Association between vascular endothelial growth factor gene polymorphisms and survival in hepatocellular carcinoma patients


  • Sun-Young Kong,

    1. Center for Clinical Services, Department of Laboratory Medicine, Research Institute & Hospital, National Cancer Center, Goyang-si, Gyeonggi-do, Republic of Korea
    Search for more papers by this author
  • Joong-Won Park,

    Corresponding author
    1. Center for Liver Cancer, Research Institute & Hospital, National Cancer Center, Goyang-si, Gyeonggi-do, Republic of Korea
    • Center for Liver Cancer, National Cancer Center, 809 Madu-dong, Ilsan-gu, Goyang-si, Gyeonggi-do, 411-769, Republic of Korea
    Search for more papers by this author
    • fax 82-31-920-1520

  • Jung An Lee,

    1. Center for Liver Cancer, Research Institute & Hospital, National Cancer Center, Goyang-si, Gyeonggi-do, Republic of Korea
    Search for more papers by this author
  • Jung Eun Park,

    1. Center for Liver Cancer, Research Institute & Hospital, National Cancer Center, Goyang-si, Gyeonggi-do, Republic of Korea
    Search for more papers by this author
  • Kyung Woo Park,

    1. Center for Cancer Prevention and Detection, Research Institute & Hospital, National Cancer Center, Goyang-si, Gyeonggi-do, Republic of Korea
    Search for more papers by this author
  • Eun Kyung Hong,

    1. Center for Liver Cancer, Research Institute & Hospital, National Cancer Center, Goyang-si, Gyeonggi-do, Republic of Korea
    Search for more papers by this author
  • Chang-Min Kim

    1. Center for Liver Cancer, Research Institute & Hospital, National Cancer Center, Goyang-si, Gyeonggi-do, Republic of Korea
    Search for more papers by this author

  • Potential conflict of interest: Nothing to report.


Vascular endothelial growth factor (VEGF) plays an important role in angiogenesis and progression of tumor, including hepatocellular carcinoma (HCC), and elevated VEGF levels in serum and tissues have been known to be related with poor prognosis in patients with HCC. However, the effect of such polymorphisms of the VEGF gene on HCC prognosis has not been elucidated. In the present study, we investigated the association between VEGF gene polymorphisms and HCC patient prognosis. The study involved 416 HCC patients treated at the National Cancer Center Korea from November 2000 to December 2005. The median patient age was 57 years, and 328 patients (78.8%) were men. A total of 19 polymorphisms were analyzed, and the hazard ratios (HRs) for genotypes and haplotypes were determined in terms of risk for overall survival using Cox proportional hazard regression analysis. Of the 19 alleles, 7 showed no heterozygous allele. PHASE analysis identified a total of 36 haplotypes. The −2578 to −1498 region of the VEGF gene showed a strong linkage disequilibrium (correlation coefficient, r2 = 0.91; Lewontin's D′, D′ = 0.982). The adjusted HRs were 0.67 [95% confidence interval (CI), 0.46 to 0.99] for −634CC genotype carriers and 0.57 (95% CI, 0.36 to 0.92) for homozygous haplotype 1 (Ht1: CCGAGCCC at −2578/−1203/−1190/−1179/−1154/−634/−7/+936) carriers compared with noncarriers. Conclusion: These findings suggest that VEGF polymorphisms may be significant prognostic indicators for HCC patients. (HEPATOLOGY 2007.)

Angiogenesis is essential for tumor growth, invasion, and metastasis.1 Tumors release angiogenic factors that stimulate endothelial cell migration, proliferation, and capillary formation. Newly formed vessels promote tumor growth by supplying oxygen and nutrients, and by removing catabolites. HCC is a typical hypervascular tumor, and a radiology finding of an arterial hypervascular pattern is a diagnostic criterion for HCC.2 HCC is the fifth most common malignancy worldwide3 and has been classically considered a neoplasm with a dismal prognosis. The majority of HCC patients are diagnosed at advanced stages including having vascular invasion and no chance for cure with surgical treatment and local ablation therapy.

