Gene Expression Profiling of Fixed Tissues Identified Hypoxia-Inducible Factor-1α, VEGF, and Matrix Metalloproteinase-2 as Biomarkers of Lymph Node Metastasis in Hepatocellular Carcinoma Free

Hepatocellular carcinoma (HCC) has an extremely poor prognosis and is the fifth most common cancer in the world (1), making it the third leading cause of cancer-related mortality (2). Lymph node metastases (LNM), which contribute to the risk of cancer death (3), are found in approximately 7.45% of HCC patients (4). We have reported that LNM in HCC is sensitive to external beam radiotherapy (5), but the quality of life and survival of patients who undergo this procedure remains unsatisfactory.

LNMs are correlated with advanced disease and poorer patient survival in HCC; however, the mechanisms underlying the development of LNM in HCC are poorly understood. Screening molecules associated with LNM in HCC may enable the development of biomarkers to identify high-risk patients, so steps could be taken to prevent the development of LNM after curative resection. Although we have found that upregulation of VEGF-C and nuclear CXCR4 in HCC is correlated with LNM (6), identifying effective biomarkers for LNM in HCC requires further study.

Gene expression profiling by microarray is a powerful tool for cancer biomarker discovery. Formalin-fixed, paraffin-embedded (FFPE) tissue specimens are stored in large numbers in pathology laboratories (7), thus the ability to carry out gene expression profiling in FFPE specimens should greatly facilitate the correlation of expression profiles with clinical outcomes in both prospective and retrospective studies (8). High-throughput expression profiling of archived FFPE clinical samples of HCC can be achieved with the cDNA-mediated annealing, selection, extension, and ligation (DASL) assay (9). We used this approach to identify biomarkers for predicting LNM in HCC.

We included 36 HCC patients after curative resection at Zhongshan Hospital in whom LNMs (but no additional metastases in other parts of the body) were detected during follow-up. We used age, gender, hepatitis B surface antigen (HBsAg), hepatitis C antibody (HCV-Ab), α-fetoprotein, alanine transaminase, liver cirrhosis, Child–Pugh score, tumor differentiation, tumor size, tumor number, tumor encapsulation, vascular invasion, and TNM cancer stage to match the LNM group with a non-LNM (NLNM) group; these factors did not differ significantly between the 2 groups. The matched NLNM control group consisted of 36 HCC patients in whom neither NLNM nor other metastases were detected during a 5-year follow-up after curative resection. The 72 matched pairs of intratumoral and peritumoral FFPE HCC tissue specimens were obtained from Zhongshan Hospital between February 1999 and March 2004. Paraffin blocks were selected only on the basis of the availability of suitable FFPE tissue and complete clinicopathologic and follow-up data for the patients. Clinicopathologic characteristics are summarized in Supplementary Table S1.

The clinical significance of the candidate biomarkers was evaluated in an independent cohort of 309 consecutive HCC patients, none of whom had distant metastasis before surgery. To exclude the possibility of extrahepatic spread, all patients received a chest x-ray and abdominal ultrasonography before surgery, and bone scanning was done if the patient reported bone pain. If extrahepatic spread was suspected, computed tomography and/or MRI were used to verify whether extrahepatic spread had occurred. Archival specimens from these patients were obtained from Zhongshan Hospital between August 1999 and September 2004. All 309 patients underwent curative resection for HCC, defined as complete macroscopic removal of the tumor (10). Additional inclusion criteria were as follows: HCC diagnosis based on pathology, no prior anticancer treatment, suitable FFPE tissue specimens, and complete clinicopathologic and follow-up data. HCC was classified as Child–Pugh class A in 306 patients and Child–Pugh class B in 3 patients. Tumor stage was determined according to the 2002 International Union against Cancer TNM classification system and the Barcelona Clinic Liver Cancer (BCLC) staging system (2010 version). The histologic grade of tumor differentiation was assigned by the Edmondson grading system. Tumor size was based on the largest dimension of the tumor specimen. The extent of vascular invasion was determined by microscopic examination of the resected specimen. Detailed clinicopathologic features are provided in Table 1. The study protocol was approved by the Zhongshan Hospital research ethics committee. Informed consent was obtained from each patient in accordance with this committee's regulations.

