Clonal consolidation and transcriptional re-programming define non-small cell lung cancers resistant to immune checkpoint inhibitors

Analysis of pre-treatment non-small cell lung cancers (NSCLCs) has helped identify predictors of response to immune checkpoint inhibitors (ICI), including low PD-L1 expression, targetable drivers alterations ( EGFR , ALK ), and STK11 / KEAP1 mutations. To-date, however, few samples at treatment resistance have been studied, and consequently less is known about the mechanisms contributing to ICI resistance.Methods: Building on our previous analysis Stand Up 2 Cancer-Mark Foundation (SU2C-MARK) Cohort, we identified patients with on or post-treatment samples and performed whole exome (WES) and/or RNA-sequencing, based on tissue availability and quality. Genomic and transcriptomic features in post vs pre-treatment samples were compared. Subclonal evolution in paired samples was assessed using PhylogicNDT, and immune phenotypes were inferred using published gene signatures and deconvolution methods (CIBERSORTx).Results: Building on the original n=393 cohort, 41 patients with on and/or post-treatment samples were identified, with n=45 WES and n=35 RNA-seq on-treatment/post-treatment samples passing QC, for a total of n=445 samples included in the updated cohort, SU2C-MARKv2. N=28 patients had paired samples across treatment time, ranging from 2-5 time points. Only 3 samples had acquired mutations in B2M or JAK2 ; more commonly, post-treatment tumors showed persistence of clones containing STK11 , KEAP1 , and/or ARID1A / SMARCA4 alterations. The proportion of subclonal alterations decreased at resistance, suggesting elimination of passenger-rich antigenic subclones (median 7.9% vs1.9%, p<0.001). Acquired copy number loss in antigen presentation genes in 6p21 ( STAT1/HLA/TAP1/TAP2 ) were observed in 5 treatment pairs, and amplification in 9p24.1 (JAK2/PD-L1/PD-L2) in 4 pairs. In the transcriptional space, resistant tumors demonstrated increased expression of gene sets associated with tumor-intrinsic biology, including MYC, oxidative phosphorylation, and EMT, and decrease in DNA repair genes (ATM). Alterations in immune gene sets were most prominent in on- rather than post-treatment specimens, with increase in T, B and myeloid cell signatures in both responding and non-responding tumors, though numbers were low (n=5). Immune clustering into hot, intermediate, and cold phenotypes confirmed an increase in hot tumors in on-treatment specimens, while post-treatment tumors had predominantly ‘intermediate' immune phenotype.Conclusions: Integrated genomic/transcriptomic analysis of the expanded SU2C-MARKv2 cohort suggests that ICI resistance emerges through elimination of immunogenic subclones and selection for resistant subclones that maintain immune-suppressive transcriptomic phenotypes. Further integration of spatial and single-cell data will define the tumor-immune architecture and identify potential treatment targets.