Gut CD4+ T cell phenotypes are a continuum molded by microbes, not by TH archetypes

Abstract

CD4⁺ effector lymphocytes (T_eff) are traditionally classified by the cytokines they produce. To determine the states that T_eff cells actually adopt in frontline tissues in vivo, we applied single-cell transcriptome and chromatin analyses to colonic T_eff cells in germ-free or conventional mice or in mice after challenge with a range of phenotypically biasing microbes. Unexpected subsets were marked by the expression of the interferon (IFN) signature or myeloid-specific transcripts, but transcriptome or chromatin structure could not resolve discrete clusters fitting classic helper T cell (T_H) subsets. At baseline or at different times of infection, transcripts encoding cytokines or proteins commonly used as T_H markers were distributed in a polarized continuum, which was functionally validated. Clones derived from single progenitors gave rise to both IFN-γ- and interleukin (IL)-17-producing cells. Most of the transcriptional variance was tied to the infecting agent, independent of the cytokines produced, and chromatin variance primarily reflected activities of activator protein (AP)-1 and IFN-regulatory factor (IRF) transcription factor (TF) families, not the canonical subset master regulators T-bet, GATA3 or RORγ.

Extended Data Fig. 1. scRNAseq of Teff under normal conditions

a) Quality control plots (per-cell number of unique reads vs number of transcripts detected) for the scRNAseq data from total colonic CD4⁺ T cells (data from Fig. 1a). b) Same plots as (a), for CD4⁺ QC of scRNAseq data from total colonic CD4⁺ T cells of germ-free and SPF mice. c) SMART-SEQ2 single-cell data from colon T memory cells (from Miragaia et al.{11059}). Aggregate expression of Th-specific genesets (defined as for Fig. 1) are overlayed on the tSNE.

Extended Data Fig. 2. scRNAseq of Teff under infectious conditions.

a) tSNE representation of all CD4⁺ T cells in the scRNAseq data from the parallel infection experiment of Fig. 2. Left panel: each color represents cells from a different infection condition. Tregs, naive Tconvs, cycling cells and Teffs are circled; right panel: expression of key genes. b) UMAP representation of Teff cells from the same experiment, colored by condition; Right panels: Overlay of T_H genesets (per Fig. 2). c) Data from the same parallel-infection experiment as Fig. 2c and displayed using the same tSNE coordinates, highlighted with aggregate expression of T_H signature genes from ref{9203}. d) Expression of key cytokines and transcription factors in the same scRNAseq data as Fig. 2c. e) Independent parallel infection experiment. Samples were not hash-tagged, and processed in parallel encapsulations, and cell data were aligned by canonical correlation analysis (CCA) for tSNE representation, color-coded by sample. Right: expression of Th-specific genesets, defined as for Fig. 2c.

Extended Data Fig. 3. Different clustering approaches and signatures do not parse out the data into T_H subsets.

a) KNN clusters shown on hash-tagged tSNE. Percentages of cells corresponding to each signature in each KNN cluster are shown in the table. b) Biscuit clusters shown on hash-tagged tSNE. Percentages of cells corresponding to each signature in each Biscuit cluster are shown in the table. c) Backspin clusters shown on hash-tagged tSNE. Percentages of cells corresponding to each signature in each Backspin cluster are shown in the table. d) Overlay of pathogenic T_H17 signatures from{9915, 8270}. Left panel: all Teff; right panel: only Il17a⁺ Teff. e) Overlay of Citrobacter T_H17 signature from{11132} on the tSNE plot.

