Antigen perception in T cells by long-term Erk and NFAT signaling dynamics

Significance

To counter diverse pathogens, T cells mount distinct responses to varying peptide–major histocompatibility complex ligands (pMHCs). They perceive the affinity of pMHCs for the T cell receptor (TCR), which reflects its foreignness, as well as pMHC abundance. By tracking signaling responses in single living cells to different pMHCs, we find that T cells can independently perceive pMHC affinity vs dose and encode this information through the dynamics of the extracellular signal-regulated kinase (Erk) and nuclear factor of activated T cells (NFAT) signaling pathways downstream of the TCR. These dynamics are jointly decoded by gene regulatory mechanisms to produce pMHC-specific activation responses. Our work reveals how T cells can elicit tailored functional responses to diverse threats and how dysregulation of these responses may lead to immune pathologies.

Abstract

Immune system threat detection hinges on T cells’ ability to perceive varying peptide–major histocompatibility complex (pMHC) antigens. As the Erk and NFAT pathways link T cell receptor engagement to gene regulation, their signaling dynamics may convey information about pMHC inputs. To test this idea, we developed a dual reporter mouse strain and a quantitative imaging assay that, together, enable simultaneous monitoring of Erk and NFAT dynamics in live T cells over day-long timescales as they respond to varying pMHC inputs. Both pathways initially activate uniformly across various pMHC inputs but diverge only over longer (9+ h) timescales, enabling independent encoding of pMHC affinity and dose. These late signaling dynamics are decoded via multiple temporal and combinatorial mechanisms to generate pMHC-specific transcriptional responses. Our findings underscore the importance of long timescale signaling dynamics in antigen perception and establish a framework for understanding T cell responses under diverse contexts.

The immune system can mount tailored responses to different threats while sparing healthy tissues. This remarkable selectivity stems from T cells’ ability to obtain information about the threat from peptide antigens presented on peptide–major histocompatibility complex ligands (pMHCs) (1, 2). pMHCs that bind the T cell receptor (TCR) with high affinity derive preferentially from foreign sources due to a sharp affinity threshold for negative selection (3, 4). On the other hand, lower-affinity pMHCs can derive either from foreign sources or self. The dose of pMHCs presented to the T cell yields complementary information: High doses of pMHCs can convey pathogen virulence or infection severity or simply reflect high abundance of self-antigen.

Despite the importance of antigen sensing for immunity, how T cells perceive pMHC affinity and dose to generate tailored functional responses remains incompletely understood. Early studies suggested that T cells respond to high-affinity pMHCs while ignoring high doses of low-affinity pMHCs (5–10). This selectivity was proposed to result from kinetic proofreading in the TCR signaling pathway (6, 11–19), which enables selective responses to pMHCs with long binding lifetimes. However, while there is compelling evidence for kinetic proofreading (20, 21), T cells do not ignore low-affinity pMHC ligands (22–25). More recent studies indicate that T cells enact a common gene activation program amid different pMHC affinities, albeit with varying kinetics (26, 27). pMHC differences can tune these gene programs (26, 27), thereby modulating downstream outcomes (28–31), with their dose and affinity exerting distinct regulatory effects (32–35). These findings suggest that the TCR signaling network can sense pMHC affinity and dose to translate this information into distinct functional outcomes.

As the Erk and NFAT signaling pathways are primary links between TCR engagement and the regulation of gene transcription (36), they are well positioned to convey pMHC information in the cell. Upon pMHC engagement, the TCR activates a series of intracellular signaling molecules, leading to activation of both Erk and NFAT pathways. Both pathways activate within minutes upon antigen encounter and, over these short timescales, appear to do so in an all-or-one, “digital” manner, with activated cells showing similar signaling levels regardless of pMHC affinity or dose (6, 37–39). However, T cells can be exposed to pMHCs for up to 24 h after initial encounter in vivo (40, 41) through stable contacts with dendritic cells in lymph nodes. Over these longer timescales, it is unclear whether Erk and NFAT signaling remains uniform to varying pMHC inputs or whether they may show variable responses to different pMHC inputs only upon prolonged exposure. Indeed, signaling pathways, including both Erk and NFAT, have been shown to convey input information in other systems through different long-term dynamics, including pulsatile and oscillatory behaviors (42–51). However, roles for their long-term dynamics in T cell antigen perception remain unclear, as most TCR signaling measurements, to our knowledge, have been performed over short timescales (<3 h).

NFAT and Erk signaling leads to nuclear accumulation and activation of the NFAT transcription factor (TF) and AP-1 family TFs respectively (52). As these TFs can show different modes of cooperation by which they regulate their target genes, they could enable cells to translate pMHC-specific Erk and NFAT signaling activities to distinct transcriptional responses. NFAT and AP-1 can bind to composite DNA binding sites for both TFs cooperatively (53, 54), a mode of action that underlies regulation of many T cell activation and effector genes (Irf4 and Il2ra). Disruption of NFAT:AP-1 impairs T cell effector responses and results in an inability to clear chronic infections and tumors (55). At the same time, NFAT can work independently from AP-1, a mode of regulation for exhaustion- or anergy-associated genes (e.g., Pdcd1) (53, 56–58). These different cis-regulatory mechanisms may allow T cells to generate distinct genomic programs to different pMHC inputs.

