Stochasticity in cancer immunotherapy stems from rare but functionally-critical Spark T cells

Cancer immunotherapies trigger highly variable responses in patients and in genetically identical mice models. To assess the intrinsic stochasticity of these therapies, we performed thousands of well-controlled ex vivo immunoassays. We show that leukocyte responses and tumor cytotoxicity are highly variable at the macroscopic level and statistically distributed as a shifted Poisson process. Stochastic activation of a rare subpopulation of T cells (so-called Spark T cells) coupled to a paracrine IFN-?- driven positive feedback account for this measured “noise” in immunotherapeutic reactions. We integrated these quantitative insights in a custom-designed machinelearning pipeline to analyze immune reactions with single-cell resolution; this led us to phenotypically and functionally identify the Spark T cells in murine naïve T cells, and in human T cell blasts, as prepared for adoptive T cell therapy. We then demonstrate their relevance in explaining variable outcomes in cancer immunotherapies. Overall design: scATACseq : Nuclei of sorted CD5lowCD11ahigh and CD5highCD11alow T cell subpopulations from OT-1 Rag1-/- mice were isolated following the nuclei isolation protocol for Single Cell ATAC Sequencing (Low Cell Input) provided from 10x Genomics.