Genome-wide Analysis of Gene Expression in Human Hepatocellular Carcinomas Using cDNA Microarray

INTRODUCTION

Primary HCC 3 is one of the most common malignancies in the world. Despite development of novel therapeutic methods in recent years, prognosis of advanced HCC remains very poor. Major risk factors for HCC are chronic hepatitis resulting from infection with HBV or HCV, and exposure to various exogenous carcinogens including aflatoxin B1 (1) . Molecular approaches have recently revealed involvement of altered TP53, CTNNB1 (β-catenin), and/or AXIN1 genes in hepatocarcinogenesis (2 , 3) . However, these genetic changes do not precisely reflect the biological nature of cancer cells or the clinical characteristics of individual HCC patients. Like other cancers, HCCs manifest diverse clinicopathological and biological phenotypes including grade of differentiation, proliferation rate, ability to invade vessels, potential for metastasis, sensitivity to chemotherapeutic agents, and so on. Hence, analysis of expression profiles of a large number of genes in clinical HCC materials is an essential step toward clarifying the detailed mechanisms of hepatocarcinogenesis and discovering target molecules for the development of novel therapeutic drugs.

cDNA microarray technology, which enables investigators to obtain comprehensive data with respect to gene-expression profiles, is progressing rapidly. Several studies have already demonstrated the usefulness of this technique for identifying novel cancer-related genes and for classifying human cancers at the molecular level (4 , 5) .

In this paper, we report the identification of genes the expression of which has been altered during hepatocarcinogenesis through the use of a genome-wide cDNA microarray containing 23,040 genes. Expression profiles of these genes in 20 primary HCCs fell into three categories that correlated well with the infection status and type of hepatitis virus. Analyses of these profiles along with clinicopathological data also facilitated identification of genes associated with tumor differentiation and vessel invasiveness. This large body of information not only furthers an understanding of the mechanisms of hepatocarcinogenesis but also reveals novel features of known genes and identifies additional biological factors involved in liver cancer.

MATERIALS AND METHODS

Patients and Tissue Samples.

Primary HCCs and corresponding noncancerous liver tissues were obtained with informed consent from 20 patients who underwent hepatectomy. Patient profiles were obtained from medical records. Serologically, 10 cases were hepatitis B surface antigen-positive and 10 cases were HCV-positive. No cases with coinfections of HBV and HCV were included in this study. Histopathological classification was performed according to the Edmondson grading system; clinical stages were determined according to the Union International Contre Cancer TNM classification. No significant differences were seen between HBV-positive and HCV-positive status with respect to age, sex, grade of differentiation, vessel invasion, or tumor stage.

cDNA Microarrays.

We fabricated a “genome-wide” cDNA microarray with 23,040 cDNAs selected from the UniGene database of the National Center for Biotechnology Information. The cDNAs were amplified by reverse transcription-PCR using poly(A)+RNA isolated from various human organs as templates; lengths of the amplicons ranged from 200 to 1100 bp without repetitive or poly(A) sequences. The PCR products were spotted in duplicate on type-7 glass slides (Amersham) using an Array Spotter Generation III (Amersham). Each slide contained 52 housekeeping genes, to normalize the signal intensities of the different fluorescent dyes.

RNA Preparation, Hybridization, and Acquisition of Data.

Frozen specimens were serially sectioned in 10-μm slices and stained with H&E to define the analyzed regions. To avoid cross-contamination of cancer and noncancerous cells, we prepared these two populations by laser-captured microdissection. Total RNA was extracted from each population and then amplified using Ampliscribe T7 Transcription Kit (Epicentre Technologies). The preparation of probes, hybridization, and scanning was performed as described previously (6) . The fluorescence intensities of Cy5 (nontumor) and Cy3 (tumor) for each target spot were adjusted so that the mean Cy5 and Cy3 intensities of 52 housekeeping genes for each slide were equal.

Validation of Data.

