3D chromatin reprogramming primes human memory T_H2 cells for rapid recall and pathogenic dysfunction

Abstract

Memory T cells provide long-lasting defense responses through their ability to rapidly reactivate, but how they efficiently “recall” an inflammatory transcriptional program remains unclear. Here, we show that human CD4⁺ memory T helper 2 (T_H2) cells carry a chromatin landscape synergistically reprogrammed at both one-dimensional (1D) and 3D levels to accommodate recall responses, which is absent in naive T cells. In memory T_H2 cells, recall genes were epigenetically primed through the maintenance of transcription-permissive chromatin at distal (super)enhancers organized in long-range 3D chromatin hubs. Precise transcriptional control of key recall genes occurred inside dedicated topologically associating domains (“memory TADs”), in which activation-associated promoter-enhancer interactions were preformed and exploited by AP-1 transcription factors to promote rapid transcriptional induction. Resting memory T_H2 cells from patients with asthma showed premature activation of primed recall circuits, linking aberrant transcriptional control of recall responses to chronic inflammation. Together, our results implicate stable multiscale reprogramming of chromatin organization as a key mechanism underlying immunological memory and dysfunction in T cells.

Introduction

T lymphocytes use a near-infinite repertoire of antigen-specific receptors to generate adaptive immune responses against pathogens and malignant cells (1). CD4⁺ T helper (T_H) cell subsets orchestrate adaptive immunity through the production of specialized cytokines (2): T_H1 cells produce interferon-γ (IFN-γ) to combat intracellular bacteria and viruses; T_H2 cells secrete interleukin-4 (IL-4), IL-5, IL-9, and IL-13 to expel parasites; and T_H17 cells synthesize IL-17 to defend against extracellular bacteria or fungi. During T cell responses, a fraction of activated cells adopt a memory phenotype and return to quiescence. These long-lived memory T cells retain an irreversible molecular imprint that enables them to mount a secondary recall response to the same antigen faster and greater in magnitude than the primary response of a naive cell (3). Memory T cells thus provide long-lasting immunological protection, forming the foundation for vaccination strategies (4). Moreover, pathogenic dysfunction of memory T cells has been implicated in major diseases, including autoimmunity [T_H17 cells (5)] and asthma [T_H2 cells (6, 7)].

Functionally distinct cellular states and phenotypes emerge from specific gene expression programs, which are controlled at the epigenomic level. Epigenomes consist of one-dimensional (1D) features such as chromatin accessibility and posttranslational histone modifications that either facilitate or impede access of DNA binding transcription factors (TFs) to gene regulatory elements such as promoters and enhancers (8, 9). Epigenomes adopt a nonrandom 3D organization of chromatin that is tightly linked to the control of gene expression (10–14). Examples include clustering of genomic regions with similar levels of transcriptional activity in chromosome compartments (i.e., active genes tend to reside in the so-called A compartment, repressed genes in the B compartment), the formation of spatially insulated genomic neighborhoods called topologically associating domains (TADs), and enhancer-promoter interactions preferentially occurring within these TADs.

Human T cell activation involves coordinated changes in 1D and 3D chromatin structure as well as gene expression (15–22), resulting in phenotypic changes characterized by metabolic reprogramming and inflammatory cytokine production. Memory T cells have a distinct transcriptome and epigenome as compared with their naive counterparts (15, 16, 23–29), indicating that the molecular basis of the memory T cell response is anchored in mechanisms of transcriptional control (30). Whereas priming of the chromatin landscape in resting memory cells has been implicated in facilitating rapid recall (27, 28, 31–33), we still lack detailed insight into the epigenomic circuitry that endows human memory T cells with their ability to rapidly activate specific sets of inflammatory genes. Furthermore, the role of 3D genome conformation for T cell memory has not been systematically addressed (22). Here, we applied a multiscale 1D/3D epigenomic approach to comprehensively dissect the molecular program driving rapid recall in primary human memory T_H2 cells. Our analyses reveal functionally distinct epigenomic priming mechanisms underlying rapid recall, which are stably stored in memory T cells at the level of transcription, chromatin accessibility, and 3D genome folding. We also find that this distinct reprogramming of 1D/3D chromatin organization in memory T_H cells is exploited by AP-1 TFs for efficient transcriptional reactivation and is dysregulated in resting memory T_H2 cells from patients with asthma.

Results

An experimental system to identify epigenomic drivers of T cell memory

To systematically identify epigenomic mechanisms underlying rapid recall in memory T cells, we interrogated gene expression, chromatin structure, and 3D genome conformation dynamics during early memory T cell activation (Fig. 1, A and B). Contrasting memory and naive T cells allows for separation of events that are specific to memory T cell activation from those that are shared with naive cells that lack memory of a prior stimulation. We focused on naive CD4⁺ T cells (CD4⁺CD127⁺CD25⁻CD45RA⁺) and memory T_H2 cells (CD4⁺CD127⁺CD25⁻CD45RA⁻CCR6^{− -} CXCR3⁻CXCR5⁻CCR4⁺) isolated using fluorescence-activated cell sorting (FACS) from peripheral blood samples of healthy donors (Fig. 1C). T cells were either analyzed as resting, unstimulated cells (“−”) or stimulated in vitro for 24 hours using bead-linked anti-CD3/CD28 antibodies (“+”) to activate T cell receptor (TCR) signaling and costimulatory pathways.

Fig. 1. Identification of distinct transcriptional modules linked to rapid recall in human CD4⁺ memory T cells.

(A) Potential epigenomic drivers of rapid recall investigated in this study. (B) Experimental system used to investigate the epigenomic underpinnings of rapid transcriptional recall in human memory (“mem.”) T_H2 cells. Cells were stimulated with bead-linked anti-CD3 and anti-CD28 antibodies to mimic generic TCR and CD28 co-receptor activation. (C) Flow cytometry gating strategy used to isolate human naive CD4⁺ T cells and memory T_H2 cells from peripheral blood. (D) mRNA expression levels of signature T_H cytokine genes in the indicated resting (−) and stimulated (+) cells. Data represent measurements from three independent biological replicates. (E) PCA of scaled gene expression values for all genes expressed in at least one sample type (n = 12,284 genes). (F) Heatmap showing mRNA expression levels for genes within three recall-associated gene modules (example genes are indicated). Boxplots on the right indicate averaged expression values (z-score scaled) for each module in each of the four indicated T cell samples. P values: Kruskal-Wallis test corrected for multiple testing. ns, not significant. (G) Pathway enrichment analysis of genes belonging to the three modules. GTPases, guanosine triphosphatases; ER, endoplasmic reticulum; JAK, Janus kinase. (H) Network of biological pathways associated with recall-associated module genes. Nodes represent individual pathways (P < 0.05) that are connected and clustered on the basis of functional similarity. Nodes are represented as pie charts: The size of a pie is proportional to the total number of hits; pie charts are color-coded on the basis of the module origin of the associated genes. Circos plot indicates connections between the three modules at the pathway level, with gray lines linking genes that belong to the same enriched pathway. (I) mRNA gene expression (top) and protein levels (as measured by flow cytometry, bottom) of selected module members.

Gene expression was measured by RNA-sequencing (RNA-Seq), and genome-wide 3D chromosome conformation was assessed using in situ Hi-C; chromatin accessibility information was obtained from available assay for transposase-accessible chromatin using sequencing (ATAC-Seq) data from precisely matched T cell populations (15). Examining expression levels of signature T_H cell cytokines revealed a strong and reproducible induction of canonical T_H2 cytokine genes (e.g., IL4 and IL5) in memory T_H2 cells, whereas signature T_H1/T_H17 genes (e.g., IFNG and IL17A) remained silent (Fig. 1D and fig. S1A). Conversely, activated naive CD4⁺ T cells only produced IFNG mRNA and showed virtually undetectable T_H2 cytokine gene expression (Fig. 1D and fig. S1A), confirming that our experimental setup accurately and reproducibly captures the recall ability of memory T_H2 cells.

Transcriptional modules linked to rapid recall in CD4⁺ memory T cells

Differential gene expression analysis revealed widespread transcriptional changes after 24 hours of stimulation, including well-known markers of T cell activation such as IL2RA (encoding the high-affinity IL-2 receptor subunit CD25) and MYC (fig. S1B). Although naive CD4⁺ T cells lack any recall ability, they showed a quantitatively and qualitatively similar transcriptional response to stimulation as memory T_H2 cells, i.e., 7997 and 7368 differentially expressed genes, respectively, of which ∼80% were shared between naive and memory cells (Fig. 1E and fig. S1, C and D). Thus, large-scale transcriptional rewiring shortly after activation is highly comparable in naive and memory CD4⁺ T cells, suggesting that the transcriptional program associated with memory recall consists of a restricted number of genes.

