Peanut allergy oral immunotherapy drives single-cell multi-omic changes in peanut-reactive T cells associated with sustained unresponsiveness

Single-cell multi-omics analysis of pr CD4⁺ cells from individuals in the POISED study

The POISED trial design³ is illustrated in Fig. 1a. Participants were randomly assigned to placebo or active treatment groups to be built up and maintained on 4,000 mg peanut protein until week 104 then discontinued peanut dosing (peanut-0 group) or given 300 mg peanut protein daily (peanut-300 group) after week 104. DBPCFCs to 4,000 mg peanut protein were done at BL and week 104 for all participants and at weeks 117, 130, 143 and 156 for participants who passed the DBPCFC at the last time point. Excluding 16 participants who dropped out of the study, 80 of 81 participants in the active treatment group passed the DBPCFC at week 104 and were desensitized, whereas only one of 23 participants in the placebo group passed the week 104 DBPCFC. Twenty-one of 51 (41.2%) peanut-0 group desensitized participants achieved SU, defined as passing DBPCFC at week 117 after 13 weeks of peanut avoidance after OIT. The remaining 30 participants in the peanut-0 group who passed DBPCFC at week 104 but failed at week 117, were defined as only desensitized.

In this trial, we analyzed peripheral blood mononuclear cells (PBMCs) from 31 participants of POISED, selected for having complete time course sampling and sufficient cell numbers available: 23 active participants, who underwent OIT and passed DBPCFC at week 104 (14 peanut-0 participants, of whom seven passed (SU) and seven failed DS at week 117, and nine peanut-300 participants, of whom five passed and four failed DBPCFC at week 117); eight placebo participants; and five age-matched and sex-matched controls without PA (Fig. 1a and Supplementary Tables 1 and 2). We stimulated PBMCs with peanut solution for 18 h and sorted pr CD4⁺ T cells (CD4⁺CD137⁺ or CD4⁺CD154⁺) and non-reactive (nr) CD4⁺ T cells (CD4⁺CD154⁻CD137⁻)^9,16. Sorted cells were processed using the BD Rhapsody platform for single-cell RNA sequencing (scRNA-seq) with a focus on 432 immune-relevant target transcripts (399 commercially predefined target panel complemented with 33 targets of interest; Supplementary Table 3) and 42 immune-relevant single-cell antibody sequencing (scAb-seq) markers for protein level detection (Supplementary Table 4), as well as single-cell paired TCRα and TCRβ sequencing (Fig. 1b).

After quality control and removal of multiplets and undetermined cells, we recovered 160,336 cells with high-quality sequencing data. The resulting transcriptomics and protein expression data were separately processed using reciprocal principal component analysis (PCA) to remove batch effects and were integrated using weighted-nearest neighbor (WNN) analysis, an unsupervised framework enabling integrative analysis of multiple modalities, after dimensionality reduction¹⁵. Cells were visualized using uniform manifold approximation and projection (UMAP)¹⁷. Three major populations could be readily distinguished using unsupervised clustering: pr T_eff CD4⁺ T cells (CD4⁺CD154⁺); pr T_reg CD4⁺ T cells (CD4⁺CD137⁺CD154⁻); and nr CD4⁺ T cells (CD4⁺CD154⁻CD137⁻) (Extended Data Fig. 1). Using the same approach, we subclustered pr T_eff and T_reg CD4⁺ T cell subsets into 14 subclusters based on integrated RNA and protein expression data using the WNN method and annotated based on their expression profiles (Fig. 1c,d and Supplementary Table 5): T_H2a-like cells (PTGDR2^high, GATA3^high, IL4^high, IL5^high, IL13^high, HPGDS^high, CD27^low); conventional T_H2 (T_H2conv)-like cells (GATA3^high, IL4^high, IL5^high, IL13^high, IL9^high); T_fh-like cells (CXCR5^high); T_H1 and T_H17-like cells (IL17F^high, CCL20^high, RORC^high, IFNG^high, TBX21^high); T_H22 and T_H17-like cells (IL22^high, CCR6^high); T_H1 cytotoxic T lymphocyte (CTL)-like cells (THBS1^high, NKG7^high, GNLY^high, GZMB^high, GZMH^high, PRF1^high, CXCR3^high); two subtypes of activated T_eff (act1 and act2) cells (IFITM3^high, IFITM2^high, TNFSF10^high, IL7R^high); two subtypes of pr T_reg cells, that is, T_reg_act1 (CCR6^high, ENTPD1^high) and T_reg_act2 (SELPLG^high); and four subtypes of pr naive cells (TNa_1, TNa_2, TNa_3, TNa_4).

