Research ResourceSINGLE CELL ANALYSIS

Mixed-effects association of single cells identifies an expanded effector CD4+ T cell subset in rheumatoid arthritis

See allHide authors and affiliations

Science Translational Medicine  17 Oct 2018:
Vol. 10, Issue 463, eaaq0305
DOI: 10.1126/scitranslmed.aaq0305
  • Fig. 1 MASC overview.

    Single-cell transcriptomics or proteomics are used to assay samples from cases and controls, such as immunoprofiling of peripheral blood. The data are then clustered to define populations of similar cells. Mixed-effects logistic regression is used to predict individual cell membership in previously defined populations. The addition of a case-control term to the regression model allows the user to identify populations for which case-control status is significantly associated.

  • Fig. 2 Diversity of CD4+ memory T cells before and after stimulation.

    (A) t-SNE projection of 50,000 resting CD4+ memory T cells sampled equally from RA patients (n = 24) and controls (n = 26). DensVM identified 19 populations in this dataset. (B) Same t-SNE projections as in (A) colored by the density of cells on the SNE plot or the expression of the markers labeled above each panel. (C) t-SNE projection of 52,000 CD4+ stimulated memory T cells sampled equally from RA patients (n = 26) and controls (n = 26). Cells were stimulated for 24 hours with anti-CD3/anti-CD28 beads. (D) Same t-SNE projections as in (C) colored by the density of cells on the SNE plot or the expression of the markers labeled above each panel. (E) Heatmap showing mean expression of indicated markers across the 19 populations found in resting cells. (F) Heatmap showing mean expression of indicated markers across the 21 populations found after stimulation. Protein expression data are shown after arcsinh transformation. All markers but CD4 and CD45RO were used to create t-SNE projections and perform clustering.

  • Fig. 3 MASC identifies a population that is expanded in RA.

    (A and B) ORs and association P values were calculated by MASC for each population identified the resting (A) and stimulated (B) datasets. The yellow dashed lines indicate the significance threshold after applying the Bonferroni correction for multiple testing. (C) Flow cytometry dot plot of gated memory CD4+ T cells from a single RA donor shows the gates used to identify CD27 HLA-DR+ memory CD4+ T cells (blue quadrant). (D) Flow cytometric quantification of the percentage of CD27 HLA-DR+ cells among blood memory CD4+ T cells in an independent cohort of seropositive RA patients (n = 39) and controls (n = 27). Statistical significance was assessed using a one-tailed t test after assessing normality with a Shapiro-Wilk test (P > 0.52).

  • Fig. 4 CD27 and HLA-DR expression specifically mark the expanded population.

    (A) Plot of the Kullback-Liebler (KL) divergence for each marker comparing cluster 18 to all other cells (gray) in both the resting dataset (red) and the stimulated dataset (blue). (B) Density plots showing expression of the five markers most different between cluster 18 cells (resting, red; stimulated, blue) and all other cells in the same dataset (black line). (C) Left: t-SNE projection of clusters identified in resting dataset. Middle: Same t-SNE projection, with cells gated as CD27 HLA-DR+ colored in red. Right: F-measure scores were calculated for the overlap between gated cells and each cluster in the resting dataset. (D) Left: t-SNE projection of clusters identified in stimulated dataset. Middle: Same t-SNE projection, with cells gated as CD27 HLA-DR+ colored in red. Right: F-measure scores were calculated for the overlap between gated cells and each cluster in the stimulated dataset.

  • Fig. 5 CD27 HLA-DR+ memory CD4+ T cells are expanded in the blood and joints of patients with active RA.

    (A) Flow cytometric quantification of the frequency of CD27 HLA-DR+ memory CD4+ T cells in 18 RA patients before starting a new medication, plotted against change in cell frequency after 3 months of new therapy. Treatment significantly reduced CD27 HLA-DR+ cell frequency as determined by a Wilcoxon signed-rank test. (B) Flow cytometric quantification of the percentage of memory CD4+ T cells with a CD27 HLA-DR+ phenotype in cells from seropositive RA synovial fluid (n = 8) and synovial tissue (n = 9) compared to blood samples from RA patients and controls. Blood sample data are the same as shown in Fig. 3D. Significance was assessed using one-tailed t test after determining normality with a Shapiro-Wilk test (P > 0.52) and applying a Bonferroni correction for multiple testing.

