Protein Expression Profiling and Molecular Classification of Gastric Cancer by the Tissue Array Method

Abstract

Gastric cancer is heterogeneous clinically and histologically, and prognosis prediction by tumor grade or type is difficult. Although previous studies have suggested that frozen tissue-based molecular classifications effectively predict prognosis, prognostic classification on formalin-fixed tissue is needed, especially in early gastric cancer. We immunostained 659 consecutive gastric cancers using 56 tumor-associated antibodies and the tissue array method. Hierarchical cluster analyses were done before and after feature selection. To optimize classifier number and prediction accuracy for prognosis, a supervised analysis using a support vector machine algorithm was used. Of 56 gene products, 27 survival-associated proteins were selected (feature selection), and hierarchical clustering identified two clusters: cluster 1 and cluster 2. Cluster 1 cancers were more likely to have intestinal type, earlier stage, and better prognosis than cluster 2 (P<0.05). In 187 early gastric cancers (pT1), cluster 2 was associated with the presence of metastatic lymph nodes (P=0.026). Kaplan-Meier survival curves stratified by pathologic tumor-lymph node metastasis revealed that cluster 2 was associated with poor prognosis in stage I or II cancer (P<0.05). Support vector machines and genetic algorithms selected nine classifiers from the whole data set, another nine classifiers for stage I and II, and eight classifiers for stage III and IV. The prediction accuracies for patient outcome were 73.1%, 88.1%, and 76%, respectively. Protein expression profiling using the tissue array method provided a useful means for the molecular classification of gastric cancer into survival-predictive subgroups. The molecular classification predicted lymph node metastasis and prognosis in early stage gastric cancer.