To dissect clonotypic features among 72,491 pr CD4⁺ T cell phenotypes, we retrieved TCRα and TCRβ sequences from 65,032 pr CD4⁺ T cells with 40,897 of them having paired TCRα and TCRβ sequences. Analysis revealed distinct clonal lineages among the 14 clusters of pr CD4⁺ T cells, with minimal overlap between subclusters sharing similar phenotypes (Fig. 1e). Clonally expanded cells within an individual were identified based on common α-chain and β-chain complementarity-determining region 3 (CDR3) nucleotide sequences with the same nucleotide sequences of the V, DJ and C region (the most stringent way to define clonotype¹⁸), with low levels observed among pr naive cells and high levels among T_H2a-like and T_H1 CTL-like cells (Fig. 1f,g). Peanut stimulation upregulated CD154 on pr naive-like cells at 18 h without inducing memory transition or clonal expansion, suggesting early activation or bystander response. As these cells showed limited association with clinical outcomes (Supplementary Figs. 1 and 2), we did not focus on this population in this study.

Peanut OIT reduces T_H2-related phenotype and enhance T_H1 cytotoxic phenotypes

To investigate pr T cell phenotypic changes during OIT, we first checked subcluster frequencies within pr T_eff memory cells and found that the frequency of T_H2a-like cells among pr T_eff memory cells decreased significantly (Fig. 2a; q = 0.02); the frequency of T_H2conv-like, T_H22 and T_H17-like, and T_eff-act2 cells each showed a decreasing trend (Fig. 2a; q = 0.09, q = 0.09 and q = 0.1, respectively). Notably, the frequency of T_H1 CTL-like, and T_H1-like and T_H17-like, cells increased significantly (Fig. 2a; q = 0.02 and q = 0.02, respectively); T_fh-like cells showed a nonsignificant increase during OIT in active participants (Fig. 2a; q = 0.07). Placebo participants had no significant changes in frequency in any subcluster between BL and week 104 (Fig. 2a). Additionally, we noted that participants with PA compared to participants without PA had higher frequencies of T_H2conv-like cells (Extended Data Fig. 2a; P = 0.017). The frequency of pr T_H2a-like cells among pr T_eff memory cells was not significantly different between groups (Extended Data Fig. 2b), probably because individuals without PA have fewer pr CD4⁺ T cells among total CD4⁺ T cells¹⁹ and because of the small number of participants without PA who were evaluated. In addition, T_H2a-like clonotypes showed less clonal expansion in individuals without PA compared to participants with PA (Extended Data Fig. 2c).

Peanut-specific and Ara h 2-specific IgE significantly decreased (Extended Data Fig. 3a; P = 0.00098 and P = 0.012, respectively), while the peanut-specific IgG4-to-IgE ratio and Ara h 2-specific IgG4-to-IgE ratio (Extended Data Fig. 3a; P = 2.4 × 10⁻⁷ and P = 4.8 × 10⁻⁷, respectively) each significantly increased during OIT in the active but not in the placebo group (Extended Data Fig. 3b). We examined the relationship between antibody measurements and subcluster frequencies among pr T_eff memory cells at the combined BL and week 104 time points in active and placebo participants (Extended Data Fig. 3c). The frequencies of T_H2conv-like and T_H2a-like cells were positively correlated with peanut-specific and Ara h 2-specific IgE levels and negatively correlated with IgG4-to-IgE ratios. In contrast, T_H1 CTL-like cell frequencies showed the opposite pattern, positively correlating with IgG4-to-IgE ratios and negatively correlating with IgE levels. T_H1 CTL-like, and T_H1-like and T_H17-like, frequencies were positively correlated with each other and inversely correlated with T_H2-like subsets (Extended Data Fig. 3d). The T_H2conv-like, T_H2a-like and T_H1 CTL-like subcluster frequencies at the BL time point also showed similar patterns of correlation with each other and with the BL serological measurements (Extended Data Fig. 3e,f).

