Purpose: To estimate gastric cancer patients’ prognosis more comprehensively,we tried to develop a prognostic scoring system using a cDNA microarray.

Experimental Design: RNA was extracted from tumor/normal paired samples of 43 patients with gastric cancer, and cDNA microarray hybridization was performed.

Results: We selected 78 genes that were differentially expressed between aggressive and nonaggressive groups with respect to five conventional pathological factors. Next, we determined a coefficient for each gene. Thereafter a prognostic score was calculated by summing-up the value for each gene. It ranged from −47 to 201 with a median of 114. There were two peaks in its distribution. Ten of 11 patients who were alive with no evidence of recurrence >5 years after the operation showed a score of <100 points, whereas all 19 patients who died of disease showed >100 points. In 13 patients who were alive but the follow-up time was <5 years, 2 of the 3 patients with >100 points revealed recurrent disease during the follow-up.

Conclusions: These findings demonstrate that such a system with cDNA microarray can contribute to the comprehensive analysis of malignant behavior of the tumor and may provide accurate information on prognosis.

Gastric cancer is one of the ten most frequent cancers in the world, and it is the second most common in Japan. Recent improvements in diagnostic tools and methods have facilitated detection of early gastric cancer; however, there still exist patients whose tumor is advanced at the time of diagnosis (1). Prognosis depends on the stage of disease. Patients with stage I disease have a good prognosis, and those with stage IV disease show a very poor prognosis (1). It is difficult to predict the prognosis of each patient with stage II or III disease because these stages include patients with either a good or a poor outcome. In other words, these stages include high-grade and low-grade malignant tumors. To differentiate the malignant potential appropriately, many efforts have been made to select prognostic markers.

Several interesting molecules have been reported with respect to the prognosis of gastric cancer patients. These include cell cycle regulation factors such as p27 or proliferating cell nuclear antigen, matrix proteinases, cell adhesion molecules such as E-cadherin, autocrine motility factors, angiogenic factors, growth factors, oncogenes, tumor suppressor genes such as p53, and several others (1, 2, 3, 4, 5, 6). The expression of these molecules has been studied at the protein level by immunohistochemistry or at the RNA level by molecular methodologies such as Northern blot or reverse transcription-PCR. The interpretation of the immunohistochemical results is often difficult and sometimes different between investigators. On the other hand, it is more quantitative based on Northern blot or reverse transcription-PCR analysis. However, these methods cannot evaluate many genes at once. In addition, only one or a few selected molecules will not define the whole characteristics of a tumor. Thus, it has been recommended to study the malignant potential of a tumor from the viewpoint of the total expression profile of many genes.

Recent excellent advances in the cDNA microarray technique that can investigate gene expression systematically enable us to visualize gene expression profiles in human tumors (7, 8, 9). This technique can show another interesting possibility; it may be equivalent to multiple competitive Northern blot analyses to evaluate the tumor:normal ratio of many genes simultaneously. The purpose of this study was to establish a prognostic scoring system for gastric cancer using cDNA microarray. Our strategy was to: (a) select the differentially expressed genes between two different groups for each of the five conventional pathological factors; (b) determine a coefficient of expression for each gene; and (c) calculate the prognostic score. Our prognostic scoring system was revealed to be very useful in the determination of an individual’s prognosis.

Tissue Specimens.

Gastric cancer tissues, along with corresponding normal gastric tissues from the same patient, were excised during surgery from 43 patients from 1994 to 1998. All grossly dissected tissue specimens were frozen in liquid nitrogen and kept at −90°C. Informed consent was obtained from the patients. Clinical and pathological descriptions were made according to the criteria outlined by the Japanese Research Society for Gastric Cancer. The stage of disease was as follows: 11 stage I, 10 stage II, 15 stage III, and 7 stage IV. Five of the stage IV patients showed liver metastasis and two showed peritoneal dissemination at surgery, and a simple gastrectomy was performed. The interval of follow-up after operation ranged from 93 days to 7 years 11 months with a mean of 4 years and 7 months.

