Development of a SOMAmer (slow off-rate modified aptamer)-based assay to detect NSCLC.

Abstract

7601 Background: Support for low-dose helical computed tomography (CT) screening for lung cancer in a high risk population has emerged from the National Lung Screening Trial (NLST). However, the low specificity of CT raises concerns regarding the cost and potential morbidity associated with resection of benign nodules. Non-invasive lung cancer biomarkers may serve as a useful complement to imaging, providing a simple means to further clarify the diagnosis of suspicious pulmonary nodules. SOMAmer-based chip technology was previously used to identify a panel of serological biomarkers capable of accurately classifying NSCLC. Here we assess the analytical and clinical performance of an automated assay that uses SOMAmer reagents and qPCR to simultaneously quantify a subset of these markers (n=12). Methods: Biomarker levels were measured in sera from 43 subjects with stage I-III NSCLC and 63 long-term smoker controls. qPCR results were analyzed by relative (ΔΔCT) quantitation, which uses calibrator serum and a set of normalizers. Linear mixed-effects models were fit to identify key sources of analytical variability (variance components), including plate-to-plate, within-sample, and between sample effects. Clinical performance for distinguishing NSCLC from control sera was established by a random forest predictive model. Results: Median within-sample %CV for the 12 markers was 6.7%. Variance components analyses suggest that, on average, 5.7% of the variance for a given biomarker was due to within-sample effects, 5.2% to plate-to-plate effects, and 86.2% to between-sample effects. A 10-marker random forest model exhibited an AUC [95% CI] of 0.915 [0.905, 0.924], classifying NSCLC from control sera with 79% sensitivity and 90% specificity. Conclusions: We have developed a highly reproducible automated assay designed for the clinical laboratory. Combining SOMAmer-based protein capture and qPCR-based quantification, the assay monitors the levels of 12 potential lung cancer markers. An algorithm-based model incorporating 10 of these markers showed good accuracy for distinguishing NSCLC from control sera. Further studies are warranted to evaluate the performance of the test in classifying pulmonary nodules.