- © 2008 by American Society of Clinical Oncology
Distinct Gene Expression–Defined Classes of Gastrointestinal Stromal Tumor
- Umio Yamaguchi,
- Robert Nakayama,
- Kazufumi Honda,
- Hitoshi Ichikawa,
- Tadashi Hasegawa,
- Miki Shitashige,
- Masaya Ono,
- Ayako Shoji,
- Tomohiro Sakuma,
- Hideya Kuwabara,
- Yasuhiro Shimada,
- Mitsuru Sasako,
- Tadakazu Shimoda,
- Akira Kawai,
- Setsuo Hirohashi and
- Tesshi Yamada
- From the Chemotherapy Division and Cancer Proteomics Project; Cancer Transcriptome Project, National Cancer Center Research Institute; Orthopaedic Surgery, Gastrointestinal Oncology, Gastric Surgery, and Clinical Laboratory Divisions, National Cancer Center Hospital; BioBusiness Group, Mitsui Knowledge Industry, Tokyo; and the Department of Surgical Pathology, Sapporo Medical University School of Medicine, Sapporo, Japan
- Corresponding author: Tesshi Yamada, MD, PhD, Chemotherapy Division and Cancer Proteomics Project, National Cancer Center Research Institute, 5-1-1 Tsukiji, Chuo-ku, Tokyo 104-0045, Japan; e-mail: tyamada{at}gan2.res.ncc.go.jp
Abstract
Purpose The majority of gastrointestinal stromal tumors (GIST) can be cured by surgery alone, but relapse occurs in 20% to 40% of cases. GISTs are considered to invariably arise through gain of function KIT or PDGFA mutation of the interstitial cells of Cajal (ICC). However, the genetic basis of the malignant progression of GISTs are poorly understood.
Patients and Methods The expression levels of 54,613 probe sets in 32 surgical samples of untreated GISTs of the stomach and small intestine were analyzed with oligonucleotide microarrays. The representative GeneChip data were validated by real-time reverse transcriptase polymerase chain reaction and immunohistochemistry.
Results Unbiased hierarchical clustering consistently separated the 32 cases of GIST into two major classes according to tumor site. The two major classes were further separated into novel subclasses, which were significantly correlated with various pathological prognostic parameters, the frequency of metastasis (P < .05), and clinical outcome. Immunohistochemical analysis of 152 independent patients with gastric GISTs revealed that the expression of dipeptidyl peptidase IV (T-cell activation antigen CD26) protein was significantly associated with poorer overall and disease-free survival (P < .00001).
Conclusion CD26 appears to be a reliable biomarker of malignant GISTs of the stomach. The postoperative recurrence rate of CD26-negative cases was as low as 2.0% (two of 102). Therefore, postoperative follow-up of such patients might be made less intensive. CD26 may play an important role in the malignant progression of gastric GISTs and serve as a therapeutic target.
INTRODUCTION
Gastrointestinal stromal tumors (GISTs) are an established human tumor entity characterized by distinct clinical, genetic, and histopathological features.1-3 The overall frequency of GISTs are estimated to be no more than 10 to 20 cases per million in Western countries,1 but GISTs comprise the majority of primary mesenchymal tumors of the gastrointestinal tract. Approximately 60% to 70% of GISTs arise in the stomach, 20% to 30% in the small intestine, and 5% in the colon and rectum.1,3 On the basis of similarities in immunohistochemical and ultrastructural features, it is considered that GISTs arise from interstitial cells of Cajal (ICC) or their precursor cells.4 More than 80% of GISTs have gain of function mutations of the KIT proto-oncogene that encodes the c-Kit (CD117) receptor tyrosine kinase,5 and one third of GISTs without KIT mutation carry reciprocal mutations in the PDGFRA gene that encodes platelet-derived growth factor receptor α (PDGFRA) tyrosine kinase.6,7
GISTs show a wide spectrum of clinical courses. The majority of cases can be cured by surgical resection alone, but 20% to 40% of cases relapse during the postsurgical follow-up.8-10 Distant metastasis to the liver is the most common manifestation of recurrence,10 and our previous experience indicates that the 5-year and 10-year survival rates after grossly curative surgery are 81.7% and 67.4%, respectively.8 Many pathological criteria based on tumor site, size, cell type, degree of necrosis,8,10-12 mitotic rate, Ki-67 immunoreactivity (MIB1 labeling) as well as their combinations have been proposed for predicting the outcome of patients with GISTs. The National Institutes of Health convened a workshop in 2001, and a consensus (risk category) was proposed to estimate the relative risk of GISTs based on tumor size and mitotic count.11 However, the cutoff values for these criteria have been determined empirically, and subjective assessments by skilled pathologists are inevitable. Therefore, it is necessary to identify an objective biomarker for recurrence of GISTs with a high positive or negative predictive value.
