A three-miRNA signature as promising non-invasive diagnostic marker for gastric cancer

Abstract

Background: Despite the declining incidence of gastric cancer, mortality rate remains high due to late presentation. We aimed to evaluate the sensitivity of miRNA as a diagnostic marker for gastric cancer in the circulation.

Methods: Plasma samples from 3 independent groups comprise 123 gastric cancer patients and 111 healthy controls for miRNA profiling from microarray screening.

Results: Microarray data showed that 25 miRNAs were upregulated in gastric cancer patients and 6 highly expressed miRNAs (miR-18a, miR-140-5p, miR-199a-3p, miR-627, miR-629 and miR-652) were selected for validation. In an independent validation set, levels of miR-627, miR-629 and miR-652 were significantly higher in gastric cancer patients than healthy controls (P <0.0001). An algorithm with improved sensitivity and specificity as gastric cancer classifier was adopted and validated in another random set of 15 plasma samples. Results showed that combination of 3 miRNAs obtained the highest area under curve, with a cut-off at 0.373, with a sensitivity of 86.7% and a specificity of 85.5%.

Conclusion: This study revealed a three-miRNA signature as a promising classifier for gastric cancer, and greatly enhances the feasibility of circulating miRNAs as non-invasive diagnostic marker for this disease.

Fig. 1

An overview of the workflow of the study design

Fig. 2

Validation of plasma (a) miR-18a, (b) miR-140-5p, (c) 199a-3p, (d) miR-627, (e) miR-629 and (f) miR-652 levels in training set (TS). Expression levels were normalized to U6. Box plots of six miRNAs in plasma of gastric cancer patients (n = 50) and healthy controls (n = 50) The boxes mark the interval between the 25^th and 75^th percentiles, and the lines inside the box denote the medians. The whiskers represent the interval between the 10^th and 90^th percentiles. Statistically significant differences were analyzed using Mann–Whitney test. Receiver-operating characteristic (ROC) curve analysis of miRNA for discriminating gastric cancer patients from healthy controls

Fig. 3

Selection and validation of (a) miR-18a, (b) miR-140-5p, (c) 199a-3p, (d) miR-627, (e) miR-629 and (f) miR-652 in an independent validation set (VS). Box plots and ROC curve analysis of miRNAs in the plasma of gastric cancer patients (n = 58) and healthy controls (n = 46). The boxes mark the interval between the 25^th and 75^th percentiles, and the lines inside the box denote the medians. The whiskers represent the interval between the 10^th and 90^th percentiles. Statistically significant differences were analyzed using Mann–Whitney test. Receiver-operating characteristic (ROC) curve analysis of miRNA for discriminating gastric cancer patients from healthy controls

Fig. 4

Expression levels of (a) miR-627, (b) miR-629 and (c) miR-652 in paired gastric tumor tissues and adjacent non-tumor counterparts (n = 36). Expression levels were normalized to U6. Statistically significant differences were analyzed using Wilcoxon test. Receiver-operating characteristic (ROC) curve analysis of miRNA for discriminating gastric cancer patients from healthy controls

Fig. 5

Validation of gastric cancer classifier in different validation sets. a Box plot and ROC curve analysis of combining 3 miRNAs in random set (RS) with gastric cancer patients (n = 15) and healthy controls (n = 15). b Box plot and ROC curve analysis of combining 3 miRNAs in TSVS with gastric cancer patients (n = 108) and healthy controls (n = 96). The boxes mark the interval between the 25^th and 75^th percentiles, and the lines inside the box denote the medians. The whiskers represent the interval between the 10^th and 90^th percentiles. Statistically significant differences were analyzed using Mann–Whitney test

Fig. 6

Correlation between gastric cancer classifier and tumor stage in plasma of gastric cancer patients (n = 123)