Skip to main content
Fig. 7 | BMC Biology

Fig. 7

From: Time-resolved transcriptome and proteome landscape of human regulatory T cell (Treg) differentiation reveals novel regulators of FOXP3

Fig. 7

Molecular profiling reveals known and novel Treg regulators. Novel ‘candidate’ molecules putatively involved in FOXP3 induction were selected, along with ‘known’ Treg regulators as control (see text and Additional file 1: Figure S7a, b). ac Expression profile of novel iTreg candidate molecules. IKZF4 (Eos) and FOXP3 are shown for comparison. Labels as in Figs. 1 and 5. a Candidate gene mRNA counts were rlog-transformed, and row-normalized z-scores are displayed (blue: low, red: high expression). b iTreg candidate protein expression (log2R). Grey cells: the protein was not detected in the respective sample. c iTreg candidate gene expression from an additional, completely independent RNA-Seq dataset is displayed and analyzed as in (a). iTregs were cultured with TGF-β + ATRA + serum, and additional time points were measured. d, e In silico validation of novel candidate molecules for iTreg classification. d Linear discriminant analysis (LDA) with all possible combinations of two genes out of ‘37 candidates’, ‘37 known’ Treg, or 349 nodes in the iTreg subnetwork (‘349 iTreg’) lists, followed by group classification. Grey boxplots show the cross-validated accuracy of classifiers regarding discrimination of Mock-stimulated (G02) vs. iTregs generated with either protocol (G03–G06) in the Main dataset. White boxplots show results from LDA analyses performed in the same way, but for discrimination of Mock-stimulated cells (G02a) vs. iTregs (G04a) in the independent dataset. Whiskers: min. to max. Value; + mean value; n = 1332, 1332 and 121,452 pairs for ‘37 candidates’, ‘37 known’, and ‘349 iTreg’, respectively. The adjusted p value was ≤ 0.0001 by Kruskal–Wallis test with Dunn’s multiple comparison test for each vs. each boxplot. e Example of two candidate genes (top classifiers) which separate Mock cells from iTregs in the Main dataset with 100% accuracy (0% error) in LDA

Back to article page