Vascular endothelial growth factor (VEGF) is a major driver of physiological and pathological angiogenesis.4 VEGF is believed to play an important role in various liver diseases, including HCC. An animal model has shown that VEGF augments HCC development and metastasis,5 and VEGF overexpression has been reported in HCC and surrounding liver.6–10 Circulating VEGF concentrations have been found to increase according to HCC stage,11 and serum VEGF concentration has been suggested to be an independent predictor of HCC recurrence and overall survival.12–14

Because no survival advantages were obtained from systemic chemotherapy in patients with advanced HCC,15 there is an absolute need to develop new therapeutic agents. Recently, the potential benefits of targeting the VEGF pathway in the treatment of advanced HCC with bevacizumab and sorafenib were reported.16, 17 Therefore, the necessities of understanding the effect of VEGF genetic polymorphisms on the clinicopathological phenotype of HCC are growing.

The human VEGF gene is localized on chromosome 6p21.3 and contains 8 exons.18 Alternative exon splicing of the human VEGF pre-mRNA produces 6 VEGF isoforms. More than 30 single nucleotide polymorphisms (SNPs) have been identified in this gene, and some polymorphisms have been reported to be risk factors for breast and prostate cancer.19, 20

It is not known whether the presence of VEGF polymorphisms affects the prognosis of HCC patients. In the present study, we investigated the association between VEGF gene polymorphisms and HCC patient prognosis.


α-FP, α-fetoprotein; BCLC, Barcelona-Clinic-Liver-Cancer; CI, confidence interval; CT, computed tomography; HR, hazard ratios; Ht, haplotype; MVD, microvessel density; SNP, single-nucleotide polymorphism; UICC, International Union Against Cancer (Union Internationale Contre le Cancer); VEGF, vascular endothelial growth factor; WBC, white blood cell.

Patients and Methods


Of the 1078 patients diagnosed with HCC at the National Cancer Center (Korea) from November 2000 to December 2004, we included 416 patients (328 males and 88 females; median age 57 years, age range 27 to 97 years) who gave informed consent in the present study and were followed to December 2005.

The HCC diagnostic criteria were based on the local guidelines proposed by the Korea Liver Cancer Study Group and the National Cancer Center (Korea),21 which is similar to the European criteria.2 We gave a diagnosis of HCC when a patient had 1 or more risk factors (i.e., HBV or HCV infection, or cirrhosis) and 1 of the following: >400 ng/mL serum α-fetoprotein (α-FP) and at least 1 positive finding following examination using spiral computed tomography (CT), contrast-enhanced dynamic MRI, or hepatic angiography (n = 178); or <400 ng/mL serum α-FP level and at least 2 positive findings following CT, MRI, or hepatic angiography (n = 238). A positive HCC finding using dynamic CT or MRI is indicative of arterial enhancement followed by venous washout in the delayed portal/venous phase. According to the local guidelines, we performed histopathological diagnosis for cases that did not fulfill all of the clinical noninvasive diagnostic criteria of HCC. We performed pathologic diagnosis in 104 (n = 50 surgical specimens, n = 54 biopsy specimens) of the 416 patients.

We reviewed clinicopathological features such as sex, age, HCC etiology, tumor size, number of tumors, nodal invasion, metastasis, existence of vascular invasion (hepatic or portal vein invasion according to radiology), tumor type (well-defined versus ill-defined according to radiology), performance status by the World Health Organization classification, Child-Pugh classification, treatment modalities, and the results of clinical laboratory tests from medical records. We used both modified International Union Against Cancer (UICC; Union Internationale Contre le Cancer) stages and Barcelona-Clinic-Liver-Cancer (BCLC) stages for staging.22, 23

Based on Child-Pugh classification and modified UICC staging,21 we initially treated patients with 1 of 6 treatment modalities: transarterial chemoembolization, liver resection, radiofrequency ablation, systemic chemotherapy, radiotherapy, or conservative management. We followed patients to December 2005, by which time 276 deaths had been recorded.

