a, viSNE plots of mass cytometry of RA synovial tissue CD4⁺ T cells as inFig. 1a from 2 additional donors.b, Gating strategy to identify synovial tissue PD-1^hi CD4⁺ T cell populations by mass cytometry.c, viSNE plots of flow cytometry of paired RA synovial fluid and blood memory CD4⁺ T cells.d, Gating strategy to identify synovial fluid PD-1^hi memory CD4⁺ T cells by flow cytometry.e, Examples of gating used to sort memory CD4⁺ T cell populations from patient samples.f, Detection of CXCR5 mRNA by RT–PCR in sorted memory CD4⁺ T-cell populations from synovial tissue (n = 3 donors, 2 of which provided sufficient PD-1^hi CXCR5⁺ cells for analysis), synovial fluid (n = 3 donors, 1 of which provided sufficient PD-1^hi CXCR5⁺ cells for analysis), and blood (n = 2 donors). Purple boxes indicate PD-1⁻ and PD-1^hi CXCR5⁺ cells sorted from human tonsil as controls. Lines inf indicate mean for synovial or blood samples.

a, Mean expression of MHC II and ICOS in memory CD4⁺ T cell populations from synovial tissue (n = 10), synovial fluid (n = 9), and blood (n = 42) from patients with seropositive RA. Mean ± s.d. shown.b, Flow cytometric detection of PD-1 and CXCR5 expression on blood memory CD4⁺ T cells.c–e, Frequency of PD-1^hi cells that co-express MHC II or ICOS (c), PD-1^hi CXCR5⁺ cells (d) or cells with intermediate PD-1 expression (e) within memory CD4⁺ T cells from blood of patients with seropositive RA (n = 42), seronegative RA (n = 16), spondyloarthropathy (SpA,n = 11), and non-inflammatory controls (n = 35).f, Correlation between age or disease duration and blood PD-1^hiCXCR5⁻ cell frequency in seropositive RA patients (n = 38).g, PD-1^hiCXCR5⁻ cell frequencies in seropositive RA patients segregated based on sex or medication usage (n = 38).h, Correlation between serum anti-CCP antibody titer and blood PD-1^hiCXCR5⁻ cell frequency in all RA patients (n = 53, black line,P = 0.0049) or in only anti-CCP antibody⁺ patients (n = 29, green line,P = 0.48).i, Correlation between fold change in CDAI and fold change in PD-1^hiCXCR5⁻ cell frequency in patients 3 months after addition of a new RA medication (n = 23; methotrexate, 11; anti-TNF, 4; abatacept, 4; tocilizumab, 2; tofacitinib, 2).j, Frequency of PD-1^hi T-cell subpopulations in blood before and after RA treatment escalation in 18 patients with reduced disease activity after therapy. Median ± interquartile range (c–e); mean ± s.d. (a, g) shown. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001 by Mann–Whitney (c, g), Kruskal–Wallis (d, e), Wilcoxon test (j). Inf, h, i,P values calculated by Spearman correlation.

a, RT–PCR for intracellular regulators in memory CD4⁺ T cell populations from RA synovial fluid (n = 5 or 6 donors).b, RT–PCR for cytokines (n = 10 donors, 6 RA patients (black), 4 controls (grey)) and intracellular regulators (n = 4 or 5 donors) in memory CD4⁺ T cell populations from blood.c, Cytokine and transcription factor mRNA expression in blood PD-1^hi memory CD4⁺ T cell populations divided according to CXCR5 expression, relative to PD-1⁻ cells (n = 6 donors).d, Flow cytometric quantification of BCL6 and BLIMP1 in blood PD-1^hi memory CD4⁺ T cell subpopulations sorted according to chemokine receptor expression, then stimulatedin vitro for 2 days with anti-CD3/CD28 beads. Representative data from 1 of 3 experiments using cells from different donors. Median ± interquartile range (a,b); mean ± s.d. (c,d) shown. *P < 0.05, **P < 0.01, ***P < 0.001 by Friedman’s test, compared to PD-1^–MHC-II⁻ group (a,b).

a, Gating of blood PD-1^hi memory CD4⁺ T cells in mass cytometry analyses.b, Flow cytometric detection of FOXP3 and PD-1 in blood memory CD4⁺ T cells from patients with RA (black,n = 5) and controls (grey,n = 3).c, Flow cytometric detection of inhibitory receptors on blood CXCR5⁻ memory CD4⁺ T cells. Data from 1 of 3 patients with RA with similar results.d, Sorting strategy for PD-1^hiCXCR5⁻ and PD-1^hiCXCR5⁺ cell populations for RNA-seq.e, Hierarchical clustering T-cell subpopulations sorted as ind, with clustering based on expression of T_FH-associated genes measured by RNA-seq.f, Chemokine receptor expression on blood memory CD4⁺ T cells from patients with RA (black) or controls (grey) by flow cytometry. Mean ± s.d. shown. *P < 0.05, **P < 0.001, ***P < 0.0001 by Kruskal–Wallis test compared to PD-1⁻ cells (b) or Wilcoxon test (f).

a, Flow cytometry of CXCR5 and CCR2 on gated PD-1^hi memory CD4⁺ T cells from blood.b, Expression of CXCR5 and CCR2 on indicated sorted PD-1^hi T-cell populations 7 days afterin vitro stimulation with anti-CD3/CD28 beads.c,d, Percentage of cells from each sorted PD-1^hi population that expressed CXCR5 or CCR2 on day 2 (c) or day 7 (d) afterin vitro stimulation. Naive CD4⁺ T cells are shown as control. Mean ± s.d. shown (n = 3 donors from 3 separate experiments).

a, Flow cytometric quantification of SLAM, SLAMF5, and SLAMF6 expression on memory CD4⁺ T cells (n = 10 donors; 5 patients with RA, 5 controls).b, Quantification of frequency of memory B cells with plasma cell markers after co-culture with PD-1^hiCXCR5⁺CD4⁺ T cells with addition of blocking antibodies against SLAMF5 and/or SLAMF6.c, IgG quantification by ELISA in co-cultures of memory B cells with PD-1^hiCXCR5⁻ or PD-1^hiCXCR5⁺ T cells with addition of blocking antibodies against SLAMF5 and/or SLAMF6. Forb,c, 1 of 3 experiments with similar results (n = 3 replicates shown). Mean ± s.d. shown. *P < 0.05, **P < 0.01, ***P < 0.001 by Kruskal–Wallis compared to PD-1⁻CXCR5⁻ (a) or isotype control (b,c).d, Immunofluorescence microscopy of CD20 (green), CXCR5 (red), and PD-1 (blue), in seropositive RA synovial tissue. Arrows point to PD-1^hiCXCR5⁻ cells adjacent to B cells. Scale bar, 50 μm.