Extended Data Fig. 4. Neural Network prediction of IFN-γ and Il17-producing phenotypes

a) A Keras neural network was trained to use as input the expression of 500 most variable genes in Teff single-cell RNAseq data to predict Ifng or Il17a expression in each cell. Loss as a function of training epochs plotted here. Note the overfitting beyond 10 epochs (representative of >50 independent training runs with random 80/20 training/test). b) Accuracy of DNN-predicted cytokine expression by individual Teff cells, relative to their actual expression in the test scRNAseq data (non-expressing cells were not included as input, since there is uncertainty as to their real nature given drop-out frequencies in scRNAseq data). Numbers shown represent the range observed in 10 independent training runs (with different training/test sets). c) Contribution of each transcript to the prediction of Il17a or Ifng expression, as score in the Integrated Gradients, comparing the model learned in two independent runs. A positive score indicates influence on predicting Il17a expression, a negative score influence in predicting Ifng expression.

Extended Data Fig. 5. Th-associated genes are not the main drivers of Teff heterogeneity.

a) Distribution of Top 6 PCs of Teffs from all hash-tagged samples, with cell cycle genes regressed out. Genes that are Th-associated are highlighted. b) Co-expression of key cytokines across all samples. Mean Pearson gene:gene correlation of cytokine genes across all samples. Only significantly correlated cytokines are colored (p<0.05, χ² test). Significant P values: Il4/Il13 6.3X10⁻³, Il4/Il5 1.8X10⁻⁹⁸, Il5/Il13 5.5X10⁻¹²⁹, Il17a/Il17f 1.3X10⁻⁴. c) Coregulated gene modules in Teff single-cells. Gene:gene correlation between 588 most variable genes was calculated independently within each condition/infection of the single-cell datasets, then averaged between conditions. 16 gene modules were determined by Affinity Propagation within this matrix, annotated at right. d) Overlay of average expression of these gene modules on Teff tSNE (per 2c) with barplots showing genes with highest mean correlation (full list in Supplementary Table 3).

Extended Data Fig. 6. Unique clonotypes are not restricted to a T_H type and do not diversify over time.

a) Quantification of flow cytometry data on cells from mouse LP at different timepoints of infection; Left: Proportion of CD4⁺ T cells within total CD45⁺; Middle: Proportion of Teff (CD44^hi Foxp3⁻) within total CD4⁺ T; Right: Proportion of IFN-γ⁺ cells within total CD4 T. b) Cell numbers per scRNAseq clustering by day post infection. Treg clusters were identified as Foxp3⁺, naive cluster as Foxp3- Ccr7⁺ and Teff clusters as Foxp3- Cd44⁺. c) Left: UMAP as in 5a, showing two groups of cell clusters: cells taken from mice after day 10 are colored in red, and cells taken prior to day 7 are colored in blue. Right: DEG analysis on top 20 differentially expressed genes between the two cluster groups. Asterisks represent genes that overlap with genes that are higher in Teff after Salmonella infection in Fig. 3a. d) Bar graph representing proportions of cells belonging to singlet clones (clones that appear only once) or expanded clones (clones that appear more than once) in each of the clusters defined in S6b, grouped by day post infection. e) Median Euclidean distances between cells within the same clonotype across the top 10 clonotypes for each timepoint. Euclidean distance was calculated based on the top 1000 variable genes. Each color dot represents a unique clonotype, and the size of the dot signifies the number of cells within each clonotype.

Extended Data Fig. 7. The unexpected MyT subset.

a) Flow cytometric analysis (gated CD4⁺TCRβ⁺FOXP3⁻ Teff) cells from colonic LP of Salmonella infected mice. b) Volcano plot of bulk RNAseq from colonic Teff sorted as in C (LP of Salmonella infected mice). Genes highlighted in red belong to the myeloid genes listed in B.

Figure 1:. The transcriptional landscape of CD4⁺ T cells in the colon

a) scRNAseq analysis of total colonic LP CD4⁺ T cells of SPF mouse (computed from 658 most variable genes). Top: tSNE representation, color-coded by KNN cell clusters, identified based on expression of prototypic transcripts (bottom). b) scRNAseq analysis of total colonic LP CD4⁺ T cells of Germ-free and SPF mice. Top: tSNE representation, color-coded by cell of origin. Marked clusters are identified based on expression of prototypic transcripts (bottom). c) tSNE representation, restricted to CD4⁺ Teff selected in Fig. 1A (tSNE computed from 584 most variable genes). Right: overlay of combined expression of prototypic T_H genesets (Supplementary Table 2). d) Heatmap of Teff divided into two clusters by KNN clustering. Representative genes overexpressed in each cluster are shown.