We next evaluated cluster-specific feature gene module scores in pr T_eff memory CD4⁺ T cells to better capture both subcluster frequency and phenotype changes. Module scores reflect the expression of key subtype-defining genes (Supplementary Table 5 and Methods) and allow assessment of functional shifts even when cell frequencies remain stable. We found that during active OIT, but not placebo treatment, the module scores of T_H2a-like, T_H2conv-like, T_H22-like and T_H17-like, T_eff-act1 and T_eff-act2 cells significantly decreased, while T_H1 CTL-like, T_H1-like and T_H17-like, and T_fh-like cells significantly increased in pr T_eff memory cells (Fig. 2b,c). The gene module score and subcluster frequency analysis showed congruent patterns; they probably represent different but complementary views of the same underlying biological signal. Notably, although the frequency of T_H2conv-like cells showed a nonsignificant downward trend, the T_H2conv gene module score significantly decreased during OIT, suggesting phenotypic suppression beyond changes in frequency. Differentially expressed gene (DEG) analysis further revealed reduced expression of the type 2 cytokine interleukin-9 (IL-9) and the proliferation cytokine interleukin-2 in T_H2conv-like cells during OIT in the active but not in the placebo group (Fig. 2d,e). At the individual level, T_H2 cytokine module scores (IL4, IL5, IL9, IL13) in T_H2conv-like cells were significantly reduced during OIT (Fig. 2f; P = 0.006) but not in the placebo group. While IL-9 expression also decreased in T_H2a-like cells during OIT (Fig. 2d,e), their overall T_H2 cytokine module score decrease did not reach significance (Fig. 2g). However, when considering all pr T_eff memory cells, the T_H2 cytokine module score significantly decreased during OIT (Fig. 2h; P = 0.0011), supporting suppression of T_H2-related phenotypes beyond the decrease of cell numbers in T_H2 cell subsets.

To track pr CD4⁺ T cell clonal frequency changes more broadly, we performed bulk TCRβ sequencing (TCRβ-seq) on PBMCs from the same participants and time points. The summed frequency of BL pr T_eff memory CD4⁺ clones significantly decreased from BL/week 0 to week 104 in active but not placebo participants (Extended Data Fig. 4a; P = 0.011). BL T_H2a and T_H2conv clones also declined during active OIT, which is consistent with their role in allergy pathogenesis and treatment response (Extended Data Fig. 4b). In contrast, BL T_H1 CTL-like clones did not significantly expand in the bulk data at week 104, suggesting that the increased T_H1 CTL cell frequency detected in single-cell data reflects either newly generated clones during OIT or potential expansion of small BL clones below detection (Extended Data Fig. 4b).

To test another key question, whether OIT induces phenotypic shifts in T cell clones, we analyzed clones (defined by identical TRAV/TRAJ/TRBV/TRBD/TRBJ and CDR3α and CDR3β sequences) present at both BL and week 104. We detected 738 and 946 clones in active and placebo group across time points, with most clones maintaining consistent phenotypes. However, compared to the placebo group, fewer T_H2a-like clones persisted as T_H2a, with more shifting to T_H1 CTL-like or T_H1-like and T_H17-like states in active participants (Extended Data Fig. 4c,d). Analysis of BL T_H2a-like clones still detectable at week 104 in active participants confirmed this shift, with decreased T_H2a module scores (Fig. 2i, left; P = 0.0034) and increased T_H1 CTL module scores (Fig. 2i, right; P = 0.044), suggesting a transcriptional shift from a T_H2a-like phenotype toward a T_H1 CTL-like state induced by OIT. In contrast, BL T_H2a-like clones in the placebo participants showed increased T_H2a module scores (Fig. 2j, left; P = 0.0053) and no change in T_H1 CTL module scores (Fig. 2j, right; P = 0.65).

