31	43804508	5.77×10−4	4.65×10−2	2.20×10−3
rs4561519	KIF2B	17q22	49256702	1.90×10−1	9.26×10−3	3.61×10−3
rs729565	CNTNAP2	7q35	146668586	3.70×10−1	1.02×10−2	6.49×10−3
rs4726849	CNTNAP2	7q35	146704681	3.40×10−1	3.23×10−2	1.90×10−2
rs10500181	CNTNAP2	7q35	146704487	3.60×10−1	3.75×10−2	2.30×10−2
rs12532315	CNTNAP2	7q35	146730215	5.40×10−1	4.95×10−2	4.04×10−2
rs3818605	PTPRA	20p12	2788773		6.12×10−3
rs9898643	KIF2B	17q22	49254648		7.49×10−3
rs3806523	SH3BP4	2q37.1	235523908		1.33×10−2
rs1155569	RPL38	17q23	61082421		2.54×10−2
rs1057090	MCPH1	8p23.1	6466450		2.80×10−2
rs17138848	COMMD10	5q23.1	115447202		4.22×10−2
rs730819	PTPRA	20p12	2793130		4.28×10−2

Accessible at http://www.ncbi.nlm.nih.gov/SNP/. SNP, single-nucleotide polymorphism.

Table III.

Nine SNPs with combined allele P<0.05 and FPRP <0.20.

		Allele P-value
dbSNP rsID[a]	Associated gene	Phase 1	Phase 2	Combined	ORallele	95% CI	FPRP
rs2120243	VEPH1	1.37×10−4	3.32×10−4	2.00×10−6	1.76	1.39–2.22	<0.001
rs1350171	FZD4	1.68×10−4	9.28×10−4	6.48×10−6	1.66	1.33–2.07	0.014
rs1048338	FZD4	1.58×10−4	1.44×10−3	1.11×10−5	1.64	1.31–2.04	0.046
rs2212522	L3MBTL4	5.23×10−8	4.81×10−2	7.91×10−5	1.57	1.25–1.97	0.019
rs7116140	FZD4	2.14×10−4	1.20×10−2	1.76×10−4	1.51	1.22–1.88	0.039
rs4480667	PCDH9	6.60×10−1	1.92×10−4	3.02×10−4	1.52	1.21–1.90	0.037
rs4417097	PRMT6	8.78×10−5	2.78×10−2	4.27×10−4	1.48	1.19–1.85	0.057
rs9893681	LHX1	4.46×10−4	2.82×10−2	9.31×10−4	1.65	1.22–2.21	0.159
rs4561519	KIF2B	1.90×10−1	9.26×10−3	3.61×10−3	1.52	1.14–2.02	0.176

Accessible at http://www.ncbi.nlm.nih.gov/SNP/. SNP, single-nucleotide polymorphism; FPRP, false-positive report probability; CI, confidence interval.

Genes associated with the 9 susceptibility loci

rs2120243 (C>A) maps within the fourth intron of ventricular zone expressed PH domain homolog 1 [VEPH1; HUGO Gene Nomenclature Committee (HGNC) ID: 25735]. VEPH1 maps at 3q24–25 and covers 273.88 kb.

Frizzled homolog 4 (FZD4; HGNC ID: 4042) is a member of the frizzled gene family, maps at 11q14.2 and covers 9.73 kb. The majority of frizzled receptors are coupled to the β-catenin signaling pathway. It has been reported FZD4 is associated with numerous types of cancer (37). In the present study, 3 SNPs (rs1048338, rs7116140 and rs1350171) were identified in the downstream 12–23 kb of FZD4. Haploview version 4.1 1 (The Broad Institute) revealed that the 3 SNPs are in one strong linkage disequilibrium (LD) block (Fig. 1). Using the haplotype analysis software PLINK (pngu.mgh.harvard.edu/purcell/plink/), it was demonstrated that the haplotype exhibited significantly different distributions between cases and controls (P<0.01).

Figure 1.