Figure 2:. Variation in Teff transcriptomes shows continuous distribution that is not dictated by “T_H subsets”

a) Schematic of hash-tagging experiment. Mice were infected with different pathogens, their colonic LP cells extracted, labeled with hash-tagging antibodies, sorted as CD4⁺ T cells and processed as a single batch on the 10X chromium controller. Sample demultiplexing was done computationally. b) Flow cytometric confirmation of the intestinal infections after intracellular staining for the cytokines shown (gated on CD4⁺TCRβ⁺FOXP3⁻CD44^hi cells). c) tSNE representation of Teff scRNAseq data from mice under different infection conditions (computed from 930 variable genes). Left: color-coded by condition/infection; right: overlay of combined expression of prototypic TH genesets. d) Dendrogram of Euclidean distances between cells in the scRNAseq dataset of Fig. 2c, splitting cells that express Ifng or Il17a in each of the infection conditions. e) Hartigan’s Dip test applied to whole colonic CD4⁺ T cells from SPF mouse (top), or only to Teff from Salmonella-infected (middle) or Citrobacter-infected mice (bottom). MyT and cycling cells were taken out for this analysis. f) Expression of commonly-used markers of T_H subsets. Top: RNA expression in the scRNAseq data (overlaid on the tSNE from 2c). Bottom: Protein expression by flow cytometry in CD4⁺ Teff (gated CD4⁺TCRβ⁺FOXP3⁻CD44^hi cells) from Salmonella-infected mice.

Figure 3:. Teff phenotypes are distinguishable by infection rather than by T_H type

a) Heatmap of transcripts most differentially expressed within Ifng⁺ (top) or Il17a⁺ (bottom) Teff from Citrobacter or Salmonella infected mice. b) Sort of IL17a-expressing CD4⁺ CD44⁺ Teff cells from Il17a-IRES-gfp reporter mice (infected or not with Salmonella or Citrobacter) for expression profiling by ultra-low input RNAseq. c) PCA analysis of datasets from b. d) Volcano plot of bulk RNAseq data from C, comparing IL-17A⁺ Teff from Salmonella or Citrobacter infected mice. Red and blue highlights: transcripts found differentially expressed by scRNAseq.

Figure 4:. Repeated clonotypes can adopt different phenotypes, and do not diverge over time

a) UMAP representation of Teff from mouse LP at different timepoints post-infection with Salmonella. MyT cells are circled. b) Cells from different time points, cytokine-producing cells highlighted as shown c) Representative examples of clonotypes with unique CDR3 identified by scTCR sequencing (non-germline N and P nucleotides shown). d) Numbers of Il17a-, Ifng-, or both (DP) -expressing cells within the 10 most frequent clonotypes identified in each individual timepoint. e) Median Euclidean distances between cells within the same clonotype across the top 10 clonotypes for each timepoint. Euclidean distance was calculated based on T_H genes from Supplementary Table 2. Clonotypes are color-coded, and the size denotes the number of cells that express each clonotype.

