Genome-Wide Analysis of Gene Expression in Intestinal-Type Gastric Cancers Using a Complementary DNA Microarray Representing 23,040 Genes

INTRODUCTION

Although its incidence is declining, probably as a result of environmental changes, gastric cancer still appears to be the second leading cause of cancer death in the world (1) . Surgery is the mainstay in terms of treatment because chemotherapy remains unsatisfactory. Gastric cancers at an early stage can be cured by surgical resection, but prognosis of advanced gastric cancers remains very poor. Therefore, the development of novel therapeutic modalities is an issue of great clinical importance.

Histological studies have classified gastric carcinomas into two distinct groups, namely the intestinal (or differentiated) type and the diffuse (or undifferentiated) type (2) , having different features with regard to epidemiology, etiology, pathogenesis, and biological behavior. The intestinal type occurs more commonly in elderly people and has better prognosis, but diffuse-type gastric cancer is seen in relatively younger individuals, without preference for either sex, and displays a more invasive phenotype with a serious clinical course. Intestinal-type gastric cancer is presumed to result from atrophic gastritis, followed by progression to intestinal metaplasia and/or dysplasia (3) , but the precursor lesion of the diffuse-type tumor is not known.

Epidemiological and experimental studies have revealed that a high intake of smoked, salted, and nitrated foods and a low intake of vegetables and fruits increase the risk of gastric cancer and also that Helicobacter pylori infection is a risk factor for the disease (4) . Molecular investigations have provided evidence that multiple genetic alterations are involved in gastric tumorigenesis. Loss of heterozygosity (LOH) is observed frequently at loci on chromosomes 1p, 5q, 7p, 12q, 13q, 17p, 18q, and Y (5) . Genetic alterations and/or amplification of oncogenes including KRAS2 (c-K-ras), CTNNB1 (β-catenin), ERBB2 (c-erbB-2), FGFR2 (K-sam), CCNE1 (cyclin E), and HGFR (c-MET) play roles in some gastric cancers, and inactivation of tumor suppressor genes such as p53, RB, APC, DCC, and/or CDH1 (E-cadherin) is often a factor as well (6) . In fact, germ-line mutation in CDH1 is responsible for disease in a subset of patients with familial gastric cancer, who usually suffer from diffuse-type tumors (7) . On the other hand, mutations in APC or CTNNB1 are observed preferentially in intestinal-type tumors. Although genetic changes in the genes just mentioned have been intensively studied, altered expression of larger numbers of genes in gastric cancer tissues remains to be disclosed.

A critical factor affecting the prognosis of most solid tumors in humans is lymph node metastasis, an independent risk factor for recurrence of gastric cancer. A few genes are already known to be associated with this process; for example, amplification of c-erbB-2 (8) , enhanced expression of VEGF-c (9) , or reduced expression of nm23 or E-cadherin (10 , 11) correlate with metastasis to lymph nodes. However, the molecular mechanisms involved remain unclear.

In this article, we report a genome-wide analysis of gene-expression profiles of intestinal-type gastric cancer tissues obtained by laser-capture microdissection; RNAs from the tumor cells were hybridized to a cDNA microarray containing 23,040 genes. In this manner, we identified a large set of genes with altered expression in intestinal-type cancers as well as a small number associated with lymph node metastasis.

MATERIALS AND METHODS

Patients and Tissue Samples.

Primary gastric cancers and corresponding noncancerous gastric mucosae were obtained with informed consent from the patients who underwent gastrectomy. Patient profiles were obtained from medical records. Histopathological classification of each tumor, performed according to the Lauren’s classification (2) , diagnosed all of the samples as intestinal-type adenocarcinomas. Clinical stage was determined according to the Union International Contre Cancer tumor-node-metastasis classification. The 20 gastric cancer tissues initially analyzed included 18 advanced (T₂-T₄) and two early (T₁) cases. The advanced category included nine node-positive and nine node-negative tumors. No significant differences were seen between node-positive and node-negative patients with respect to age, sex, depth of tumor, or tumor grade. All of the samples were immediately frozen and embedded in TissueTek OCT medium (Sakura, Tokyo, Japan) and stored at −80°C until used for microarray analysis.

Laser-Capture Microdissection, Extraction of RNA, and T7-based RNA Amplification.

Cancer cells and noncancerous gastric epithelium (∼2 × 10⁴ cells from each sample) were selectively collected from the preserved samples using laser-capture microdissection (12) . Extraction of total RNA and T7-based amplification were performed as described previously (12) . Aliquots (2.5-μg) of aRNA 3 from each cancerous and noncancerous tissue were labeled with Cy3-dCTP and Cy5-dCTP, respectively.

cDNA Microarray and Analysis of Data.

