Mining novel biomarkers for prognosis of gastric cancer with serum proteomics

Abstract

Background: Although gastric cancer (GC) remains the second cause of cancer-related death, useful biomarkers for prognosis are still unavailable. We present here the attempt of mining novel biomarkers for GC prognosis by using serum proteomics.

Methods: Sera from 43 GC patients and 41 controls with gastritis as Group 1 and 11 GC patients as Group 2 was successively detected by Surface Enhanced Laser Desorption/ionization Time of Flight Mass Spectrometry (SELDI-TOF-MS) with Q10 chip. Peaks were acquired by Ciphergen ProteinChip Software 3.2.0 and analyzed by Zhejiang University-ProteinChip Data Analysis System (ZJU-PDAS). CEA level were evaluated by chemiluminescence immunoassay.

Results: After median follow-up periods of 33 months, Group 1 with 4 GC patients lost was divided into 20 good-prognosis GC patients (overall survival more than 24 months) and 19 poor-prognosis GC patients (no more than 24 months). The established prognosis pattern consisted of 5 novel prognosis biomarkers with 84.2% sensitivity and 85.0% specificity, which were significantly higher than those of carcinoembryonic antigen (CEA) and TNM stage. We also tested prognosis pattern blindly in Group 2 with 66.7% sensitivity and 80.0% specificity. Moreover, we found that 4474-Da peak elevated significantly in GC and was associated with advanced stage (III+IV) and short survival (p < 0.03).

Conclusion: We have identified a number of novel biomarkers for prognosis prediction of GC by using SELDI-TOF-MS combined with sophisticated bioinformatics. Particularly, elevated expression of 4474-Da peak showed very promising to be developed into a novel biomarker associated with biologically aggressive features of GC.

Figure 1

Survival curve for all included GC patients, good-prognosis and poor-prognosis GC patients. The media survival time (months) for all included GC patients (n = 54), poor- prognosis (n = 25) and good-prognosis GC patients (n = 25) was 23, 12 and not reached, respectively. There was significantly statistical difference between poor-prognosis and good-prognosis groups (Log-rank test p = 0.00).

Figure 2

The areas under Receiver Operating Characteristic (ROC) curves for prognosis pattern and CEA (A), detection pattern and CEA (B), stage pattern and CEA (C).

Figure 3

Representative expression of the peak at 4474 Da (red) in prognosis pattern. Peak at 4474 Da was significantly higher in poor-prognosis GC (upper panel), compared with good-prognosis GC (lower panel) in biomarker mining set. Wilcoxon Rank Sum p = 0.04.

Figure 4

Representative expression of the peak at 4474 Da (red) in blind test set for prognosis pattern. Peak at 4474 Da was high expressed in poor-prognosis GC (upper panel), compared with good-prognosis GC (lower panel) in blind test with 5 good-prognosis and 6 poor-prognosis GC patients.

Figure 5

Representative expression of the peak at 4474 Da (red) in detection pattern. Peak at 4474 Da was significantly higher in GC (lower panel), compared with non-cancer controls (upper panel). Wilcoxon Rank Sum p < 0.001.

Figure 6

Representative expression of the peak at 4474 Da (red) in stage pattern. Peak at 4474 Da was significantly higher in stage III+IV GC (lower panel), compared with stage I/II GC (upper panel). Wilcoxon Rank Sum p = 0.01.