To assess the reproducibility of the normalized intensity ratios, we compared the log₂(Cy3:Cy5 intensity ratio) of the 52 housekeeping genes between different slide sets. When the difference between normalized logarithmic ratios from two experiments was less than 1.0, we defined the data as reproducible. The reproducibility was more than 90% when the intensities of Cy3 and Cy5 were both above 25,000.

Classification of 20 HCCs According to Gene Expression Profiles.

We applied the hierarchical clustering method to both genes and samples. To obtain reproducible clusters, we used only selected genes that passed the cutoff filter (both Cy3 and Cy5 signals greater than 25,000 in more than 80% cases examined). The analysis was performed using web-available software (“Cluster” and “TreeView”) written by M. Eisen. 4 Before applying the clustering algorithm, the fluorescence ratio for each spot was first log-transformed; then the data for each sample were centered to remove experimental biases.

Identification of Genes Responsible for Clinicopathological Factors.

We first arranged the relative expression of each gene (Cy3:Cy5 intensity ratio) into one of four categories: up-regulated (ratio, >2.0), down-regulated (ratio, <0.5), unchanged (ratio, between 0.5 and 2.0), and not expressed (or slight expression but under the cutoff level for detection). We used these categories to detect changes in expression that were common among samples as well as specific to a certain subgroup. To detect differentially expressed genes, we recorded the number of samples in each category within each subgroup, for each gene. Then we calculated the U values of Mann-Whitney tests, which measured how the sample distributions between subgroups overlap. The number of samples within each group is counted and, according to the order of the category, the number of overlapped samples is incorporated into the U value. A small U shows that the sample distribution of the two groups is clearly separated, e.g., commonly up-regulated in the HBV group and down-regulated in the HCV group. We applied a hierarchical clustering algorithm to all of the selected genes using hamming distance (edit distance).

RESULTS AND DISCUSSION

Identification of Genes That Were Differently Regulated in HCCs.

To identify genes generally involved in hepatocarcinogenesis, we compared expression profiles between 20 HCCs and their corresponding noncancerous liver tissues by means of cDNA microarray. We excluded individual data when Cy3 and Cy5 signals were <25,000 because data were not reliable for genes giving low signal intensities (see “Materials and Methods”). When we applied a cutoff signal:intensity ratio of cancer:noncancer at 2.0 165 genes including 69 ESTs were selected as being up-regulated in 75% or more of the cases examined (Table 1) ⇓ ⇓ . This list of up-regulated genes contained MAP4K1 as well as MAPK3, suggesting that activation of the MAPK pathway is a common feature of hepatocarcinogenesis. Interestingly, expression of several genes associated with mitosis, including CDC23, TUBG1, CBX1, CKS1, PCTK1, PSMD8, CSE1L, TTK, and PLK1, was commonly increased in cancer cells. As a cell-cycle modulator, CDC23 is a known component of the anaphase-promoting complex (APC) and leads to metaphase/anaphase transition through degradation of cyclin B. TUBG1 (γ-tubulin) and CBX1 participate in centrosome formation (7 , 8) ; CKS1 and PCTK1, encoding cdc2/cdc28 kinases, are essential for activation of the anaphase-promoting complex. PSMD8 (26S proteasome subunit p31) is reportedly responsible for activation of these kinases (9) . Others have reported that CSE1L, TTK, and PLK1 are associated with formation of the mitotic spindle (7 , 10) and that PLK1 can affect the number of centrosomes when exogenously expressed (11) ; overexpression of PLK1 has been correlated with poor prognosis in a subset of human cancers (12) . Our comprehensive expression data for these genes may account for a high incidence of chromosomal instability in HCC, and they suggest that promotion of the mitotic process is generally involved in hepatocarcinogenesis. Therefore, regulation of these mitosis-associated genes either by chemotherapeutic agents or by gene delivery might be an effective therapeutic strategy for HCCs.

