Background: Gastric cancer (GC) is the third leading cause of cancer deaths in the world. It is highly heterogeneous. Many molecular therapies for GC have entered clinical trials but, apart from trastuzumab, apatinib and ramucirumab, all have failed. One important reason is that insufficient attention is paid to the underlying subtypes and characteristics of GC, especially the diffuse-type gastric cancer (DGC) according to the Lauren classification with worst clinical outcomes. Aim: Here we firstly investigated formalin-fixed paraffin-embedded (FFPE) samples of DGC to establish clinically relevant molecular classification based on proteomics analysis. Also, we tried to generate a suitable classifier of DGC that can guide patient therapy. Methods: We screened a total of 2548 cases retrospectively, who underwent GC resection at Beijing Cancer Hospital from October 2006 to December 2011. We used a fast mass spectrometry workflow for proteome profiling. Finally we carried out proteome profiling of 99 DGC paired tumor-nearby tissues from FFPE sections. Median overall survival of the whole population was 55.0 months. Proteome profiling data from these samples were used to develop a subtype prediction model. We used consensus clustering to identify molecular subtypes based on differentially expressed proteins. The pathway enrichment was performed by GSEA, and the prediction classifier was generated by elastic-net machine learning. Kaplan-Meier survival analysis and Cox regression multivariate analysis were used. Results: A total of 8201 gene products were identified in this study, and 1249 differential expressed proteins between tumor and nearby-normal tissue was detected (FDR q-value < 0.01 by SAM). Tumor upregulated proteins mostly enriched into pathways including RNA processing, epithelial-mesenchymal transition (EMT), immune response and inflammation related pathways. Tumor downregulated proteins mostly enriched into metabolic pathways such as oxidative phosphorylation pathway. Based on proteome profiling alone, DGC can be subtyped into 3 major classes (PX1-3) that exhibit distinct proteome features and correlate with distinct clinical outcomes. PX1 (31 patients) exhibits RNA processing proteins and associates with the best prognosis; PX2 (26 patients) exhibits highly expressed cell cycle features, and the patients have poorer prognosis than those with cluster1 but better prognosis than those with cluster3; PX3 (42 patients) features EMT and the worst prognosis. We built a classifier of 12 marker proteins that can stratify DGC patients into these 3 subtypes, opening a door for protein classification in clinical application and intervention. Conclusion: Our study demonstrated that proteome profiling alone from FFPE samples was able to subtype DGC into 3 protein subtypes that were linked to distinct patterns of molecular alterations and prognosis. The prediction model need to be further verified in more clinical cohorts.
Matrix metalloproteinases and their inhibitors in gastric cancer