Together, these data indicate that OIT attenuates pathogenic T_H2a-like clonotypes and suppresses T_H2-associated phenotypes in both T_H2a-like and T_H2conv-like cells, while promoting T_H1 CTL cell frequency, T_H1-associated phenotypes and also potentially inducing phenotype transitions from T_H2a to T_H1 cytotoxic states.

Peanut OIT leads to T_H2a-like cell clonal contraction and T_H1 CTL-like cell expansion

In addition to phenotype analysis, we also examined TCR clonal expansion, defining expanded clones as clonotypes presented in more than two cells in a sample (Fig. 3a). Among active participants, the proportion of expanded T_H2a-like cells significantly declined during OIT, both relative to total pr T_eff memory T cells (Fig. 3b; P = 0.0083) and within the T_H2a-like population (Fig. 3c; P = 0.025). In contrast, clonally expanded T_H1 CTL-like cells increased over time (Fig. 3b, P = 0.076; Fig. 3c, P = 0.044). These results indicate that peanut OIT drives T_H2a-like clonal contraction alongside T_H1 CTL-like expansion, suggesting that the relative increase in T_H1 cytotoxic phenotypes is not solely the arithmetic consequence of decreases in the proportion of T_H2 phenotype cells.

Subtle phenotypic changes in pr T_reg cells during OIT

T_reg cells are important in immune tolerance, but we did not observe significant frequency or phenotype changes in pr T_reg subclusters during OIT (Extended Data Fig. 5a–d). DEG analysis of pr T_reg cells from active participants between week 104 and BL revealed only two significantly altered genes: CXCR3 (average log₂(fold change) = 0.406, P_adj = 0.044) was upregulated, and IER3 (average log₂(fold change) = −0.468, P_adj = 0.002) was downregulated (Extended Data Fig. 5e). Notably, CXCR3 expression was also higher in the active group than in the placebo group at week 104 (average log₂(fold change) = −0.539, P_adj = 6.04 × 10⁻⁶) (Extended Data Fig. 5f). Additionally, the frequency of BL pr T_reg clones in the total PBMC repertoire remained stable across all time points (Extended Data Fig. 4a). These findings align with the hypothesis that CD137⁺ pr T_reg cells are not primarily affected by OIT.

Distinct pr T_eff memory CD4⁺ cells in SU and DS individuals

Next, we investigated the phenotypes associated with SU or DS outcomes in pr T_eff memory CD4⁺ cells. While subcluster frequencies did not significantly differ between SU and DS at the individual level, we observed distinctive gene expression profiles between SU and DS participants at BL, week 104 and week 117 (Fig. 4a,b). Specifically, SU participants had lower T_H2conv-like module scores at BL, lower T_fh-like module scores at week 117, higher T_H22-like and T_H17-like module scores at BL, and higher T_H1 CTL-like module scores across all time points.

DEG analysis on pr T_eff memory CD4⁺ cells between SU and DS participants at BL, week 104 and week 117 (Fig. 4c–e) further revealed that the key feature genes related to cytotoxicity phenotype (for example, PRF1, GNLY, NKG7, GZMB, CST7) were higher in the SU than in the DS group at BL and week 117; T_H2-related genes (for example, IL13, IL4, IL5) were higher in the SU than in the DS group at BL. Gene set variation analysis (GSVA)²⁰ confirmed upregulation of pathways related to cytotoxicity signaling in pr T_eff memory CD4⁺ cells in SU participants at all three time points, while pathways related to asthma were lower in SU participants at BL compared to DS participants (Fig. 4f–h). To evaluate whether differences in gene expression between SU and DS participants differed between OIT time points, we carried out correlation analysis of gene expression effect sizes (average log₂(fold change)) between SU and DS participants for each pair of time points. This analysis revealed that SU versus DS differences in gene expression were more highly correlated between BL and week 117 (R = 0.62) than between BL and week 104 (R = 0.46) or between week 104 and week 117 (R = 0.50) (Extended Data Fig. 6), suggesting that gene expression differences at week 104 after continuous allergen exposure are of limited duration.