RNA Isolation and cDNA Microarray.

The total RNA was extracted from the specimens as described previously (10). Poly(A)+ mRNA was prepared from total RNA using oligotex TM-MAG mRNA purification kit (TaKaRa Shuzo Co., Ltd., Shiga, Japan). The fluorescently labeled cDNA probes were prepared as described. One-μg aliquots of RNA from cancer tissue and normal corresponding tissue were labeled using the RNA Fluorescence Labeling Core Kit (TaKaRa Shuzo Co., Ltd.) with Cy3-dUTP and Cy5-dUTP (Amersham), respectively, in each paired case.

We used the commercially available cDNA microarray (IntelliGene Human Cancer chip; TaKaRa Shuzo Co., Ltd.) that contains 425 selected known human genes that are considered to be cancer related. These include oncogenes, tumor suppressor genes, growth factors, apoptosis-related genes, matrix proteinase genes, angiogenesis-related genes, drug-resistant genes, and so on. The internal control genes included 14 genes such as β-actin, ATP synthase, or glyceraldehyde-3-phosphate dehydrogenase. Labeled probes were mixed with hybridization solution (6× SSC, 0.2% SDS, 5× Denhardt’s solution, and 0.1 mg/ml denatured salmon sperm DNA). After hybridization for 14 h at 42°C, the slides were twice washed in 2× SSC and 0.1% SDS for 30 min at 55°C, washed in 2× SSC and 0.1% SDS for 5 min at 65°C, and washed in 0.05× SSC for 5 min at room temperature. The slides were scanned using the Affymetrix 418 (Affymetrix).

Data Analysis.

The signal intensity of hybridization was evaluated photometrically by the ImaGene computer program (BioDiscovery) and normalized to the averaged signals of housekeeping genes. A cutoff value for each expression level was automatically calculated according to the background fluctuation. The fluctuation can be estimated as the variance of the log ratio of Cy3:Cy5 minus the variance of the log ratio of Cy3:Cy5 of highly expressed genes.

We selected five conventional prognostic factors (6, 11) and divided the cases into two groups for each factor as shown in Table 1. The average expression intensity of each gene was calculated in each group and compared between the two groups. Then we picked up the differentially expressed genes with a significant difference that was defined by the Student’s t test as mentioned below. The representative selected genes related to the depth of invasion and lymph node metastasis are shown in Table 2. The frequency of pick-up of each gene was evaluated as a coefficient. This means that if the gene was selected as differentially in all five prognostic factors, its coefficient was evaluated as five. If it was picked up in relation to three prognostic factors, its coefficient was three. If the gene expression was inversely correlated with the malignant potential, a minus point was given in this case.

Statistical Analysis.

The average expression intensity of each gene was compared between the two groups for each of the five pathological factors, and the statistical difference was evaluated by Student’s t test.

The average expression intensity of each gene was calculated in each group and compared between the two groups. Table 2 shows the selected genes that were differentially expressed with a significant difference between the two groups with respect to depth of tumor invasion (Table 2A) and lymph node metastasis (Table 2B). Similarly, the differentially expressed genes were selected with respect to tumor size, histological growth pattern, and liver metastasis (data not shown).

Table 3 shows a coefficient for each gene ranging from plus 5 to minus 3. Interestingly, 347 of 425 evaluated genes were not differentially expressed between the two groups in any factor, and a coefficient was therefore 0. There was a significantly different expression in the remaining 78 genes between the two groups. Four genes were regarded as plus 5 because they appeared in all of the five factors; the next eight genes were regarded as plus 4 because they appeared in four factors. Similarly, 19 genes were scored as plus 3, 9 genes were plus 2, 20 genes were plus 1, 15 genes were minus 1, and 3 genes were minus 3. There was no gene scored as minus 2. To calculate the prognostic score using the 78 differentially expressed genes, the following formula was considered:

\[{{\sum}}XnYn\ (n\ {=}\ 1,\ 2,\ 3,\ 4,\ {\ldots},\ 77,\ 78)\]

X means the log value of the expression ratio (Cy3:Cy5) for gene Xn. Y means a coefficient for gene Xn.