DNA Isolation and Genotyping

We isolated genomic DNA from white blood cells (WBCs) of peripheral blood using a Puregene DNA Purification kit and following the manufacturer's instructions (Gentra System, Minneapolis, MN). We determined DNA concentrations using a DU 800 spectrophotometer (Beckman Coulter, Fullerton, CA). The study examined 19 VEGF polymorphic sites following a review of a genetic database ( and the literature.24–26 We particularly focused on the 5′ untranslated region and promoter region, given reports that SNPs in this region regulate VEGF expression via alternative initiation of transcription and internal initiation of translation.27, 28 We amplified the 5′ untranslated region and promoter regions of the VEGF gene using primers that targeted the polymorphic sequences. The PCR reaction mixture (10 μL) contained 1.0 μL 10× PCR buffer (Takara, Tokyo, Japan), 0.7 μL 2.5 mM each deoxyribonucleotide triphosphate (Takara), 0.3 μM each primer (Bioneer Corp., Chungwon, Korea), 0.5 U Taq DNA polymerase (Takara), and 1 μL (0.5 μg) genomic DNA. The thermal cycler (Biometra T Gradient PCR, Gottingen, Germany) protocol was as follows: 35 cycles of 30 seconds denaturation at 95°C, 30 seconds annealing at 60°C (up to 65°C), and 30 seconds extension at 72°C. There was a 5-minute preincubation at 95°C before commencing a cycle, and a 10-minute additional extension at 72°C after completion of the cycles. We incubated amplified DNA (1.5 μL) with 2 U shrimp alkaline phosphatase and 5 U exonuclease I (USB Corp., Cleveland, OH) at 37°C for 15 minutes. We inactivated the enzymes by incubation at 80°C for 15 minutes, after which the DNA was denatured at 95°C for 15 minutes. We determined the presence of a PCR product using agarose gel electrophoresis. We performed cycle sequencing using a BigDye Terminator Cycle Sequencing Ready Reaction kit v3.0. (Applied Biosystems, Foster City, CA) and an automated ABI Prism 3100 Genetic Analyzer (Applied Biosystems).

We performed the allelic discrimination of the VEGF gene polymorphisms at the +936 position (rs3025039) using an ABI PRISM 7900 Sequence Detection Systems apparatus (Applied Biosystems) and the fluorogenic 5′ nucleases assay with Taqman Minor Groove Binder probes. Wild-type Taqman probes were FAM-labeled and recessive probes were VIC-labeled. The final volume for each reaction was 5 μL, consisting of 2.5 μL Taqman Universal PCR Master Mix (Applied Biosystems), 0.25 μL 20× Taqman probe, and 2.5 μL genomic DNA. The PCR protocol involved an initial denaturation step at 95°C for 10 minutes, and 40 cycles of 92°C for 15 seconds and 60°C for 1 minute. We measured fluorescent signals at 60°C. Each genotype assay run included at least 2 negative controls that contained the same reaction mixture excluding DNA.

After sequencing and genotyping, we determined the Hardy-Weinberg equilibrium using χ2 tests. We analyzed the haplotype using PHASE software v2.1.29, 30 We determined linkage disequilibrium between VEGF gene polymorphisms using the Haploview program v3.2 (

Plasma and Serum VEGF Assay

We obtained plasma and serum samples at the time of HCC diagnosis, prior to treatment. We processed blood samples within 6 hours of collection and were stored at −80°C after centrifugation until assayed. We determined VEGF plasma and serum concentrations in 416 and 414 patients, respectively, using the Human VEGF Immunoassay Quantikine (R&D Systems, Inc., Minneapolis, MN) kit and Power Wave HT and KC4 v3.0 computer software (Bio-TEK, Winooski, VA).

Microvessel Density and VEGF Expression on Tissue

Among 104 paraffin-embedded tissue materials, only 63 specimens (50 surgical specimen, 13 biopsy specimen) were available for additional microvessel density (MVD) and VEGF immunostaining. The surrounding liver served as an internal control for each tumor. We performed immunohistochemical stains using the streptoavidin-biotin complex technique after microwave antigen retrieval. In brief, we pretreated deparaffinized 5-μm sections for 32 minutes with CC1 buffer, stained them with monoclonal antibody CD34 QBEnd10 clone (1:500 dilution; Novocastra, Newcastle, UK), pretreated them for 15 minutes in pH 6.0 citrate buffer in the microwave, and stained them with monoclonal antibody VEGF (1:100 dilution; BD Pharmingen, San Diego, CA), respectively, using a Benchmark XT autoimmunostainer (Ventana Medical Systems, Inc., USA). We did the negative control using normal rabbit immunoglobulin G (IgG) instead of primary antibody.