In this study, we investigated the role for Erk and NFAT signaling dynamics in antigen perception, specifically seeking to measure signaling pathway dynamics in living T cells over daylong timescales over which T cells contact antigen-presenting cells and over which the gene regulatory responses unfold (40). To do so, we developed a dual-pathway fluorescent reporter mouse strain that concurrently reveals Erk and NFAT signaling activity in the same cell. We then developed an in vitro imaging assay to quantify the first 30 h of Erk and NFAT signaling dynamics in CD8⁺ T cells responding to alterations in pMHC affinity and dose. We then use RNA sequencing to investigate how these dynamics generate specific affinity- and dose-dependent gene programs and propose a model to explain how diverse input-dependent gene expression arises from these signaling activities.

Results

“Dual-Pathway Reporter (DPR)” Mouse Enables Continuous Measurements of Erk and NFAT Activity in Living T Cells.

To quantify Erk and NFAT signaling dynamics in living T cells, we developed a reporter system to concurrently measure the activities of both pathways in live CD8⁺ T cells (Fig. 1A). This system consists of 1) the N-terminal regulatory domain of mouse NFATc2 (residues 1 to 399) fused to mRuby3 (39, 59, 60); 2) an Erk kinase-translocation reporter (Erk-KTR) (47) fused to BFP; and 3) histone 2B (H2B) fused to iRFP for nuclear segmentation. We generated a DPR mouse strain with this reporter cassette inserted into the Rosa26 locus downstream of a loxP-flanked STOP codon (Rosa26^LSL-DPR). Rosa26^{LSL-DPR/LSL-DPR} mice were crossed to the P14 transgenic TCR strain to enable stimulation with altered peptide ligands (APLs) for the cognate gp_33-41 peptide from Lymphocytic Choriomeningitis Virus (LCMV) complexed to H-2D^b MHC-1 (61). DPR⁺/P14⁺ mice were then crossed to a Cre-ERT2 strain and treated with tamoxifen for reporter induction (SI Appendix, Fig. S1A).

Fig. 1.

Erk and NFAT signaling varies with pMHC affinity and dose over long timescales. (A) Schematic of the DPR mouse strain used to quantify Erk and NFAT signaling dynamics in live T cells. The NFAT reporter translocates from the cytoplasm to the nucleus when active. Erk-KTR translocates from the nucleus to cytoplasm when active. To assess pMHC-dependent signaling, we used various doses of two APLs that bind the P14 TCR with ~eightfold difference in binding affinity. (B) Representative cell images of reporter T cells responding to a saturating dose of agonist pMHC over the first hour of stimulation, compared to a negative control cell treated with NFAT (Cyclosporin A, CsA) and Erk (Trametinib, MEKi) inhibitors. Nuclear and cytoplasmic segmentation masks are shown in solid and dashed lines, respectively. (Scale bars, 5 µm.) (C) P14 Jurkat T cell dose–response curve based on CD69 upregulation 4 h after stimulation with each of the two APLs. The Y axis is MFI normalized to the EC₅₀ for each APL. (D) Mean ± 99% CI for Erk and NFAT activity over the first 30 h of stimulation with each of the five pMHC conditions from ~180 k cell images from two independent replicates with 265 to 818 cells (~600 average) per data point. (E) A total of 220 single-cell traces per condition from the data in D were clustered on their scaled dual-pathway signaling activities from 2 to 15 h and partitioned into 8 clusters. The proportions of cells that comprise each cluster, and the mean response of each cluster is shown to the Right of the heatmap. Black dashed lines indicate the mean activity for both pathways across all cells and timepoints. On the far right are representative cell images for four broad types of combinatorial Erk and NFAT responses classified from the eight clusters indicated in parentheses. Nuclear and cytoplasmic segmentations are shown in solid and dashed lines, respectively. (Scale bar, 5 µm.)

To measure Erk and NFAT dynamics in DPR⁺/P14 (reporter) CD8⁺ T cells responding to pMHCs of different doses and affinities, we developed a quantitative imaging assay where T cells are cultured on purified, surface-immobilized pMHCs (62, 63) to ensure that they experience uniform, defined inputs and then imaged at regular intervals with time-lapse microscopy (Fig. 1A). We stimulated reporter T cells with a high-affinity gp_33-41 pMHC variant (M9C, K_D = 2.3 μM) (64), imaged them at 4-min intervals for 1 h, and quantified reporter localization (Fig. 1B and SI Appendix, Fig. S1B). Both Erk and NFAT activated within minutes (SI Appendix, Fig. S1 C–E), consistent with their rapid activation upon TCR engagement (37, 39), and remained active for the duration of the observation. Activation was abolished by MEK (Trametinib; MEKi) and Calcineurin (Cyclosporin A; CsA) inhibitor treatment to block activation of Erk and NFAT, respectively (Fig. 1B), providing a baseline for reporter inactivity.

T Cells Perceive pMHC Affinity and Dose Using Long-Term Erk and NFAT Signaling Dynamics.