Imatinib mesylate (STI-571/Gleevec; Novartis Pharma, Basel, Switzerland), which selectively inhibits a group of tyrosine kinase receptors including KIT and PDGFRA, has been proven to be effective for the management of recurrent and unresectable GISTs.13,14 However, the effect of imatinib mesylate varies depending on the domains of KIT and PDGFRA affected by the mutations.15 Imatinib treatment is generally safe, but serious events such as gastrointestinal and intraabdominal hemorrhage have been reported.16,17 Furthermore, drug-refractory tumor cells develop due to second mutations of KIT during continuous therapy.18 Although several clinical studies are currently underway to investigate the efficacy of emerging kinase inhibitors,19,20 it is necessary to identify a new target molecule other than KIT or PDGFR.
In this study, we analyzed a well-characterized cohort of GIST cases in order to clarify the genomic alterations associated with the malignant progression of this tumor and to identify a biomarker that might be applicable to the prediction of outcome in patients with GISTs.
PATIENTS AND METHODS
Tumor Samples
All of the samples were obtained surgically at the National Cancer Center Hospital (Tokyo, Japan) between July 1972 and November 2005. Fresh frozen tumor specimens of 32 cases of GISTs of the stomach and small intestine were used for GeneChip (Affymetrix, Santa Clara, CA) analysis, and formalin-fixed paraffin-embedded tissue sections of 152 other cases of gastric GIST cases were used for independent validation. The study protocol for collection of tumor samples and clinical information was approved by the institutional review board, and patients provided written informed consent authorizing the collection and use of the tumor samples for research purposes.
Clinicopathological Assessment and Mutation Analysis
Immunohistochemistry for c-Kit, CD34, and Ki-67 was performed as described previously.21,22 Mitotic score was determined by counting the number of mitotic figures in 10 consecutive high-power fields (HPF; ×400). Score 1 was ≤ 5 per 10 HPF, and score 2 was > 5 per 10 HPF. MIB1 labeling index (LI) was assigned as index 1 (< 10% MIB1-positive cells) and index 2 (≥ 10% MIB1-positive cells). Tumor grade was defined as grade 1 (index 1 and no tumor necrosis) and grade 2 (index 2 or tumor necrosis). Risk group was defined as low-risk group (grade 1 and tumor size of < 5.0 cm) and high-risk group (grade 1 and tumor size of ≥ 5.0 cm or any grade 2). Risk category was defined as described previously.11 The mutational status of the KIT and PDGFRA genes was determined as described previously.23
GeneChip Analysis
Total RNA was extracted with IsoGen lysis buffer (Nippon Gene, Toyama, Japan) and purified with a RNeasy Mini kit (Qiagen, Hilden, Germany). We used GeneChip Human Genome U133 Plus 2.0 arrays (Affymetrix) to analyze the mRNA expression levels of 54,613 probe sets corresponding to more than 38,000 human UniGene Clusters in accordance with the manufacturer's protocols. The background correction, probe summarization, and normalization of all the GeneChip data were performed with the Microarray Analysis Suite 5 algorithm, and the processed values of all probe sets were then log-transformed for subsequent analyses, using the ArrayAssist 4.0 software package (Stratagene, La Jolla, CA).
Hierarchical clustering analysis was performed with centered values of the Pearson's correlation coefficient and Ward's linkage method. Clustering analysis was performed by biostatisticians (A.S., T.S., H.K.) who were blinded to the clinicopathological data.
