KEAP1 and TP53 frame genomic, evolutionary and immunological subtypes of lung adenocarcinoma with different sensitivity to immunotherapy.

Abstract Introduction The connection between driver mutations and efficacy of immune-checkpoint inhibitors (ICIs) is the focus of intense investigations. In lung adenocarcinoma (LUAD), KEAP1/STK11 alterations have been tied to immunoresistance. Nevertheless, the heterogeneity characterizing immunotherapy efficacy suggests the contribution of still unappreciated events. Methods Somatic interaction analysis of top-ranking mutant genes in LUAD was carried out in the AARC project GENIE (N=6208). Mutational processes, intratumor heterogeneity, evolutionary trajectories, immunological features and cancer-associated signatures were investigated exploiting multiple datasets (AACR GENIE, TCGA, TRACERx). The impact of the proposed subtyping on survival outcomes was assessed in two independent cohorts of ICI-treated patients: the tissue-based sequencing cohort (Rome/MSKCC/DFCI, tNGS cohort, N=343) and the blood-based sequencing cohort (OAK/POPLAR trials, bNGS cohort, N=304). Results Observing the neutral interaction between KEAP1 and TP53, KEAP1/TP53-based subtypes were dissected at the molecular and clinical level. KEAP1 single-mutant (KEAP1 SM) and KEAP1/TP53 double-mutant (KEAP1/TP53 DM) LUAD share a transcriptomic profile characterized by AKR gene overexpression, which are under the control of a productive super-enhancer with NEF2L2-binding signals. Nevertheless, KEAP1 SM and KEAP1/TP53 DM tumors differ by mutational repertoire, degree of intratumor heterogeneity, evolutionary trajectories, pathway-level signatures and immune microenvironment composition. In both cohorts (bNGS and tNGS), KEAP1 SM tumors had the shortest survival, the KEAP1/TP53 DM subgroup had intermediate prognosis matching that of pure TP53 LUAD, whereas the longest survival was noticed in the double-wild-type group. Conclusions Our data provide a framework for genomically-informed immunotherapy, highlighting the importance of multi-modal data integration to achieve a clinically exploitable taxonomy.