Previous studies have shown that both Erk and NFAT pathways can activate in a digital, all-or-none manner in response to varying pMHC inputs (37, 65), at least shortly after activation. However, as T cells can form stable contacts with dendritic cells for up to 24 h after initial contact (40, 41), they may respond divergently to differing pMHC affinity and dose only over these longer timescales. To test this, we measured Erk and NFAT signaling activity in T cells over 30 h of stimulation with five pMHC conditions that vary in affinity or dose. We used both gp33-M9C (KD = 2.3 µM) and a lower affinity partial agonist, gp33-L6F (K_D = 19 µM) (64), which shows a similar rate for TCR binding but has a shorter binding lifetime (66). To assess T cells’ ability to sense pMHC affinity, we measured the dose–response to these two pMHCs by measuring expression of the T cell activation marker CD69 (Fig. 1C and SI Appendix, Fig. S2 B and C) and chose saturating doses of each pMHC (2 and 20 pmol M9C vs. 20 and 80 pmol L6F), which yield equivalent CD69 activation potencies (3). To assess T cells’ ability to sense pMHC dose, we chose a dose of M9C (0.2 pmol) just below that needed for half-maximal CD69 activation. We then stimulated reporter CD8⁺ T cells with each of these five pMHC conditions and acquired images at 1-h intervals for 30 h (Movies S1–S5). We quantified the average Erk and NFAT activities in single cells (Fig. 1D and SI Appendix, Fig. S2E) and extracted Erk and NFAT activity traces for ~1,500 cells over the first 15 h (Fig. 1E).

Upon stimulation by different pMHC inputs, both pathways activated, as expected, and reached similar maximal levels of activity within the first hour after pMHC encounter (Fig. 1D and SI Appendix, Fig. S1D). Lowering pMHC dose by one hundred-fold did not reduce maximal Erk and NFAT activity levels but reduced the rate of NFAT activation (SI Appendix, Fig. S1E), consistent with prior studies showing that signal strength modulates the speed of all-or-none T cell activation (26, 27). However, after several hours of continued stimulation, divergent signaling dynamics emerged in a pMHC-dependent manner (Fig. 1 D and E). Mean Erk activity dropped after 8 to 20 h of stimulation with L6F but remained high for M9C at both high and low doses. In contrast, mean NFAT activity dropped rapidly for all doses of L6F, as well as for low dose of M9C, but remained persistently higher with higher doses of M9C. These signaling differences were not due to differences in pMHC stability, as both L6F and M9C were found to show negligible reduction in activity over these timescales when plate-immobilized (SI Appendix, Fig. S2A). Furthermore, similar pMHC-dependent differences were also observed in nonreporter T cells immunostained for endogenous ppErk or nuclear-localized NFAT1 (SI Appendix, Fig. S2D), indicating that our reporter system faithfully captures endogenous Erk/NFAT signaling states in T cells.

We next determined the dynamic features of Erk and NFAT signaling in single cells that encode information about pMHC affinity and dose. To do so, we performed hierarchical clustering on 1,100 single-cell signaling responses from the 5 different pMHC conditions and identified eight clusters that showed distinct long-term signaling dynamics (Fig. 1E). While some cell clusters showed a monotonic decay in the activity of both pathways (clusters 5 to 8), others showed subsequent pulses of activity following the initial activation peak in either NFAT (cluster 3), Erk (cluster 4), or both pathways (clusters 1 and 2). Different clusters were enriched for cells stimulated with different pMHC inputs, suggesting a role for these pulsed dynamics in encoding pMHC affinity and dose information. Thus, we quantified the incidence and dynamic features of Erk and NFAT pulsing in single cells and analyzed how they vary across different pMHC conditions (Fig. 2 A–E and SI Appendix, Fig. S2F).

Fig. 2.

T cells independently encode pMHC affinity and dose information through late Erk and NFAT pulsatile dynamics. (A) Quantification of pulsatile dynamics from single cells. (B) Probabilities of NFAT and/or Erk pulses occurring for each pMHC condition. (C) Quantification of the fraction of cells from each pMHC condition that exhibited one or more pulses of signaling activity following the initial activation peak. (D) Quantification of early and late integrated activities from single cells. (E) 2D scatter plots of scaled early and late integrated activities for Erk and NFAT in each cell, separated by pMHC condition with colors matching those in C. Black diamond is the mean for each condition. n = 1,510 total cells. (F) A random forest machine learning classification model was used to make predictions on the input pMHC condition given an Erk and NFAT signaling response in single cells. (G) Precision matrices from random forest models trained on either the full dual-pathway time series (2 to 15 h) or the late time series (8 to 15 h) and used to predict the pMHC input in single cells. Color bars on each axis represent three different pMHC inputs that vary in affinity or dose. (H) Accuracy of random forest models trained on each 3-h time-series window (Left) or on varying early or late durations of the time series (Right). The full and late time-series model accuracies from G are indicated.

We found that high doses of M9C elicited subsequent Erk and NFAT pulses with high probability. Most cells showed one or two subsequent pulses, though additional pulses or slow oscillations could not be ruled out due to the limited time window for observation. Equivalently high doses of L6F elicited substantially lower probabilities of subsequent Erk and NFAT pulses (Fig. 2 B and C). In contrast, lowering M9C dose selectively reduced the likelihood of NFAT pulsing, while maintaining high Erk pulse probabilities. As an alternate measure of late pulsatile activity, we also quantified late integrated Erk and NFAT activities in single cells (Fig. 2D). We observed similar pMHC affinity and dose-dependent trends in late integrated activities as we did with pulse probabilities (Fig. 2E), while the early integrated activities were similar across all pMHC inputs.

To evaluate the extent to which the late Erk and NFAT activities enable discrimination between different pMHC inputs, we trained a series of random forest classification models to distinguish between three pMHC dose and affinity regimes (Fig. 2 F and G and SI Appendix, Fig. S2 G–I) (43). A random forest model was trained with either full Erk and NFAT time-series data, or varying temporal windows within these time series. We then used hold-out data to assess model performance and determine the time points that convey the most pMHC information. The full time-series model was reasonably accurate in distinguishing pMHC inputs, with high precision for high doses of high- or low-affinity pMHC inputs (75% or 78.6%) and reasonable precision for low-dose, high-affinity pMHC inputs (58.6%). Models trained on later time windows showed higher accuracy compared to those trained on earlier time windows (Fig. 2H), with Erk and NFAT activity solely from 8 to 15 h being able to discriminate pMHC inputs with comparable precision and accuracy to the full time series (Fig. 2 G and H).

