4 mM of AC-

or BC-primer and 0.4 mM of AV- or BV-specific primer. After an initial denaturation step of 10 min at 94°C, the reactions were subjected to 40 cycles of PCR (94°C HDAC inhibitor for 30 s, 58°C for 40 s, 72°C for 50 s), followed by a final extension step of 5 min at 72°C. Runoff products were purified using Sephadex gel and filter plates (Multiscreen, Millipore, Billerica, MA, USA) before they were sequenced using fluorescent chain-terminating inhibitors (BigDye Terminator v1.1 kit) and an automated capillary sequencer (ABI Prism 3700 DNA Analyzer, Applied Biosystems). CDR3α and CDR3β definitions as well as AV and BV nomenclature are according to IMGT (http://imgt.cines.fr). Cytokines were selected for cluster analysis on the basis of their recognized contribution to characterize both known and potential CD4+ T-cell subsets. Cytokine secretion levels of PMA and calcium ionophore-activated CD4+ T cells were determined ex vivo at the single-cell level using a BD LSRII apparatus. Fluorescence intensity values were directly extracted from the corresponding Flow Cytometry Standard (FCS) files

using Flow Cytometry Standard Extract Utility (Earl F. Glynn, Stowers Institute for Medical Research, KS, USA) and analyzed using Ward’s method (see below). T-cell clone clustering was based on cytokine ELISA measurements in culture supernatants. In that case, the molar concentration of each cytokine measured was expressed as the percentage of the six measured cytokines produced by a given

DAPT nmr T-cell clone and normalized in order to express results independently of their measurement scale. Agglomerative hierarchical cluster analysis according to Ward’s 46 was performed using Reverse transcriptase the JMP7 software (SAS Software, NC, USA). The optimal number of clusters was identified according to the largest distance change between successive junctions of the dendrogram plot. Validity and reproducibility of the classification obtained with hierarchical cluster analysis was assessed using non-hierarchical k-means cluster analysis, in which the optimal number of clusters identified through hierarchical cluster analysis was pre-specified. Reproducibility of the classifications obtained with both hierarchical and non-hierarchical clustering was assessed by determination of the kappa value. Differences between groups and clusters were tested using Mann-Whitney U-test (unpaired), Wilcoxon signed rank test (paired) and Kruskal-Wallis test. All tests were two-sided and a p-value <0.05 was considered statistically significant. This study was supported by Inserm, by the Centre d’Investigations Biologiques (C.I.B.) Pitié-Salpêtrière, by the Université Pierre et Marie Curie “EMERGENCE” program and by the European FP6 “ATTACK” program (Contract: LSHC-CT-2005-018914).Authorship contributions: M.L., M.H., D.D., C.P., J.P., K.D., M.S. and D.S. performed research, J.P., H.Y., and L.A. contributed vital new reagents and analytical tool, S.B., M.K. and Z.A.