Fig. 2: Unsupervised cell clustering classifies CAR variants based on distinct transcriptional phenotypes.

a UMAP embedding and unsupervised cell clustering based on scRNA-seq data of the pooled CAR T cell library following functional activation (co-culture with cognate antigen-expressing SKBR3 tumor cells). Shown are data from 7,244 variant-assigned CAR T cells with random subsampling restricting it to a maximum of 250 cells per CAR construct. In addition, 500 cells for TCR-negative (TCR-) and unstim-CD28-CD3Z negative control samples were included. On the right, UMAP feature plots show the distribution of CD4- and CD8-expressing cells. b Dot plot showing the expression of a selection of T cell marker genes across clusters found in a. Color and size indicate differences in expression levels between groups. c Cluster enrichment observed for the top 40 most represented CAR variants (variants with at least 50 cells assigned), benchmark controls (CD28-CD3Z and 4-1BB-CD3Z) and negative controls (TCR- and unstim-CD28-CD3Z). Variants are ordered by a confidence score based on the number of available cells. The top panel displays the CD4 and CD8 counts for each group and the bottom panel the fraction of cells found in each of the clusters from a. d Distribution of the unstimulated controls and the benchmark controls within the UMAP embedding of a. On the right a UMAP embedding plot highlights the CICs. In all UMAP plots, a red line depicts the region where there was no enrichment of unstimulated control samples. e Bar plot showing the distribution of CAR unstimulated control cells between the different clusters. f Bar plot depicting the percentage of cells per CAR variant that belong to the 5 different CICs. Source data for panels c, e, f are provided as a Source Data file.