Consistently, models trained on late pulse features or late integrated signaling activities showed higher precision in pMHC discrimination compared to a model trained on early signaling activities (SI Appendix, Fig. S2 G and I). Taken together, these findings indicate that combined late Erk and NFAT signaling dynamics convey sufficient information to enable discrimination between pMHC affinity and dose.

T Cell Activation Genes Exhibit Distinct Modes of Decoding Erk and NFAT Signaling.

Genes downstream of TCR signaling integrate AP-1 and NFAT inputs in distinct ways. Therefore, tuning the activation of each TF through the differential long-term Erk and NFAT dynamics we observed could plausibly lead to distinct gene expression programs. To explore this possibility, we used RNA sequencing to identify gene programs that depend on Erk and NFAT signaling at long timescales. We subjected CD8⁺ T cells to strong stimulation (2 pmol M9C) to elicit maximal signaling, then treated cells with either MEKi and/or CsA after 9 h (Fig. 3A), when signaling activities diverge for different pMHC inputs (Fig. 1). Live imaging confirmed signaling inhibition as expected (Fig. 3B). Cells were then subjected to bulk RNA sequencing after 30 h of stimulation and analyzed for differential gene expression to identify targets of late Erk and/or NFAT signaling. This analysis identified 1,089 differentially expressed genes (DEGs) regulated by Erk and/or NFAT signaling, with substantial overlap between their target genes (SI Appendix, Fig. S3 A and B). We then classified each gene as being strongly or weakly regulated based on their magnitude of expression fold change upon inhibition (Fig. 3C). Using measured Erk and NFAT signaling activities after inhibitor treatment (Fig. 3B), we fit a multiple linear regression model to individual genes in the RNA-seq dataset, obtaining sensitivity coefficients for the dependence of their expression levels on Erk activity (β_E), NFAT activity (β_N), or the product of their activities (β_EN) (Fig. 3F and SI Appendix, Fig. S3C).

Fig. 3.

Target genes of Erk and NFAT show distinct modes for decoding signaling activity. (A) Live imaging and bulk RNA sequencing of CD8⁺ T cells treated with MEKi, CsA, both inhibitors, or no inhibitors (DMSO only–vehicle control). (B) Mean ± SD from 96 to 110 single-cell Erk and NFAT activity traces in response to each perturbation. (C) DEGs from bulk RNA-seq were identified based on a q-value threshold of 0.001 and classified as being strongly or weakly regulated by NFAT and/or Erk relative to a maximum fold change from any treatment of 1.8. (D and E) Hierarchical clustering of genes strongly (D) and weakly (E) regulated by Erk and NFAT signaling. Classification of regulatory modes was determined by distinct patterns of expression change due to inhibitor treatments. Genes of interest are annotated from each regulatory mode. “A” = Activated, “R” = Repressed, “I” = Incoherent, “s” = strong, and “w” = weak. (F) Boxplots of β coefficients from multiple linear regression models and grouped by regulatory mode. (G) Characterization of five regulatory modes. Example genes for each mode are shown with the linear model prediction of their expression with changing levels of Erk and NFAT signaling. (H) Representative pairwise correlations between β_EN vs. β_E and β_EN vs. β_N for the A1s and A3s genes. All DEGs are shown in gray dots with genes of the respective regulatory mode shown in red. Linear regressions are shown in black. Pearson correlation coefficients are indicated. trans TF partner redirection is required to reproduce the observed β coefficient correlations between single and combinatorial signaling sensitivities.

Analysis of Erk/NFAT target genes using hierarchical clustering and linear regression (Fig. 3 D–G) revealed a spectrum of regulatory modes by which distinct clusters of genes differentially integrate Erk and NFAT signaling inputs. We defined regulatory modes for genes that are activated (“A”) or repressed (“R”) upon stimulation and as being strongly (“s”) or weakly (“w”) dependent on Erk/NFAT signaling (Fig. 3 D–H). Up-regulated genes could be either primarily activated by Erk (A1s, A1w, A2s, and A2w) or by NFAT (A3s, A3w, A4s, and A4w) (Fig. 3 D and E), with these two subsets having highest sensitivity coefficients for Erk (β_E) and NFAT (β_N) in the linear models (Fig. 3F). Erk-biased genes include the AP-1 subunits (Fos and Fosb) and known target genes (Egr1 and Tnf), as expected (67), whereas genes activated primarily by NFAT included key regulators of effector function (Ccl3, Ccl4, and Ifng), as well as inhibitory receptors that are associated with suppression of T cell function (Pdcd1, Lag3, and Nr4a3), consistent with known roles of NFAT in driving these gene programs (56). Interestingly, while Erk and NFAT work cooperatively in some modes (A2s, A3s, and A3w) and show a positive sensitivity coefficient for combined Erk/NFAT signaling (β_EN > 0), they antagonize one another in other modes (A1s and A4s; β_EN < 0). Repressed target genes, which include canonical memory and self-renewal genes such as Tcf7, Il7r, and Sell, also showed a range of Erk and NFAT dependencies and regulatory modes (R1s-R3s, R1w-R3w, I).