We also looked for down-regulated genes and found 170 (including 75 ESTs) that were underexpressed in 65% or more of the HCCs examined (Table 2) ⇓ ⇓ when we applied a cutoff intensity ratio of cancer:noncancer at 0.5. The majority of the down-regulated genes encoded hepatocyte-specific gene products (e.g., complement species, amyloid, and albumin) and detoxification enzymes (cytochrome P-450 and metallothionein families), reflecting de-differentiation of cancer cells. Regarding retinoid metabolism, LY6E and RBP1, both of which appear to play roles in retinoid-induced differentiation (13 , 14) were repressed, as was IGFBP3, which also is involved in the retinol-mediated inhibition of HCC development (15) . Because retinoid is an accepted therapy to encourage differentiation of cells in acute promyelocytic leukemia and is thought to help prevent development of HCC (16) , reduced expression of these genes may play a crucial role in hepatocarcinogenesis.

We identified 69 ESTs that were frequently up-regulated and 75 that were frequently down-regulated, which indicated that a large number of genes of unknown function are also involved in hepatocarcinogenesis.

Classification of HCCs by Gene Expression Profiles.

We further investigated whether clinical HCCs could be classified into groups on the basis of their gene-expression profiles. For this purpose, we used the hierarchical clustering method. To obtain reproducible clusters, we selected 4,531 genes that passed the cutoff filter (both cy3 and cy5 signals greater than 25,000). The overall expression patterns across 20 HCC samples are shown in Fig. 1 ⇓ . The analyses resulted in the clustering of identical genes spotted on different positions into adjacent rows, indicating the reliability of the expression data. The 20 HCCs examined fell into three groups, as the dendrogram shows.

To clarify the factors responsible for this classification, we carried out Spearman rank-correlation tests and examined clinicopathological factors including tumor differentiation, hepatitis-virus infection, TNM classification, vascular invasion, intrahepatic metastasis, and gender of the patients (data not shown). However, only the type of hepatitis virus correlated closely with these clusters (P = 0.0079). Therefore, HBV-positive and HCV-positive HCC may result from distinct mechanisms and be different in character as a consequence of differently expressed genes.

Identification of Genes Related to HBV-positive or HCV-positive Status.

To identify genes responsible for the differences between HBV-positive and HCV-positive tumors, we performed Mann-Whitney tests and found that 19 known genes and 21 ESTs showed significantly different expression patterns between these two groups. Among the 19 known genes (Fig. 2) ⇓ , seven (GPX2, CYP2E, EPHX1, AKR1C4, FMO3, UGT1A1, and UGT2B10) encode key molecules for activating chemotherapeutic drugs or detoxifying xenobiotic carcinogens.

Most carcinogens are metabolized by Phase I modification enzymes that generate activated intermediates that are then detoxified by Phase II conjugation enzymes (17) . Phase I enzymes CYP2E, AKR1C4, EPHX1, and FMO3 convert several pro-carcinogens to activated metabolites. For example, dimethylnitrosamine is activated by CYP2E, and polycyclic aromatic hydrocarbons are activated by EPHX1 and AKR1C4 (18, 19, 20) . In our study, we observed increased expression of genes encoding these four enzymes exclusively in HCV-positive HCCs, which may suggest that their enhanced expression leads to a greater contribution of carcinogenic metabolites to the mechanisms of HCV-specific hepatocarcinogenesis.

On the other hand, expression of UGT1A1, UGT2B10, and GPX2 was preferentially repressed in HBV-positive HCCs (UGT1A1 was reduced in 8 of 10 HBV-positive HCCs examined), but expression levels of these genes were unchanged in most HCV-positive HCCs. In accordance with our observations, Strassburg et al. (21) have shown decreased expression of UGT1A1 in HCCs as well as in hepatic adenomas, implicating UGT1A1 in an early step of hepatocarcinogenesis. UGT1A1 and UGT2B10 catalyze Phase II conjugation reactions, which are frequently related to detoxification of the active forms of carcinogens. GPX2, a major form of glutathione peroxidase in liver, functions as an antioxidant, and decreased glutathione peroxidase activity in HCCs has been reported elsewhere (22) . Hence, reduced activities of these enzymes may reflect enhanced exposure of hepatocytes to activated carcinogens or radicals. Our results suggest that decreased expression of detoxification enzymes may be involved especially in the mechanisms of HBV-specific hepatocarcinogenesis. Furthermore, because UGT1A1 also catalyzes glucuronidation of SN-38, an active form of irinotecan (23) , HBV-positive HCCs may show greater sensitivity to irinotecan than do HCV-positive HCCs. Different expression patterns among detoxification enzymes should provide information for optimizing the choice and/or the dosage of anticancer drugs for treating HCC patients on an individual basis.

