Transcriptome-based molecular subtyping of non-small cell lung cancer may predict response to immune checkpoint inhibitors.

OBJECTIVES: We set out to investigate whether transcriptome-based molecular subtypes in lung adenocarcinoma and lung squamous cell carcinoma are predictive of the response to programmed cell death 1 blockade. METHODS: Molecular classification of non-small cell lung cancer was performed by unsupervised clustering of mRNA sequencing data from 87 lung adenocarcinoma and 101 lung squamous cell carcinoma specimens, and molecular subtypes were characterized according to their immunogenomic determinants. A prediction algorithm of molecular subtypes was applied to 35 patients with non-small cell lung cancer treated with programmed cell death 1 blockade to test its association with treatment response (GSE93157; the Barcelona cohort). RESULTS: Unsupervised hierarchical clustering of transcriptome sequencing data in lung adenocarcinoma and lung squamous cell carcinoma revealed 3 and 2 distinct clusters, respectively. Cluster 1 in each histology had a higher expression of immune regulatory molecules, increased cytolytic activity, higher interferon-γ signature, and more abundant infiltration of immune cells. Cluster 1 and other cluster(s) in lung adenocarcinoma and lung squamous cell carcinoma had immunologically-hot and immunologically-cold tumor-immune microenvironments, respectively. Immunologically-hot cluster 1 subtype is hereafter referred to as "good-tumor-immune microenvironments" and the other subtypes as "bad-tumor-immune microenvironments." The "good-tumor-immune microenvironments" subtype in lung adenocarcinoma included a high fraction of CD8 T cells and memory B cells, but a low fraction of regulatory CD4 T cells and tumor-associated myeloid cells. Forward and backward application of our molecular subtyping to the Barcelona cohort revealed that transcriptome-based molecular subtyping is significantly associated with response to programmed cell death 1 blockade. CONCLUSIONS: Molecular stratification by transcriptome sequencing data in non-small cell lung cancer identifies distinct immunomolecular subtypes that predict the response to programmed cell death 1 blockade.