Peanut oral immunotherapy differentially suppresses clonally distinct subsets of T helper cells

Abstract

Food allergy affects an estimated 8% of children in the United States. Oral immunotherapy (OIT) is a recently approved treatment, with outcomes ranging from sustained tolerance to food allergens to no apparent benefit. The immunological underpinnings that influence clinical outcomes of OIT remain largely unresolved. Using single-cell RNA-Seq and paired T cell receptor α/β (TCRα/β) sequencing, we assessed the transcriptomes of CD154+ and CD137+ peanut-reactive T helper (Th) cells from 12 patients with peanut allergy longitudinally throughout OIT. We observed expanded populations of cells expressing Th1, Th2, and Th17 signatures that further separated into 6 clonally distinct subsets. Four of these subsets demonstrated a convergence of TCR sequences, suggesting antigen-driven T cell fates. Over the course of OIT, we observed suppression of Th2 and Th1 gene signatures in effector clonotypes but not T follicular helper-like (Tfh-like) clonotypes. Positive outcomes were associated with stronger suppression of Th2 signatures in Th2A-like cells, while treatment failure was associated with the expression of baseline inflammatory gene signatures that were present in Th1 and Th17 cell populations and unmodulated by OIT. These results demonstrate that differential clinical responses to OIT are associated with both preexisting characteristics of peanut-reactive CD4+ T cells and suppression of a subset of Th2 cells.

Trial registration: ClinicalTrials.gov NCT01750879.

Keywords: Allergy; Immunology; T cells.

Figures

Figure 1. Peanut-reactive T cells decrease in frequency over the course of OIT.

(A) OIT study design, sample processing, and patient cohorts. CD3+CD4+CD45RA– memory T cells were sorted by FACS as CD154+CD137+/– (CD154+), CD154–CD137+ (CD137+), or CD154–CD137–. (For clinical outcomes and patient information, see Methods and Supplemental Tables 1 and 2.) (B) Representative flow plots of cells from 1 patient at 1 time point (n = 12 patients total). (C) Percentage of CD4+ memory T cells at each time point that were CD154+ (left) or CD137+ (right) in peanut-stimulated PBMC cultures from patients in the treatment group. *P < 0.05 (adjusted), by paired Wilcoxon rank sum test. AV, avoidance; BL, baseline; BU, buildup; MN, maintenance; PL, placebo; PT, partial tolerance; TF, treatment failure; TO, tolerance.

Figure 2. Peanut-reactive T cells from patients undergoing OIT have diverse and distinct transcriptional signatures.

(A) 2D UMAP visualization of all single-cell transcriptomes (n = 134,129 cells), colored by sorted subset and time point (left) and by patient and clinical group (right). (B) Top differentially expressed genes between the sorted subsets. Each column represents the scaled average gene expression of cells from a single patient. Genes were selected using a receiver operating characteristic (ROC) test. (C) Selected gene modules discovered using sparse PCA, labeled with module number and a proposed descriptor. For each module, the relative weights of each contributing gene and the module score of all cells overlaid on the UMAP coordinates are shown.

Figure 3. Gene modules for Th function are associated with clonal expansion and expression in activated cells.

(A) Clonal size of TCRα sequence (left) or TCRβ sequence (right) for all cells with paired TCR recovery, overlaid onto UMAP coordinates. Clonal size is defined as the number of cells sharing a TCR sequence. (B) Diversity (normalized Shannon index) of TCRβ repertoires of each sorted subset. Each data point represents the repertoire for 1 patient at 1 time point (CD137+: n = 41; CD154+: n = 44; CD154–CD137–: n = 23). (C) Distribution of TCRβ clonal sizes, within each sorted subset. Cells within each sorted subset were downsampled to equal numbers before clonal sizes were calculated. (D) Percentage of TCRβ sequences shared between time points and sorted subsets. The percentage shared is defined as the number of unique TCRβ sequences detected in both conditions, divided by the geometric mean of the number of unique TCRβ sequences in each of the 2 conditions. Sequences from all patients with samples in all 3 conditions (n = 6 patients) were pooled. (E) Mean clonal size and fold change in mean module scores (compared with module-expressing CD154–CD137– cells) in CD154+ cells expressing each gene module. Each data point represents a single gene module. Cells were classified as “expressing” each module or not, relative to background expression (see Methods). Clonal size was calculated with respect to all cells in the data set. ****P < 0.0001 (adjusted), by unpaired Wilcoxon rank-sum test. Data represent combined data from all patients at all time points (A–C and E).