TCR clonal analysis revealed higher clonal expansion in the T_H1 CTL-like subcluster in pooled SU participant data compared to pooled DS participant data at week 104 and week 117, and in the T_H2a subcluster in the DS group than in the SU group at BL (Fig. 4i). In the individual participant-level analysis, the frequency of expanded pr T_H2a-like cells within the T_H2a cluster or among pr T_eff memory cells was lower in SU compared to DS participants at BL (Fig. 4j, P = 0.054; Fig. 4k, P = 0.054), while the frequency of expanded pr T_H1 CTL-like cells within the pr T_H1 CTL cluster or among pr T_eff memory cells showed a nonsignificant trend toward higher values in SU compared to DS participants at week 117 (Fig. 4j, P = 0.37; Fig. 4k, P = 0.17).

In SU participants, T_H2a-like clonotypes persisting from BL to week 104 showed a significant decrease in T_H2a module score (Fig. 4l; P = 0.04) and a trend toward increased T_H1 CTL module score (Fig. 4l; P = 0.052), with no changes in DS participants (Fig. 4m). DEG analysis of T_H2a-like cells between SU and DS participants revealed no changes at week 104, but at BL and week 117. SU cells upregulated cytotoxic/T_H1 CTL-related genes (for example, NKG7, GNLY, PRF1, TBX21), while DS cells expressed higher T_H2 cytokines (for example, IL4, IL5, IL13) (Extended Data Fig. 7a,b). DEG patterns between SU and DS mirrored differences between the T_H1 CTL-like and T_H2a-like subsets (Extended Data Fig. 7c,d), indicating a phenotypic shift rather than a difference in subset abundance. Supporting this, we found that the average T_H1 CTL module score in T_H2a-like cells was consistently higher in SU participants, reaching statistical significance at week 117 at the individual level (Extended Data Fig. 7e; P = 0.022) and trending upward over time in SU but not in DS participants. These results suggest that a greater potential for T_H2a-like cells to adopt a T_H1 CTL-like phenotype may contribute to the achievement of SU.

Lower BL T_H2 gene signatures are associated with higher likelihood of SU outcome

Given the higher prevalence of the T_H2-related phenotype in DS compared to SU participants, we calculated the average module scores of T_H2-related cytokine gene sets and Kyoto Encyclopedia of Genes and Genomes asthma pathway gene sets (CD40, CD40LG, FCER1G, IL13, IL3, IL4, IL5, TNF, shown in Fig. 5a) in pr T_eff cells for each participant at BL, week 104 and week 117. Both the T_H2 cytokine and asthma pathway gene module were relatively elevated in DS participants at BL (Fig. 5b; P = 0.073 and P = 0.053, respectively). These features declined more noticeably in DS participants over time, potentially because of their initially higher levels, and was more comparable between SU and DS participants by week 104, with a modest trend toward remaining slightly higher in the DS group at week 117. Receiver operating characteristic (ROC) analysis showed that the T_H2 cytokine module score at BL had the highest potential for distinguishing SU from DS (Fig. 5c; area under the curve (AUC) = 0.7959, P = 0.0639) and was significantly correlated with the final clinical outcome (peanut cumulative tolerated dose (CTD) during food challenge at week 117) (Fig. 5d; R = 0.57, P = 0.033). Similarly, the asthma pathway module score at BL had the highest potential for distinguishing SU from DS after OIT (Fig. 5e; AUC = 0.8163, P = 0.0476) and was significantly correlated with the final clinical outcome (Fig. 5f; R = 0.68, P = 0.007). Together, these results highlight that BL T_H2-related activity is associated with SU outcomes.

Higher cytotoxicity gene signatures after OIT are linked to higher likelihood of SU outcome