All patients were observed until January 2009; median follow-up was 55.6 months (range, 3.5–120.1 months). Follow-up procedures and diagnosis of LNMs were described in our previous study (6). Briefly, ultrasonographic examination of the liver and lymph nodes of the abdomen and laboratory tests (liver function, α-fetoprotein, and hematologic parameters) were carried out every 3 months by doctors who had no knowledge of the study. If LNM was suspected, computerized tomography scanning or MRI was done immediately. Metastatic lymph nodes appeared as hypoechoic masses on ultrasonography, high-signal intensities on T2-weighted MRI, or hypodensities with a faint rim of hyperdensity on contrast computerized tomography. Time to recurrence was calculated from the date of the operation to the date of recurrence. Treatment modalities after relapse were administered according to a uniform guideline as previously described (5, 11, 12). Briefly, when a diagnosis of LNM was made, external beam radiotherapy was focused on the involved lymph node. Other site relapses received radiotherapy, interventional therapy, or surgery.

Total RNA was extracted from a total of 144 FFPE specimens of intratumoral and matched peritumoral tissues of the HCC patients (patients with LNM, n = 36; patients with NLNM, n = 36). Hematoxylin and eosin (H&E)-stained slides were screened to identify optimal intratumoral and peritumoral tissues for analysis. Total RNA was purified from the tissue specimens by using the High Pure RNA Paraffin Kit (Roche Diagnostics) according to the manufacturer's protocol. Two to four 1-mm cores were removed from each block, placed in 800 μL Hemo-De, and mixed overhead several times. Proteinase K digestion time was 12 hours for each sample. RNA samples were stored at −80°C until use.

For quantitative reverse transcriptase PCR (qRT-PCR) analysis, RNA (200 ng) was first converted to cDNA. PCR amplification was carried out in a 50-μL reaction containing 5 μL cDNA template, 500 nmol/L each primer, and 25 μL SYBR Green PCR Master Mix (Applied Biosystems). Primers used to amplify the RPL13a gene were: forward (5′-GTACGCTGTGAAGGCATCAA-3′) and reverse (5′-GTTGGTGTTCATCCGCTTG-3′). PCR amplification consisted of an initial enzyme activation step at 95°C for 12 minutes, followed by 40 cycles of 95°C for 20 seconds, 54°C for 20 seconds, and 72°C for 60 seconds. RNA with a cycle threshold (C_t) value 28 or less was used for the DASL assay.

DASL experiments were done to identify genes that were differentially expressed between the LNM and NLNM groups of matched intratumoral and peritumoral tissues. The DASL assay was done by using the Human Cancer Panel DASL Assay (Illumina Inc.), which targets 502 known cancer-related genes. The data were normalized with GenomeStudio Gene Expression Module (v1.5.4 Illumina Inc.). Fluorescence intensities from Cy3 and Cy5 dyes were averaged for each probe, and the expression level of each gene was computed as the average intensity of 3 probes. Images were extracted and fluorescence intensities were detected on a BeadArray Reader. Scanned data were uploaded into GenomeStudio for further analysis.

A tissue microarray (TMA) was constructed as described previously (13–15). H&E-stained slides were screened to identify optimal intratumoral and peritumoral tissue for analysis. TMA slides were then constructed (in collaboration with the Shanghai Biochip Company Ltd.) by using samples from the cohort of 309 HCC patients. By using the paraffin-embedded HCC tissue samples, 2 punch cores (longest dimension, 1.0 mm) were collected from nonnecrotic areas of tumor foci. We also collected peritumoral tissue within 10 mm of the tumor to capture possible direct effects from the tumor. TMA sections (4 μm) were constructed with the 309 pairs of intratumoral and matched peritumoral samples.

The immunohistochemistry protocols are described elsewhere (13–15). Primary antibodies were mouse antihuman monoclonal antibodies against hypoxia-inducible factor-1α (HIF-1α), VEGF (Santa Cruz Biotechnology), and matrix metalloproteinase-2 (MMP-2; R&D Systems). TMA slides were incubated with primary antibodies overnight at 4°C and then washed. The components of the EnVision Plus Detection System (EnVision/HRP/Mo; Dako) were applied, and reaction products were visualized by incubation with 3,3′-diaminobenzidine. Sections were dehydrated, counterstained with hematoxylin, and mounted. Negative controls were prepared in the same way but without primary antibodies.