Figure 5:. The chromatin states of Teffs are found on a continuum

a) scATACseq of total LP CD4⁺ T cells from Salmonella-infected mice. Top: UMAP representation, with Tregs and naive Tconv identified based on gene activity at prototypic loci (bottom). b) Cell chromatin scores for Rorc and Tbx21 loci, computed from accessibility of expression-correlated OCRs. Top: score in in vitro differentiated Th0, T_H1 and T_H17 cells. N=4 biological replicates for each condition. Centre, median; box limits, first and third percentiles; whiskers, 1.5× interquartile range (IQR). Bottom: scores for each cell in the scATACseq data from (a) (Teff only) in a UMAP plot. c) Aggregated coverage maps around Rorc and Tbx21 loci in Teffs split based according to their chromatin score at each locus (shown at left); arrows: location of the best expression-correlated OCRs used to compute the scores. d) Over-representation in each Teff cell (data from a) of TF-binding motifs in accessible chromatin (chromVAR bias-corrected TF motif deviation scores) for classic master regulators (UMAP framework from b) e) Combined variability across the Teff scATACseq data for OCRs that contain motifs for different TFs (blue: null distribution for permuted dataset). TF families across the ranking shown at right. f) TF motif deviation scores per Teff cell (as in d) for FOS and IRF4 motifs.

Figure 6:. Transcriptional and functional validation of Teff continuity

a) Experiment schematic. Surface markers with continuous distribution in the scRNAseq are selected, and cells stained with corresponding antibodies for flow cytometry. tSNE plots are computed from the cytometry data, from which sorting gates are set to prepare cells for transcriptional and functional analysis. b) Gene expression of selected surface markers in colonic Teffs from Salmonella infected mice (scRNAseq plot from 2c). c) Flow cytometry tSNE generated from fluorescence intensities of CD4⁺ Teffs stained for these markers. d) Sorting strategy, corresponding to the poles of the flow tSNE from C. The tSNE positions of the sorted cells are shown at right. e) Heatmap comparing differentially expressed genes in the bulk RNAseq profiling of populations A, B and C, sorted in D. Hierarchically clustered and row-mean normalized. f) Multiplex ELISA comparing secretion of cytokines and chemokines from populations A, B and C, where each bar is an independent biological replicate. *: cytokines with significant differences (at p <0.05) between any two populations per paired student t-test (IL-17A: A vs B p=4.0 X10⁻⁴, IL-22: A vs B p=1.3 X10⁻² and A vs B p=3.6X10⁻², IFN-γ: A vs B p=1.5X10⁻², CCL-5: A vs C p=2.7X10⁻²).

Figure 7:. Novel Teff populations

a) The ISG-T subset. Left: Interferon-type-I signature overlayed on the Teff tSNE. Right: Genes overexpressed in cluster ISG-T overlayed on top of genes upregulated in CD4⁺ T cells upon IFN-α or IFN-γ administration. b) scRNAseq expression data of genes in Crtam⁺ cluster. c) Volcano plot from RNA-seq of sorted CRTAM⁺ vs CRTAM⁻ colon Teff cells; over/under-expressed genes in the Crtam⁺ T cluster in scRNAseq data are shown in red and blue, with significance of overlap. d) Expression in MyT cells of genes overlayed on the general tSNE plot of Fig. 2c. top panels: typical myeloid cell transcripts; bottom: typical T cell transcripts. e) FoldChange histograms of myeloid-specific genes. Left, in myeloid vs CD4⁺ T cells (ImmGen RNAseq data); right: in MyT vs other colon Teff (Salmonella-infected, data from Fig. 2c). X axis on logarithmic scale. f) Contour plot representing RNA and protein expression in the single-cell data from Fig. 5 (x-axis: normalized scRNAseq; y-axis; raw CITE-seq counts) for MHCII (top) or CD14 (bottom). Individual cells are represented by dots, and are colored by their classification based on unsupervised clustering. g) Experiment schematic. Bone marrow from WT CD45.1 and CD45.2 H2-Ab1^−/− was mixed and transferred to irradiated CD45.1/CD45.2 hosts. After 8 weeks mice were infected with Salmonella and 13 days later the WT or KO LP CD4⁺ T cells were sorted for RNAseq. h) Top: schematic representation of the WT or KO H2-Ab1 loci [neomycin resistance gene inserted into the second exon]. Bottom: position of RNAseq reads in colonic CD4⁺ Teff stemming from WT or H2-Ab1 KO stem cells in mixed bone marrow chimeras infected with Salmonella.