Fabrication of the cDNA microarray slides, hybridization, washing, and detection of signals were carried out as described previously (12) . The fluorescence intensities of Cy5 (nontumor) and Cy3 (tumor) for each target spot were adjusted so that the mean Cy3:Cy5 ratios of 52 housekeeping genes were equal to one. Because data derived from low signal intensities are less reliable, we first determined cutoff values for signal intensities on each slide so that all of the filtered genes had greater S:N (signal to noise) ratios of Cy3 or Cy5 than three, and we excluded genes for further analysis when both Cy3 and Cy5 dyes gave signal intensities lower than the cutoff. To estimate the range of expression ratio within which the expression change could be considered as fluctuation in noncancerous cells, we compared expression profiles of noncancerous epithelial cells from five patients. Because 90% of expression ratios in noncancerous cells fell within the range of 1.73 and 0.50, we categorized genes into three groups according to their expression ratios (Cy3:Cy5): up-regulated (ratio, ≥2.0); down-regulated (ratio ≤0.5); and unchanged expression (ratios, between 0.5 and 2.0). Genes with Cy3:Cy5 ratios >2.0 or <0.5 in more than 75% of the cases examined were defined as commonly up- or down-regulated genes, respectively. The significance of altered expression of each gene was also evaluated by calculating the Ps using a random permutation test as described previously (12) . A permutation P of <0.01 was considered to be significant.

Quantitative RT-PCR.

We selected three commonly up-regulated genes (SOX9, NME1, PLAB) and a gene with altered expression between node-positive and node-negative tumors (NEDD8) and examined their expression levels by applying the real-time PCR technique (TaqMan PCR; Applied Biosystems, Foster City, CA). The glutaminyl-tRNA synthetase (QARS) gene served as an internal control because it showed the smallest Cy3:Cy5 fluctuation in our experiments. The TaqMan assay was carried out with the same aRNAs used for array analysis, and those from 11 additional samples that were not used for array analysis, according to the manufacturer’s protocol. The PCR process was started at 95°C for 10 min, then underwent 40 cycles at 95°C for 15 s and 60°C for 1 min. The sequences of primers and probes were as follows: QARS forward primer, 5′-GGTGGATGCAGCATTAGTGGA-3′ and reverse, 5′-AAGACGCTCAAACTGGAACTTGTC-3′; probe, 5′-VIC-CTCTGTGGCCCTGGCAAAACCCTT-TAMRA-3′; NME1 forward primer, 5′-CAGAGAAGGAGATCGGCTTGTG-3′ and reverse, 5′-CTTGTCATTCATAGATCCAGTT-3′; probe, 5′-FAM-CACCCTGAGGAACTGGTAGATTACACGAGC-TAMRA-3′; PLAB forward primer, 5′-GTGCTCATTCAAAAGACCGACA-3′ and reverse, 5′-GGAAGGACCAGGACTGCTCATAT-3′; probe, 5′-FAM-TTAGCCAAAGACTGCCAC-TAMRA-3′; SOX9 forward primer, 5′-TGCAAGCATGTGTCATCCA-3′ and reverse, 5′-AGCAATCCTCAAACTCTCTAGCC-3′; probe, 5′-FAM-CTCTGCATCTTCTCTTGGAGTG-TAMRA-3′; and NEDD8 forward primer, 5′-AGGTGGTCTTAGGCAGTGATGG-3′ and reverse, 5′-TGGCTATGGTGTCCCAGAGAGT-3′; probe, 5′-FAM-CTTTACCCTGTCGCTCATAATGAGGCATCA-TAMRA-3′.

Identification of Differentially Regulated Genes and Development of Prediction Scores.

A random permutation test was carried out to identify “predictor” genes that showed significant differences in mean expression level (Cy3:Cy5) between node-positive and node-negative tumors, as reported previously (13) . A permutation P of <0.01 was considered to be significant. To select a set of genes by which two classes were best discriminated, we used discriminant analysis (14 , 15) . In forward stepwise discriminant analysis, a model of discrimination was built step-by-step. At each step, all of the expression ratios (Cy3:Cy5) were reviewed and evaluated to determine which one would contribute most to the discrimination of the two classes. The candidate genes were included in the prediction model, and the process was repeated until any inclusion did not contribute to the improvement of the prediction model. The analysis determined a discriminant coefficient (kj) of each predictor gene (j) and a constant value (C = −1.945). We calculated the “predictive score (χ)” of each sample (i) by the following formula: where rij is the expression ratio (Cy3:Cy5) of gene j of sample i. Statistical analyses were performed with the SPSS software package (SPSS Inc., Chicago, IL).

