Bioinformatic immunogenomic profiling of melanoma biopsies to study mechanisms of resistance to immunotherapy

SEGOVIA, MERCEDES; PEREZ SAEZ, JUAN MANUEL; HILL, MARCELO; MAHMOUD, YAMIL; BALZARINI, MÓNICA; MARIÑO, KARINA VALERIA; FERNÁNDEZ, ELMER; VEIGAS, FLORENCIA; ROCHA, DARÍO; RABINOVICH, GABRIEL A.; GIROTTI, MARÍA ROMINA

Rosario, Santa Fe

Congreso; Congreso Argentino e Internacional de Oncología Clínica; 2019

Targeted therapies and immunotherapies have revolutionized melanoma treatment. However, responses are not universal andpatients who initially respond to therapy develop resistance. We implemented computational tools to analyze public datasetswith the aim of studying predictive biomarkers of response and resistance to immunotherapy. We analyzed single-cell and bulkRNA-Seq gene expression profiles from biopsies of melanoma patients before and during treatment with anti-PD-1. Our analysisshow that inflammasome activation is required for anti-PD1 response and the expression of inflammasome-related genescorrelate in tumor-infiltrating macrophages. We used CIBERSORT, a machine learning deconvolution approach based on SVR, toinfer the immune cell composition from bulk gene expression profiles. We found that patients who respond to anti- PD1 therapypresent increased proportion of effector immune populations during treatment that correlates with inflammasome activation.However, this method tends to overestimate the amount of cell types present in the sample thus underestimating trueproportions, biasing the results. We therefore developed MIXTURE, a new algorithm that overcomes such limitationsimplementing a noise-constrained recursive feature selection for SVR. We evaluated MIXTURE over simulated and benchmarkdata sets overcoming limitations of competitive methods in terms of accuracy and estimation bias. By applying MIXTURE onmelanoma biopsies, we identified associations between the immune infiltrate and outcome that were not revealed by previousmethods such as a decreased proportion of immunosuppressive M2-macrophages in responders to anti-PD1 therapy. Theanalysis of other public datasets showed associations between the tumor immune infiltrate composition, survival, andintratumoral heterogeneity. MIXTURE is available for the wider scientific community for Python and R.