Fig. 1 A shows the distribution of the prognostic score in all 43 patients. The score ranged from −47 to 201 with a median of 114 points. There were two peaks around 50 and 130 points. The cutoff point for the prediction of the clinical outcome of the patients was selected as 100, arbitrarily based on the observed distribution of scores.

Twenty patients had <100 points, and 23 patients had >100 points. Fig. 1 B shows the correlation between the prognostic score, stage of disease, and prognosis. The patients were divided into three groups. Group A included patients who were alive with no evidence of recurrent disease >5 years after operation, group B included patients who were alive but followed <5 years after operation, and group C included patients who died of recurrence of their disease after operation. Very interestingly, the prognostic scores of 10 of 11 patients in group A were <100 points, whereas those of all 19 patients in group C were >100 points. One stage I patient in group C died of liver metastasis, and the tumor showed the rare histology of hepatoid carcinoma that is known to produce α-fetoprotein and have a very poor prognosis (12). Group B included 3 patients with >100 points and 10 with <100 points. Two of the 3 with >100 points showed recurrent disease; one stage II patient recurred in the paraaortic lymph nodes 3 years after operation, and the other stage III patient recurred in the liver 2.5 years after operation. None of the 10 patients with <100 points showed recurrent disease at the time of this writing.

cDNA microarray is a powerful tool to monitor the complete transcriptional profile of hundreds to ten thousands of genes at once (7, 8, 9). This technology has a strong impact not only on the basic biology of genetic information but also on clinical practice, and it is anticipated to produce more accurate disease diagnosis and more appropriate therapy. For example, Golub et al.(7) reported that acute myeloid leukemia and acute lymphoblastic leukemia could be classified by DNA microarray without previous knowledge of these classes. Alizadeh et al.(8) reported that the molecular classification of diffuse large B-cell lymphomas by DNA microarray could identify previously undetected and clinically significant subtypes. Regarding hepatocellular carcinoma, Okabe et al.(9) demonstrated that DNA microarray could differentiate hepatitis B-related carcinomas from hepatitis C-related ones. These studies demonstrate that the gene expression pattern as determined by DNA microarray is useful for differential diagnosis that has been difficult in routine histopathological examinations.

Our study demonstrates another useful application of cDNA microarray in clinical care. It has been difficult to predict the recurrence, metastasis, or prognosis of the individual gastric cancer patient, especially those with an intermediate stage, by the conventional staging system. Thus, it is desirable to estimate the grade of malignancy of the tumor as a score. Our study demonstrates that it is possible to do so by cDNA microarray. The prognostic score in this case demonstrated that patients whose score was >100 points showed a poor prognosis, whereas those whose score was <100 points showed a good prognosis. In addition, this scoring system was able to evaluate a patient who showed an unexpected outcome; one patient with stage I disease died of liver metastasis 37 months after operation. The score of this patient was 137, indicating a poor prognosis, although his stage of disease was early. As shown in Fig. 1,B, the prognostic scoring system could differentiate high-risk patients in an intermediate stage of II or III. Two of three patients in group B (Fig. 1 B) showed recurrent disease, and the tumors of these patients showed a score of >100. These patients were difficult to identify using the conventional staging system.

The strategy for our scoring system was: (a) to select the differentially expressed genes between two different groups for each of the five conventional pathological factors; (b) to determine a coefficient for each gene; and (c) to calculate the prognostic score. The methods described here may be considered tentative and may be further refined. Variables include which conventional pathologic factors should be selected and what the most suitable method to determine a coefficient for each selected gene is. These problems will be resolved, and the most suitable method or analysis will be optimized by further studies in the near future.

