Purpose: Although endoscopic surveillance accompanied with pathological examination of tissue biopsies remains the gold standard for diagnosing asymptomatic gastric cancer (GC) patients, associated costs and its invasive nature render it inadequate as a screening approach. Development of less invasive tests is needed for surveillance and detection of early stage GCs - when the disease is still treatable. Over the last decade, tumor-derived miRNAs in peripheral blood are emerging as promising disease biomarkers. Herein, for the first time, we have conducted a comprehensive miRNA expression profiling, followed by bioinformatic and statistical analysis to establish a novel serum-based miRNA signature for the diagnosis of patients with GC.
Experimental Design: We analyzed tissue miRNA expression profiles in three patient cohorts (TCGA: n = 477; GSE23739: n=82 and GSE33743: n=43) in an insilico discovery step, during which the robustness of candidate biomarkers was tested and validated. The expression of this miRNA panel was subsequently examined in 50 matched pairs of GC and normal mucosa tissues. The performance of this miRNA signature was evaluated in a serum training cohort of 268 patients (GC=218; endoscopically negative patients (ENP)=50). Using a stepwise logistic regression model, the panel was further refined to accommodate for differences between tissue and serum miRNA expression levels, and this circulating miRNA signature was validated in another independent 359 patient cohort (GC=292; ENP=67).
Results: Initial in silico candidate selection resulted in the identification of 7 differentially expressed miRNAs in GC patients (miR-18a, 21, 181a, 181b, 196a, 196b, 146b), and a combined expression panel yielded remarkable robustness for distinguishing GC vs. normal mucosa tissues (AUC=1). These results were validated in two independent publicly available datasets (TCGA: AUC=0.94, GSE33743: AUC=0.97), as well as GC tissues (AUC=0.98). We next examined the performance of this 7-miRNA panel in a serum training cohort, and refined this panel to include three miRNAs (miR-18a, 181b, 196b: AUC=0.87, sensitivity=87.6%, specificity=70.0%).We thereafter successfully evaluated and validated the performance of this panel in an independent patient cohort (AUC=0.82). Intriguingly, this panel distinguished Stage-I GC patients from ENP (AUC=0.80) indicating its effectiveness and clinical usefulness for noninvasive detection of GC. Furthermore, we discovered that this signature was significantly superior in distinguishing GC from ENP, compared to conventional clinical tumor markers, CEA and CA19-9.
Conclusions: Using a systematic and comprehensive biomarker discovery, prioritization and validation approach, we for the first time, have identified and developed a novel serum-based miRNA signature that offers a promise for noninvasive, early detection of GC.