  • Fig. 6 Transcriptomic characterization of CD27 HLA-DR+ memory CD4+ T cells identified a TH1-skewed cytotoxic phenotype.

    RNA-seq characterization of CD27 HLA-DR+ (DR+27) cells and six related CD4+ T cell populations: TN, Treg, TCM, and three populations of TEM [CD27+ HLA-DR (DR27+), CD27+ HLA-DR+ (DR+27+), and CD27 HLA-DR (DR27)]. (A) PCA plot (top) and PC1 gene loadings (bottom) of 90 samples from the seven CD4+ T cell populations. Cells were colored on the PCA plot according to known cell type. Normal confidence ellipses at 1 SD were plotted for each cell type. The 300 most positive and 300 most negative PC1 gene loadings for each cell type were averaged and plotted in the heatmap. Genes relevant to the CD27 HLA-DR+ population were labeled. (B) Gene set enrichment analysis (GSEA) was performed on all genes, ranked on their PC1 loadings. Two significantly enriched gene sets, NK signature (GSE22886 NAIVE CD4 T CELL VS NK CELL DN) and TEM signature (GSE11057 NAIVE VS EFF MEMORY CD4 T CELL), are shown. (C) Distribution of log-scaled expression of six canonical TH1 genes: CCR5, CIITA, CXCR3, IFNG, TBX21 (Tbet), and TNF. Populations are ordered by PC1 loading values, with CD27 HLA-DR+ population highlighted in red. TPM, transcripts per million. (D) Distribution of log-scaled gene expression of six canonical cytotoxic genes: GNLY, GZMA, GZMB, GMZK, NKG7, and PRF1. Populations are ordered by PC1 loading values, with the CD27 HLA-DR+ population highlighted in red. Reported P values in (C) and (D) correspond to a linear model of gene expression against ordered cell type (as an ordinal variable), with P values adjusted for multiple testing by the Benjamini-Hochberg procedure. (E) Cytokine expression determined by intracellular cytokine staining of peripheral effector memory CD4+ T cells after in vitro stimulation with PMA/ionomycin. The percentage of cells positive for each stain is plotted for CD27+ HLA-DR and CD27 HLA-DR+ subsets. Each dot represents a separate donor (n = 12; 6 RA patients and 6 controls, except for the quantification of granyzme A and perforin, where n = 6; 3 RA patients and 3 controls). Statistical significance was assessed using a Wilcoxon signed-rank test.

  • Table 1 Clinical characteristics of patient samples used.

    Mean ± SD is shown. Parentheses indicate percentages. “Other biologics” includes rituximab, tofacitinib, and abatacept. ACPA, anticitrullinated protein antibody; CDAI, clinical disease activity index; CRP, C-reactive protein; NA, not applicable; TNF, tumor necrosis factor.

    Mass cytometry cohortFlow cytometry cohort
    ControlsCasesControlsCases
    Blood cross-sectional
    cohorts
    Number26262739
    Age57 ± 1566 ± 961 ± 1458 ± 14
    Female19 (73)20 (77)18 (64)30 (77)
    ACPA- or RF-positiveNA22 (85)NA39 (100)
    CRP (mg/liter)NA8.6 ± 16.9NA9.8 ± 17.9
    CDAINA9.3 ± 4.4NA13.7 ± 7.4
    Methotrexate018 (69)018 (46)
    Anti-TNF010 (38)016 (41)
    Other biologics05 (19)09 (23)
    Longitudinal
    cohort
    Blood longitudinal cohortNumber18
    Age49 ± 17
    Female17 (94)
    ACPA- or RF-positive18 (100)
    CDAI before17.6 ± 9.3
    CDAI after6.3 ± 4.2
    Started methotrexate7
    Started anti-TNF4
    Started other biologic7
    Patient#1#2#3#4#5#6#7#8#9
    Synovial tissue donorsAge575476464679626352
    SexFFFFFFMMF
    CRP (mg/liter)2588111719136676
    CDAI1491715212559NA
    MethotrexateNoYesNoNoNoNoNoYesNo
    Biologic therapyYesYesYesYesYesNoNoNoNo
  • Table 2 Overview of the subsets found to be significantly expanded in RA.

