Abstract
BACKGROUND: Tumor immune infiltration is closely associated with clinical outcome in lung cancer. We aimed to develop an immune signature to improve the prognostic predictions of lung adenocarcinoma (LUAD). METHODS: We applied "Cell type Identification by Estimating Relative Subsets of RNA Transcripts" method to quantify the fraction of 22 leukocyte cells from six public microarray datasets. Four datasets from GPL570 were treated as the training cohort and two datasets from GPL96 and GPL10379 as the validation cohorts. An immune risk score (IRS) based on leukocyte cell fraction was established by least absolute shrinkage and selection operator cox regression model. RESULTS: IRS consisting of 6 types of leukocytes was constructed in the training dataset. In the training cohort (520 patients), the IRS stratified patients into high-IRS group (215 patients) and low-IRS group (305 patients) with significant differences in overall survival (OS) (HR: 2.77, 95% CI 2.08-3.06). Multivariate analysis including age, gender, stage, IRS and tumor purity revealed the IRS to be an independent prognostic factor in all datasets (training: HR: 10.71, 95% CI 5.72-20.07; validation-1: HR 2.68, 95% CI 1.15-6.27; validation-2: HR 3.71, 95% CI 1.33-10.33); all p < 0.05). IRS was significantly positively correlated to the expression levels of PD1, PDL1, CTLA and LAG3 (all p < 0.001). When integrated with clinical characteristics including stage and age, the composite immune and clinical signature presented with improved prognostic accuracy than IRS (mean C-index 0.66 vs. 0.60). CONCLUSIONS: The proposed immune-clinical signature could predict OS in patients with LUAD effectively.