We determined VEGF expression by cytoplasmic staining intensity as 0 (negative), 1 (weak positive), 2 (moderate positive), and 3 (strong positive). We assessed the tumor and adjacent liver tissue separately.

We counted the MVD as follows: we examined the CD34 stained slides under ×100 magnification to select the highest vascular density areas within the tumor. We captured 5 areas under ×200 magnification (0.21 mm3/captured area). When the tumor vessel had a large and elongated lumen, the regular counting method might not accurately reflect vascular density. Thus, we used the method of counting every 40 μm length of lumen as 1 vessel.32 We counted strong CD34-positive capillaries including single endothelial cells. We calculated the average number of vessels in 5 areas.

Statistical Analysis

We used Fisher's exact test or the Wilcoxon rank sum test to determine associations between VEGF gene genotypes and haplotypes, and the following clinical variables: age, HCC etiology, tumor size, number of tumors, nodal invasion, metastasis, existence of vascular invasion, tumor type (well-defined vs. ill-defined), performance status, modified UICC stage, BCLC stage, and Child-Pugh class. For ordered variable, we performed a test for trend (P trend).

The primary outcome was overall survival. The endpoint for overall survival analysis was death related to HCC. We calculated survival time as the time from cancer diagnosis to the study endpoint. We used the Kaplan-Meier method and log-rank test to compute overall survival rates. We applied the Cox regression model to evaluate the effect of each clinical variable, and the VEGF genotype or haplotype on overall survival. We calculated hazard ratios (HRs) for significant genotypes and haplotypes with adjustments for clinical variables important for survival, such as age, tumor size, number of tumor, vascular invasion, tumor type, nodal invasion, metastasis, performance status, and Child-Pugh class. Additionally, we performed survival analysis according to significant genotype stratified by BCLC stage.

We determined the relationship between plasma or serum VEGF levels and clinical variables, genotypes, and tissue VEGF expression using Kruskal-Wallis analysis. We used Pearson correlation analysis to determine the relationship between circulating VEGF levels, and both tissue MVD and laboratory parameters such as prothrombin time and albumin, total bilirubin level, α-FP level, WBC level, hemoglobin level, and platelet level.

We used STATA software v9.1 (StataCorp LP, College Station, TX) for statistical analysis. All tests were based on a 2-sided probability.


Patient Characteristics and Prognostic Factors for Overall Survival

Patient characteristics and prognostic factors for overall survival in 416 patients with HCC are shown in Table 1. The major prognostic factors for survival were age at diagnosis, tumor size, the number of tumors, portal or hepatic vein invasion, tumor type, performance status, modified UICC stage, BCLC stage, and Child-Pugh classification. In terms of the initial treatments, the overall survival rates were 33.0% for transarterial chemoembolization (number of deaths/number of treated patients = 187/279), 84.0% for liver resection (8/50), 66.7% for radiofrequency ablation (4/12), 6.7% for systemic chemotherapy (14/15), 8.3% for radiotherapy (11/12), and 2.1% for conservative management (47/48). Patients undergoing surgical resection as the initial treatment (n = 50) showed better prognosis due to the disease being at an early stage and there being good liver function.

Table 1. Hazard Ratios of Clinical Characteristics for Overall Survival in 416 Patients with HCC
CharacteristicsClassNumber of CasesNumber of EventsHR (95% CI)*P*
  • *

    The HR (95% confidence interval [CI]) and P value for each clinical characteristic of overall survival were calculated using Cox proportional regression analysis. Death was defined as death associated with HCC. †Vascular invasion was defined as positive when either portal vein invasion or hepatic vein invasion were present, which was observed radiologically.

