Skip to main content
Fig. 5 | Journal of Hematology & Oncology

Fig. 5

From: Treatment with soluble CD24 attenuates COVID-19-associated systemic immunopathology

Fig. 5

Subcluster analysis of peripheral blood NK cells in COVID-19 patients: activation following SARS-CoV-2 infection is dampened by CD24Fc treatment. CD56+ cells (n = 783,623) from HD (n = 17) and COVID-19 (n = 22) patients were clustered using an unbiased multivariate t-mixture model, which identified 12 sub-clusters that reflect statistically distinct CD56+ T cell activation states. The relative similarity of each cell and cell cluster on the two-dimensional UMAP space were visualized with a 10% downsampling (A). Using median expression of flow cytometry markers, a cluster-by-marker heatmap was generated to characterize subsets (B) and visualize individual marker expression patterns on UMAP space (C). To understand effect of SARS-CoV-2 infection on NK population dynamics, we compared UMAP plots (D) and cluster frequencies (E; cluster 1, p = 0.04; cluster 2, p = 0.003; cluster 9, p = 0.002; cluster 12, p = 0.03; p < 0.01 for clusters 4–6, 8 and 11) of HD vs. baseline COVID-19 samples. D2, 4, 8 samples from placebo and CD24Fc-treated groups were visualized using contour plots to represent density of cells throughout regions of the UMAP space (F, white arrows indicate visual changes between CD24Fc vs. placebo). The cluster population dynamics as fold change over baseline in each group was shown (G; sample distribution described in Fig. 1; p < 0.001 for cluster 1, 3–12). To better characterize the activation status of NK cells, a subset of markers (TOX, GZMB, KLRG1, Ki-67, LAG-3) was linearly transformed to create a univariate cell-level activation score (H), where highly activated cell clusters (such as cluster 11) had highest activation scores (I). A GLMM was fit to the longitudinal cell-level activation scores to assess the effect of CD24Fc on activation scores over time (J). The p values in E and J were calculated using Wilcoxon rank-sum and Kenward–Roger method, respectively. *p < 0.05; **p < 0.01; ***p < 0.001

Back to article page