Intriguingly, across all modes, we observed an inverse relationship between the sensitivity of combined Erk/NFAT signaling sensitivity (β_EN) and the sensitivities of Erk (β_E) or NFAT (β_N) alone (Fig. 3H and SI Appendix, Fig. S3 D and E). This observation suggests that when both Erk and NFAT are present, they may act together in a way that prevents one another from acting alone. In other systems (68, 69), TFs can be redirected from one set of binding sites to another by a trans-acting binding partner, raising the possibility that AP-1 and NFAT could redirect each other from sites involving other partners to those where they bind together. To evaluate the plausibility of such a mechanism, we analyzed a series of mathematical models relating NFAT and AP-1 levels in the nucleus at steady-state to their binding to cis-regulatory elements at target gene loci and the resultant rates of transcription (SI Appendix, Mathematical Appendix, section 1). We found that the observed range of sensitivities to single and combined inputs for different target genes ( β_E, β_N vs. β_EN) cannot be readily explained by having composite AP-1:NFAT binding sites alone (Fig. 3H—model 1) but most likely reflect binding of these factors at multiple single and composite sites, which could each vary in their transcriptional activity (Fig. 3H—model 2). However, having multiple sites alone would give rise to uncorrelated single and combined input sensitivities if these sites act independently from each other to control transcriptional activity. In contrast, when AP-1 and NFAT can redirect one another from single to composite sites, negative relationships between the single ( β_E, β_N), and combined (β_EN) input sensitivities arise (Fig. 3H—model 3), which agrees with experimental data. Importantly, because partner redirection occurs in trans and affects TF availability across all target genes, our model predicts that negative correlations would arise across a range of cis-regulatory architectures with different input sensitivities, and indeed, this is what we observed.

T Cells Adopt Distinct Activation Programs in Response to Variable pMHC Affinity and Dose.

pMHC inputs of different affinity and dose may elicit distinct gene programs, and do so via decoding of pMHC-dependent Erk and NFAT signaling. To test this hypothesis, we performed single-cell RNA sequencing (scRNA-seq) on CD8⁺ T cells cultured with the pMHC conditions from imaging experiments (Fig. 1) using a standard 10× Genomics platform. We analyzed 11,412 cells from five pMHC conditions, as well as a nonstimulated (rest) control. We then performed UMAP dimensionality reduction and clustering of the cells (70, 71) (Fig. 4A) using the 1,089 Erk and NFAT target genes identified from bulk RNA-seq (Fig. 3).

Fig. 4.

pMHCs of varying affinity and dose elicit distinct T cell functional programs via decoding of Erk and NFAT signaling. (A) UMAP of the 1,089 Erk and NFAT target genes from CD8⁺ T cells stimulated for 30 h with five different pMHC conditions. Cells are clustered using Leiden community detection and colored by cluster. (B) Cells from each pMHC condition are colored on the UMAP. The horizontal bars below each UMAP show the proportion of cells from the respective pMHC condition that belong to each of the eight UMAP clusters identified in A. The colors correspond to the cluster colors in A. (C) GOI expression overlaid on UMAP, and their scaled (z-score) aggregate expression by UMAP cluster shown as bar plots. Bar colors correspond to the cluster colors in A. (D) Hierarchical clustering of a single-cell gene–gene coexpression matrix of 309 Erk and NFAT target genes. The matrix is colored by the LASSO regression β coefficients and indicates the magnitude and direction of coexpression between two genes in single cells (i.e., correlation). Eight coexpression clusters are defined along the diagonal and termed “gene modules”. (E) From Left to Right: Scaled aggregate expression by pMHC condition for each gene. Color bars correspond to the pMHC color scheme used in B. Normalized effect of pMHC affinity and dose on gene expression as determined by differential expression tests in Monocle3. β_N, β_E, and β_EN values are from Fig. 3F for each gene. (F) Summary statistics for β coefficients by gene module. t test P-values are indicated for mean β coefficients significantly different from zero. (G) Enrichment of regulatory modes identified in Fig. 3 within each gene module, with P-values determined by Fisher’s exact test. (H) Precision matrices from random forest models trained on either the 154 genes comprising the eight gene modules (D and E) or the top 154 DEGs from the entire scRNA-seq dataset and used to predict the pMHC input in single cells.

Cells subject to different pMHCs conditions localized to different regions of the UMAP, indicating the presence of pMHC-specific gene programs (Fig. 4B and SI Appendix, Fig. S4 A–C). Rested cells resided in a cluster far away from all activated cells (cluster 1), reflecting the large transcriptome changes during activation. Among activated cells, those exposed to high-affinity pMHC (M9C) acquired an effector-like state (clusters 5 to 8), with upregulation of canonical effector genes (Irf4 and Il2ra) and downregulation of genes associated with memory and lymph node homing (Tcf7, Sell, and Klf2) (Fig. 4C and SI Appendix, Fig. S4D). Interestingly, cells exposed to higher doses of this pMHC showed additional gene expression changes including higher levels of inhibitory receptors (Pdcd1 and Lag3) (clusters 7 and 8). Enhanced inhibitory receptor expression at elevated doses of high-affinity pMHC may represent a protective mechanism to attenuate T cell responses and limit immunopathology during a severe infection or challenge (72). In contrast, cells exposed to low-affinity pMHC (L6F) resided in a state with less effector function and greater retention of memory programming (clusters 2 to 4). Strikingly, these cells showed enhanced expression of a distinct set of inhibitory receptors (Ctla4 and Cd47) as well as Il2rb (CD122), a marker of suppressive CD8⁺ T cells (73, 74). These inhibitory receptors up-regulated by low-affinity pMHCs may play distinct roles compared to those elevated by high doses of high-affinity pMHCs and may primarily serve to suppress autoimmune attack on healthy tissues bearing low-affinity self-antigens (75). Taken together, these results show that different pMHC dose and affinity conditions elicit different transcriptomic states in T cells with distinct effector and suppressive programs.

pMHC Affinity- and Dose-Dependent Gene Programs Reflect Decoding of Erk and NFAT Signaling.