 Female88560.9 (0.7-1.2)0.400
Age (years)≤551801291.0 
 >552361420.6 (0.5-0.8)<0.001
Etiologies of HCCHBV3302201.0 
 HCV42250.8 (0.5-1.2)0.220
 Alcohol24130.7 (0.4-1.3)0.242
 Others20131.0 (0.6-1.8)0.976
Liver cirrhosisAbsent1911231.0 
 Present2251481.1 (0.8-1.4)0.549
Portal vein invasionAbsent3081701.0 
 Present1081012.3 (2.0-2.5)<0.001
Hepatic vein invasionAbsent3882461.0 
 Present27241.6 (1.4-1.9)<0.001
Vascular invasion†Absent3011651.0 
 Present1151065.3 (4.1-6.9)<0.001
Size of tumor (cm)≤246161.0 
 >2, ≤5177991.9 (1.1-3.3)0.015
 >5, ≤10124944.1 (2.4-7.0)<0.001
 >1069628.1 (4.6-14.1)<0.001
Number of tumor masses1157691.0 
 2-3109701.7 (1.2-2.3)0.003
 ≥41501323.9 (2.9-5.3)<0.001
Tumor typeWell-defined2991631.0 
 Ill-defined1171084.4 (3.4-5.6)<0.001
Performance status074361.0 
 141271.7 (1.1-2.9)0.029
 22952021.8 (1.3-2.6)0.001
 3663.3 (1.4-7.8)0.008
Modified UICC stageI32101.0 
 II101351.1 (0.5-2.2)0.788
 III1521033.0 (1.6-5.8)0.001
 IVb615711.2 (5.7-22.1)<0.001
BCLC stageVery early50Not reached 
 Intermediate133893.0 (2.1-4.4)<0.001
 Advanced14013011.5 (7.9-16.7)<0.001
 Terminal131314.4 (7.7-27.3)<0.001
CTP classificationA3121791.0 
 B92792.6 (2.0-3.4)<0.001
 C13135.2 (2.9-9.2)<0.001

VEGF Genotypes and Clinicopathological Association

Genotype frequencies are shown in Table 2. All genotype frequencies followed the Hardy-Weinberg equilibrium, other than polymorphisms at −1190, −1179, and −1154. There was strong linkage disequilibrium (correlation coefficient, r2 = 0.91; Lewontin's D′, D′ = 0.982) from −2578 to −1498, and we observed a single block using Haploview analysis. In contrast, the linkage between −2578 and −1154 polymorphism was much weaker (r2 = 0.349, D′ = 0.644), and we identified no significant linkage disequilibrium between −2578 and +936 (r2 = 0.035, D′ = 0.232) (Fig. 1).

Table 2. Frequencies of VEGF Polymorphisms in 416 Patients with HCC
Locationrs NumberGenotypeNumber of PatientsPercent
−2549 18–bp/Deletion22052.9
−2429 G insertion22052.9
  G insertion/deletion16639.9
  G deletion307.2
−1203 CC39394.4
−1001 GG416100.0
−627 GG416100.0
Figure 1.

The linkage disequilibrium (LD) pattern for VEGF gene in 416 patients with HCC. Haplotype block structure, as depicted by Haploview, is shown. The numbers in box indicate the pairwise D′ value and the box shading change (white to black) according to increasing D′. Black represents strong LD.

We identified a total of 36 haplotypes, with only 5 occurring at a frequency greater than 5% (Table 3). We analyzed these 5 haplotypes in terms of associations with clinical characteristics or laboratory parameters. The haplotypes with 1% to 5% frequency are as follows: haplotype 6 (Ht6), ACAAGGTC at −2578/−1203/−1190/−1179/−1154/−634/−7/+936 locus (4.2%); Ht7, ACAAGGTT (4.1%); Ht8, CCAAACCC (2.7%); Ht9, CCGCGGCC (1.9%); Ht10, CCGAGGCT (1.7%); Ht11, CTGCGGCC (1.6%); Ht12, CCAAGCCC (1.4%); Ht13, CCAAGGCC (1.4%); and Ht14, CCAAAGCT (1.3%).

Table 3. Frequencies of VEGF Haplotypes in 416 Patients with HCC
Haplotype (−2578/−1203/−1190/−1179/−1154/−634/−7/+936)Haplotype Frequency (%)*
  • *

    Only haplotypes observed at a frequency of greater than 5% are shown.