Positive staining was measured with a computerized image system composed of a Leica CCD camera DFC420 connected to a Leica DM IRE2 microscope (Leica Microsystems Imaging Solutions Ltd.). Photographs of 10 representative fields were captured by the Leica QWin Plus v3 software; identical settings were used for each photograph. Immunohistochemical labeling was assessed on a compound microscope by 3 of the authors who were blinded to patient outcome. We randomly selected 10 fields (magnification, 400×; 100 cells/hpf) and counted 1,000 cells in each core (16). The percentage of positive cells was determined by each observer, and the average of the 3 scores was calculated. Both nuclear and cytoplasmic levels of HIF-1α expression were evaluated: low expression was defined as nuclear staining in less than 10% of cells and/or weak cytoplasmic staining; high expression was defined as nuclear staining in 10% or more of cells and/or distinct strong cytoplasmic staining (17). Cytoplasmic VEGF staining was categorized as follows: low expression, less than 25% of cells; high expression, 25% or more of cells (18). Cytoplasmic MMP-2 staining was categorized as follows: low expression, less than 50% of cells; high expression, 50% or more of cells (19).

Statistical analyses were done with SPSS 16.0 Software (SPSS). The data were censored at the last follow-up for living patients. Qualitative variables were compared by Fisher's exact test and quantitative variables were analyzed by Pearson's correlation test. Relationships between the expression of these proteins and LNM and recurrence were evaluated as censored time-to-event data by using a log-rank test and Cox regression model. Two-tailed P < 0.05 was judged to be significant. Microarray data were normalized by using average algorithm implemented in GenomeStudio software ver.3 (Illumina, Inc.; according to GenomeStudio Gene Expression Module v1.0 User Guide). Differential gene expression analysis was done by using Illumina custom method. Correction for multiple hypothesis testing was conducted with the use of false discovery rate.

RNA was isolated from all 144 (100%) FFPE blocks. Testing by qRT-PCR revealed that 110 of the 144 (76.4%) samples were suitable for DASL analysis (C_t ≤ 28). The 110 samples consisted of RNA from 28 pairs of matched intratumoral and peritumoral tissues from the LNM group, 23 pairs of matched intratumoral and peritumoral tissues from the NLNM group, and 8 unmatched samples from 6 intratumoral tissues and 2 peritumoral tissues. To eliminate the effect of clinicopathologic factors other than LNM on gene expression, we used age, gender, HBsAg, HCV-Ab, α-fetoprotein, alanine transaminase, liver cirrhosis, Child–Pugh score, tumor differentiation, tumor size, tumor number, tumor encapsulation, vascular invasion, and TNM stage to match the LNM group with the NLNM group, and verified that there were no statistical differences between the two groups (Table 2). We selected RNA from 20 of the 28 pairs of matched intratumoral and peritumoral tissues in the LNM group and from 20 of the 23 pairs of matched intratumoral and peritumoral tissues in the NLNM group, for a total of 80 RNA samples (40 HCC patients) for the DASL assay.

We identified 83 genes that were differentially expressed between the LNM and NLNM groups (Supplementary Table S2). In intratumoral tissue of the LNM group, 17 genes were overexpressed and 19 genes were underexpressed compared with the NLNM group. In peritumoral tissue of the LNM group, 25 genes were overexpressed and 22 genes were underexpressed compared with the NLNM group (for microarray datasets see http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE28248, accession numbers GSE28248). Heatmaps are shown in Figure 1. On the basis of previous reports (17, 20–23), we chose HIF-1α, VEGF, and MMP-2 for further analysis. Immunohistochemistry and TMA analyses of these proteins were conducted with an independent cohort of 309 HCC patients to evaluate the ability of these potential biomarkers to predict LNM.

HIF-1α was observed in the cytoplasm and nucleus of tumor cells and hepatocytes, whereas VEGF and MMP-2 were localized primarily in the cytoplasm (Fig. 2). In intratumoral tissue, 85 (27.5%) of the 309 specimens showed high HIF-1α expression and 224 (72.5%) showed low HIF-1α expression; in peritumoral tissue, 29 (9.3%) specimens showed high HIF-1α expression and 280 (90.7%) showed low HIF-1α expression. With regard to VEGF, in intratumoral tissue, 104 (33.7%) specimens showed high expression and 205 (66.3%) showed low expression; in peritumoral tissue, 21 (6.8%) specimens showed high expression and 288 (93.2%) showed low expression. With regard to MMP-2, in intratumoral tissue, 91 (29.4%) specimens showed high expression and 218 (70.6%) showed low expression; in peritumoral tissue, 49 (15.9%) specimens showed high expression and 260 (84.1%) showed low expression.