The DNA chip used in this study contained 425 cancer-related genes that have been reported in the literature. Our study demonstrated that ∼80% of these genes were not differentially expressed between aggressive and nonaggressive groups of gastric cancer. This means that not a large number of genes appear to participate in determining the grade of malignancy. Microdissection of exclusively tumor elements could, however, modify this conclusion. There were four genes that were selected in all five pathological factors. Of these, three genes such as matrix metalloproteinase-7(13), secreted protein, acidic, cysteine-rich (osteonectin) (14), and transforming growth factor-β3(15) have been reported to be correlated with tumor progression or metastasis in several cancers including gastric cancer. Especially we have demonstrated that matrix metalloproteinase-7 is a good indicator of aggressive behavior of the tumor (13, 16, 17), and this study confirmed our hypothesis. The molecules that are known to correlate with tumor progression or metastasis, such as fibronectin precursor, carcinoembryonic antigen-related cell adhesion molecule 6, IGF binding protein 3, or proliferating cell nuclear antigen were selected in four factors. On the other hand, thrombospondin 2 is an angiostatic factor reported to be suppressed in highly metastatic cancers; its expression was inversely correlated with tumor progression in colon or lung cancers (18, 19). Thus, the selection of thrombospondin 2 in this study of gastric cancer needs further study. However, recently, Lee et al.(20) reported similar results to those of our study in that it was overexpressed in gastric cancer compared with normal gastric tissue.

In conclusion, this study demonstrates that the prognosis of individual gastric cancer patients can be predicted by cDNA microarray technique. Such analysis will be important to both patients and their doctors in terms of life planning and treatment planning, respectively.

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.

1

This work was supported in part by the Grant-in-Aid for Scientific Research on Priority Areas of Cancer (12215116, 11671251, 12218227), the Grant-in-Aid for Scientific Research (B) (12557100, 12470241) and (C) (12213101, 12671232, 12670166), and the Grant-in-Aid for Exploratory Research (13877188), the Ministry of Education, Culture, Sports, Science and Technology of Japan. This work was also supported by the Uehara Memorial Foundation, Naito Foundation, Casio Science Promotion Foundation, Foundation for Promotion of Cancer Research in Japan, and Sagawa Foundation for Promotion of Cancer Research.

Fig. 1.

A, distribution of prognostic scores. The score ranged from −47 to 201 with a median of 114. Note the two peaks around 50 and 130 points. B, correlation between the prognostic score, stage of disease, and prognosis. Group A included patients who were alive with no evidence of recurrent disease >5 years after the operation, group B included patients who were alive but followed <5 years after the operation, and group C included patients who died of recurrence of their disease after operation.

Fig. 1.

A, distribution of prognostic scores. The score ranged from −47 to 201 with a median of 114. Note the two peaks around 50 and 130 points. B, correlation between the prognostic score, stage of disease, and prognosis. Group A included patients who were alive with no evidence of recurrent disease >5 years after the operation, group B included patients who were alive but followed <5 years after the operation, and group C included patients who died of recurrence of their disease after operation.

Close modal
Table 1

Pathological factors

Tumor size  
 ≥5 cm 12 
 <5 cm 31 
Depth of invasion  
 Up to muscular layer 16 
 Subserosa or beyond 27 
Histological growth pattern  
 Expansive pattern 22 
 Infiltrative pattern 21 
Lymph node metastasis  
 Absent 15 
 Present 28 
Liver metastasis  
 Absent 38 
 Present 
Tumor size  
 ≥5 cm 12 
 <5 cm 31 
Depth of invasion  
 Up to muscular layer 16 
 Subserosa or beyond 27 
Histological growth pattern  
 Expansive pattern 22 
 Infiltrative pattern 21 
Lymph node metastasis  
 Absent 15 
 Present 28 
Liver metastasis  
 Absent 38 
 Present 
Table 2