    RA proportion reflects the fraction of cells in the subset that were from RA donors. The 95% CI is shown next to the OR.

    ConditionDescriptionSubsetRA proportionPORTest
    RestingHLA-DR+, CD27180.6365.9 × 10−41.9 (1.3–2.7)MASC
    StimulatedHLA-DR+, CD27180.6191.3 × 10−31.7 (1.2–2.2)MASC
    Flow cytometry replicationGated
    HLA-DR+, CD27
    NANA4.4 × 10−2NAOne-tailed t test
    Meta-analysisHLA-DR+, CD27NANA2.3 × 10−4NAStouffer’s Z-score method

Supplementary Materials

  • www.sciencetranslationalmedicine.org/cgi/content/full/10/463/eaaq0305/DC1

    Fig. S1. MASC type 1 error.

    Fig. S2. t-SNE projection density.

    Fig. S3. CIM analysis of clustering approaches.

    Fig. S4. DensVM clustering of elbow plots.

    Fig. S5. Marker expression distribution plots for DensVM clusters.

    Fig. S6. Association permutation testing and cluster alignment.

    Fig. S7. PhenoGraph and FlowSOM clustering.

    Fig. S8. Association testing with Citrus.

    Fig. S9. Flow cytometry and RNA-seq gating strategies.

    Fig. S10. CD27 and HLA-DR expression in flow cytometry cohort.

    Fig. S11. CD4+ TEM populations in a clinical response cohort.

    Fig. S12. CD27 HLA-DR+ frequency and clinical characteristics.

    Fig. S13. RNA-seq analysis of CD4+ T cell subsets.

    Fig. S14. Flow cytometry expression quantification.

    Fig. S15. Using a neural network autoencoder to cluster mass cytometry data.

    Table S1. Panel design for mass cytometry experiments.

    Table S2. MASC analysis of the 19 clusters identified in the resting dataset.

    Table S3. MASC analysis of the 21 clusters identified in the stimulated dataset.

    Table S4. GSEA of genes differentially expressed in CD27 HLA-DR+ cells.

  • The PDF file includes:

    • Fig. S1. MASC type 1 error.
    • Fig. S2. t-SNE projection density.
    • Fig. S3. CIM analysis of clustering approaches.
    • Fig. S4. DensVM clustering of elbow plots.
    • Legend for fig. S5
    • Fig. S6. Association permutation testing and cluster alignment.
    • Fig. S7. PhenoGraph and FlowSOM clustering.
    • Fig. S8. Association testing with Citrus.
    • Fig. S9. Flow cytometry and RNA-seq gating strategies.
    • Fig. S10. CD27 and HLA-DR expression in flow cytometry cohort.
    • Fig. S11. CD4+ TEM populations in a clinical response cohort.
    • Fig. S12. CD27 HLA-DR+ frequency and clinical characteristics.
    • Fig. S13. RNA-seq analysis of CD4+ T cell subsets.
    • Fig. S14. Flow cytometry expression quantification.
    • Fig. S15. Using a neural network autoencoder to cluster mass cytometry data.
    • Table S1. Panel design for mass cytometry experiments.
    • Table S2. MASC analysis of the 19 clusters identified in the resting dataset.
    • Table S3. MASC analysis of the 21 clusters identified in the stimulated dataset.
    • Table S4. GSEA of genes differentially expressed in CD27 HLA-DR+ cells.

    [Download PDF]

    Other Supplementary Material for this manuscript includes the following:

    • Fig. S5 (.pdf format). Marker expression distribution plots for DensVM clusters.

Stay Connected to Science Translational Medicine

Navigate This Article