As shown in Supplementary Table S3, high intratumoral HIF-1α expression was correlated with vascular invasion (P = 0.027), advanced TNM stage (P = 0.046), and BCLC stage (P = 0.020). Advanced TNM stage and BCLC stage were also correlated with high intratumoral VEGF expression (P < 0.001 and P = 0.019, respectively) and high intratumoral MMP-2 expression (P = 0.037 and P = 0.015). Intratumoral HIF-1α expression was correlated with intratumoral VEGF expression (r = 0.159, P = 0.005) and with intratumoral MMP-2 expression (r = 0.286, P < 0.001). Intratumoral VEGF expression was correlated with intratumoral MMP-2 expression (r = 0.231, P < 0.001). The expression levels of these proteins in peritumoral tissues were not correlated, and peritumoral protein expression was not correlated with clinicopathologic features.

Of the 309 HCC patients, survival rates were 50.2% (LNM-free survival), 45.3% (disease-free survival), and 51.0% (overall survival) at 5 years. Thirty-one patients developed LNM: 11 of 31 (35.5%) patients had solitary LNM and 20 of 31 (64.5%) patients had multiple LNM. Of these 31 patients with LNM, 9 (29.0%) patients had portal LNM, 10 (32.3%) patients had peripancreatic LNM, and 12 (38.7%) patients had para-aortic LNM. As shown in Supplementary Figure S1, Kaplan–Meier analyses revealed that patients with high intratumoral HIF-1α expression (P = 0.018), high VEGF expression (P = 0.017), high MMP-2 expression (P = 0.003), and the combination of high intratumoral HIF-1α, VEGF, and MMP-2 (P = 0.001) were more likely to develop LNM. Cox regression univariate analyses of candidate predictors showed that vascular invasion (P = 0.025), intratumoral levels of HIF-1α (P = 0.030), VEGF (P = 0.048), MMP-2 (P = 0.016), and the combination of intratumoral HIF-1α, VEGF, and MMP-2 (HIF-1α low expression + VEGF low expression + MMP-2 low expression vs. HIF-1α high expression and/or VEGF high expression and/or MMP-2 high expression; P = 0.007) were associated with LNM (Table 3). Cox regression multivariate analyses identified the following independent risk factors for LNM: high intratumoral HIF-1α expression (HR = 2.281, 95% CI: 1.124–4.630, P = 0.022), high VEGF expression (HR = 2.289, 95% CI: 1.131–4.634, P = 0.021), high MMP-2 expression (HR = 2.812, 95% CI: 1.389–5.693, P = 0.004), and the combination of intratumoral HIF-1α, VEGF, and MMP-2 (HR = 3.320, 95% CI: 1.430–7.708; P = 0.005).

For patients who developed metastases that did not involve the lymph nodes, Kaplan–Meier and log-rank test results indicated that patients with high intratumoral HIF-1α expression (P = 0.039) were more likely to develop NLNMs; however, no difference in the rate of NLNMs was observed between patients with high intratumoral VEGF versus low VEGF expression (P = 0.313) or high MMP-2 versus low MMP-2 expression (P = 0.059). Rates of overall metastasis were higher in patients with high HIF-1α expression (P = 0.008), high VEGF expression (P = 0.031), and high MMP-2 expression (P = 0.001).

In this study, we analyzed FFPE specimens of intratumoral and peritumoral tissues from HCC patients to investigate differential gene expression between HCC with LNM and HCC without LNM (NLNM). The quality of RNA obtained from FFPE specimens is not satisfactory for traditional gene expression analysis, and the storage of frozen tissues is usually short, so they are rarely used to clinical correlation analysis. We therefore used a DASL assay to analyze 502 genes known to be involved in carcinogenesis, and showed that this technology is able to carry out gene expression profiling with RNA extracted from archived FFPE specimens. We identified 83 genes that were differentially expressed between the LNM and NLNM groups.

HCC patients have a poor prognosis, and extrahepatic metastasis is a primary cause of poor quality of life and low survival rates. The rate of LNM in extrahepatic metastases of HCC is 33.8% (24); however, the clinical significance of LNM and its regulation in HCC are unclear. Although the biology of LN-specific metastasis of HCC is poorly understood, gene expression profiling of primary tumors may be useful to predict survival and the risk of metastasis (25, 26). It has been reported that genes favoring metastasis progression are initiated in the peritumoral liver microenvironment of HCC (9, 13). Further, Lee and colleagues reported that frozen HCC tissue can provide genome-wide information that contributes to an improved understanding of molecular alterations during LNM in HCC (27). In this study, we evaluated gene expression in primary tumors and in the peritumoral liver microenvironment and identified 83 genes that were differentially expressed between the LNM and NLNM groups. Analysis of an independent cohort of 309 HCC patients showed that intratumoral protein expression of HIF-1α, VEGF, and MMP-2 combined can predict LNM.