RESULTS

Identification of Commonly Up- or Down-Regulated Genes in Intestinal-Type Gastric Cancers.

To clarify mechanisms underlying carcinogenesis of the intestinal type of gastric cancer, we first searched for genes commonly up- or down-regulated in this type of tumor. A cDNA microarray analysis of more than 20,000 genes in 20 tumors identified 61 genes (including 13 of unknown function) that were up-regulated in more than 75% of the cases examined (Table 1) ⇓ . We also identified 63 genes (including 22 of unknown function) that were down-regulated in 75% or more of the samples examined (Table 2) ⇓ .

Commonly up-regulated elements included genes associated with signal-transduction pathways (GFRA2, HGF, HRH1, PLEK2, PLAB, and PROCR, genes encoding transcription factors (NFIL3, LHX1, SOX9, IRF7, HOXB7 and HSF4), and genes involved in various metabolic pathways (SCD, CHST1, LPAP, and PRPS1), transport systems (TFRC, SLC2A1, SLC16A1, SLC16A2, and SLC25A4), cell proliferation (MIA), antiapoptosis (BCL2), protein translation and processing (EIF3S9, HSPA9B, HSPCB, and RPL10), DNA replication and recombination (RPA3, RUVBL1, and PRDKC), or other functions (SERPING1 and HRG).

Among the commonly down-regulated genes were some that are specific to gastric mucosa and involved in lipid metabolism (MTP, APOB, and APOA4), carbohydrate metabolism (KHK, ADH3, ALDH3, FBP1, ADH1, ALDOB, and MGAM), drug metabolism (CYP2C9, CYP3A7, and CYP3A5), carbon dioxide metabolism (LOC56287 and CA2), defense response (TFF1 and TFF2), or transport of small molecules or heavy metals (ATP2A3, GIF, ATP4B, and MT1E).

On examining the reliability of data obtained from our microarray analysis, we found that reproducibility was >85% when we excluded genes with signal intensities lower than the cutoff values. To verify the microarray data more quantitatively, we selected three commonly up-regulated genes (NME1, SOX9, and PLAB) and a gene the expression level of which was significantly altered between the node-positive and -negative tumors (NEDD8) and performed quantitative RT-PCR using 18 pairs of aRNA samples used for array analysis and nine additional pairs that were not used for the array analysis. Both results were very similar to the microarray data for all four genes (Fig. 1) ⇓ , supporting the reliability and rationality of our strategy.

Identification of Genes Associated with Lymph Node Metastasis.

Because metastasis to lymph nodes is a critical step in tumor progression, we searched for genes associated with that event. We compared expression profiles in nine node-positive cases with those of nine node-negative tumor samples and identified 12 genes that were expressed differently (P of <0.01) by a random-permutation test (Table 3) ⇓ . Nine of the 12 genes were relatively up-regulated (DDOST, GNS, NEDD8, LOC51096, CCT3, CCT5, PPP2R1, and two ESTs) and three were down-regulated (UBQLN1, AIM2, and USP9X) in node-positive tumors.

Development of Predictive Scores for Lymph Node Metastasis.

We developed an equation to achieve a scoring parameter for the prediction of lymph node metastasis. Among the 12 genes with statistically significant differences in expression between node-positive and node-negative tumors, a forward stepwise discriminant analysis identified five as independent predictors (Table 3) ⇓ . The predictive score was calculated using the expression profiles of these five genes and their discriminant coefficients, as described in “Materials and Methods.” As Fig. 2 ⇓ shows, this scoring system correctly separated node-positive tumors from node-negative tumors. The robustness of the classification was validated by the leave-one-out cross-validation method using the 12 genes (14 , 15) , i.e., by training on all but one of the samples and applying the resulting model to predict the classification for the sample that is left out. All 18 samples that were left out were correctly classified by the method (data not shown). Furthermore, we obtained nine additional gastric cancer samples and examined their predictive scores. As shown in Fig. 2 ⇓ , all nine test cases (four node-positive and five node-negative cases) were correctly assigned to one of the two classes, suggesting the usefulness of the predictive score.

