A miRNA-based diagnostic model predicts resectable lung cancer in humans with high accuracy

Abstract

Lung cancer, the leading cause of cancer death worldwide, is most frequently detected through imaging tests. In this study, we investigated serum microRNAs (miRNAs) as a possible early screening tool for resectable lung cancer. First, we used serum samples from participants with and without lung cancer to comprehensively create 2588 miRNAs profiles; next, we established a diagnostic model based on the combined expression levels of two miRNAs (miR-1268b and miR-6075) in the discovery set (208 lung cancer patients and 208 non-cancer participants). The model displayed a sensitivity of 99% and specificity of 99% in the validation set (1358 patients and 1970 non-cancer participants) and exhibited high sensitivity regardless of histological type and pathological TNM stage of the cancer. Moreover, the diagnostic index markedly decreased after lung cancer resection. Thus, the model we developed has the potential to markedly improve screening for resectable lung cancer.

Fig. 1. Workflow of lung cancer and non-cancer controls used for developing a diagnostic model.

Serum samples were obtained from 1698 lung cancer, 207 non-cancer patients in National Cancer Center biobank, and 1998 non-cancer participants in Yokohama Minoru Clinic. After excluding 132 samples from National Cancer Center biobank and 27 samples from Yokohama Minoru Clinic, the sample set was divided into two groups: the discovery set (N = 416) and validation set (N = 3328).

Fig. 2. Strategy for the selection of candidate miRNAs for lung cancer diagnosis.

a Flow diagram of the (1) miRNA expression analysis and (2) development of the diagnostic model. b A principal component analysis map for 208 lung cancer samples and 208 non-cancer samples with 406 miRNAs. c Heat map showing the differences in miRNA expression levels between 208 lung cancer and 208 non-cancer control samples.

Fig. 3. Development of a discrimination model to distinguish lung cancers from non-cancer participants.

a Receiver operating characteristic (ROC) curves for detecting lung cancer patients using miR-17-3p selected as the best single miRNA model in the discovery set. b Receiver operating characteristic (ROC) curves for detecting lung cancer patients using two miRNAs selected as best discrimination model in the discovery set. c Diagnostic index levels of miR-17-3p and two miRNAs selected as best discrimination model among lung cancer, non-cancer 1 and non-cancer 2. A diagnostic index score ≥ 0 indicated lung cancer and a diagnostic index score < 0 indicated the absence of lung cancer. model(miR-7-3p): (0.491213) × miR-17-3p−2.49845, model(miR-1268b): (−3.55955) × miR-1268b + 34.53362, model(miR-6075): (2.17625) × miR-6075−18.78233, model(miR-1268b + miR-6075): (−3.56049) × miR-1268b + (1.99039) × miR-6075 + 16.7999.

Fig. 4. Diagnostic performance of the discrimination model in the validation set.

Receiver operating characteristic (ROC) curves for detecting lung cancer patients using two miRNAs selected as best discrimination model.

Fig. 5. Diagnostic performance of the discrimination model at different stages and histological types of lung cancer.

a Diagnostic performance of the two selected miRNAs at different pathological TNM stages in the validation set. The diagnostic index showed high performance for all stages. Each positive rate is shown in the plot. b Diagnostic performance of the two selected miRNAs at different histological types in the validation set. The diagnostic index showed high performance for all histological types. Each positive rate is shown in the plot.

Fig. 6. Comparison of diagnostic indexes between preoperative and postoperative serum samples from lung cancer patients.

Diagnostic index levels of miR-17-3p, miR-1268b, miR-6075, and two-miRNA panel (miR-1268b and miR-6075) were decreased after lung cancer resection. A diagnostic index score ≥ 0 indicated lung cancer and a diagnostic index score < 0 indicated the absence of lung cancer.