Several genotypes, including −2578AA, −1190AA, −1154AA, and −634GG, showed an association with larger tumor size (P trend = 0.011, 0.007, 0.041, and 0.045, respectively). The −1154AA genotype correlated with invasive characteristics such as a higher number of tumors (P trend = 0.049), nodal invasion (P = 0.014), and ill-defined tumor type (P = 0.014). The +936TT genotype was associated with nodal invasion (P = 0.006). The −2578AA (P trend = 0.015 and 0.008, respectively), −1154AA (P trend = 0.007 and 0.004, respectively), and −634GG genotypes (P trend = 0.015 and 0.025, respectively) were associated with both of the modified UICC and BCLC stages.

The most common haplotype was haplotype 1 (Ht1, CCGAGCCC at −2578/−1203/−1190/−1179/−1154/−634/−7/+936), and we found this to be associated with a smaller tumor size, a lower modified UICC stage, and BCLC stage (P trend =0.007, 0.011, and 0.047, respectively). The increasing copy number of haplotype 3 (Ht3, ACAAAGCC at −2578/−1203/−1190/−1179/−1154/−634/−7/+936) showed a correlation with nodal invasion and an ill-defined tumor type (P trend = 0.006 and 0.005, respectively), and individuals with 2 Ht3 copies had a larger tumor size (P trend = 0.047).

VEGF Genotypes and Survival

Table 4 shows the association between VEGF genotype and haplotype with overall survival. In terms of overall survival, the −2578AA and −1190AA genotypes were associated with high HRs of 1.69 [95% confidence interval (CI), 1.08 to 2.63] and 1.35 (95% CI, 1.01 to 1.81), respectively. The −634CC genotype was associated with a low HR of 0.60 (95% CI, 0.41 to 0.87), which was lower than the HR for the −634GG genotype. Patients with 2 Ht1 copies had a lower HR of 0.53 (95% CI, 0.33 to 0.83) than patients with none or a single Ht1 copy (P = 0.006) (Fig. 2). We found no other alleles to be associated with survival.

Table 4. Association Between VEGF Genotypes and Haplotypes and Overall Survival in 416 Patients with HCC
GenotypeNumber of patientsNumber of deathsHR (95% CI)*P value*
  • *

    HRs (95% confidence interval [CI]) and P values for each genotype or haplotype for overall survival were calculated using Cox proportional regression analysis. Death was defined as death associated with HCC. †Haplotypes (Ht) 1 showed the alleles CCGAGCCC at −2578/−1203/−1190/−1179/−1154/−634/−7/+936. “Others” indicates haplotypes other than the one indicated.

 CA1661111.19 (0.93–1.54)0.163
 AA30231.69 (1.08–2.63)0.020
 GA104671.17 (0.87–1.57)0.290
 AA91661.35 (1.01–1.81)0.046
 GC1991330.93 (0.72–1.21)0.596
 CC70370.60 (0.41–0.87)0.008
 Ht1/Others1551040.94 (0.73–1.21)0.646
 Ht1/Ht145210.53 (0.33–0.83)0.006
Figure 2.

Overall survival after diagnosis of hepatocellular carcinoma in patients with each −2578, −1190, and −634 genotype, and haplotype 1 (Ht1: CCGAGCCC at −2578/−1203/−1190/−1179/−1154/−634/−7/+936) of the VEGF gene. Survival according to each (A) −2578 genotype, (B) −1190 genotype, (C) −634 genotype, and (D) survival in patients homozygous for Ht1. P value was calculated using a log-rank test.

We recalculated The HRs for each −2578, −1190, −634, and Ht1 following adjustments for clinical characteristics. This analysis showed that only the −634 genotype and Ht1 were associated with overall survival, with the adjusted HRs being 0.67 (95% CI, 0.46 to 0.99) for −634CC genotype carriers and 0.57 (95% CI, 0.36 to 0.92) for Ht1 homozygote carriers compared with noncarriers (Table 5). We stratified the overall survival of patients according to VEGF −634 genotypes by each BCLC stage; they are represented in Fig. 3. The predictive power of genotype for survival was not statistically apparent.

Table 5. Cox Multivariate Regression Analysis of Potential Factors for Overall Survival in 416 Patients with HCC
Cox modelsVariablesAdjusted HR (95% CI)*P*
  • *

    HR (95% confidence interval [CI]) and P values for the −634 genotype and haplotype 1 (Ht1) for overall survival were adjusted according to important clinical characteristics. In the adjusted Cox model analysis, the −634 genotype and Ht1 were analyzed separately, and other clinical characteristics were included simultaneously.