Haploview revealed the linkage disequilibrium block contained rs1048338, rs7116140 and rs1350171. The block covers 12 kb containing 19 SNPs. SNPs, single-nucleotide polymorphisms.

rs4480667 is located in the 125 kb upstream of protocadherin 9 (PCDH9; HGNC ID: 8661) and maps in a 24 kb block (38). PCDH9, which belongs to the protocadherin gene family, maps at 13q14.3-q21.1 and covers 927.62 kb. It has been reported that CNV of PCDH9 may be associated with glioblastoma as a tumor suppressor gene (TSG) (39). Patch 1.0 revealed that allele A of rs4480667 (A>G) created a new binding site for the transcription factors C-Ets-1 (HGNC ID: 3488) and Elf-1 (HGNC ID: 3316). These two transcription factors have been reported to be associated with several cancer types, including lung cancer and HCC (40).

rs9893681 maps at the upstream 6.4 kb of the LIM homeobox 1 (LHX1) gene. LHX1 (HGNC ID: 6593) maps at 17q12 and covers 7.15 kb. It has been reported the CNV of LHX1 is associated with gastric cancer (41). Patch 1.0 indicated that allele A of rs9893681 (T>A) created a new binding site for transcription factor Fushi tarazu (FTZ gene). Liu et al (42) reported that FTZ regulated the ubiquitin E3 ligase complex factor Speckle-type POZ protein (SPOP; HGNC ID: 11254), which mediated degradation of the Jun-kinase phosphatase, thereby inducing tumor necrosis factor/Eiger-dependent apoptosis (42). The human homolog of FTZ, nuclear receptor subfamily 5 group A member 2 (NR5A2; HGNC ID: 7984) has been shown to play critical roles in various cancer types (43,44).

rs4417097 is located in a region of gene desert, the nearest gene being protein arginine methyltransferase 6 (PRMT6; HGNC ID: 18241). rs4417097 maps in the upstream 799kb of PRMT6. A previous GWAS reported that PRMT6 gene is involved in acquired immune deficiency syndrome (45). Another study reported that thrombospondin-1 was a transcriptional repression target of PRMT6, and suggested that neutralizing the activity of PRMT6 could inhibit tumor progression (46).

rs4561519 is a nsSNP of kinesin family member 2B (KIF2B; HGNC ID: 29443). KIF2B has been proposed to participate in the process of microtubule-based movement. Three protein function prediction software packages: SIFT (sift.jcvi.org/), Polyphen (genetics.bwh.harvard.edu/pph2/) and SNPs3D (snps3d.org/), predicted that rs4561519 is deleterious to the protein's function (47,48).

L(3)mbt-like 4 (L3MBTL4) maps on chromosome 18, at 18p11.3. It covers 460.53 kb, from 6405235 to 5944705 [National Center for Biotechnology Information (NCBI) 36, March 2006; ncbi.nlm.nih.gov/IEB/Research/Acembly/av.cgi?c=geneid&org=9606&l=91133], on the reverse strand. rs2212522 maps to 53.9 kb downstream of L3MBTL4. Functionally, this gene has been proposed to participate in various processes, including cell adhesion, platelet activation and regulation of transcription. The protein is also predicted to have molecular functions (transcription factor activity and zinc ion binding) and to localize to various compartments (integrin complex, cytoplasm, extracellular space and the nucleus). L3MBT has been reported to be a TSG in Drosophila (49). In addition, one study reported that L3MBTL4 was associated with HCC (50).

Gene gene interactions

In order to find the gene-gene interactions, we selected all SNPs with combined Pallele<0.05 and SNPs with Pallele<0.05 in phase 2 and searched for the nearest genes using the NCBI Map Viewer (www.ncbi.nlm.nih.gov/mapview/). A total of 109 SNPs associated with 84 genes were identified. Protein-protein interaction networks were then explored using STRING software (string-db.org/); 80 proteins were recognized and analyzed. By adding 20 protein notes automatically with STRING, many crucial proteins were identified, including EGFR, ERBBs, FYN, CDC42, PIK3R1, ARHGEF, INS, INSR, CDC42 and PRKCQ (Fig. 2). These genes involved a number of important signaling pathways involved in carcinogenesis, such as transforming growth factor β, insulin/phosphoinositide 3 kinase (PI3K) and Wnt/β-catenin.

Figure 2.

Protein network configuration by STRING.