Having identified global differences in the transcriptional programs of T cells stimulated with different pMHC inputs, we asked whether these differences arose from the decoding of long-term Erk and NFAT signaling. To do so, we sought to identify genes coexpressed in single cells, as they are likely regulated by the same upstream TFs (71, 76). We used a regression model to compute pairwise gene–gene expression correlations (71), followed by hierarchical clustering on the regression coefficients to identify groups of coexpressed genes we term gene modules (Fig. 4D). We found eight gene modules that were either activated (modules 1 to 6; M1 to M6) or repressed (modules 7 to 8; M7 to M8) by pMHC stimulation and which varied in both the direction and magnitude of regulation in response to different inputs. Notably, activated modules could be up- or down-regulated with increasing pMHC affinity (M1 to M4 vs. M5 to M6) or further up-regulated at greater pMHC dose (M4).

To test the role of long-term Erk and NFAT signaling dynamics in generating these pMHC-dependent gene expression patterns, we analyzed the single and combinatorial Erk and NFAT dependencies (β_E, β_N, and β_EN) of genes in each module obtained by signaling perturbation measurements (Fig. 4 E, Right; Fig. 4F and SI Appendix, Fig. S4F), as well as gene module enrichment in different Erk/NFAT regulatory modes (Figs. 3 D and E and 4G). We found that modules up-regulated with pMHC affinity (M1 to M4) showed positive dependencies on Erk/NFAT regulation, whereas those down-regulated with affinity (M5 to M8) showed negative dependencies (Fig. 4 E and F). These results are consistent with pMHC affinity-dependent differences arising due to stronger Erk and NFAT signaling (Fig. 2E). As increasing pMHC dose further increases NFAT but not Erk signaling (Fig. 2), we predict that modules with strong pMHC dose-dependencies also show greater reliance on NFAT signaling compared to Erk signaling. Indeed, module M4, which showed the greatest pMHC dose sensitivity (SI Appendix, Fig. S4E), also showed strong sensitivity to NFAT and the greatest sensitivity to combinatorial NFAT/Erk activity compared with Erk activity (β_N > β_E; β_EN > β_E; β_N > 0; β_EN > 0) (Fig. 4F). Furthermore, this module was enriched for genes exhibiting an NFAT-biased mode for cooperative regulation (Fig. 3–mode “A3s” and 4G), thus identifying composite NFAT:AP-1 regulation as a regulatory mode for sensing pMHC dose. Similar NFAT-dominant regulation was also observed for the moderately dose-dependent module M2 (β_N > β_E; β_N > 0), though, interestingly, this module showed considerably weaker sensitivity to combined Erk/NFAT regulation (β_EN). In contrast, module M3, which was strongly up-regulated with increasing affinity but showed no dose-dependency, showed the greatest sensitivity to Erk signaling (β_E), and was also enriched for genes exhibiting an Erk-biased cooperative mode of regulation (Fig. 3–mode “A2s”). This finding aligns with the pMHC affinity dependence and dose independence of this module, given that late Erk activity is the largest signaling difference between high- and low-affinity pMHC (Figs. 1 and 2).

Additionally, some modules showed weaker sensitivities to Erk and/or NFAT signaling, yet they showed distinctive dependencies on pMHC inputs (M1, M5, and M8), suggesting that additional trans-factors activating downstream of pMHC inputs may be responsible for the observed pMHC-dependent expression patterns. To evaluate the contribution of Erk and NFAT targets vs. targets likely regulated by other factors in defining pMHC-specific transcriptomes, we trained random forest classification models on the expression of either the 154 genes that comprise the eight Erk- and NFAT-regulated gene modules (Fig. 4 D and E), or the top 154 DEGs from the entire scRNA-seq dataset, of which 113 are not in the eight gene modules and 42 are not Erk/NFAT targets (i.e., not affected by inhibitors from Fig. 3) (Fig. 4H). These models show similar precision in distinguishing pMHC inputs, and also show comparable precision to models trained on late Erk and NFAT signaling responses (Fig. 2G). Taken together, these results indicate that Erk and NFAT signaling plays a major role in conveying pMHC information within the cell and that pMHC affinity- and dose-specific T cell transcriptional programs arise, at least in part, from decoding of long-term signaling dynamics by Erk and NFAT target genes.

Distinct cis-Regulatory Mechanisms Underlie the Decoding of pMHC-Dependent Erk and NFAT Signaling.