Genes related to pathological factors

A. Genes related to depth of tumor invasion
Expression of tumor:normal ratioDeep: Shallow ratio
Shallow invasionDeep invasion
Matrix metalloproteinase-7 1.27 5.10 4.02 
Keratin 6B 0.75 2.24 2.99 
Thrombospondin 2 2.29 5.67 2.47 
Matrix metalloproteinase-10 1.05 2.32 2.22 
IGF                  a                  binding protein 3 1.02 2.21 2.17 
Transforming growth factor3 4.85 9.53 1.97 
Matrix metalloproteinase 1 1.25 2.42 1.94 
Secreted protein, acidic, cysteine-rich 3.29 6.04 1.83 
CEA-related cell adhesion molecule 6 2.14 3.90 1.83 
Chondroitin sulfate proteoglycan 2 1.80 3.20 1.78 
A. Genes related to depth of tumor invasion
Expression of tumor:normal ratioDeep: Shallow ratio
Shallow invasionDeep invasion
Matrix metalloproteinase-7 1.27 5.10 4.02 
Keratin 6B 0.75 2.24 2.99 
Thrombospondin 2 2.29 5.67 2.47 
Matrix metalloproteinase-10 1.05 2.32 2.22 
IGF                  a                  binding protein 3 1.02 2.21 2.17 
Transforming growth factor3 4.85 9.53 1.97 
Matrix metalloproteinase 1 1.25 2.42 1.94 
Secreted protein, acidic, cysteine-rich 3.29 6.04 1.83 
CEA-related cell adhesion molecule 6 2.14 3.90 1.83 
Chondroitin sulfate proteoglycan 2 1.80 3.20 1.78 
B. Genes related to lymph node metastasis
Expression of tumor:normal ratio Lymph node metastasisPositive/Negative ratio
NegativePositive
Matrix metalloproteinase-7 1.54 4.70 3.06 
Thrombospondin 2 1.84 5.64 3.06 
Fibronectin precursor 5.05 14.15 2.80 
Matrix metalloproteinase-10 0.86 2.30 2.67 
Transforming growth factor3 3.97 9.91 2.49 
IGF binding protein 3 0.94 2.15 2.27 
Secreted protein, acidic, cysteine-rich 2.74 6.13 2.24 
Collagen, type I, α2 3.81 8.07 2.12 
Matrix metalloproteinase 1 1.17 2.35 2.02 
PDGF receptor, β polypeptide 1.39 2.58 1.86 
B. Genes related to lymph node metastasis
Expression of tumor:normal ratio Lymph node metastasisPositive/Negative ratio
NegativePositive
Matrix metalloproteinase-7 1.54 4.70 3.06 
Thrombospondin 2 1.84 5.64 3.06 
Fibronectin precursor 5.05 14.15 2.80 
Matrix metalloproteinase-10 0.86 2.30 2.67 
Transforming growth factor3 3.97 9.91 2.49 
IGF binding protein 3 0.94 2.15 2.27 
Secreted protein, acidic, cysteine-rich 2.74 6.13 2.24 
Collagen, type I, α2 3.81 8.07 2.12 
Matrix metalloproteinase 1 1.17 2.35 2.02 
PDGF receptor, β polypeptide 1.39 2.58 1.86 
a

IGF, insulin-like growth factor; CEA, carcinoembryonic antigen; PDGF, platelet-derived growth factor.