Previous studies have detected hundreds of differentially expressed genes between naive and memory T cells (30). They, however, did not investigate recall-associated transcriptional activation, which we define here as stimulation-induced genes that show increased expression in memory T cells compared with naive T cells before and/or upon early activation (“recall genes”). To identify recall genes, we used several clustering and filtering strategies (see Methods and fig. S1, E and F). Of the 742 differentially expressed genes between resting naive CD4⁺ T cells and resting memory T_H2 cells (fig. S1, C and D), we extracted 150 stimulation-induced genes that were already up-regulated in resting memory T_H2 versus resting naive CD4⁺ T cells and that were further induced upon T cell activation (recall module I; Fig. 1F and fig. S1E). Pathway enrichment analyses revealed that these genes were functionally involved in glycolytic metabolism (e.g., GAPDH and PGK1), lymphocyte activation (e.g., CD58 and IL2), apical protein localization (e.g., RAB8B), and cytoskeletal organization (e.g., ACTB and RHOG) (Fig. 1G). Earlier work provided evidence that memory T cells exhibit elevated glycolysis and increased cortical actin content to facilitate rapid recall (34, 35), validating our experimental approach. In addition, we identified several genes encoding molecules not previously implicated in T_H cell recall, including the long noncoding RNA CYTOR, the C-type lectin-like receptor CLECL1, the histone 3 lysine 27 (H3K27) methyltransferase EZH2, and several AP-1 family TFs (i.e., BATF, FOSL2, and CREM).

Clustering of all stimulation-induced genes in both cell types revealed two additional modules of recall genes that showed similar expression in both resting T cell subsets but are poised for more rapid reactivation in memory T_H2 cells (recall modules II and III; Fig. 1F and fig. S1F). Module II consists of 230 genes strongly induced only in activated memory T_H2 cells, whereas module III genes (n = 504) were up-regulated in both cell types but substantially higher in memory T_H2 cells (Fig. 1F). Module II genes showed higher overall inducibility than module III genes (average: 3.7 versus 2.4 log₂ fold change, respectively) (fig. S1G). Recall module II genes were poorly expressed in resting memory cells [median level of 1.1 reads per kilobase per million (RPKM)] and encoded all canonical T_H2 cytokines as well as critical immunomodulatory molecules (e.g., ICOS and CTLA4). Module III genes were enriched for more diverse yet highly relevant biological functions, which included cell cycle control (e.g., CDK5), lipid metabolism (e.g., ACOX1), and lymphocyte activation (e.g., IL2RA, IRF4, and SLAMF1) (Fig. 1, F and G). Network analysis of enriched biological pathways revealed that the three recall gene modules (data file S1) were to a large extent functionally interconnected (Fig. 1H). Flow cytometry analyses confirmed similar dynamics of selected factors from the three modules at the protein level (Fig. 1I). All three recall gene modules showed similar expression dynamics in resting and activated memory T_H1 or T_H17 cells (fig. S1H) (15), indicating that they are shared across CD4⁺ memory T cell subsets. Applying an additional set of stringent filtering criteria identified recall genes with T_H2-specific activation dynamics (n = 27, see Methods), but many more showed stronger induction in all three T_H subsets as compared with naive cells (“T_H-shared” recall genes, n = 199) (fig. S1I).

Last, we also defined genes associated with transcriptional silencing upon T cell stimulation and that show decreased expression in memory T_H2 cells compared with naive T cells before and/or upon early activation. Four clusters of such “inverse recall” genes emerged (n = 281 genes in total), all showing lower expression already at baseline in memory T_H2 cells (fig. S2, A and B). These genes were enriched in biological pathways relevant for T cell biology, including known regulators of T cell stemness, homeostatic proliferation, and metabolic reprogramming such as BACH2, TCF7, and LEF1 (fig. S2C).

Together, these results show that resting memory CD4⁺ T_H2 cells maintain increased expression of a highly restricted set of stimulation-induced genes (recall module I). These do not include prototypical regulators of T_H2 cell effector function, which are instead organized in two larger recall gene modules (II and III) that show no differential expression at baseline but are more robustly induced upon TCR stimulation in memory T_H cells as compared with naive T_H cells. In addition, a small number of stimulation-repressed genes encoding for regulators of T cell homeostasis are expressed at lower levels in memory T_H2 cells, many of which are silenced more efficiently in memory cells upon activation.

Extensive epigenomic priming of effector loci in resting memory T_H cells

Chromatin accessibility landscapes analyzed using ATAC-Seq can be exploited to identify potential regulators of transcriptional programs underlying specific immune cell states (15, 24, 36). In line with transcriptional changes, both naive CD4⁺ T cells and memory T_H1, T_H2, and T_H17 cells showed similar capacities for rapid genome-wide expansion of chromatin accessibility (∼2-fold increase in total ATAC-Seq peaks; fig. S3, A and B). Principal components analysis (PCA) not only confirmed the major effect of stimulation on the epigenome but also separated resting naive from memory T cells (Fig. 2A). Clustering of differentially accessible sites across all cell states (n = 32,847 peaks, see Methods) revealed various patterns of ATAC-Seq signal dynamics, including several clusters of naive-specific peaks (Fig. 2B and fig. S3, C and D).

Fig. 2. Extensive priming of chromatin accessibility in resting human CD4⁺ memory T cells.

(A) PCA of normalized accessibility values for all reproducibly detected ATAC-Seq peaks (present in at least one sample type, n = 96,102 peaks). (B) Heatmap showing normalized ATAC-Seq signals of indicated peak clusters in the indicated resting (−) and stimulated (+) cells. Line graph indicates averaged signal across all peaks within the three memory-specific clusters. (C) Pie chart depicting distances of recall-associated ATAC-Seq cluster peaks in human T_H2 cells to the nearest transcription start site (TSS). Red line indicates proportion of peaks outside of promoter regions. (D) Pathway enrichment analysis of genes closest to ATAC-Seq peaks from the indicated clusters. (E) Overlap between genes linked to recall-associated ATAC-Seq peaks (“Primed genes”) and genes from the three recall-associated transcriptional modules (see Fig. 1). (F) Genome browser screenshot showing ATAC-Seq profiles at the primed IL3 locus in resting (−) and stimulated (+) naive or memory T_H2 cells. Boxplot depicts IL3 mRNA levels in samples from three independent donors. (G) Heatmaps with RNA-Seq (left) and ATAC-Seq (right) signals across the indicated sample types for epigenomically primed genes with a statistically significant expression advantage after stimulation. Example genes are indicated.

We detected several clusters of sites that either were uniquely present in memory T_H cells—irrespective of stimulation status— or gained chromatin accessibility upon stimulation and were already accessible in resting memory T_H2 cells. We refer to these clusters shared across all memory T_H subsets (memory-specific) or specific to T_H2 cells as “recall-associated” accessible chromatin regions (n = 3025 peaks in total) (Fig. 2B and fig. S3, C and D). Recall-associated accessible chromatin was predominantly (88.9%) localized to distal regulatory elements and linked to genes enriched for effector cell–associated functions such as lymphocyte activation, cytokine production, and cell adhesion (Fig. 2, C and D, and fig. S3E). A substantial fraction of genes from the three recall modules (22 to 44%; P < 0.0001, Fisher’s exact test) was associated with a stimulation-induced ATAC-Seq peak already present in resting T_H memory cells, as exemplified by IL3 (Fig. 2, E and F). Similar results were obtained when using T_H2-specific and T_H-shared recall gene definitions (27 and 78%, respectively; P < 0.0001) (fig. S3F). A small group of T_H-shared recall genes with critical roles in memory T cell activation (e.g., IRF4, ICOS, and MAF) showed epigenomic priming in both a T_H-shared and a T_H2-specific fashion (fig. S3F), indicating that memory T_H cells may use a combination of shared and subset-specific regulatory elements to facilitate rapid gene induction. Using additional stringent filtering criteria (see Methods), we identified 130 genes that were primed at the level of chromatin accessibility in resting memory cells and had a significant expression advantage specifically after stimulation. These included genes encoding key inflammatory cytokines (e.g., IL3, IL9, and IL13), costimulatory molecules (e.g., ICOS, OX40, and GITR), the high-affinity IL-2 receptor subunit (IL2RA), and the MAF AP-1 family TF (Fig. 2G). Around 46% (P < 0.0001, Fisher’s exact test) of stimulation-repressed inverse recall genes could be linked to ATAC-Seq peaks that were uniquely present in naive T_H cells (fig. S2D; using the “naive-specific” clusters from fig. S3C).

Specific chromatin features have previously been associated with epigenomically primed (or “poised”) loci (37, 38), including H3K4 dimethylation (H3K4Me2), H3K27Me3, RNA polymerase II (RNAPII), and the H2A.Z histone variant. Conversely, these regions are generally low in H3K27Ac, a hallmark of transcriptional activity (37). We profiled H3K4Me2 in resting naive or memory CD4⁺ T cells from three independent donors using ChIPmentation (39) and obtained H3K27Me3, RNAPII, H2A.Z, and H3K27Ac chromatin immunoprecipitation followed by high-throughput sequencing (ChIP-Seq) data from published studies (see Methods). In line with bona fide priming, stimulation-induced regions accessible in resting memory T_H cells showed low levels of H3K27Ac [obtained from total T_H memory cells (40)], which increased upon activation (Fig. 3A and fig. S4, A and B). In contrast, H3K27Ac levels at naive-specific ATAC-Seq peaks were not induced after stimulation of memory T_H cells (Fig. 3A). Recall-associated accessible chromatin showed no enrichment for H3K27Me3 and was marked by H2A.Z in both naive and memory CD4⁺ T cells (fig. S4, C and D). However, we observed strong and specific enrichment for H3K4Me2 and RNAPII at these sites in memory cell subsets compared with naive cells (Fig. 3, B and C, and fig. S4E).

