Page 66 - The autoimmune hypothesis of narcolepsy and its unexplored clinical features M.S. Schinkelshoek

64
Chapter 4
Integrated data analysis
Gating of CD45+ live, single cells from the data derived from single cell acquisition was performed using FlowJoTM, version 10. The gating strategy can be found in Supplementary figure 4.1. After sampletagging, hyperbolic ArcSinh transformation with a cofactor of 5, these CD45+ cells were subjected to dimensionality reduction analysis using Cytosplore (Höllt et al., 2016). The overview level of a 6-level hierarchical stochastic neighbor embedding (HSNE) analysis with default perplexity (30) and iterations (1000) was used to identify major immune cell lineages in the merged CD45+ cells of all samples (18.9 x 106 cells). Gaussian mean shift (GMS) unsupervised clustering was performed in Cytosplore and an algorithm was used to merge clusters that showed high similarity in ArcSinh5-transformed marker expression. For further zooming into the data, we selected cells based on visible clusters, selecting clusters derived by using the GMS clustering. HSNE analyses were repeated until clusters consisted of a maximum of 0.5x106 landmarks (van Unen et al., 2017). Subsequently, hierarchical clustering on cell frequencies was performed in Matlab using Spearman’s rank correlation, as previously described (de Vries et al., 2019). Immune cell clusters that were most strongly correlated based on Spearman’s ρ, were assessed for differences in antibody staining. When differences in antibody staining were deemed negligible, the subsets were merged. This process was repeated until differences in antibody staining yielded phenotypically distinct cell clusters. We have not been able to account for technical variation by adding a peripheral blood mononuclear cell (PBMC) reference sample to every mass cytometry experiment. Therefore, batch effects were controlled for in the final set of immune cell clusters.
Statistics
The primary analysis concerned the comparison between the immune cell composition of NT1 patients with recent onset and HLA-DQB1*06:02-matched healthy controls. Secondary analyses were performed between NT1 patients with recent onset and those with onset longer than two years ago. All results are shown as mean ± standard error of the mean (SEM). Differences in relative frequencies of the major immune cell lineages between participant groups were assessed by one-way ANOVA with Bonferroni post-hoc comparisons. For differences between immune cell clusters within major immune cell lineages between participant groups two-tailed Mann-Whitney U tests for unpaired samples were used. Corrections for multiple comparisons were made when