For −634 genotype   
 Age >55 years0.65 (0.50–0.84)0.001
 Tumor size >2, ≤5 (cm)1.39 (0.80–2.40)0.239
 Tumor size >5, ≤10 (cm)1.87 (1.06–3.31)0.031
 Tumor size >10 (cm)2.71 (1.45–5.06)0.002
 Number of tumors = 2–32.06 (1.44–2.94)<0.001
 Number of tumors ≥42.90 (2.09–4.01)<0.001
 Nodal invasion1.87 (1.26–2.79)0.002
 Metastasis2.27 (1.59–3.23)<0.001
 CTP classification B2.23 (1.66–2.98)<0.001
 CTP classification C6.42 (3.17–13.00)<0.001
 Performance status 11.18 (0.70–1.99)0.530
 Performance status 21.38 (0.96–1.99)0.086
 Performance status 30.69 (0.22–2.10)0.512
 Vascular invasion1.87 (1.26–2.79)0.001
 Ill-defined tumor type1.97 (1.41–2.76)<0.001
 −634GC0.97 (0.73–1.28)0.809
 −634CC0.67 (0.46–0.99)0.046
For haplotype 1 (Ht1)   
 Age >55 years0.65 (0.50–0.84)0.001
 Tumor size >2, ≤5 (cm)1.37 (0.79–2.36)0.258
 Tumor size >5, ≤10 (cm)1.86 (1.05–3.27)0.033
 Tumor size >10 (cm)2.72 (1.46–5.07)0.002
 Number of tumors = 2–31.99 (1.39–2.83)<0.001
 Number. of tumors ≥42.86 (2.06–3.98)<0.001
 Nodal invasion1.84 (1.23–2.74)0.002
 Metastasis2.22 (1.57–3.15)<0.001
 CTP classification B2.29 (1.71–3.06)<0.001
 CTP classification C6.51 (3.23–13.15)<0.001
 Performance status 11.20 (0.72–2.02)0.485
 Performance status 21.42 (0.98–2.04)0.061
 Performance status 30.67 (0.22–2.05)0.487
 Vascular invasion1.91 (1.31–2.78)0.001
 Ill-defined tumor type1.99 (1.42–2.78)<0.001
 Ht1/Others0.88 (0.67–1.16)0.356
 Ht1/Ht10.57 (0.36–0.92)0.021
Figure 3.

Overall survival of patients with HCC according to VEGF −634 genotypes in each BCLC stage: (A) very early, (B) early, (C) intermediate, (D) advanced, and (E) terminal. P value was calculated using a log-rank test.

VEGF Genotypes and VEGF Expression Levels

The median (range) of plasma and serum VEGF levels were 14.0 (0 to 1070.5) pg/mL and 244.3 (0 to 3169.0) pg/mL, respectively. Serum VEGF concentrations differed according to tumor size (P < 0.001), number of tumors (P < 0.001), nodal invasion (P < 0.001), metastasis, (P < 0.001), presence of vascular invasion (P < 0.001), tumor type (P < 0.001), modified UICC stage (P < 0.001), and BCLC stage (P < 0.001). Plasma VEGF levels showed also similar associations, other than for the number of tumors (P = 0.062), vascular invasion (P = 0.215), and tumor type (P = 0.541). Neither VEGF serum nor plasma levels differed according to any VEGF genotype, even when we stratified data by stage. Serum VEGF levels correlated plasma VEGF (r = 0.30) levels, platelet levels (r = 0.55), WBC levels (r = 0.42), and α-FP level (r = 0.28).

In 63 tissue specimens, the frequencies of VEGF expression were 10 cases (15.9%) for negative, 19 cases (30.2%) for 1+, 28 cases (44.4%) for 2+, and 6 cases (9.5%) for 3+. The median of MVD was 61 (range, 20 to 151). No VEGF genotype showed a correlation with either VEGF expression or MVD in tissue specimens (P > 0.05)


The present study found that VEGF polymorphisms were associated with HCC prognosis. Patients with −634CC or who were Ht1/Ht1 homozygotes showed increased overall survival [adjusted HRs; 0.67 (95% CI, 0.46 to 0.99) and 0.57 (95% CI, 0.36 to 0.92), respectively]. In addition, polymorphisms in the VEGF gene promoter region were associated with aggressive clinical features such as larger tumor size, increased number of tumors, nodal invasion, ill-defined tumor type, modified UICC stage, and BCLC stage. The association between VEGF polymorphisms and overall prognosis was evident even after adjusting for clinical characteristics, indicating that the −634 genotype and the VEGF gene haplotype may be independent prognostic factors for HCC. Although more functional studies on the genotype effect are necessary, these findings may be useful information for the clinical therapeutic trials of targeting VEGF pathway.