Fig. 3. Specific TFs govern the recall-associated chromatin landscape in human CD4⁺ memory T cells.

(A) Fold changes (log₂ scale) in H3K27Ac levels upon TCR stimulation of total CD4⁺ memory T cells for naive-specific (n = 652) or recall-associated (mem-sp.1 cluster, n = 1173) ATAC-Seq peaks. (B) Heatmap representations of averaged H3K4Me2 (top) or RNA polymerase II (RNAPII, bottom) ChIP-Seq signal at the indicated recall-associated ATAC-Seq peak clusters in different resting T cell subsets. (C) Genome browser screenshots showing H3K4Me2, ATAC, and RNAPII signals at selected recall-associated ATAC-Seq peaks. Numbers on top indicate distance to TSS in kilobase. (D) Top three overrepresented TF binding motifs based on P values for the indicated recall-associated ATAC-Seq peak clusters. Percentages indicate proportion of sites in which motif was detected [versus background (vs. bg.)]. (E) Scaled mRNA levels (z-score) of motif-associated TF genes significantly up-regulated in memory T_H2 cells compared with naive CD4⁺ T cells, either at resting state (*P < 0.05) or after activation (^#P < 0.05). Boxplots depict RPKM values of selected genes from the heatmap. (F) TF occupancy patterns at all recall-associated ATAC-Seq peaks (n = 3025 peaks) in stimulated CD4⁺ memory T cells. Inset bar graph shows total fraction of peaks bound the indicated TF. (G) Expression analysis of recall (IL4 and IL5) and general activation (CD40LG and HPRT) genes in resting (−) and stimulated (+) memory T_H2 cells with or without T-5224 AP-1 inhibitor before treatment (n = 3 independent biological replicates). (H) Flow cytometry analysis of MAF protein levels in resting (−) and stimulated (+) memory T_H2 cells. Bottom two tracks show cells in which MAF was disrupted using CRISPR-Cas9 with two different crRNAs, preventing MAF induction. (I) Expression analysis of recall (IL4 and IL9) and general activation (HPRT) genes in resting (−) and stimulated (+) memory T_H2 cells pretreated with Cas9 and a nontargeting control (ctrl) or MAF crRNA. (J) Proportion of indicated superenhancer (SE) categories that overlap with recall-associated ATAC-Seq peaks, including P values indicating statistical significance of the overlaps shown. (K) Heatmap showing mRNA expression levels for genes linked to memory T_H2-specific SEs (example genes are indicated). (L) Genome browser screenshot (top: H3K4Me2 ChIPmentation, bottom: ATAC-Seq) of the SE landscape at the CD28-CTLA4-ICOS locus. (M) Boxplots depicting gene expression dynamics of CD28, CTLA4, and ICOS. Statistical tests used: (A, G, and I) Mann-Whitney U test; (D and J) Fisher’s exact test; (E) DESeq2 Wald test.

We next conducted motif enrichment analysis to identify TFs associated with recall-associated chromatin priming in memory CD4⁺ T cells. Whereas naive-specific accessible chromatin was enriched for CTCF and TCF factor motifs (fig. S4F) (41), all three memory-specific clusters were strongly enriched for AP-1, RUNX, and ETS family motifs (Fig. 3D). The T_H2-specific cluster was uniquely enriched for GATA alongside RUNX and ETS motifs (Fig. 3D), in line with an important role for the T_H2 lineage–determining TF GATA3 (42) at these sites. RNA-Seq revealed that nine members from the AP-1, RUNX, or ETS TF families were up-regulated in resting or stimulated T_H2 cells compared with naive cells (Fig. 3E and fig. S4G), compatible with roles in maintaining recall-associated chromatin priming (e.g., BATF, FOSL2, and RUNX3) and/or in exploiting these sites once memory T_H2 cells are activated (e.g., BATF3, MAF, and ETV6). ChIP-Seq experiments (43, 44) revealed extensive binding of GATA3 and in particular AP-1 TFs (i.e., 50% of all sites) at recall-associated accessible chromatin sites in activated memory T_H cells (Fig. 3F). To test the functional relevance of AP-1 family TFs for recall, we treated memory T_H2 cells with a selective small-molecule inhibitor of AP-1 TF binding [T-5224 (45)] before stimulation. Treatment with T-5224 blunted transcriptional recall of selected recall genes (i.e., IL4 and IL5), whereas the effect on early induction of general T cell activation markers HPRT and CD40LG was limited (Fig. 3G). We next used CRISPR-Cas9 genome editing in resting memory T_H2 cells to mutate the coding sequence of the MAF gene, encoding an AP-1 family TF implicated in T_H2 specification (46). Ribonucleoprotein complexes of Cas9 and guide RNAs successfully abrogated MAF protein production (Fig. 3H), blunting transcriptional recall of several type 2 cytokine genes but not the early induction of general stimulation-responsive genes (Fig. 3I and fig. S4H).

In summary, resting memory CD4⁺ T cells carry a defined set of epigenomically primed gene regulatory elements that mark a restricted set of genes for rapid recall. Upon activation, these “recall enhancers” are targeted by lineage-determining (e.g., GATA3) and signal-responsive TFs (e.g., the AP-1 factor MAF) to enable robust induction of inflammatory gene expression.

Chromatin priming is linked to superenhancer modules

Superenhancers have emerged as clusters of multiple regulatory elements that ensure robust expression of key cell identity genes (9). We defined superenhancers in naive and memory T_H2 cells on the basis of H3K4Me2 abundance (47) to detect both active and primed superenhancer elements in resting T cells. H3K4Me2 superenhancer–associated genes in T_H2 memory cells were robustly expressed (>3-fold higher than the average expressed gene) and included signature T cell genes such as BCL11B, CD3E, and IL7R (fig. S4I). Compared with naive CD4⁺ T cells, resting T_H2 memory cells showed an expanded superenhancer landscape (1.8-fold more superenhancers; fig. S4I). Stringent filtering (>2-fold H3K4Me2 enrichment versus naive CD4⁺ T cells) yielded 83 T_H2 memory–specific superenhancers associated with 80 unique genes, which are involved in critical processes linked to memory T cell function. These included enhanced mitochondrial respiration (e.g., OGDH), migratory capacity (e.g., CCR4), activation (e.g., ICOS and TIGIT), and T_H2 cell identity (e.g., GATA3 and PRMD1) (fig. S4J).

More than 80% of T_H2-specific superenhancers colocalized with recall-associated ATAC-Seq peaks (versus 11% for naive-specific superenhancers; P < 0.0001, Fisher’s exact test), revealing wide-spread chromatin priming at superenhancers in resting memory T_H cells (Fig. 3J and fig. S4K). Many T_H2-specific superenhancer-associated genes showed recall-associated transcriptional dynamics (Fig. 3K and fig. S4L). This is illustrated by the CD28-CTLA4-ICOS locus encoding three surface receptors that control T cell activity (48). Several superenhancers already established in resting memory T_H2 cells, including a T_H2-specific one downstream of ICOS, show increased accessibility at stimulation-associated open chromatin sites specifically in resting T_H2 cells (Fig. 3L). In accordance with transcriptional priming in memory cells, TCR stimulation more efficiently induced CD28-CTLA4-ICOS expression in T_H2 cells (Fig. 3M). Collectively, these results indicate that primed superenhancers control the activation-inducible expression of a small set of recall genes that are critical for memory T cell function.

Memory T cells exhibit distinct patterns of chromosome compartmentalization

Changes in 3D genome organization are intimately connected to gene regulatory dynamics and can facilitate or hamper transcriptional activation (10). Hence, we reasoned that 3D genome folding may provide memory T cells with an additional priming mechanism for efficient recall. For this, genome topology was profiled at high-resolution (5 kb; using ∼1 billion valid interaction read pairs per condition) in resting and stimulated naive CD4⁺ and memory T_H2 cells using in situ Hi-C with excellent overall reproducibility (fig. S5A; see data file S2 for dataset statistics).