Table 3

Coefficient and genes related to malignancy

GenesCoefficient
Matrix metalloproteinase-7 (matrilysin
Secreted protein, acidic, cysteine-rich (osteonectin
Transforming growth factor3 
Thrombospondin 2 
Proliferating cell nuclear antigen 
CEAa-related cell adhesion molecule 6 
Fibronectin precursor 
IGF binding protein 3 
Chondroitin sulfate proteoglycan 2 (versican
IFN-induced transmembrane protein 2 (1–8D
Connective tissue growth factor 
Collagen, type III, α1 (Ehlers-Danlos syndrome type IV
Transforming growth factor, β-induced, 68kD 
PDGF receptor, β polypeptide 
Matrix metalloproteinase-1 (interstitial collagenase
Vimentin 
Matrix metalloproteinase-2 (gelatinase A, 72kD gelatinase
Cell division cycle 2, G                  1                  to S and G                  2                  to M 
Matrix metalloproteinase-10 (stromelysin 2
Antioxidant protein 1 
Plasminogen activator, urokinase 
Cathepsin B 
Signal transducer and activator of transcription 1, 91kD 
v-Ki-ras2 Kirsten rat sarcoma 2 viral oncogene homologue 
Collagen, type XVIII, α1 
Collagen, type I, α2 
α2-Macroglobulin 
Apoptosis inhibitor 4 (survivin
Platelet/endothelial cell adhesion molecule (CD31 antigen
Rho-associated, coiled-coil containing protein kinase 1 
CD9 antigen (p24
Stromal cell-derived factor 1 
Laminin, α4 
Heparan sulfate proteoglycan 2 (perlecan
Hexabrachion (tenascin C, cytotactin
Tissue inhibitor of metalloproteinase-3 
Laminin receptor 1 (67kD, ribosomal protein SA
Ia-associated invariant γ-chain 
Matrix metalloproteinase-3 (stromelysin 1, progelatinase
Nidogen (enactin
Colony stimulating factor 1 receptor 
Rho GDP dissociation inhibitor (GDI) β 
Tubulin, α, brain-specific 
Plasminogen activator, tissue 
Plasminogen activator, urokinase receptor 
Retinoblastoma-binding protein 6 
Cyclin D2 
Lactate dehydrogenase A 
Keratin 6B 
CD72 antigen 
Cyclin-dependent kinase 4 
Human mRNA for SB class II histocompatibility antigen α-chain 
Profilin 1 
GTP-binding protein 
Homo sapiens clone 24703 β-tubulin mRNA, complete cds 
Collagen, type IV, α1 
Topoisomerase (DNA) I 
Envoplakin −1 
Leukocyte antigen-related protein −1 
Matrix metalloproteinase-15 (membrane-inserted−1 
IGF binding protein 2 (36kD−1 
Ubiquitin-conjugating enzyme E2M −1 
K-sam −1 
Ras homologue gene family, member D −1 
Thyroid hormone receptor interactor 6 −1 
Rho GTPase activating protein 1 −1 
Cyclin-dependent kinase 5, regulatory subunit 1 (p35−1 
Integrin, β4 −3 
Interleukin 8 −3 
Serine protease inhibitor, Kunitz type 1 −3 
GenesCoefficient
Matrix metalloproteinase-7 (matrilysin
Secreted protein, acidic, cysteine-rich (osteonectin
Transforming growth factor3 
Thrombospondin 2 
Proliferating cell nuclear antigen 
CEAa-related cell adhesion molecule 6 
Fibronectin precursor 
IGF binding protein 3 
Chondroitin sulfate proteoglycan 2 (versican
IFN-induced transmembrane protein 2 (1–8D
Connective tissue growth factor 
Collagen, type III, α1 (Ehlers-Danlos syndrome type IV
Transforming growth factor, β-induced, 68kD 
PDGF receptor, β polypeptide 
Matrix metalloproteinase-1 (interstitial collagenase
Vimentin 
Matrix metalloproteinase-2 (gelatinase A, 72kD gelatinase
Cell division cycle 2, G                  1                  to S and G                  2                  to M 
Matrix metalloproteinase-10 (stromelysin 2
Antioxidant protein 1 
Plasminogen activator, urokinase 
Cathepsin B 
Signal transducer and activator of transcription 1, 91kD 
v-Ki-ras2 Kirsten rat sarcoma 2 viral oncogene homologue 
Collagen, type XVIII, α1 
Collagen, type I, α2 
α2-Macroglobulin 
Apoptosis inhibitor 4 (survivin
Platelet/endothelial cell adhesion molecule (CD31 antigen
Rho-associated, coiled-coil containing protein kinase 1 
CD9 antigen (p24
Stromal cell-derived factor 1 
Laminin, α4 
Heparan sulfate proteoglycan 2 (perlecan
Hexabrachion (tenascin C, cytotactin
Tissue inhibitor of metalloproteinase-3 
Laminin receptor 1 (67kD, ribosomal protein SA
Ia-associated invariant γ-chain 
Matrix metalloproteinase-3 (stromelysin 1, progelatinase
Nidogen (enactin
Colony stimulating factor 1 receptor 
Rho GDP dissociation inhibitor (GDI) β 
Tubulin, α, brain-specific 
Plasminogen activator, tissue 
Plasminogen activator, urokinase receptor 
Retinoblastoma-binding protein 6 
Cyclin D2 
Lactate dehydrogenase A 
Keratin 6B 
CD72 antigen 
Cyclin-dependent kinase 4 
Human mRNA for SB class II histocompatibility antigen α-chain 
Profilin 1 
GTP-binding protein 
Homo sapiens clone 24703 β-tubulin mRNA, complete cds 
Collagen, type IV, α1 
Topoisomerase (DNA) I 
Envoplakin −1 
Leukocyte antigen-related protein −1 
Matrix metalloproteinase-15 (membrane-inserted−1 
IGF binding protein 2 (36kD−1 
Ubiquitin-conjugating enzyme E2M −1 
K-sam −1 
Ras homologue gene family, member D −1 
Thyroid hormone receptor interactor 6 −1 
Rho GTPase activating protein 1 −1 
Cyclin-dependent kinase 5, regulatory subunit 1 (p35−1 
Integrin, β4 −3 
Interleukin 8 −3 
Serine protease inhibitor, Kunitz type 1 −3 
a