Real-Time Reverse-Transcriptase Polymerase Chain Reaction
For cDNA synthesis, 5 μg of total RNA was reverse transcribed by random priming with Superscript II reverse transcriptase (Invitrogen). The gene-specific TaqMan primers and probes were designed by Applied Biosystems (Foster City, CA). Real-time reverse-transcriptase polymerase chain reaction (RT-PCR) was carried out using the ABI Prism 7000 Sequence Detection System (Applied Biosystems). The comparative Ct values were normalized to that of glyceraldehyde 3-phosphate dehydrogenase.24
Immunohistochemistry of CD26
Goat antihuman CD26 antibody (AF1180) was purchased from R&D Systems (Minneapolis, MN). Immunoperoxidase staining of formalin-fixed and paraffin-embedded tissue sections using the avidin-biotin complex was performed as described previously.25 Immunohistochemical results were judged by three investigators (T.Y., K.H., U.Y.) without awareness of the clinical information. Endothelial cells of blood vessels served as internal positive controls. Tumors that showed any degree of CD26 staining were classified as positive.
Statistical Analysis
Estimates of overall and disease-free survival were computed using the Kaplan-Meier method using the StatFlex statistical software package version 5.0 (Artec, Osaka, Japan). Overall survival was calculated from the day of diagnosis until death or until the end of follow-up. Disease-free survival was calculated from the day of diagnosis until the day of relapse or death as a result of disease, whichever came first. Differences between survival curves were assessed for statistical significance with the log-rank test. Other statistical tests were performed using tools available in the R statistical package (version 2.0.1; http://www.r-project.org/).
RESULTS
Classification of GISTs Into Four Subclasses Based on Global Gene Expression
The clinicopathological, immunohistochemical, and genetic characteristics of the 32 cases of GIST used in the GeneChip analysis are presented in Appendix Table A1, online only.
To grasp the overall gene expression pattern, we first performed unsupervised analysis of all 54,613 probe sets. Hierarchical clustering separated the 32 GISTs into two principal classes, each of which was further divided into two subclasses (Appendix Fig A1, online only). To eliminate probes that had little or no variation across samples (probes that were not working well), we next selected a set of 21,214 probes showing intensity differences of more than 23-fold between the maximum and minimum signals across the 32 samples and repeated the same unsupervised analysis. Hierarchical clustering separated the 32 samples into the same four subclasses except for one sample (case 28; Fig 1). We further confirmed the stability of this gene expression–defined clustering by eliminating probe sets with intensity differences of less than 24-fold (6,231 probe sets), 25-fold (2,907 probe sets), and 26-fold (1,380 probe sets; data not shown).
Four distinct gene expression–defined subclasses of gastrointestinal stromal tumors. Unsupervised hierarchical clustering separated 32 GIST cases into four subclasses based on the expression levels of the 21,214 probe sets of the GeneChip Human Genome U133 Plus 2.0 array (Affymetrix, Santa Clara, CA). Case numbers correspond to those of Appendix Table A1. The clinicopathological characteristics of the 32 cases in the four subclasses are indicated by red and green rows as follows: tumor site (red, small intestine/green, stomach), mitotic score (red, score 1/green, score 2), MIB1 labeling index (red, index 1/green index 2), tumor grade (red, grade 1/green, grade 2), risk group (red, low grade/green, high grade), risk category (red, low and intermediate risk/green, high risk), necrosis (red, absent/green, present), metastasis (red, absent/green, present), KIT mutation (red, absent/green, present), mutated exon of KIT (red, other than exon 11/green, exon 11), and tumor size (red, < 5 cm/green, ≥ 5.0 cm).