Figure 4. Peanut-reactive Th subtypes are clonally distinct and exhibit TCR convergence.

(A) UMAP visualizations of Th1- (n = 7,609 cells), Th2- (n = 7,877 cells), and Th17-scoring cells (n = 7111 cells). Clusters are annotated by their putative identity. (B) Scatter plots of the average expression of IL5 and IL4 in Tfh2-like cells or Th2A-like cells (for each patient at each time point) and peanut-specific IgE titers. Linear fit, Spearman’s correlation (rs, n = 34), and adjusted P values are shown. (C) Fraction of TCRβ clonotypes belonging to each subset. The fraction is defined as the number of cells of a TCRβ CDR3 sequence (column) detected in each Th subset, divided by the total number of cells within the clonotype. Clonotypes were randomly downsampled to visualize a comparable number from each subset. (D) Differentially expressed genes in each Th subset. Genes were selected using an ROC test and manual curation. Each row represents the scaled average gene expression in 1 patient. (E) TCR distance analysis of TCR sequences. The x axis represents bins of increasing pairwise TCR distance, calculated using TCRdist, and the y axis represents the likelihood of pairs of cells at a given TCR distance to be of the same Th subset, normalized to the prior probability of any 2 cells belonging to that subset (see Methods). ****P < 0.0001 (adjusted), ***P < 0.001 (adjusted), and **P < 0.01 (adjusted), by 2-sided χ2 proportion test with 1 degree of freedom. The total number of pairs within each TCR distance and subset is indicated above each data point. The red asterisk indicates that no pair of TCR sequences with the respective TCR distance bin was found in the respective subset. Error bars represent 85% binomial CIs. Data were combined from all patients at all time points (A–E).

Figure 5. Th1 and Th2 effector, but not Tfh-like, subsets are suppressed by OIT.

(A) Mean Th2, Th1 and Th17 gene module expression over time within Th2, Th1, and Th17 clones (see Methods), respectively, in each treatment group patient at each time point. (B) Fractional expression of Th2, Th1, and Th17 modules within clonotypes of Th subtypes over time. Fractional clonal expression is defined as the proportion of cells within each clonotype expressing their respective module (see Methods). Each data point represents the cells of an individual expanded clonotype from 1 patient at 1 time point. Patients in the placebo group were excluded. (C) Degree of suppression in Th2A-like clones by clinical group. The ratio of mean Th2 module expression in Th2A-like clones from each patient was calculated between buildup and maintenance. Spearman’s rho = 0.74; *P < 0.05. n = 9. Spearman’s test was performed to determine the correlation between ratio and outcome within the treatment group (assigning 2 for tolerance, 1 for partial tolerance, and 0 for treatment failure to represent the ordinal relationship between treatment groups). (D) PC1 score for CD154+ cells by outcome. PCA was performed using the 50 gene modules as features and all CD154+ cells at baseline as the input data (see Methods). Each data point represents the mean PC1 score for all CD154+ cells from a single patient at a single time point. Black-outlined data points represent the baseline time point. (E) Top 5 gene module loadings in PC1. Bar heights represent the magnitude of each contribution to PC1. (See Supplemental Figures 4 and 5 for further details on each gene module.) *P < 0.05 (adjusted), **P < 0.005 (adjusted), and ****P < 0.0005 (adjusted), by paired (A) or unpaired (B and D) Wilcoxon rank-sum test.

Figure 6. Treg phenotypes are not significantly modulated by OIT.

(A) Average expression of Treg module (gene module 1), FOXP3, and IL10 by patient and time point within Treg clones. Each data point represents the mean expression of all Treg clones for a given patient at a given time point. (B) UMAP visualization of all Tregs (data from all patients at all time points), colored by cluster assignment and labeled by putative cluster identity. (C) Differentially expressed genes in each Treg cluster. Genes were selected using an ROC test and manual curation. Each row represents the scaled average gene expression in 1 patient. (D) Average expression of the Treg module (top), FOXP3 (middle), and IL10 (bottom) by patient and time point within clones of each Treg cluster, colored by clinical group. Adjusted P values were calculated by paired Wilcoxon rank sum test (A and D).