To gain insights into the mechanisms integrating Erk and NFAT signaling inputs to generate pMHC-specific expression patterns, we analyzed mathematical models of candidate cis-regulatory mechanisms for two well-studied T cell activation genes, Pdcd1 (encoding PD-1) and Il2ra (encoding CD25), to account for experimentally measured Erk and NFAT signaling responses and gene expression patterns. These genes were chosen for their distinct pMHC affinity and dose dependencies, as well as their inclusion in distinct modules (M4 for Pdcd1; M1 for Il2ra,), allowing us to explore a range of cis-regulatory mechanisms for decoding signaling dynamics (Fig. 5A). This approach utilizes quantitative measurements of both processes to identify and constrain models of cis-regulatory element function, allowing us to determine whether the modeled mechanisms can explain experimental observations, and to determine the plausible parameter regimes under which these mechanisms operate. We used flow cytometry to measure PD-1 and CD25 levels in T cells activated by M9C and L6F at different doses, as above (Fig. 5 B–D and SI Appendix, Fig. S5). We also perturbed cells with Erk and NFAT inhibitors after 9 h of stimulation (Fig. 5B–”Late”). However, because some cis-regulatory elements can control activation timing without affecting maintenance of gene expression (69, 77–79), as observed for Il2ra (77), we also perturbed signaling at the onset of stimulation (Fig. 5B–”Early”), as Erk and NFAT may play distinct roles in controlling activation vs. maintenance of gene expression. Based on these measurements, we developed two ordinary differential equation (ODE) models describing the regulation of Pdcd1 and Il2ra in response to time-varying NFAT and AP-1 levels. We then performed least-squares fitting of these models to experimental data, then used best-fit model parameters to perform ODE simulations of gene expression over the time course of signaling.

Fig. 5.

Temporal and combinatorial cis-regulatory mechanisms work together to decode Erk and NFAT signaling. (A) scRNA-seq UMAP (from Fig. 4A) overlaid with Pdcd1 and Il2ra expression. (B) CD8⁺ T cells were stimulated with various pMHC conditions, treated with MEKi and/or CsA or none (DMSO vehicle control) immediately (early) or 9 h (late) after stimulation, and then assayed for PD-1 and CD25 expression after 16 (early) or 30 (late) h. (C and D) Expression of PD-1 and CD25 in T cells following variable stimulation and signaling perturbation shown in Β. Experimental results are compared with our model predictions. (E) Proposed NFAT and AP-1 cis-regulatory mechanisms controlling PD-1 (Pdcd1) and CD25 (Il2ra) expression. Mathematical simulations for each pMHC input are shown for each gene, depicting the levels of active NFAT and AP-1 TFs, and the resulting expression of PD-1 and CD25.

This analysis revealed an array of regulatory mechanisms that collaborate in cis and trans to enable input-specific expression of these two genes downstream of Erk and NFAT signaling. PD-1 expression is predicted to depend on the concerted action of two elements, both of which are required for the observed dose, affinity, and inhibitor dependencies in its expression: 1) an element that binds NFAT without AP-1 and 2) a composite NFAT:AP-1 binding site that acts independently (Fig. 5E). This predicted bipartite architecture contrasts with an alternative where partnerless NFAT may play a dominant role in Pdcd1 regulation (56), and highlights how target genes may utilize multiple elements that bind TFs in different combinations to construct input-specific responses. CD25 expression is also predicted to require two separate elements, but that have distinct regulatory functions: 1) a timing element consisting of an NFAT:AP-1 composite binding site and 2) an element controlling expression levels that receives input from NFAT and an undetermined factor X (Fig. 5E). As the timing element is required for initiation but not maintenance of expression, this system enables hysteresis in its regulation, allowing target genes to activate in response to prior signaling exposure. Interestingly, after activation, CD25 expression is regulated not only by a third, unidentified trans-factor but also by NFAT whose effects are revealed only upon attenuation of Erk signaling and presumably occur by NFAT redirection by AP-1 at other gene loci (Fig. 5D and SI Appendix, Fig. S5B). These results highlight how trans-mechanisms for TF partner redirection could generate additional dependencies for decoding combinatorial signaling states generated by diverse signaling inputs.

Discussion

Using an in vitro platform to follow individual T cells responding to varying pMHC inputs, we find that T cells independently encode pMHC affinity and dose information using the long-term (9+ h) dynamics of these two pathways. Through transcriptomic profiling, we find that Erk and NFAT signaling dynamics are decoded to generate pMHC affinity and dose-specific gene programs, with multiple mechanisms by which these inputs are integrated at the cis-regulatory level. This work reveals a previously unknown role for long-timescale signaling in antigen encoding and highlights the combinatorial effects of multiple signaling pathways in encoding distinct signaling input features.

Our findings indicate that T cells initially activate uniformly and elaborate pMHC-specific signaling and gene regulatory responses only after prolonged contact with various pMHC inputs. This shared molecular program for early activation, also observed in other studies (26, 27), may enable T cells to prepare for a potential response before fully determining the nature of the threat. Such early response initiation may allow for rapid responses if a threat is indeed present, which may be critical as T cells reach the point of first cell division only ~30 h after pMHC encounter (80). The lag in pMHC dose and affinity-dependent responses may be due to time-delayed negative feedback on signaling resulting in reduced sensitivity to pMHC inputs, such that only high-affinity pMHC can overcome this feedback. Indeed, T cell signaling is subject to multiple layers of negative feedback (6, 81, 82). Some of these negative feedback loops, particularly those involving transcription of checkpoint inhibitory molecules (83, 84), may act on extended time scales similar to those we observed. Alternatively, the pMHC-specific response lag may reflect a need for T cells to integrate pMHC information over extended durations for accurate input discrimination (82, 85, 86). In this picture, late divergent Erk and NFAT signaling levels would result from prior pMHC exposure along with immediate signaling inputs, potentially through cytokine feedback loops that could provide additional inputs into these pathways (22, 82).