HIF-1α is a transcription factor that regulates critical pathways involved in tumor growth and metastases. It is a positive factor for cancer growth, and its increased activation is correlated with an aggressive phenotype in a model of epidermal carcinogenesis (28, 29). Intratumoral HIF-1α is correlated with poor prognosis and enhanced metastatic potential (30, 31). Hamaguchi and colleagues reported that activation of an HIF-1α–regulated glucose metabolism is closely related to the aggressive phenotype of HCC (32). In this study, we found that high HIF-1α expression in intratumoral tissue was correlated with LNM. Although a previous study reported that HIF-1α protein expression is associated with tumor differentiation and metastases in HCC (33), we did not observe a relationship between HIF-1α expression and tumor differentiation in the present study. Overexpression of HIF-1α in esophageal squamous cell carcinoma has been reported to correlate with LNM and advanced pathologic stage (17). Similarly, we found that intratumoral HIF-1α overexpression correlates with LNM in HCC.

VEGF belongs to the platelet-derived growth factor family and is the most important known inducer of angiogenesis and vessel permeability. VEGF induces migration of endothelial cells and formation of new vessels; overexpression of VEGF has been associated with tumor progression and poor prognosis in several tumors, including breast cancer (34), prostate cancer (35), colorectal cancer (31), and renal cell carcinoma (36). Furthermore, VEGF expression in salivary gland carcinoma correlates with LNM (22). In this study, we found that intratumoral overexpression of VEGF was associated with LNM in HCC. This finding is consistent with a study by Lee and colleagues, who observed that HCC cell invasiveness was regulated by VEGF (37).

Tumor invasion and metastasis are key steps in cancer progression and involve the degradation of basement membranes and subsequent remodeling of the extracellular matrix. MMPs are thought to play a central role in these processes; MMP-2 is of particular importance in tumor progression because it is capable of cleaving type IV collagen. MMP-2 is correlated with poor survival in prostate cancer (38), HCC (39), and breast cancer (40). In addition, upregulation of MMP-2 protein expression has been observed in cervical cancer tissues and is associated with LNM (23). In this study, we found that MMP-2 is an independent prognostic factor for LNM in HCC and may be a useful predictive marker for LNM in HCC.

Hypoxia-driven VEGF expression is considered to be a principal inducer of angiogenesis in tumors, and HIF-1α regulates VEGF transcription. In this study, we found that intratumoral HIF-1α was correlated with intratumoral VEGF, suggesting that the HIF-1α/VEGF pathway is involved in the development of LNM in HCC. We also observed that upregulation of HIF-1α, VEGF, and MMP-2, as assessed by immunohistochemistry, correlated with LNM. Furthermore, the combination of these 3 proteins was found to be an independent risk factor for developing LNM. Our finding that the combination of HIF-1α, VEGF, and MMP-2 may serve a molecular predictor for LNM is consistent with previous reports of HIF-1α (41), VEGF (42), and MMP-2 (43) as potential biomarkers of LNM in other tumors. Our findings also indicate that the HIF-1α/VEGF signaling pathway may be associated with MMP-2 expression, although the precise mechanism underlying the association between these proteins and HCC with LNM requires further study. In this study, we found that intratumoral HIF-1α expression was also correlated with NLNM, but VEGF and MMP-2 were not. Thus HIF-1α may play a more important role in HCC metastasis.

In conclusion, we analyzed RNA from FFPE tissue specimens with the DASL assay and identified 83 genes that were differentially expressed between LNM and NLNM in HCC. This study suggests that intratumoral expression of HIF-1α, VEGF, and MMP-2 can predict LNM in patients with HCC and may provide a prognostic indicator for clinical decision making.

No potential conflicts of interest were disclosed.

The authors thank Wei-Zhong Wu and Xiao-Dong Zhu (Liver Cancer Institute, Zhongshan Hospital) for expert technical assistance.

The work was supported by grants no. 30973500 from the National Natural Science Foundation of China and Youth Science Foundation of Zhongshan Hospital.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.