Clinicopathological Significance of the Gene-Expression–Defined Subclasses
We named the two principal classes separated by unsupervised analysis of the 21,214 probe sets as class B (for bowel) and class G (for gastric), because all tumors of the small intestine were clustered into class B, and all tumors of the stomach were clustered into class G (Fig 1). The four subclasses were designated as class B1, class B2, class G1, and class G2 (from left to right in Fig 1). The subclasses were found to be associated with the known prognosis-relevant clinicopathogical variables (Fig 1). Fisher's exact test showed that there were significant differences between class B1 and class B2 as well as between class G1 and class G2 in the frequency of mitotic score, MIB1 LI, tumor grade, risk group, and metastasis (P < .05; Appendix Table A2, online only). There was no significant difference in the presence of KIT mutation, mutated exon of KIT, tumor size, cell type, sex, or expression of c-Kit or CD34 (Table A2 and data not shown). Mitotic score, MIB1 LI, tumor grade, risk group, and metastasis did not remain significantly different (P < .05) between class B1 and class B2, when Holm's adjustment of P values was applied for dealing with the multiple testing situation.26
Appendix Figure A2A and A2B shows the Kaplan-Meier plots for disease-free survival of patients in the subclasses. The gene expression–defined clusters clearly separated the patients into those with good outcome (class B1 and class G1) and those with poor outcome (class B2 and class G2; P < .005). Remarkably, none of the patients in class B1 or class G1 died during follow-up period of 108 months.
Heterogeneous Gene Expression of Malignant GISTs
The correlation coefficient values of 21,214 probe sets between all the combinations of the 32 GIST cases were calculated, and are presented as a pseudocolored heat map in Figure 2. There were high similarities of overall gene expression within cases of class B1 and within cases of class G1, but not within cases of class B2 or within cases of class G2 (Fig 2). The average correlation coefficient values were significantly different between class B1 and class B2 (P < .01, Welch's t-test) as well as between class G1 and class G2 (P < .001; Table 1). These findings suggest that genomic diversity increases significantly during the malignant progression of GISTs.
Heterogeneous gene expression of malignant gastrointestinal stromal tumors. The correlation coefficient value for each pair is shown in pseudo color according to the scale at the bottom. Red indicates higher correlation, and blue indicates lower correlation.
Heterogeneous Gene Expression of Malignant Gastrointestinal Stromal Tumors
Gene Expression Changes Associated With Malignant Progression of GISTs
There were 122 probe sets whose expression was increased in class B1 compared with class B2, and 400 probe sets whose expression was increased in class G1 compared with class G2 (Appendix Fig A3A, online only). There were 97 probe sets whose expression was increased in class B2 compared with class B1, and 321 probe sets whose expression was increased in class G2 compared with class G1 (Fig A3B). Only eight probe sets (eight UniGene clusters) were commonly increased in class B1 and class G1 relative to each respective counterpart (Fig A3A and Appendix Table A3), and 12 probe sets (12 UniGene clusters) were commonly increased in class B2 and class G2 relative to each respective counterpart (Fig A3B and Appendix Table A4), suggesting that the genomic alterations promoting malignant progression differ between small intestinal GISTs and gastric GISTs.
We conducted real-time RT-PCR analysis of 20 representative genes differentially expressed between class G1 and class G2 to validate the results of the GeneChip analysis. Appendix Figure A4 represents 10 of these 20 genes.
CD26 Is a Significant Prognostic Factor of Gastric GISTs
Among the 400 probe sets whose expression was significantly increased in class G2 compared with class G1, we noticed that the DPP4 (dipeptidyl peptidase IV) gene (which encodes the CD26 protein) was ranked in the first, second, third, and fifth places (Appendix Table A8, online only). Immunohistochemistry of 21 gastric GIST cases for which specimens were available revealed there were 12 CD26-positive (Fig 3A, 3B, 3D, and 3E) and nine CD26-negative cases (Fig 3C and 3F). The expression of CD26 protein appeared to be correlated well with gene expression–defined classes except for one case (case 2; Fig 3G). The disease-free and overall survival of patients with CD26-positive GISTs was worse than that of patients with CD26-negative GISTs (P < .05; Appendix Fig A5, online only). Appendix Table A5 (online only) presents the relationship between CD26 expression and gene expression–defined subclasses of small intestinal and gastric GISTs.
Correlation of dipeptidyl peptidase IV (DPP4) mRNA with CD26 protein expression. (A-C) Hematoxylin and eosin and (D-F) immunoperoxidase staining of (A, B, D, E) CD26-positive and (C, F) CD26-negative gastric gastrointestinal stromal tumors. The arrowhead indicates vascular endothelium, and arrows indicate CD26-positive inflammatory cells. (G) Relative DPP4 mRNA expression level was determined by real-time reverse transcriptase polymerase chain reaction (in arbitrary units). The bar at the bottom indicates CD26-positive (black) and CD26-negative (gray) cases.