DISCUSSION

The development of microarray technology has facilitated the analysis of expression levels of thousands of genes in a single experiment. This technology is a powerful tool for analyzing genes the expression of which can be correlated with pathological phenotypes of various tumors. Revised classifications of cancer types can now be proposed on the basis of altered expression of multiple genes in tumor tissues. In addition, analyses of gene expression profiles not only have disclosed specific patterns that may reflect prognosis and drug sensitivity of tumor cells but have revealed the identity of genes involved in malignant transformation, progression, and/or metastasis of tumors (12 , 13) . Recently two groups reported results of expression-profile analyses in gastric cancers; one group used an oligonucleotide array representing 6800 genes to examine expression in scirrhous-type gastric cancer cell lines derived from tumors that had shown aggressive properties (metastasis to lymph node and/or peritoneum; Ref. 16 ). The other group analyzed expression profiles in xenografts of human intestinal-type and diffuse-type gastric tumors, as well as in two clinical samples, using a cDNA array consisting of 1174 genes (17) . Therefore, ours is the first genome-wide study of gene expression in microdissected cells from intestinal-type gastric cancers.

The work reported here disclosed commonalities among intestinal-type gastric cancers through the analysis of expression of more than 20,000 genes. Genes that were commonly altered in the tumors that we examined represented widely diverse functions. Several that had been cited by others for association with gastric carcinogenesis, such as ERBB2, EGFR, and CCNE, were not included in our list because the frequency of their up-regulation in our experiments did not fit our defined criterion for commonly up-regulated genes (i.e., frequency of 75% or more). For instance, ERBB2, EGFR, and CCNE were reported to be overexpressed in 20, 50, and 20% of intestinal gastric cancers, respectively (18) , and in our study, those genes showed expression ratios of >2) in only 45, 62.5, and 10% of the tumors, respectively.

Some of the up-regulated genes had been already reported to be involved in carcinogenesis or cell proliferation. Those included genes associated with signal transduction: HGF, which is known to be the ligand of receptor-type tyrosine kinase; and c-MET, which was reported to be up-regulated in various cancers including gastric cancers (19 , 20) . Overexpression of HGF and MET genes might activate the HGF/MET signaling pathway in an autocrine manner and play a crucial role in gastric carcinogenesis. Altered expression of transcription factors including NFIL3, LHX1, and HOXB7 was reported in leukemia (21, 22, 23, 24) and solid tumors (23) . In addition, the list of up-regulated genes includes genes related to intracellular metabolism, DNA replication, and protein synthesis and processing, a result that might reflect accelerated growth and/or cell division. Interestingly, genes related to blood coagulation [PROCR (25) , SERPING1, and HRG (26)] were also up-regulated in cancer cells as well. Because coagulopathy is a common complication among cancer patients, the administration of drugs inhibiting these targets may help to relieve patients with gastric cancer from coagulation defects.

On the other hand, 63 genes, including 22 functionally unknown genes, were down-regulated in more than 75% of the gastric carcinomas we examined. This list includes genes involved in the metabolism of carbohydrates, lipids, and drugs, or in the transport of small molecules. Several genes having specific functions in gastric epithelium were down-regulated as well; many of those encode products associated with absorption of nutrients or barriers against bacteria in the intestinal lumen. Hence down-regulation of these genes may reflect “de-differentiation” during carcinogenesis.

Metastasis to lymph nodes is one of the most useful prognostic factors for cancer patients. Recently two groups demonstrated that VEGF-C and -D play critical roles in this process (27 , 28) . However, the complex mechanisms of metastasis cannot be fully explained by alterations in just a few genes. Therefore our identification of a set of genes that were differently expressed between node-positive and node-negative tumors should contribute to an improved understanding of the precise biophysical events. For example, among the 12 genes that showed significantly different expression levels between the two groups, DDOST and GNS are known to be involved in the metabolism of glycoproteins that are constituents of extracellular matrix and cell-surface adhesion molecules. Therefore DDOST and/or GNS may mediate a process that modifies proteins associated with cell-adhesion or invasion. In addition, AIM2 (absent in melanoma), a putative tumor suppressor gene, was down-regulated in our node-positive group compared with node-negative tumors. Further analysis of these genes may clarify their roles in metastasis.

Predictive scores based on expression levels of the five genes described here as “discriminators” should have an impact on clinical practice because these scores could separate node-positive tumors from node-negative tumors with a high probability without the need to remove lymph nodes for examination. Although verification using a larger number of cases is essential, this predictive model may become a powerful tool for clinical purposes. Moreover, our results indicate that genome-wide analysis of gene expression analysis using laser-capture microdissection techniques and cDNA microarrays can provide useful information for clarifying the mechanism of development and progression of gastric cancers and for identifying new therapeutic targets and biomarkers for this disease.