Development and validation of a gene-based classification model for pN2 lung adenocarcinoma.

Abstract

Lung adenocarcinoma (LUAD) with pathological ipsilateral mediastinal lymph node (LN) involvement (pN2) exhibits strong biological and clinical heterogeneity. Thus, it is necessary to classify the biomolecular characteristics that lead to the prognostic heterogeneity of pN2-LUAD. The clinical characteristics and bulk RNA sequencing (RNA-seq) data of 75 patients with pN2-LUAD obtained from The Cancer Genome Atlas (TCGA) database were collected as the training set. The disease-free survival (DFS) and overall survival (OS) of patients with different molecular classifications were evaluated. Next, differentially expressed genes (DEGs), biology, and immune cell infiltration in the microenvironment were analysed. Finally, DEGs in the pN2-A and pN2-B groups were included using a least absolute shrinkage and selection operator (LASSO) model, and gene signatures were selected for pN2-A/B type classification. The RNA-seq and single-nucleus RNA sequencing (snRNA-seq) data from our center (n=58) and the GSE68465 dataset (n=53) were used as the validation data sets. Patients with pN2 LUAD were classified into two distinct molecular categories (pN2-A and pN2-B) based on transcriptome information, pN2-A and pN2-B represent low-risk and high-risk patients, respectively. The survival analysis showed that pN2-A patients had significantly better DFS (P=0.0162) and OS (P=0.0105) compared to pN2-B patients. Multivariate analysis confirmed that molecular classification was an independent factor affecting the prognosis of pN2 LUAD (P=0.0038, and P=0.0024). Next, we found that compared with pN2-A stage patients, pN2-B stage patients had a higher frequency of canonical oncogenic pathway mutations and enrichments. At the single-cell level, we also found that the increase of endothelial cells and the decrease of cytotoxic T/natural killer (NK) cells led to a worse prognosis for pN2-B patients compared to pN2-A patients. Moreover, we established a reasonable gene prediction model of 18 differentially expressed genes (DEGs) to classify the pN2-A and pN2-B patients. Finally, the key above-mentioned results were confirmed using our data and the GES68645 dataset. The molecular classification of pN2 LUAD is expected to be a powerful supplement to pN2 substaging. Driver gene status and the immune microenvironment mediate different molecular types of LUAD and provide evidence for individualized treatment strategies.