Results of comparing expression profiles between HBV-positive and HCV-positive HCCs implied that hepatitis viruses affect expression of dozens of genes in HCC in a type-specific manner, invoking partly different mechanisms of carcinogenesis. Consequently, identification of genes defining virus-type-specific expression profiles would contribute to our ability to develop virus-type-dependent treatment regimens.

Identification of Genes Related to HCC Progression.

As in the multistep model of adenoma-to-carcinoma sequence accepted for colorectal tumors, HCCs are considered to develop as well-differentiated tumors and then progress to moderately-to-poorly differentiated states (24) . A comparison of expression profiles between well-differentiated tumors (Edmondson grade I; n = 7) and moderately to poorly differentiated tumors (Edmondson grade II or III; n = 13; Fig. 3A ⇓ ⇓ ) by means of Mann-Whitney test identified a total of 321 genes (including 193 ESTs) that showed different expression patterns between the two histologically divided groups. In addition to the genes encoding liver-specific proteins, they included genes associated with apoptosis and the immune system. Apoptosis-related genes including TNFSF10, TNFSF14, GADD34, CFLAR, CLU, CASP6, and phosphatidylserine receptor (25 , 26) were preferentially reduced in moderately-to-poorly differentiated tumors, implying that a reduced rate of apoptosis is a major characteristic of tumor progression. Genes associated with immune systems included MAGEC1, one of the tumor antigens recognized by CTLs, whose expression was also repressed only in moderately-to-poorly differentiated tumors. Reduced expression of genes encoding immune targets may confer a growth advantage by allowing tumor cells to escape from immune surveillance.

Furthermore, we investigated expression profiles with respect to vascular invasiveness because vascular invasion is a major factor affecting metastasis and one of the most useful predictive factors of prognosis (27) . Genes involved in vascular invasion could also represent good candidates for new therapeutic targets. We found that 151 genes (including 110 ESTs) were expressed significantly differently between noninvasive (n = 8) and invasive (n = 12) tumors (Fig. 3B) ⇓ . Among the named genes in this category, 19 were associated with both vascular invasion and tumor differentiation, but no apoptosis-related gene was among them; therefore, reduced apoptosis is likely to be correlated with tumor de-differentiation and growth, but not with vascular invasion or metastasis. Genes associated with vascular invasion contained ARHC (RhoC), which was recently reported to play a crucial role in metastasis (28) . We also found that two other small GTPase-related genes, ARHGAP8 (RhoGAP8) and ARHGEF6, were preferentially down-regulated in invasive tumors. Because the small-GTPase Rho family plays important roles in controlling cell motility and focal adhesions (29) , alterations of their signaling pathways could enhance the migratory and invasive capacity of tumor cells and induce tumor invasion and metastasis. Although its function is unknown, RhoGAP8 is thought to inhibit the Rho signaling pathway; hence, reduced expression of ARHGAP8 may also result in Rho-mediated tumor invasion. Our results suggest that controlling the Rho signaling pathway either by reducing expression of ARHC or by inducing ARHGAP8 may suppress tumor invasion and subsequent metastasis.

The genes and their products represented by the numerous ESTs of unknown function that we classified in the same clusters as genes associated with apoptosis or immunity may be useful as novel targets for drug discovery or tumor markers. Accumulation of data with respect to expression profiles of cancer specimens, clinicopathological data, sensitivity to treatment, and prognosis will not only help us to understand the precise mechanisms of carcinogenesis but also yield practical information for identifying optimized therapeutic modalities and novel therapeutic targets.