We then examined the clinical significance of CD26 protein expression in an independent validation cohort consisting of 152 gastric GISTs. The patients comprised 83 males (54.6%) and 69 females (45.4%). The average age at diagnosis was 59 years (range, 28 to 83 years), and the duration of follow-up ranged from 4 to 352 months (mean, 117 months). Follow-up computed tomography (CT) imaging was performed every 3 to 6 months. Of the 152 patients, 22 (14.5%) developed distant metastasis (14 to liver, four to peritoneum, two to bone, one to lung, and one to lymph node), seven of them were treated with imatinib mesylate. Immunohistochemically, 149 cases were positive for c-Kit, and 148 cases were positive for CD34.
Of the 152 gastric GISTs, 50 were CD26 positive (32.9%), and the remaining 102 were CD26 negative (67.1%). CD26 positivity was significantly (P < .05, Fisher's exact test) associated with tumor size, necrosis, mitotic score, MIB1 LI, tumor grade, risk group, risk category, and metastasis (Table 2). CD26 positivity was significantly associated with poor overall and disease-free survival (P < .00001; Fig 4A and 4B). The estimated overall survival rate at 10 years after surgery was 97.4% in CD26-negative patients and 69.9% in CD26-positive patients.
Correlation of CD26 expression with patient outcome in a validation cohort. (A) Kaplan-Meier analysis of overall survival of patients with CD26-positive (yellow) and CD26-negative (blue) gastric gastrointestinal stromal tumors (GISTs). (B) Kaplan-Meier analysis of disease-free survival of patients with CD26-positive (yellow) and CD26-negative (blue) gastric GIST. (C) Kaplan-Meier analysis of overall survival of patients with CD26-negative (blue), CD-26 positive and index 1 (gray), and CD-26 positive and index 2 (red) gastric GIST. (D) Kaplan-Meier analysis of disease-free survival of patients with CD26-negative (blue), CD-26 positive and index 1 (gray), and CD-26 positive and index 2 (red) gastric GIST.
Correlations Between Clinicopathological Characteristics and CD26 Expression in 152 Cases of Gastric Gastrointestinal Stromal Tumors
Almost all the CD26-negative cases were MIB1 LI index 1 (100 of 102), but the CD26-positive cases comprised a mixture of index 1 (29 of 50) and index 2 (21 of 50; Table 2). MIB1 LI is known to represent cell proliferation activity. We hypothesized that the CD26-positive cases might be further stratified by MIB1 LI. As shown in Figures 4C and 4D, the 152 gastric GIST cases were divided into three groups: CD26 negative, CD26 positive and index 1, and CD26 positive and index 2. There were significant differences in disease-free survival among these three groups (P < .01).
DISCUSSION
Several microarray analyses using smaller numbers of GIST cases had been conducted before this study.27-31 GISTs show gene expression profiles different from those of other mesenchymal tumors.27,31 The status of KIT/PDGFRA mutation has been reported to affect the global gene expression profile of GISTs.29,30 However, none of these studies investigated the clinicopathological significance of the gene expression profiles, probably because long-term follow-up (for 5 to 10 years or more) is necessary for assessing the clinical outcome of this generally low-grade malignant tumor.10
Unsupervised hierarchical clustering is a well-established statistical method that separates cases based on similarities and dissimilarities of overall gene expression.32 GISTs are considered to invariably arise through gain of function KIT or PDGFRA mutation of ICC. Most GISTs are composed of a fairly uniform population of spindle cells.3,11 Allander et al27 reported marked homogeneity in the gene expression of GISTs with KIT mutation. We assumed that low-grade GISTs constitute a uniform population and could be separated from high-grade GISTs by simple unsupervised clustering. The most principal determinant that separated the 32 GIST cases in this study was the site of tumor origin: the small intestine (class B) or stomach (class G; Fig 1), similarly to findings reported previously.29 The second most principle determinant, however, was exactly as anticipated. Low-grade GISTs constituted a population with homogeneous gene expression profiles (classes B1 and G1; Fig 2) and was separated from high-grade GISTs, which constituted a heterogeneous population (classes B2 and G2, Fig 2).