CEA, carcinoembryonic antigen; IGF, insulin-like growth factor; PDGF, platelet-derived growth factor.

We thank the pertinent collaboration for Yoshie Yoshikawa, Jyunichi Mineno, Hiroyuki Mukai, Kiyozo Asada, and Ikunoshin Kato (Biotechnology Research Laboratories, Takara Shuzo Co., Ltd., Kusatsu, Japan). All of the microarray slides were made by the company and were provided for us with a cooperation.

1
Akoh J. A., Macintyre I. M. C. Improving survival in gastric cancer—review of 5-year survival rates in English language publications from 1970.
Br. J. Surg.
,
79
:
293
-299,  
1992
.
2
Stadtlander C. T. K. H., Waterbor J. W. Molecular epidemiology, pathogenesis and prevention of gastric cancer.
Carcinogenesis (Lond.)
,
20
:
2195
-2207,  
1999
.
3
Mori M., Mimori K., Shiraishi T., Fujie T., Baba K., Kusumoto H., Haraguchi M., Ueo H., Akiyoshi T. Analysis of MT1-MMP and MMP2 expression in human gastric cancers.
Int. J. Cancer
,
74
:
316
-321,  
1997
.
4
Mori M., Mimori K., Shiraishi T., Tanaka S., Ueo H., Sugimachi K., Akiyoshi T. p27 expression and gastric carcinoma.
Nat. Med.
,
3
:
593
-593,  
1997
.
5
Mori M., Kakeji Y., Adachi Y., Moriguchi S., Maehara Y., Sugimachi K., Jessup J. M., Chen L. B., Steele G. D. The prognostic significance of proliferating cell nuclear antigen in clinical gastric cancer.
Surgery
,
113
:
683
-690,  
1993
.
6
Allgayer H., Heiss M. M., Schildberg F. W. Prognostic factors in gastric cancer.
Br. J. Surg.
,
84
:
1651
-1664,  
1997
.
7
Golub T. R., Slonim D. K., Tamayo P., Huard C., Gaasenbeek M., Mesirov J. P., Coller H., Loh M. L., Downing J. R., Caligiuri M. A., Bloomfield C. D., Lander E. S. Molecular classification of cancer: class discovery and class prediction by gene expression monitoring.
Science (Wash. DC)
,
286
:
531
-537,  
1999
.
8
Alizadeh A. A., Eisen M. B., Davis R. E., Ma C., Lossos I. S., Rosenwald A., Boldrick J. C., Sabet H., Tran T., Yu X., Powell J. I., Yang L., Marti G. E., Moore T., Hudson J., Jr., Lu L., Lewis D. B., Tibshirani R., Sherlock G., Chan W. C., Greiner T. C., Weisenburger D. D., Armitage J. O., Warnke R., Staudt L. M., et al Distinct types of diffuse large B-cell lymphoma identified by gene expression profiling.
Nature (Lond.)
,
403
:
503
-511,  
2000
.
9
Okabe H., Satoh S., Kato T., Kitahara O., Yanagawa R., Yamaoka Y., Tsunoda T., Furukawa Y., Nakamura Y. Genome-wide analysis of gene expression in human hepatocellular carcinomas using cDNA microarray: identification of genes involved in viral carcinogenesis and tumor progression.
