Immune escape and low response to immunotherapy are crucial challenges in present lung cancer treatment. In this study, we constructed a new immune-related classifier based on CXCL13/CXCR5, an important tumor microenvironment component and strongly related with the formation of tertiary lymphoid structures (TLSs) in tumor microenvironment. With the classifier, we divided patients into two main clusters and each cluster was further divided into subcluster (A1, A2, B1, B2, B3). In the later analysis, we noticed that patients in subcluster B3 had a distinct advantage over patients in A1 in survival time and immune infiltration, suggesting a more favorable response to immunotherapy. Moreover, we demonstrated the genetic and epigenetic regulation related to the subclusters and recovered four key differentially expressed genes (ERBB4, GRIN2A, IL2RA, CCND2). With several experiments, we verified the unique role of CCND2 in tumor metastasis and T cell apoptosis. Overexpressing CCND2 could significantly impair cancer cell abilities of migration and invasion and downregulate PD-1/PD-L1 signaling, which may be the cause of T cell apoptosis reduction. In the end, we constructed a regression risk model that could successfully predict ICI response. To sum up, our study established new stratification models that can successfully predict patient survival and response to ICI. And using integrative analysis of multi-omics data, four key DEGs were noticed, and CCND2, one of the four genes, was identified as a potential treatment target because of its effect in tumor metastasis and T cell apoptosis.