In order to apply the observations obtained using GeneChip analysis to clinical practice, we selected the DPP4 gene, because its expression showed the greatest significant differences between class G1 and class G2. We further validated the clinical significance of the DPP4 gene product, CD26, in a large independent cohort of gastric GIST cases (Fig 4 and Table 2). Because the postoperative recurrence rate of CD26-negative cases was as low as 2.0% (two of 102) even in this cohort, the postoperative follow-up of these patients could have been significantly less intensive. Objective assessment of CD26 expression is possible using formalin-fixed paraffin-embedded tissue specimens (Figs 3D to 3F) and can be readily incorporated into routine pathological diagnosis along with c-Kit and CD34. For these reasons, CD26 is considered to be a biomarker superior to other known prognostic parameters.
CD26 is not only a biomarker of malignant GISTs, but may also play an important role in malignant progression. CD26 is a 110-kDa cell membrane glycoprotein that belongs to the serine protease family (EC 3.4.14.5).33 It is expressed on a wide variety of cell lineages including T lymphocytes, endothelial and epithelial cells. CD26 selectively cleaves the N-terminal dipeptide from cytokines and chemokines, and modulates their function. Although the role of CD26 in tumor development is still controversial,33 an intriguing observation has been reported in a series of publications by Kotani and colleagues.34,35 Differential diagnosis of follicular carcinoma of the thyroid from follicular adenoma has been one of the most difficult tasks for surgical pathologists. CD26 expression is highly specific to carcinoma and is able to predict distant metastasis of apparently benign thyroid tumors.35 Unfortunately, CD26 expression was not associated with the outcome of small intestinal GISTs (data not shown), indicating that the molecular mechanisms behind the malignant progression of small intestinal GISTs differ from those of gastric GISTs. Further studies using cell culture and animal models are required to determine the exact biologic consequences of CD26 in GIST cells.
CD26 may serve as a therapeutic target molecule. Anti-CD26 monoclonal antibody has been shown to inhibit the growth of anaplastic large cell T-cell lymphoma both in vitro and in vivo.36 Several orally active CD26 enzyme inhibitors have been developed as a new class of antidiabetic drugs. These inhibitors are generally safe and well tolerated, and no serious adverse effect has been noticed even in elderly patients.37,38 These characteristics of CD26 inhibitors may make them suitable for long-term preventive administration to postoperative patients with GISTs.
At present, the precise molecular mechanism that induces the expression of CD26 remains to be clarified. CD26 may not be the cause of malignant progression of gastric GISTs, but its clear-cut association with the increased risk of postoperative recurrence warrants diagnostic application. It will certainly be necessary to validate our results in an independent study.
AUTHORS’ DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST
The author(s) indicated no potential conflicts of interest.
AUTHOR CONTRIBUTIONS
Conception and design: Kazufumi Honda, Miki Shitashige, Masaya Ono, Tesshi Yamada
Financial support: Tesshi Yamada
Administrative support: Akira Kawai, Setsuo Hirohashi, Tesshi Yamada
Provision of study materials or patients: Tadashi Hasegawa, Yasuhiro Shimada, Mitsuru Sasako, Tadakazu Shimoda
Collection and assembly of data: Umio Yamaguchi, Robert Nakayama, Hitoshi Ichikawa, Tadashi Hasegawa
Data analysis and interpretation: Umio Yamaguchi, Kazufumi Honda, Miki Shitashige, Masaya Ono, Ayako Shoji, Tomohiro Sakuma, Hideya Kuwabara, Tesshi Yamada
Manuscript writing: Umio Yamaguchi, Tesshi Yamada
Final approval of manuscript: Tesshi Yamada
Glossary Terms
- Dipeptidyl peptidase IV (DPP4):
- A cell membrane serine exopeptidase that cleaves dipeptides from the N terminus of proteins. DPP4 is involved in the metabolic inactivation of glucagon-like peptide-1 (GLP1).