We next assigned negative scores to B compartment and positive scores to A compartment genomic regions at 10- to 100-kb resolution [compartment scores (C-scores); Fig. 4A]. Although separated A and B compartments were detected in all conditions (fig. S5B), the overall compartmentalization strength was reduced specifically in resting memory T_H2 cells, which was normalized upon stimulation (Fig. 4B and fig. S5, B and C). PCA showed that C-scores were highly reproducible between replicates and readily distinguished naive CD4⁺ from memory T_H2 cells, both before and after stimulation (Fig. 4C). Quantitative analysis of C-score dynamics at 10-kb resolution revealed compartmental repositioning of 21.1% of the genome (ΔC-score > 0.3) across experimental groups, with a small fraction (0.13% of the genome) demonstrating very large changes (ΔC-score > 1.0) (Fig. 4D). Although stimulation had the largest effects, nearly 9000 10-kb regions showed altered compartmentalization between resting naive and memory T_H2 cells (Fig. 4D). Most differences between naive CD4⁺ and memory T_H2 cells involved C-scores increasing within the B compartment (i.e., BB to B), in line with decompaction of highly repressed chromatin (Fig. 4E and fig. S5D) (49). In contrast, stimulation predominantly triggered increased C-scores within the transcriptionally permissive A compartment in both naive and memory T cells (Fig. 4E and fig. S5D). These results show that resting memory T cells have distinct genome compartmentalization compared with naive cells, including a weaker separation of A and B compartments and a specific decompaction of heterochromatic regions.

Fig. 4. Distinct patterns of chromosome compartmentalization in human memory T_H2 cells are linked to transcriptional priming.

(A) Schematic representation of nuclear compartments A and B, including the C-score distribution used to quantify compartment association of genomic regions (i.e., from −1 = strong B to 1 = strong A). (B) Compartment strength calculated as a ratio between homotypic (AA/BB) and heterotypic (AB/BA) interactions over increasing distances. (C) PCA of genome-wide C-score values (10-kb bins). (D) Quantification of dynamic C-score bins (10-kb resolution) showing a 0.3 or greater difference in at least one comparison (n, naive CD4⁺ T cells; −, resting; +, stimulated). Assignment of dynamic bins to individual comparisons is indicated in the right bar graph. (E) Directionality of C-score changes across dynamic bins within the indicated comparisons. (F) Heatmap representation of clustered scaled C-scores (left). The six clusters showing a recall-associated increase in C-score consist of 754 regions further detailed in the right heatmap (with absolute C-score values), including examples of associated key inflammatory genes. (G) Enrichment scores and P values (Fisher’s exact test) of recall-associated ATAC-Seq peaks (see Fig. 2 for definitions), T_H2-specific or naive-specific SEs, and GATA3 binding sites in memory T_H2 cells within the 754 dynamic C-score regions. (H) Meta-plot of long-range Hi-C interactions (2- to 10-Mb window) between recall-associated (cluster mem-sp.2) or naive-specific ATAC-Seq peaks. (I) Scaled C-scores (z score) of dynamic bins (ΔC-score > 0.1 in at least one pairwise comparison) at T_H2-specific (n = 220 bins), memory-specific (n = 1285 bins), and naive-specific (n = 414 bins) ATAC-Seq peak clusters. (J) Scaled C-scores (z-score) of dynamic bins at gene regulatory elements linked to recall-associated genes from recall module I (n = 654), the combined recall modules II + III (n = 2352; see Fig. 1F), and primed genes with an expression advantage (“primed ATAC,” n = 897; see Fig. 2G). P values: Kruskal-Wallis test corrected for multiple testing. (K) Genome browser screenshot (top) showing C-score dynamics at the MAF locus, with gene expression dynamics depicted below in boxplots.

We next focused on recall-associated changes in chromosome compartmentalization. Clustering analysis identified six groups of genomic regions (n = 6888 bins) that showed elevated C-scores upon stimulation and were already increased in resting memory T_H2 cells but not in resting naive T cells (Fig. 4F). Removal of singletons and stitching together consecutive bins (see Methods) yielded 754 regions of primed nuclear repositioning toward or more firmly within the A compartment (Fig. 4F). These regions were highly enriched for recall-associated open chromatin, T_H2 memory–specific superenhancers, GATA3 binding sites, and key inflammatory genes (P < 1 × 10⁻⁶, Fisher’s exact test) (Fig. 4, F and G, and data file S3).

We observed strong spatial colocalization of recall-associated ATAC-Seq peaks at the level of compartmentalization (interactions within a 2- to 10-Mb window)—a feature specific to memory T_H2 cells and to memory-associated accessible chromatin sites (Fig. 4H and fig. S5E). Visualizing C-score dynamics at recall-associated open chromatin confirmed priming at both 1D (chromatin accessibility) and 3D (compartmentalization) levels (Fig. 4I). Regulatory elements linked to (T_H2-specific) recall genes showed stronger A-compartment membership in resting memory T_H2 cells (Fig. 4J and fig. S5F). This was particularly notable at the MAF locus, which contains 30.5% of all ΔC-score > 1.0 bins. Here, combined 1D/3D chromatin priming in resting memory T_H2 cells is likely to underlie rapid transcriptional recall of MAF upon activation (Fig. 4K), which subsequently allows MAF to induce inflammatory gene expression (i.e., Fig. 3I). Conversely, stimulation-repressed inverse recall gene promoters exhibited weaker A compartment membership in both resting and activated memory T_H2 cells as compared with their naive counterparts (fig. S2E). Thus, priming of key recall genes in resting memory T cells is associated with spatial colocalization of these genes and their enhancers inside transcriptionally permissive nuclear compartments.

Topological domain and loop reorganization associated with memory T cell function

Base-pair resolution analysis of 3D genome architecture revealed that most gene regulatory interactions occur within TADs (50). Insulation by TAD borders can potentially modulate transcriptional control in the context of memory T cell recall, e.g., by restricting or facilitating promoter-enhancer communication. We determined genome-wide insulation scores (see Methods) and identified 3647 reproducible TAD borders, of which 2184 (60%) were common to all four groups (Fig. 5, A and B). PCA using insulation score values at TAD borders readily and reproducibly distinguished resting from activated T cells (PC1) and naive from memory T_H2 cells (PC2)—even when only common borders were used (Fig. 5C). Thus, memory T_H2 cells exhibit unique global spatial genome insulation compared with naive cells already in a resting state, which is dynamically altered upon activation.

Fig. 5. TAD dynamics linked to human memory T_H2 cell function.

(A) Hi-C interaction map of the MAF locus with insulation scores and TAD borders (defined as local minima) plotted below. (B) Proportions of common (invariant in all four sample groups) and dynamic (gained or lost in at least one sample group) TAD borders. (C) PCA of insulation score values at all (left) or only common (right) TAD borders. (D) Heatmap representation of clustered scaled insulation scores at TAD borders. Patterns associated with clusters are indicated on the left; border counts for each cluster are shown on the right (bar graph). (E) Average insulation scores at dynamic TAD borders from the indicated clusters in resting (−) or stimulated (+) naive CD4⁺ T cells and memory T_H2 cells. (F) Hi-C interaction maps of the GATA3 locus from the indicated sample groups. Signal subtraction (delta) plots are shown below to highlight differences in interaction signals (left: red = higher in T_H2⁻ versus naive⁻; right: red = higher in T_H2⁺ versus T_H2⁻). Locations of genes, recall-associated ATAC-Seq peaks, and T_H2-specific SEs are indicated at the bottom. (G) Top: heatmap depicting GATA3 mRNA levels. Middle/bottom: Hi-C signal subtraction plot (red = higher in T_H2⁻ versus naive⁻) as shown in (F) (left side) but with insulation score profiles and TAD border calls (n = 4) indicated below. Note the third dynamic TAD border that is only called in naive T cells. (H) Definition of “memory TADs” as those domains that contain four or more recall-associated ATAC-Seq peaks (7% of all TADs). (I) Pathway enrichment analysis of recall-associated genes located within memory TADs. (J) Hi-C signal subtraction plots (red = higher in T_H⁻ versus naive⁻) for selected memory TADs.

Most dynamic borders were associated with activation status, although others were only detected in specific cell types or states (fig. S6A). Quantitative clustering of insulation score patterns revealed that 70% of all TAD borders showed general activation–related dynamics, whereas 26% exhibited a pattern associated with a memory phenotype or recall-associated priming in resting memory T_H2 cells (Fig. 5, D and E, and fig. S6B). Most prevalent among the latter category are borders (n = 450) that are already weakened in resting memory T_H2 cells before further loss of insulation score upon activation. A prime example hereof is the GATA3 locus, which is located within a large and highly dynamic TAD (Fig. 5F). The GATA3 TAD contains a weaker internal border in naive cells that is dissolved in memory T_H2 cells, allowing for robust interaction of GATA3 with downstream regulatory regions (including a T_H2-specific superenhancer) and transcriptional up-regulation (Fig. 5, F and G). Stimulation focuses GATA3-enhancer interactions and strengthens the downstream TAD border (Fig. 5, F and G), possibly to increase GATA3 transcriptional robustness.

Specific clusters of dynamic borders showed enrichment or depletion for recall-associated accessible chromatin sites (fig. S6C). We noticed that a small number of TADs (n = 256; 7% of total) harbor ∼50% of all recall-associated ATAC-Seq peaks (Fig. 5H). Genes within these “memory TADs” are involved in T cell activation, including many genes from the three recall-associated transcriptional modules (n = 169 of 884; P = 0.0001, Fisher’s exact test) (Fig. 5I and data file S3). In addition, memory TADs were strongly enriched for dynamic C-score bins (enrichment ratio: 13.1; P < 0.0001, Fisher’s exact test). Increased intradomain interactions were already present in many memory TADs in resting memory T_H2 cells, revealing increased topological chromatin connectivity around key recall genes and nearby regulatory elements—irrespective of their activation status (i.e., the active CCR4 or inactive IL13 gene) (Fig. 5J and fig. S6D).

