Checkpoint inhibitors (CPI) have revolutionized treatment of several tumor types, including non-small cell lung cancer (NSCLC), with unprecedented, prolonged responses. However, some patients either do not have a response or develop resistance to CPI therapy. There are currently no approved therapies to circumvent or reverse resistance to CPI, in part due to a lack of understanding of resistance mechanisms. To advance identification of patients who are most likely to benefit from CPI therapy, Vitalay Fomin and colleagues conducted research using a machine learning framework to identify genomic markers that predict clinical responses. Patient data were obtained from a large clinico-genomic database and included 1,150 patients with advanced-stage NSCLC who had received monotherapy with anti–PD-1 (programmed cell death protein 1) or anti–PD-L1 (programmed death-ligand 1) antibodies.
The study revealed core predictive genes and biological pathways associated with responses to CPIs. Mutations in single genes (BRAF, BRIP1, ASXL1, and CDKN2B) and co-occurring mutations (TP53-KEL and TP53-KRAS-NF1) correlate with CPI response and are associated with increased overall survival with CPI treatment. Resistance pathways not previously highlighted to affect CPI outcomes were also identified, including pathways related to PDGF, ESR1, YAP/TAZ, EGFR, IGF, and AGTR1 signaling. The results also highlighted that different mutations within the same gene can lead to opposite responses to therapy, although additional research is needed to understand the underlying mechanisms.
High level
In addition to identifying genomic markers to predict responses, this comprehensive machine learning analysis also provides new insights into the mechanisms of resistance and potential targets for combination therapies to overcome CPI resistance. This machine learning approach could help guide the development of new predictive biomarkers and targeted therapies against escape mechanisms, potentially improving patient stratification and treatment outcomes. It may also help to inform policy decisions on cancer treatment protocols. These parameters would be good to consider in clinical research involving CPIs moving forward, both in NSCLC and—potentially—other cancer types.
Ground level
The insights from this machine learning study are intriguing and might lead to faster identification of predictive factors to allow better personalization of therapies such as CPIs. As understanding of predictive biomarkers continues to expand, development of new therapies and care pathways will continue to be refined to further improve clinical outcomes. Close monitoring of the field for further developments is recommended, especially for intensive therapies that involve intensive monitoring and management, as patients unlikely to benefit could be identified and spared from the toxicity of these treatments.