Discussion

The success of GWAS relies on much of the risk of common diseases being due to common genetic variants; however, evidence for this hypothesis is inconclusive, and the possibility that some complex diseases are due to certain rare variants with high genetic risk cannot be excluded. Therefore, the present study investigated both common variants and relatively rare variants of the selected nsSNPs, with possibly deleterious effects on protein function, to perform a two-stage GWAS.

Chen et al (33) genotyped over 350,000 genome-wide autosomal SNPs by using Illumina Human 610-Quad BeadChips in over 6,000 Han Chinese samples from ten provinces, showing that, in the Han Chinese population, geographic matching is a good proxy for genetic matching (51). In phase 1 of the present study, 100 participants were enrolled from Qidong, China. Therefore, we considered it unnecessary to conduct a population stratification analysis. In order to increase the statistical power to search for genetic risk factors, potential confounding factors (gender, ethnicity, HBV status and age) were controlled by selecting only male HCC patients (including familial HCC patients) and male controls with HBsAg seropositivity from Qidong, where there is a high incidence of HCC. Due to the relatively small sample size (660 samples) of this GWAS, the associations did not reach previously established statistical criteria for GWAS (P<5×10−7) (52). In addition, certain researchers consider the Bonferroni adjustment too strict and not applicable to small studies (53). In the current study, by using the FPRP criterion (FPRP<0.20), 9 SNPs were identified to be associated with HCC. The majority of these SNPs were located in a gene region, including intron, promoter and coding regions. The strongest statistical evidence for an association was found in rs2120243, which maps within the fourth intron of VEPH1. According to Haploview, rs2120243 is in a 25 kb block including 19 SNPs. Teleman et al (54) identified the Drosophila melted protein as a modulator of the insulin/PI3K signaling pathway; VEPH1 in Homo sapiens and Drosophila melted protein are homologous proteins. One study has reported VEPH1 to be a cancer-associated gene in breast cancer (55). By using the online functional protein network software STRING, VEPH1 was found to be connected to ACVR1, TGFBR1, AKT1, SRC, FRAP1 and TP53 (56). Patch 1.0 software, which predicts changes in the binding sites of transcription factors, revealed that allele A of rs2120243 creates a new binding site for transcription factor retinoid X receptor α (RXR-α) (57). It has been reported that RXR-α is associated with HCC (58). In addition, a recent GWAS conducted in China reported a susceptibility locus KIF1B on 1p36.22 (rs17401966) for HBV-related HCC (18). The current GWAS discovery analysis did not reveal a consistent result for the association between rs17401966 and the development of HBV-related HCC. However, one nsSNP (rs4561519) of the KIF2B gene was associated with HCC in this study. Three protein function prediction softwares (SIFT, Polyphen and SNPs3D) all predicted that rs4561519 was deleterious to the protein's function. KIF1B and KIF2B belong to the kinesin family, which may indicate that kinesin families are associated with HCC. We further investigated the protein expression of KIF2B in chronic HBV carriers by immunohistochemistry and western blot analysis in another unpublished study. Significantly increased expression of KIF2B was detected in adjacent non-tumor liver tissues compared with that of paired HCC tissues (data not shown). The exact functional consequences of rs17401966 remain unknown. Further studies will be needed to finely map and identify the causative polymorphism and to clarify which genes drive the genetic association. Using STRING network analysis, many cancer-related genes were identified, and these were involved in various important signaling pathways associated with HCC that have been validated by other studies (59,60).

Previous studies have clearly demonstrated the existence of a subpopulation structure among the Chinese Han population along the north-south axis. In the present study, all cases and controls were residents of Qidong. Therefore, the findings should be free of adverse effects of population stratification. The sample size in this pilot two-stage GWAS was relatively small and had relatively limited statistical power to detect risk alleles with relatively low allele frequency (MAF<0.1) or genetic power (OR<1.2). Due to these limitations, the results must be interpreted and conclusions made with caution. At present, we are collecting more samples and intend to investigate certain proposed SNPs and genotypes in larger studies to replicate and validate the associations. The risk alleles and related gene functions must also be studied further.

In conclusion, by conducting two-stage GWAS and mining bioinformatic data, the present study has identified a number of potential susceptibility loci for HCC in male Chinese individuals with HBsAg seropositivity. These findings offer valuable clues in the study of hepatocarcinogenesis and may have potential clinical value. The associations and molecular mechanisms of HCC merit further research.