Gene regulatory elements can interact via focused 3D interactions often referred to as chromatin loops. Loop detection using Mustache (51) identified thousands of strong interaction foci in each condition (fig. S7A). We first focused on activation-associated loops already present in resting memory T_H2 cells as optimal candidates for priming for transcriptional recall, identifying hundreds of loops specifically associated with a memory phenotype and/or recall (Fig. 6A and fig. S7B). Anchor regions of these memory-associated loops were strongly enriched for regulatory elements, with >75% forming interactions between promoters and/or enhancers (Fig. 6B and fig. S7C). This is illustrated by the GATA3 memory TAD in which memory-specific loops connect GATA3 to primed (super)enhancers (Fig. 6C). Genes linked to memory-associated loops encoded regulators of cell adhesion, T cell activation, and the cell cycle, which showed a significant transcriptional activation advantage specifically upon memory recall (Fig. 6, D and E, and data file S3). Similar results were obtained when using a broader definition of memory-associated loops (i.e., including the 2198 and 3675 T_H2-specific loops) (fig. S7D). Regulatory connections involving these genes were formed in resting memory T_H2 cells without significant gene induction. This is exemplified by the IL9 locus, in which two memory-specific loops create an interaction domain in which IL9 is connected to downstream recall-associated regulatory elements (Fig. 6F). Upon activation, only memory T_H2 cells rapidly and markedly up-regulate IL9 expression (Fig. 6G).

Fig. 6. Chromatin loops prime genes for rapid recall in human memory T_H2 cells.

(A) Venn diagram of all loops identified in resting (−) or stimulated (+) naive CD4⁺ T cells and memory T_H2 cells. Meta-plots of Hi-C signal intensity averaged across all recall-associated loops (n = 308) in resting naive and memory T cell subsets are shown on the right. LS, loop strength (log₂ scale). (B) Combined pie chart and radar plot depicting the proportion of memory and recall-associated loops connecting enhancers (Enh./E) and/or promoters (Prom./P). The height of each pie segment indicates fold enrichment (denoted by the gray circles). (C) Hi-C interaction maps of the GATA3 locus in stimulated memory T_H2 cells. Locations of genes, recall-associated ATAC-Seq peaks, and T_H2-specific SEs are indicated below, followed by significant loops involving the GATA3 promoter detected in each condition. (D) Pathway enrichment analysis of genes (n = 838) near anchors of memory and recall-associated loops. (E) Scaled mRNA expression (z-score) of genes near anchors of memory (left, n = 430) or recall-associated (right, n = 425) loops in resting (−) or stimulated (+) naive CD4⁺ T cells and memory T_H2 cells. P values: Kruskal-Wallis test corrected for multiple testing. (F) Hi-C interaction maps and signal subtraction plot (bottom; red = higher in T_H2⁻ versus naive⁻) of the IL9 locus. H3K4Me2 ChIP-Seq and ATAC-Seq signal tracks are shown below as well as loops involving the IL9 gene specifically detected in memory T_H2 cells. (G) IL9 gene expression levels in the indicated sample types. (H) Hi-C interaction maps of the T_H2 cytokine locus. Locations of genes, the T_H2 LCR, recall-associated ATAC-Seq peaks, and H3K4Me2 peaks are indicated below (zoom in). Median read count per 10-kb bin [coverage (cov)] within the T_H2 domain is indicated. (I) Interacting read pairs detected connecting IL5 with the LCR-IL13-IL4 region (sum of 5-kb bins) in resting naive or memory T cells. P value: Mann-Whitney U test. (J) Interacting read pairs detected connecting IL5 with the T_H2 LCR, IL13, IL4, or the unrelated SHROOM1 gene across the indicated sample types. (K) Heatmap showing pathway enrichment for the indicated primed recall gene sets. Pathway cluster (n = 4) functions are summarized below.

We next examined loop sets unique to naive T cells. We reasoned that the absence of such loops in memory T_H2 cells may be linked to recall-associated transcriptional silencing or—in the case of repressive loops—facilitating transcriptional recall upon stimulation. Loops present only in naive cells irrespective of stimulation status (n = 514) were enriched for enhancer interactions (fig. S7E). Genes associated with these loops were down-regulated upon naive T_H cell stimulation and already repressed in resting memory T_H2 cells—akin to an “inverse recall” pattern (fig. S7E). In contrast, loops only detected in resting naive cells (n = 3850) were highly enriched for promoter-promoter interactions. Unexpectedly, genes linked to these loops were up-regulated upon TCR stimulation in memory T_H2 cells and enriched for general biological processes associated with cellular activation (fig. S7F). These findings indicate the existence of repressive loops connecting promoters of activation-inducible genes in naive cells, which are already decommissioned in resting memory T_H2 cells.

The signature T_H2 cytokine genes IL4-IL5-IL13 are organized as a cluster and exclusively up-regulated upon memory T_H2 cell activation (see Fig. 1D). Studies in murine T_H2 cells have identified several distal enhancers organized in a locus control region (LCR) that control cytokine gene expression through long-range interactions (52), although 3D folding of the human T_H2 cytokine locus and the relationship with memory recall remain unexplored. We observed robust yet nonpunctate 3D interactions across the entire T_H2 locus already in resting naive T_H cells, which were more abundant in memory T_H2 cells and further augmented upon memory cell activation (Fig. 6H and fig. S7G). High-resolution analysis (5 kb) showed significantly increased nuclear proximity of IL5, IL4, IL13, and the LCR in resting memory T_H2 cells as compared with naive cells (Fig. 6, I and J). These findings are consistent with priming of inflammatory genes for efficient recall through altered local 3D genome organization, i.e., by maintaining elevated promoter-enhancer interaction frequencies in the absence of activating signals. The IL9 and T_H2 cytokine loci were primed at both 1D (i.e., chromatin accessibility) and 3D (i.e., promoter-enhancer contacts) epigenomic levels. We observed widespread multiscale priming within the core T cell activation–associated recall program (Fig. 6K and fig. S7H), including all major inflammatory cytokine genes and the CD28-ICOS-CTLA4, CCR4, and MAF loci (fig. S8 and data file S1). T_H2-specific recall genes were particularly enriched for priming at the 3D level (i.e., 100% of primed genes) (fig. S7H). TF motif enrichment analyses on accessible chromatin sites within regions of recall-associated 3D genome reorganization implicated AP-1, RUNX, ETS, and IRF family proteins as well as CTCF in priming genome conformation in resting memory cells (fig. S7I).

In summary, resting memory T cells uphold a uniquely modified topological domain and loop organization. This 3D genome conformation poises genes for rapid transcriptional recall by increasing the spatial proximity between memory recall genes and their primed regulatory elements or through the loss of repressive promoter-promoter interactions.

Multiscale epigenomic priming in memory T cells is linked to asthma

We next investigated whether the recall-associated transcriptional and epigenomic signatures we identified could be linked to memory T cell dysfunction in human disease. Asthma is one of the most common chronic inflammatory disorders, with most patients exhibiting excessive pulmonary type 2 inflammation often driven by aberrant memory T_H2 cell activity (6, 7). We performed RNA-Seq on resting memory T_H2 cells from peripheral blood samples of healthy controls and patients with asthma with either mild-to-moderate or severe disease (i.e., low or high symptom burden; see Methods and table S1). We observed increased expression levels of all three recall gene modules in memory T_H2 cells from patients with asthma compared with healthy controls (Fig. 7A and fig. S9A), which were less pronounced for general activation genes (fig. S9, A and B). For example, expression of the MAF recall gene was increased in patients with severe asthma, whereas the levels of general activation marker CD40LG were unaltered (Fig. 7B). In addition, analysis of published resting memory T_H2 cell H3K4Me2 profiles from patients with asthma (53) revealed consistently higher H3K4Me2 levels at recall-associated ATAC-Seq peak clusters in asthmatics compared with healthy controls, e.g., at the primed CD82 locus (Fig. 7, C and D). In contrast, only marginal differences were observed at naive-specific accessible chromatin sites (fig. S9, C and D). Together, these findings demonstrate increased activity of the primed molecular circuits that promote rapid recall in resting disease-associated memory T_H2 cells.

Fig. 7. Multiscale epigenomic priming in human memory T_H2 cells is linked to asthma.

