Automated clustering reveals CD4(+) T cell subset imbalances in rheumatoid arthritis

BACKGROUND: Despite the report of an imbalance between CD4(+) T helper (Th) cell subsets in rheumatoid arthritis (RA), patient stratification for precision medicine has been hindered by the discovery of ever more Th cell subsets, as well as contradictory association results. OBJECTIVES: To capture previously reported Th imbalance in RA with deep immunophenotyping techniques; to compare hypothesis-free unsupervised automated clustering with hypothesis-driven conventional biaxial gating and explore if Th cell heterogeneity accounts for conflicting association results. METHODS: Unstimulated and stimulated peripheral blood mononuclear cells from 10 patients with RA and 10 controls were immunophenotyped with a 37-marker panel by mass cytometry (chemokine receptors, intra-cellular cytokines, intra-nuclear transcription factors). First, conventional biaxial gating and standard definitions of Th cell subsets were applied to compare subset frequencies between cases and controls. Second, unsupervised clustering was performed with FlowSOM and analysed using mixed-effects modelling of Associations of Single Cells (MASC). RESULTS: Conventional analytical techniques fail to identify classical Th subset imbalance, while unsupervised automated clustering, by allowing for unusual marker combinations, identified an imbalance between pro- and anti-inflammatory subsets. For example, a pro-inflammatory Th1-like (IL-2(+) T-bet(+)) subset and an unconventional but pro-inflammatory IL-17(+) T-bet(+) subset were significantly enriched in RA (odds ratio=5.7, p=2.2 x 10(-3); odds ratio=9.7, p=1.5x10(-3), respectively). In contrast, a FoxP3(+) IL-2(+) HLA-DR(+) Treg-like subset was reduced in RA (odds ratio=0.1, p=7.7x10(-7)). CONCLUSION: Taking an unbiased approach to large dataset analysis using automated clustering algorithms captures non-canonical CD4(+) T cell subset imbalances in RA blood.