In addition to the T_H2-related phenotype, we also found that pr T_H1 cytotoxicity-related features that changed during OIT could also contribute to clinical outcome. We calculated the average module scores of CD4 cytotoxicity-related genes (GZMB, PRF1, GNLY, KLRG1, NKG7, CST7, GZMH, CX3CR1; shown in Fig. 6a)²¹ and pr T_H1 CTL feature genes in pr T_eff memory cells for each participant at BL, week 104 and week 117, and found that the CD4 cytotoxicity module showed a clearer upward trend over time in SU participants, ultimately reaching significance at week 117 (Fig. 6b, left; P = 0.017). The pr T_H1 CTL feature gene module score showed a consistent trend of higher values in SU compared to DS participants across all time points; it most closely approached significance at week 117 (Fig. 5b, right; P = 0.053). These findings suggest that immune dynamics over time are relevant to the outcome. In addition, the ROC analyses showed the CD4 cytotoxicity module score at week 117 as having the highest potential for distinguishing SU from DS (Fig. 6c; AUC = 0.8776, P = 0.0181) and being significantly correlated with clinical outcome (Fig. 6d; R = −0.59, P = 0.025). Similarly, the T_H1 CTL feature gene module score at week 117 had the highest potential for distinguishing SU from DS after OIT (Fig. 6e; AUC = 0.8163, P = 0.0476) and was significantly correlated with the final clinical outcome (Fig. 6f; R = −0.57, P = 0.033). Together, these results highlight that the cytotoxicity-related phenotype after OIT is associated with SU outcomes.

Higher CD39 expression on pr T_reg cells is associated with SU

Although the subcluster frequency and cluster feature gene module score of pr T_reg cells were not significantly different between SU and DS (Extended Data Fig. 8a,b), there were immune-related DEGs in pr T_reg cells between SU and DS at week 117, week 104 and BL (Fig. 7a). ENTPD1 (encoding CD39), which has been primarily described as a T_reg marker^22,23 and is crucial in terms of immunosuppressive functions^22,23, was highly enriched in pr T_reg than in pr T_eff (Fig. 7b) cells. ENTPD1 (CD39) expression was significantly higher in pr T_reg cells in SU participants than in DS participants at BL and week 117 (Fig. 7c; average log₂(fold change) = 1.005, P_adj = 5.75 × 10⁻¹² and average log₂(fold change) = 1.84, P_adj = 2.38 × 10⁻²², respectively). Both the RNA and protein levels of ENTPD1 (CD39) in pr T_reg cells at the individual level at week 117 of the OIT could distinguish SU and DS participants (Fig. 7d; AUC = 0.9388 and AUC = 0.9592). The frequency of ENTPD1-expressing cells and CD39⁺ cells (Supplementary Fig. 3) among pr T_reg cells was significantly higher in SU than DS participants at week 117 (Fig. 7e,f; P = 0.011 and P = 0.0072). In addition, both the RNA and protein levels of ENTPD1 (CD39) at week 117 in pr T_reg cells was significantly correlated with the outcome of peanut CTD during the week 117 food challenge (Fig. 7g,h; R = −0.63, P = 0.015; R = −0.76, P = 0.0015). To explore potential suppressive mechanisms of CD39⁺ pr T_reg cells, we compared their gene expression with CD39⁻ pr T_reg cells (Fig. 7i). CD39⁺ pr T_reg cells exhibited elevated expression of LGALS1 (galectin-1), IL3RA (interleukin-3Rα) and CD27, supporting T_reg cell stability, survival and suppressive function^{24,25,26,27,28,29}, while LIF and LAG3, which are involved in interleukin-10-mediated and STAT3/RORγt-mediated suppression^30,31, were lower, suggesting it may rely on alternative suppressive pathways.

Pr clones were detected in the TCRβ repertoires from the gastrointestinal tract

To assess mucosal T cell changes, we performed bulk TCRβ sequencing on gastrointestinal (GI) biopsies from POISED participants at week 0 (n = 7) and week 104 (n = 4) (Supplementary Table 2) and identified 766 unique clones (2,471 instances) matching pr T cells from the single-cell data (Methods), broadly distributed across the upper GI tract (Supplementary Fig. 4a–d). Overall pr clone frequencies were stable, but pr T_reg cell clones trended upward in the proximal esophagus, stomach and duodenum, and T_H1 CTL-like clones in the duodenum (Supplementary Fig. 4e,f). Clones with a T_H2a-like phenotype trended upward in the stomach, while T_H2conv-like clones trended downward (Supplementary Fig. 4f), suggesting distinct dynamics in GI tissue compared to blood, or different phenotypes for the clones in the GI tract.

Taken together, these findings highlight coordinated alterations in pr CD4⁺ T cell populations during OIT. A summary of the key immune shifts associated with SU is shown in Fig. 8.