(A) Gene expression levels (z-score–scaled) of the three recall-associated transcriptional modules (see Fig. 1) as measured in peripheral blood memory T_H2 cells from healthy controls or patients with asthma with either low or high symptom burden. (B) mRNA levels of a primed recall gene (MAF) and a general activation marker (CD40LG) across the indicated groups. (C) H3K4Me2 levels (z-score–scaled) at the four recall-associated ATAC-Seq peak clusters as measured by ChIP-Seq on peripheral blood memory T_H2 cells from healthy controls or patients with asthma. (D) Genome browser screenshot of the CD82 locus showing H3K4Me2 signals in a healthy control and a patient with asthma. Inset graph below shows H3K4Me2 levels at the indicated recall-associated ATAC-Seq peak across the entire cohort (HC, healthy; AST, asthma). (E) Enrichment scores and P values for asthma-associated SNPs intersected with recall-associated or naive-specific ATAC-Seq peaks. (F) Genome browser screenshot of the CCR4-GLB1 locus, highlighting a T_H2-specific SE region containing multiple recall-associated ATAC-Seq peaks and asthma SNPs. (G) Enrichment scores and P values for asthma-associated SNPs intersected with the indicated recall-associated 3D genome features. (H) Graphical summary of this study’s main findings. See text for details. Statistical tests used: (A and B) Kruskal-Wallis test corrected for multiple testing; (C and D) Mann-Whitney U test; (E and G) Fisher’s exact test. **P < 0.01 and ****P < 0.0001.

We then intersected the location of single-nucleotide polymorphisms (SNPs) linked to asthma pathophysiology (54) with the clusters of recall-associated regions of chromatin accessibility in memory CD4⁺ T cells. Most memory-specific clusters showed statistically significant positive enrichment scores (in particular T_H2-specific peaks), which were not observed in randomly shuffled or in naive-specific peaks (Fig. 7E). T_H2-specific superenhancer regions were particularly enriched for asthma-associated SNPs (enrichment ratio: 31.79; P < 0.0001, Fisher’s exact test), as illustrated by a primed superenhancer in the CCR4-GLB1 locus in which asthma SNPs colocalized with recall-associated ATAC-Seq peaks (Fig. 7F). We detected a significant enrichment of asthma-associated variants within recall-associated and memory-specific changes in 3D genome organization (Fig. 7G), revealing links between the genetic susceptibility to asthma and epigenomic priming of memory T_H2 cells at both 1D and 3D levels.

Discussion

In this study, we show that in contrast to naive CD4⁺ T cells, human memory T_H cells maintain a distinct 1D/3D chromatin landscape that primes modules of selected recall genes for rapid transcriptional activation to facilitate near-instant immune effector responses. More specifically, our analyses indicate that efficient transcriptional recall of resting human memory T_H2 cells builds on several epigenomic hallmarks: (i) maintenance of transcriptionally permissive chromatin at stimulation-induced (super)enhancers; (ii) increased activity of a set of specific TFs at these recall enhancers, both at baseline and after activation; (iii) spatial compartmentalization of recall genes and enhancers in 3D chromatin hubs; (iv) increased connectivity of recall genes and enhancers within memory TADs; and (v) preformed chromatin loops connecting recall genes and their enhancers (Fig. 7H). In a quiescent memory T_H2 cell, these memory-specific epigenomic features position hundreds of genes in a spatial chromatin environment that is exceptionally conducive to rapid transcriptional reactivation. Efficient induction of recall genes, comprising ∼3% of the transcriptome, allows memory T_H cells to quickly adapt their metabolic, migratory, and inflammatory profile to launch specialized adaptive immune responses. Notably, we provide evidence that dysregulation of the molecular mechanisms controlling recall occurs in patients with chronic inflammatory disease.

Previous studies reported maintained active chromatin signatures in resting memory(-like) T cells at regulatory elements of inflammatory genes (27–29, 55–57). Our findings confirm and substantially extend these data by revealing that resting human memory CD4⁺ cells carry an epigenomic imprint of TCR stimulation and lineage (e.g., T_H2) specification that consists of ∼3000 genomic elements marked by accessible chromatin, H3K4Me2 deposition, and RNAPII recruitment. This epigenetic signature has previously been associated with short-term transcriptional memory in yeast, fruit fly, and human cells in response to hormones, nutrient deprivation, or cytokines (58, 59). Maintaining such a transcriptionally permissive chromatin environment at key regulatory elements controlling inflammatory gene expression—including 80% of memory-specific superenhancers in T_H2 cells—provides a mechanistic basis for rapid induction of recall module genes. Supporting this concept, murine memory T_H2 cells with reduced levels of the H3K4 methyltransferase KMT2A (also called MLL) failed to maintain H3K4Me2 levels at the T_H2 cytokine locus and showed impaired secondary responses in vivo (60).

Unexpectedly, we identified genome topology as the best global discriminator of naive and memory cell T_H cell phenotypes (i.e., in a PCA)—in both resting and activated states. Compared with naive cells, resting memory T_H2 cells exhibit a less rigid A/B compartment separation, show hallmarks of heterochromatin decompaction, and carry ∼700 genomic regions more strongly associated with transcriptionally permissive nuclear compartments. At this topological level, primed recall-associated (super)enhancers form 3D hubs by engaging in long-range interactions specifically in memory T_H2 cells. More locally, recall genes and their regulatory elements often reside in memory TADs, which compared with naive cells exhibit increased intra-TAD interaction frequencies connecting recall gene promoters to their enhancers before activation. An illustrative example of multiscale epigenomic priming is the MAF locus encoding an AP-1 family TF previously linked to the control of T_H(2) cell fate and function (46, 61). In resting memory T_H2 cells, the poorly expressed MAF gene resides in a TAD that is already positioned in the A compartment. Recall-associated distal enhancers inside the MAF memory TAD are already accessible and engage in interactions with MAF before stimulation. Hence, a specific 1D-3D chromatin organization readily explains the rapid and specific induction of MAF in memory T_H2 cells upon TCR activation.

The mechanisms that maintain and exploit the primed epigenomic landscape of human memory T cells have remained elusive. Our analyses converge on AP-1 TFs—known to act downstream of TCR signaling (62)—and lineage-determining TFs such as GATA and RUNX proteins as key regulators of rapid recall in human CD4⁺ T cells. We find that primed recall enhancers are strongly enriched for motifs of these TFs, of which several are already expressed in memory T cells before TCR stimulation (see Fig. 3). After TCR stimulation, AP-1 TFs and GATA3 occupy most of the recall-associated regulatory regions. Chemical inhibition of AP-1 TF activity or preventing MAF induction in memory T_H2 cells selectively blunted the transcriptional induction of several recall genes, partially decoupling the reactivation of memory-linked genes from the general T cell activation program—at least at the time scales chosen here (i.e., 24 hours). On the basis of these findings, we propose a model in which signal-responsive AP-1 TFs (e.g., BATF and FOSL2) cooperate with lineage-specifying TFs (e.g., GATA and RUNX proteins) to maintain a 1D/3D primed epigenome in quiescent human memory T cells, akin to the recently described role of TCF-1 in epigenomic priming of murine CD8⁺ memory T cells (63). Upon TCR stimulation, these “recall TFs” can immediately initiate transcriptional (re)activation of the three recall gene modules, which boosts their own expression and de novo activates other TFs (e.g., MAF) to maximize transcriptional output of the ∼900 recall genes. These findings are in line with studies performed in murine memory(-like) T cells, which have reported JUND, ETS1, and RUNX1 binding to sites that remained accessible after a previous episode of stimulation (29, 64, 65).

At the 3D level, we envision two broad mechanisms through which recall TFs may install primed genome topology: first, through modulating chromatin loop extrusion by CTCF and cohesin, which is critical for the formation of TAD borders and a subset of gene regulatory interactions (11, 12, 14). For example, the AP-1 factor BATF was shown to cooperate with CTCF to organize promoter-enhancer interactions at cytokine loci in murine T_H17 cells (66), and the lineage-determining TF TCF-1 was recently implicated in recruiting cohesin (67) or CTCF (41, 63) to active enhancers in naive T cells and their thymic precursors. Moreover, macrophages deficient in CTCF (68) or cohesin (69) showed impaired activation of inducible genes upon stimulation. Second, multivalent interactions between TFs and their cofactors may promote nuclear compartmentalization of regulatory regions through condensate formation (70). Our observation of 3D recall enhancer hubs forming at megabase scales in human memory T_H2 cells reveals nuclear compartmentalization of recall-associated genomic elements and regulatory proteins. Concentrating (super)enhancers, key TFs, and transcriptional machinery in the vicinity of genes targeted for recall already in a resting state, possibly through condensate formation, could greatly facilitate rapid transcriptional activation of the compartmentalized genes by recall TFs upon TCR activation. Important to note is that our results do not establish a direct causal link between genome conformation and enhanced transcriptional induction of recall genes in human memory T_H cells. For this, additional experiments aimed at precise disruption of local 3D genome folding (e.g., using genome editing) are required.

A remaining open question is how the primed 1D/3D chromatin landscape of human memory T cells is maintained across multiple cell divisions and over extended periods of time [e.g., >10 years (71)]. At the 1D level, recall TFs may be actively involved in the maintenance of accessible chromatin or the H3K4Me2 mark on primed recall enhancers, e.g., by recruiting chromatin-modifying enzymes. GATA3 was previously shown to activate transcription in a complex with KMT2A/MLL in human T_H2 cells (72), and AP-1 factor activity is important for establishing accessible chromatin during human T cell activation (44)—presumably via recruiting SWI/SNF chromatin remodeling complexes (73). The IL-2–inducible AP-1 factor JUND appeared to support the maintenance of chromatin accessibility at enhancers activated by stimulation in a murine in vitro model system of T cell priming (29, 65).

