Predicting Natural Killer Cell Behavior with Mathematical Models

Abstract

Abstract Natural killer (NK) cells are specialized lymphocytes with an innate ability to eliminate virally infected and cancerous cells, but the mechanisms that control NK cell development and cytotoxicity are incompletely understood. We hypothesize mathematical models can be used to predict how to direct NK cell development toward mature NK cells. We observed Sostdc1-knockout (KO) mice to display a partial NK cell developmental block. Building from these studies, a best fit model was simulated in which rates of transition and differentiation and cell death was determined by ordinary differential equations of deterministic compartmental models. This approach indicated that NK cell proliferation rates are not necessary to predict WT and KO population outcomes. We further hypothesize that clusters of highly cytotoxic NK cells can be identified computationally, using a combination of Ly49 receptor expression, in vitro and in vivo cytotoxicity assays using β2m−/− targets, and viSNE analysis of flow cytometry data. Sostdc1-KO NK cells also display defective cytotoxicity. We identified that KO splenic CD27+CD11b+ transitional NK cells express lower frequencies of inhibitory Ly49G2, but higher frequencies of activating Ly49H+ and D+ cells. Our viSNE results identified 3 NK cell clusters displaying a “superactive” phenotype that were virtually absent in KO mice; consistent with our observations of their hyporesponsiveness. Taken together, these data support a role for Sostdc1 in the regulation of NK cells and could provide insights into novel biological parameters to expand active NK cell numbers with high killing efficiency for immunotherapies.