Cancer Res.
,
61
:
2129
-2137,  
2000
.
10
Mori M., Barnard G. F., Staniunas R. J., Jessup J. M., Steele G. D., Chen L. B. Prothymosin-α mRNA expression correlates with that of c-myc in human colon cancer.
Oncogene
,
8
:
2821
-2826,  
1993
.
11
Mori M., Sugimachi K. Clinicopathologic studies of gastric carcinoma.
Semin. Surg. Oncol.
,
6
:
19
-27,  
1990
.
12
Chang Y. C., Nagasue N., Abe S., Kohno H., Kumar D. D., Nakamura T. α-fetoprotein producing early gastric cancer with liver metastasis—report of 3 cases.
Gut
,
32
:
542
-545,  
1991
.
13
Honda M., Mori M., Ueo H., Sugimachi K., Akiyoshi T. Matrix metalloproteinase-7 expression in gastric carcinoma.
Gut
,
39
:
444
-448,  
1996
.
14
Thomas R., True L. D., Bassuk J. A., Lange P. H., Vessella R. L. Differential expression of osteonectin/SPARC during human prostate cancer progression.
Clin. Cancer Res.
,
6
:
1140
-1149,  
2000
.
15
Li C. G., Wang J. M., Wilson P. B., Kumar P., Levine E., Hunter R. D., Kumar S. Role of transforming growth factor β3 in lymphatic metastasis in breast cancer.
Int. J. Cancer
,
79
:
455
-459,  
1998
.
16
Mori M., Barnard G. F., Mimori K., Ueo H., Akiyoshi T., Sugimachi K. Overexpression of matrix metalloproteinase-7 mRNA in human colon carcinomas.
Cancer (Phila.)
,
75
:
1516
-1519,  
1995
.
17
Yamashita K., Mori M., Shiraishi T., Shibuta K., Sugimachi K. Clinical significance of matrix metalloproteinase-7 expression in esophageal carcinoma.
Clin. Cancer Res.
,
6
:
1169
-1174,  
2000
.
18
Tokunaga T., Nakamura M., Oshika Y., Abe Y., Ozeki Y., Fukushima Y., Hatanaka H., Sadahiro S., Kijima H., Tsuchida T., Yamazaki H., Tamaoki N., Ueyama Y. Thrombospondin 2 expression is correlated with inhibition of angiogenesis and metastasis of colon cancer.
Br. J. Cancer
,
79
:
354
-359,  
1999
.
19
Oshika Y., Masuda K., Tokunaga T., Hatanaka H., Kamiya T., Abe Y., Ozeki Y., Kijima H., Yamazaki H., Tamaoki N., Ueyama Y., Nakamura M. Thrombospondin 2 gene expression is correlated with decreased vascularity in non-small cell lung cancer.
Clin. Cancer Res.
,
4
:
1785
-1788,  
1998
.
20
Lee J. H., Koh J. T., Shin B. A., Ahn K. Y., Roh J. H., Kim Y. J., Kim K. K. Comparative study of angiostatic and anti-invasive gene expressions as prognostic factors in gastric cancer.
Int. J. Oncol.
,
18
:
355
-361,  
2001
.