IL-7 and IL-15 signaling, both resulting in signal transducer and activator of transcription 5 (STAT5) TF activation, are critical for homeostasis of resting memory T cells (74). Previous work has implicated the IL-7/STAT5 axis in maintaining chromatin accessibility at a subset of primed regulatory elements in murine memory-like CD4⁺ T cells (65), although our analyses did not detect STAT motifs enriched in the recall enhancer landscape of human memory CD4⁺ T cells. Nevertheless, it remains possible that STAT5 controls a small subset of primed enhancers or acts in TF complexes without binding the DNA itself. It should be noted that alternative mechanisms of priming memory T cells—not necessarily only acting at the chromatin level—have been described and are likely to function in concert with TF-mediated epigenomic priming to enable rapid recall. For example, immediate translocation of nuclear factor κB (NFκB) proteins from the cytoplasm into the nucleus was shown to underlie the more rapid activation of Ifng in murine memory CD4⁺ T cells as compared with naive cells (75). We also detected NFκB recruitment to a small subset (∼6%) of recall enhancers after TCR activation (see Fig. 3F). Release of the translation repression of preformed mRNAs has also been implicated in rapid recall (76), as well as metabolic reprogramming and altered TCR signal transduction (77, 78).

Last, our analyses show that recall-associated gene modules and their primed regulatory elements exhibit increased biochemical activity in resting memory T_H2 cells from patients with asthma, who suffer from chronic pulmonary type 2 inflammation. Furthermore, we found that genetic variants linked to asthma pathophysiology are enriched in recall enhancers. These findings suggest that circulating quiescent memory T_H2 cells in patients with asthma may reside in a cell-intrinsic “hyperprimed” state—possibly shaped by genetics—that lowers the threshold for recall and fuels the accumulation of activated pathogenic memory T_H2 cells in the lungs. Future studies are required to validate this concept, which may offer new opportunities for therapy development and could be extended to other contexts of memory T cell dysfunction such as autoimmunity or cancer.

Materials and Methods

Study design

Human naive and T_H2 memory CD4⁺ T cells were isolated from healthy donor blood in resting (ex vivo) and activated (in vitro) states. To identify transcriptional and epigenomic signatures that predispose memory T cells to a more efficient transcriptional induction of the associated genes upon activation (i.e., recall), we used transcriptomic (RNA-Seq; n = 3 independent biological replicates), 1D epigenomic (ATAC-Seq; n = 3 or 4 independent biological replicates), and 3D epigenomic (Hi-C; n = 3 independent biological replicates) data. Recall-associated transcriptional and epigenomic signatures were confirmed using flow cytometry (n = 2 or 3 independent biological replicates) and CRISPR-Cas9 genome editing (n = 2 or 3 independent biological replicates) and by integrating published epigenomic datasets. Last, we generated bulk transcriptome data of peripheral blood memory T_H2 cells obtained from patients with asthma (n = 16) and healthy donors (n = 7), which we integrated with public H3K4me2 ChIP-Seq data and linked to known asthma-associated genetic variants.

Human T cell isolation from healthy donors and patients with asthma

Primary sources of T cells from healthy individuals in this study were buffy coats (from 500 ml of peripheral blood samples) obtained from the Sanquin Bloodbank (Amsterdam, Netherlands). Informed consent was obtained by Sanquin. For asthma patient materials used in this study, criteria for diagnosis, inclusion, and classification (79) are outlined in the Supplementary Materials and table S1. Peripheral blood mononuclear cells (PBMCs) were isolated as previously described (80). From PBMCs, CD4⁺ T cells were isolated using the human CD4⁺ T Cell Isolation Kit (Miltenyi Biotec) according to the manufacturer’s instructions, followed by FACS (using FACSAria II, BD Biosciences) to isolate resting naive CD4⁺ T cells (live single CD4⁺CD127⁺CD25^negCD45RA⁺ cells) and T_H2 cells (live single CD4⁺CD127⁺CD25^negCD45RA^negCCR6^neg-CXCR3^negCXCR5^negCCR4⁺ cells). Antibodies used for FACS are listed in table S2. Fluorescent antibody staining methods are in the Supplementary Materials.

In vitro T cell activation

Freshly isolated T cells were activated in vitro at 1 × 10⁶ cells per ml in T cell culture medium [RPMI 1640 (Gibco), 10% fetal calf serum (HyClone), 55 μM 2-mercaptoethanol (Sigma-Aldrich), 2 mM L-alanyl-L-glutamine (GlutaMAX, Gibco), 1 mM sodium pyruvate (Gibco), 0.1 mM nonessential amino acids (Gibco), penicillin/streptomycin (Gibco), 10 mM Hepes (Gibco), and recombinant human IL-2 (50 U/ml; R&D Systems)] in round-bottom 96-well culture plates using Human T-Activator CD3/CD28 Dynabeads (Thermo Fisher Scientific) in a 1:1 bead-to-cell ratio at 37°C, 5% CO₂ for 24 hours. Activator beads were not removed before further processing of samples to minimize loss of activated cells.

RNA isolation, qPCR, and RNA-Seq

RNA was extracted from ∼50,000 resting cells per condition using an RNeasy Micro kit (Qiagen) according to the manufacturer’s instructions. For quantitative polymerase chain reaction (qPCR), RNA was synthesized into complementary DNA using RevertAid H Minus Reverse Transcriptase and random hexamer primers in the presence of RiboLock RNAse inhibitor (Thermo Fisher Scientific). qPCR was performed using 7.5 μl of 2× SYBR Green Universal Master Mix (Life Technologies) and 6 pmol of both forward and reverse primers, added to a total reaction volume of 15 ul using nuclease-free water. Primer sequences used can be found in table S3. For high-throughput sequencing of RNA, libraries were prepared using Smart-seq2 methodology (81) and sequenced according to the Illumina TruSeq Rapid v2 protocol on an Illumina HiSeq2500 [single read, 51–base pair (bp) read length].

CRISPR-Cas9 genome editing and AP-1 TF perturbation

For genome editing in primary human memory T_H2 cells, we adopted the method published by Seki and Rutz (82), which we describe in more detail in the Supplementary Materials. Crispr RNAs (crRNAs) specific for MAF were designed using Benchling software (see table S3 for crRNA sequences). AP-1 factor inhibition experiments are described in the Supplementary Materials.

RNA-Seq data processing and differential gene expression analysis

Reads were aligned to the hg38 build of the human reference genome using HISAT2 (83). Raw counts and RPKMs were determined on the exons of RefSeq-annotated genes using HOMER’s [v4.11 (84)] analyzeRepeats.pl command. Differentially expressed genes [false discovery rate (FDR) < 0.1] were identified using DESeq2 (85) through HOMER’s getDiffExpression.pl command. In all downstream analyses, we only included genes that were expressed in at least one relevant group (RPKM > 1 in ≥50% of replicates). PCA was conducted using FactoMineR (86). K-means clustering was performed using PAST (87). Heatmaps, including hierarchical clustering on gene expression values, were generated using Morpheus (https://software.broadinstitute.org/morpheus/). Pathway enrichment analysis on differentially expressed genes and summarizing enriched pathways into gene-based networks were performed using Metascape (88).

Identification of recall-associated and general activation transcriptional signatures

The strategies used to define the various recall-associated and general activation transcriptional signatures identified in this study are detailed in the Supplementary Materials.

ATAC-Seq data processing, peak calling, and assignment of peaks to genes

ATAC-Seq data were previously published by Calderon et al. (15) (GSE118189). SRA files were aligned to hg38 with HISAT2. Insert sizes >500 bases were removed using SAMtools (89). Tag directories were created using HOMER’s makeTagDirectory command, after which reads aligning to the Y chromosome or mitochondrial DNA were removed, and the fragment length estimate was set to 77 as a representative estimate for all samples. A transcription start site (TSS) enrichment ratio was calculated using HOMER’s annotatePeaks.pl command (tss hg38 -size 10000 -hist 40) as the average TSS signal (−80 to +80 bp from TSS), divided by the average local background signal (−5 to −4.92 kb and +4.92 to +5 kb from TSS), to use as a measure for relative sample quality. Samples with a TSS enrichment ratio less than the mean-SD TSS enrichment ratio across all samples were excluded from further analysis. Peak calling was performed using HOMER’s findPeaks command (-region -size 100 -minDist 75). Reproducible peaks per experimental group were defined as overlapping peaks called in three of three replicates (three of four for activated memory T_H1 and activated naive CD4⁺ T cells) using HOMER’s mergePeaks command. Universal peaks were similarly identified as overlapping reproducible peaks found in all eight experimental groups using mergePeaks. HOMER was used to create bedgraph files for visualization in the IGV genome browser (90). Peaks were associated with genes using GREAT (91) with default settings (i.e., basal plus extension; proximal: 5 kb upstream, 1 kb downstream, distal: up to 1000 kb). When more than two genes were linked to a single peak, only the closest two genes were included in downstream analyses.