- Hierarchical clustering:
- An analytical tool used to find the closest associations among gene profiles and specimens under evaluation.
- c-kit:
- A member of the PDGFR family, c-kit is a tyrosine kinase receptor that dimerizes following ligand binding and is autophosphorylated on intracellular tyrosine residues.
- PDGFRA (platelet-derived growth factor alpha):
- The receptor for PDGF exists distinctly as the dimeric αα or ββ form. All dimer combinations of PDGF A and B signal through PDGFR-αα; PDGF BB signals through PDGFR-ββ; PDGF CC signals through the αα and αβ receptors; and PDGF DD signals through the ββ and αβ receptors.
- Ki67:
- A marker of proliferation, Ki67 is a protein that is expressed in the nucleus of proliferating cells. Absent only in resting cells, cells in the G1, S, G2, and M phase of the cell cycle express this marker.
Appendix
Unsupervised clustering analysis of gastrointestinal stromal tumors (GISTs). Hierarchical clustering separated 32 GIST cases into four subclasses based on the expression levels of the entire 54,613 probe sets of the GeneChip Human Genome U133 Plus 2.0 array (Affymetrix, Santa Clara, CA).
Differences in prognosis of gastrointestinal stromal tumors according to gene expression–defined subclass. (A) Kaplan-Meier analysis of disease-free survival of patients in class B1 and class B2. (B) Kaplan-Meier analysis of disease-free survival of patients in class G1 and class G2.
Different genomic alterations associated with the progression of gastric and small intestinal gastrointestinal stromal tumors. (A) Venn diagram showing eight probe sets whose intensities were commonly increased in class B1 and class G1 over each respective counterpart. (B) Venn diagram showing 12 probe sets whose intensities were commonly increased in class B2 and class G2 over each respective counterpart.
Validation of the GeneChip data (Affymetrix, Santa Clara, CA) by real-time reverse transcriptase polymerase chain reaction (RT-PCR). The expression levels of 10 representative genes whose expression differed significantly between class G1 and class G2 gastrointestinal stromal tumors in the GeneChip analysis (left) (P < .001) were confirmed by real-time RT-PCR (right; shown in arbitrary units).
Correlation of CD26 expression with patient outcome in the discovery cohort. Kaplan-Meier analysis of (A) disease-free and (B) overall survival of patients with CD26-positive (yellow) and CD26-negative (blue) gastric gastrointestinal stromal tumors.
Clinicopathological and Genetic Characteristics of the 32 Gastrointestinal Stromal Tumor Cases Used in the GeneChip Analysis
Association of the Gene Expression–Defined Classes With Known Prognostic Parameters
List of Genes Showing Commonly Increased Expression in Classes B1 and G1 in Comparison to Each Respective Counterpart
List of Genes Showing Commonly Increased Expression in Classes B2 and G2 in Comparison to Each Respective Counterpart
Association of CD26 Expression With Risk Category in the 29 Gastrointestinal Stromal Tumor Cases Subjected to GeneChip Analysis
Acknowledgments
We thank Yoshiyuki Suehara, MD, PhD, for exchanging clinical data, and Sachiyo Mitani, Yuka Nakamura, and Tomoko Umaki for technical assistance.
Footnotes
-
Supported by the Program for Promotion of Fundamental Studies in Health Sciences, conducted by the National Institute of Biomedical Innovation of Japan, the Third-Term Comprehensive Control Research for Cancer, conducted by the Ministry of Health, the Labor and Welfare of Japan and the Ministry of Education, Culture, Sports, Science and Technology of Japan, and generous grants from the Naito Foundation and the Princess Takamatsu Cancer Research Fund. U.Y. is an awardee of a Research Resident Fellowship from the Foundation for Promotion of Cancer Research (Tokyo, Japan). These fund resources played no role in designing and interpreting the results of this study.
-
Microarray data of this study have been submitted to the GEO (Gene Expression Omnibus) database (accession number GSE8167).
-
Authors’ disclosures of potential conflicts of interest and author contributions are found at the end of this article.
- Received August 29, 2007.
- Accepted May 14, 2008.