We performed survival analysis according to the VEGF −634 genotypes stratified by BCLC stage. (Fig. 3) Although the Kaplan-Meier graph showed a tendency of −634 CC homozygotes to have better survival than −634 GG or GC carriers in the intermediate stage and the advanced stage, the predictive power of genotype for survival was not statistically apparent. We think that the result from the multivariate analysis is more explanatory for prediction of survival. The weakened predictive power of genotype after stratification with the BCLC stage may be because the VEGF genotype has more association with tumor factors than liver function and performance.

Previous case-control studies suggested a correlation between VEGF genotype and cancer risk,33–35 and 2 recent studies showed an association between genotype and prognosis in breast cancer.19, 36 While the latter 2 studies were limited to breast cancer, the results were significant in terms of ethnicity and genotype. One of those studies analyzed 4 polymorphisms (−2578, −1154, −634, and +936) in Polish, German, and Swedish populations and suggested that the −634CC genotype and the −2578/−634 CC haplotype were associated with tumor aggressiveness. The other report was the Shanghai breast cancer study, which analyzed 3 polymorphic sites at positions −1498, −634, and +936 in a Chinese population, and suggested that the −1498C and −634G alleles were associated with poorer survival. Those 2 studies using different ethnic populations reported contrary findings in terms of the effect of the −634 genotype on breast cancer prognosis, and the −634CC genotype frequency was found to be 9.1% in European and 17.5% in Chinese populations.

The genotype frequencies found in the present study were similar to those reported in the Shanghai breast cancer study. The genotype frequencies of the present patients were compared to the VEGF allele frequency data from the Korean population ( and the frequencies reported for the control group of a previous study.37 This comparison showed that there was no significant difference in genotype frequencies between controls and HCC patients. These findings suggest that the effect of genotype or haplotype on cancer prognosis differs according to ethnicity.

In the present study, we found a strong linkage disequilibrium from −2578 to −1498 (r2 = 0.91, D′ = 0.982), indicating that it would be sufficient for future studies to examine only 1 polymorphism in this region. However, the alleles between −634 and +936 appear to require separate analysis.

In the present study, we found no association between VEGF levels (serum or plasma) and any VEGF gene polymorphism. This finding possibly reflects the complex effect of many clinical factors such as tumor burden on VEGF concentration. In addition, many cytokines such as interleukin, nitric oxide, and growth factors regulate VEGF levels, and there is redundancy in the angiogenesis network.4 Because the present study found no correlation between polymorphisms and circulating VEGF levels, the explanation for the correlation between genotype and survival in patients with HCC remains to be determined. Moreover, none of the VEGF polymorphisms showed a correlation with either tumor VEGF expression or MVD assessed by immunostaining on 63 specimens. This negative result may be due to unbalanced samples (50 surgical specimens; mainly in early stage, and 13 biopsies) that did not reflect all of the patients with different stages.

There is debate as to whether serum or plasma is the optimal specimen in which to measure VEGF concentrations, especially in cancer patients.38–41 The present study found that serum VEGF levels strongly correlated with the platelet count (r = 0.55), as shown in previous studies, and that the serum VEGF concentration correlated with tumor aggressiveness. Although plasma VEGF levels were associated with some tumor characteristics, the serum VEGF level may be a more useful marker in HCC patients as it was higher than the plasma VEGF level and was correlated with many clinical features.

In summary, this study investigated the relationship between VEGF gene polymorphisms and the prognosis of HCC patients. This study is the first to show that VEGF polymorphisms may be significant genetic markers for HCC prognosis.


We thank Jin Ohk Kim for her kind support with the immunohistochemical study.