ATAC-Seq peak signal normalization and differential analysis

Raw counts were determined at all reproducible peaks found in at least one of eight experimental groups using HOMER’s annotate-Peaks.pl command (hg38 -size given -raw). Raw counts were rounded down to the nearest integer. Counts at peaks were normalized by slightly adapting the standard normalization method of DESeq2. Scaling factors were determined for each sample based on the set of universal peaks with DESeq2. Next, those scaling factors were used to normalize counts at all peaks. This adaptation of DESeq2’s established normalization method allowed us to successfully normalize ATAC peak counts despite the approximately doubled number of ATAC-Seq peaks upon T cell activation and the relative lack of zeros in ATAC-Seq data (compared with RNA-Seq) at regions where there is no true signal. PCA was conducted using FactoMineR. Differentially enriched ATAC-Seq peaks were identified using DESeq2 (log₂ fold change > 1, FDR < 0.05). Clustering was performed on differentially accessible sites across all cell states (n = 32,847 peaks) using Mfuzz (92). Clusters were selected on the basis of visual inspection of patterns; high-confidence peaks (α > 0.5) were kept for downstream analyses and visualization. Histograms of the ChIP-Seq signal were generated using HOMER’s annotatePeaks. Heatmaps were generated using Morpheus, and pathway enrichment analysis on associated genes was performed using Metascape.

TF binding motif identification

HOMER’s findMotifsGenome command (-size 200 -mask) was used to search for overrepresented TF binding motifs in genomic regions. The top three ranked de novo motifs based on P value were selected for further analysis.

ChIP, ChIP-Seq, and data analysis

H3K4Me2 ChIP-Seq data were generated using the ChIPmentation protocol as previously described (39, 47). More details can be found in the Supplementary Materials. Reads were aligned to the hg38 build of the human reference genome using HISAT2. H3K4Me2-enriched regions were identified using HOMER’s findPeaks command (-region -size 1000 -minDist 2500); for superenhancers, the parameters -style super -L 1 were used instead with a negative control mock (rabbit immunoglobulin G) experiment as background. H3K4Me2 signals were quantified at recall-associated ATAC-Seq peaks or peak clusters using HOMER’s annotatePeaks.pl command. BEDTools’ (93) fisher command was used to calculate the overlap between genomic features. Similar methodology was used to analyze publicly available ChIP-Seq datasets.

Asthma-associated SNPs

Asthma-associated genetic variants were obtained from the childhood-onset asthma genome-wide association study (GWAS) reported by Ferreira et al. (54). The GWAS summary statistics file was parsed to the Functional Mapping and Annotation (FUMA) platform (94), which uses PLINK to impute missing genotypes from the 1000 Genomes reference panel (95). Default settings were used, with the exception of r² (adapted to r² < 0.8). Genomic locations of the resulting SNPs were converted from GRCh37/hg19 to GRCh38/hg38 coordinates using the LiftOver tool of the UCSC Genome Browser (96). BEDTools’ fisher command was used to calculate overlap between genomic features and candidate causal asthma SNPs.

In situ Hi-C library preparation and initial data processing

In situ Hi-C was performed as previously described (68, 97) using ∼200,000 human T cells as starting material. Libraries were sequenced on HiSeq 2500 sequencers (Illumina, paired-end, 75-bp read length) to a sequencing depth of >1.2 billion read pairs per T cell subset and activation state. Hi-C data were processed using an in-house pipeline based on TADbit (98) as previously described (68, 97). In brief, after trimming and removing poor quality reads, contact pairs were mapped using a fragment-based strategy as implemented in TADbit. Mapped reads were filtered to remove non-informative contacts (e.g., self-circle, dangling-end, PCR duplicates). Contact matrices obtained were normalized for sequencing depth and genomic biases using OneD (99). For differential interaction analysis and visualization, the resulting normalized matrices were directly subtracted.

Identification of subnuclear compartments and TADs

To segment the genome into A/B compartments, we used CscoreTool (100) at 10-kb resolution, with computed C-scores reflecting quantitative association with A (0 to 1) or B (−1 to 0) compartments for each bin. Normalized contact matrices at 50-kb resolution were used to define TADs, using the previously described insulation score method (101) with default parameters. This procedure resulted in a set of borders (i.e., local minima in insulation score) for each replicate. Across replicates, overlapping borders or borders differing by only one bin were merged to obtain a list of reproducible borders for each T cell subset and activation state. Borders overlapping or differing in location by only one bin in all four experimental groups were considered common, and the rest were considered dynamic.

Inter- and intracompartment strength measurements

Cooltools (10.5281/zenodo.5214125) was used to quantitatively segment the genome into A/B compartments on the basis of the first component of a PCA (PC1/EV1) on each of the normalized Hi-C matrices. AT content was used to assign negative and positive PC1 categories to the correct compartments. We based our analysis on the top 20% 100-kb bins showing the most extreme positive (A) or negative (B) PC1 values. We classified each bin in the genome according to PC1 percentiles and gathered contacts between each category, computing saddle strength profiles and generating saddle plots on the genomic bins.

Clustering analysis of genome-wide C-score or TAD border insulation score dynamics

For A/B compartmentalization, average C-scores per bin per experimental group were calculated. Dynamic bins were identified as bins that exhibited a ΔC-score > 0.3 in at least one pairwise comparison among our four experimental groups. Dynamic bins were grouped into 30 clusters using Mfuzz. We selected six recall-associated clusters on the basis of the average C-score of all bins in a cluster meeting the following criteria: memory T_H2⁻ > naive⁻, naive⁺ > naive⁻, and memory T_H2⁺ > memory T_H2⁻ (total n = 11,874 bins). Next, we selected bins within these clusters where the ΔC-score between memory T_H2⁻ and naive⁻ was >0.15 (n = 6888 bins). Consecutive bins were merged into larger regions, and because we do not expect single bins to follow certain dynamics without neighboring bins following similar trajectories, single bins remaining after merging were excluded. Merged regions (n = 754 regions of >10 kb) were linked to genes on the basis of intersecting TSSs (RefSeq) and the GREAT annotations of the ATAC-Seq peaks described above. Insulation scores for all dynamic TAD borders (n = 1463) were averaged across replicates and subjected to clustering using Mfuzz, with recall-associated dynamics selected on the basis of visual inspection of clusters.

Long-range interactions between ATAC-Seq peaks

Long-range interactions (2- to 10-Mb window) between ATAC-Seq peaks (from 10-kb resolution Hi-C matrices) were computed using the hicAggregateContacts command from HiCExplorer (102).

Chromatin loop detection and meta-analysis of loops

Hi-C matrices at 5-kb resolution were converted to a cool format using HiCExplorer and parsed to Mustache (51) for chromatin loop detection using default parameters. Genome-wide aggregate plots (pile-ups of normalized signals at 5-kb resolution) of loops were generated with coolpup.py (103). Signals were plotted as heat-maps ranging from −50 to +50 kb of the loop anchor coordinates.

Statistical analyses

Statistical parameters are reported in the figures and/or figure legends. Differential gene expression or peak enrichment of individual genes and peaks was calculated with the Wald test corrected for multiple testing according to the Benjamini-Hochberg procedure through DESeq2. FDR < 0.1 was considered significant for RNA-Seq, and FDR < 0.05 was considered significant for ATAC-Seq. Enrichment of TF motifs, SNPs, or ATAC peaks within genomic regions was calculated with Fisher’s exact test. For comparisons of means between groups, two-sided Mann-Whitney U tests or Benjamini-Hochberg–corrected Kruskal-Wallis tests were performed, with P < 0.05 considered significant.

Data and materials availability

High-throughput sequencing datasets generated for this study were deposited in the Gene Expression Omnibus (GEO): RNA-Seq (GSE210958), ChIP-Seq (GSE211023), and in situ Hi-C (GSE207738). Previously published datasets used in this study were RNA-Seq and ATAC-Seq data from Calderon et al. (15) for naive CD4⁺ T cells, memory T_H1 cells, memory T_H17 cells, and memory T_H2 cells—both before and after stimulation (GSE118165 and GSE118189). H3K27Ac in total CD4⁺ memory T_H cells (40) (GSE71596; healthy control samples), RNAPII in naive CD4⁺ T cells (GSM2370637) and CD4⁺ memory T_H cells (GSM2370653) (28), H2A.Z in naive CD4⁺ T cells and CD4⁺ memory T_H cells (GSE89404) (28), and H3K27Me3 in naive CD4⁺ T cells (ENCODE consortium: ENCSR637IOH) and CD4⁺ memory T_H cells (ENCODE consortium: ENCSR012KUR) (104). TF binding sites in activated memory CD4⁺ T_H(2) cells were obtained from GATA3 (43) (GSE31320) and NFAT1/NFAT2–JUNB/cFOS (AP-1)–MYC–NFκB p50 (all from GSE116695) (44). H3K4Me2 ChIP-Seq data for circulating memory T_H2 cells from healthy controls and patients with asthma were obtained from Seumois et al. (53) (GSE53646). Any code used for data analysis was derived from standard protocols of published R packages (see Materials and Methods).