Our transcriptome-wide analysis, together with in-depth modeling of specific target genes (Pdcd1 and Il2ra), suggests that T cells rely on diverse regulatory mechanisms to elaborate distinct functional responses to different pMHC inputs. Erk and NFAT can either act singly or in combination and may modulate not only the expression levels of target genes but also their activation timing through enhancer elements that control all-or-none chromatin state switching (29, 69, 77, 78). Such timing control may allow cells to temporally integrate signaling dynamics to generate stable changes in gene programming (78). Based on our findings, it is likely that additional trans-factors downstream of TCR signaling, such as NF-κB (37), work alongside NFAT and AP-1 to convey pMHC information in the cell. In future work, it will be interesting to elucidate how these transcriptional factors all work together to mediate T cell antigen sensing and responses.

In future work, it will be important to investigate roles for T cell signaling dynamics in mediating antigen perception in more physiological contexts. While our ex vivo system enables analysis of T cell signaling responses to precisely defined antigen inputs, it lacks other signals encountered by T cells during an immune response, such as costimulation or cytokines. Nonetheless, T cells stimulated with this minimal system can elaborate specific gene programs that may resemble those encountered under similar antigen contexts in vivo. Low-affinity pMHCs selectively induce certain inhibitory genes, including the immune checkpoint molecule Ctla4 (Fig. 4). These programs may suppress spurious effector responses of T cells against self-antigens to maintain tolerance, and are consistent with previous reports showing that although T cells responding to lower-affinity pMHCs are able to fully activate after antigen encounter (26, 27), they are far less efficient in mediating autoimmune destruction of healthy tissues compared to those responding to high-affinity pMHCs (87). On the other hand, high-affinity pMHCs up-regulate effector genes (e.g., Irf4), down-regulate self-renewal genes (e.g., Tcf7), and at high doses further up-regulate inhibitory receptors (e.g., Pdcd1, Lag3) distinct from those up-regulated by low-affinity pMHC. Indeed, in response to infection or tumor challenge, high-affinity T cell clones skew toward effector differentiation and exhaustion, whereas lower-affinity clones preferentially form long-lived memory cells (30–32, 88). It will be interesting to investigate how pMHC inputs are encoded by signaling dynamics under these physiological contexts and how the signaling dynamics are in turn decoded to enable optimal immune system function.

Materials And Methods

Mice.

The DPR mouse strain (Rosa26^{LSL-DPR/LSL-DPR}) was generated on a B6 background by Biocytogen Corp. using a Rosa26 targeting vector (89) containing a reporter cassette encoding Erk-KTR (47) fused to mTagBFP2, H2B fused to near-infrared fluorescent protein (iRFP), and residues 1 to 399 of mouse NFATc2 fused to mRuby3 (59). Rosa26^{LSL-DPR/LSL-DPR} mice were crossed to LCMV-specific transgenic TCR (P14) mice (90) (JAX, Strain # 004694) and to Cre-ERT2 mice (JAX, Strain # 008463), generating Rosa26^{LSL-DPR/Cre-ERT2}/TcrLCMV offspring. Eight-week-old male and female mice were injected with four doses of tamoxifen over four consecutive days (91) and used for imaging experiments 4 to 12 wk later. Eight- to sixteen-week-old Rosa26^{LSL-DPR/LSL-DPR}/TcrLCMV male and female mice were used for nonimaging experiments (SI Appendix, Fig. S1A). All mice were used in accordance with IACUC guidelines for the University of Washington.

pMHC-Coated Plates.

Stimulation plates were prepared as previously described (62) with the addition of RetroNectin and anti-LFA1 to reduce T cell motility and enable long-term tracking (92). The molar quantities of pMHC indicated in the text and figures correspond to the total amount of biotinylated monomeric H-2D^b MHC, loaded with either KAVYNFATM (M9C) or KAVYNLATC (L6F) (64) and provided by the NIH tetramer facility, that was incubated in each well.

T Cell Stimulation.

CD8⁺ T cells were isolated from RBC-depleted splenocytes from mice described above using the Miltenyi CD8⁺ negative selection kit (Miltenyi Biotec, Cat # 130-104-075), yielding a combination of naive and antigen-experienced (memory) resting cells. Reporter-expressing CD8⁺ T cells were sorted using a BD Aria III and seeded into pMHC-coated plates in AIM-V media (Thermo Fisher Scientific, Cat # 12055091) supplemented with 50 µM β-mercaptoethanol.

Live-Cell Imaging.

Time-lapse images were acquired using a 63× (0.75 NA) glycerol objective on a Leica DMi8 equipped with hardware autofocus and a 37 °C, 5% CO2 incubation chamber. Fluorescence images were acquired in the BFP (405 Ex; 440/40 Em), RFP (561 Excitation; 600/50 Em), and near-IR (640 Ex, 700/75 Em) channels every hour using an LED illuminator (SpectraX, Lumencor).

Flow Cytometry.

Cells were stained in Fc blocking solution (2.4G2 supernatant) containing 1 µl anti-CD69, 0.5 µl anti-CD25, or 1 µl anti-CD279 (PD-1) per sample for 20 min at 4 °C in the dark and analyzed on an Attune Nxt cytometer (Thermo Fisher Scientific). Antibodies are listed in SI Appendix.

Bulk RNA Sequencing.

Cells were stimulated with 2 pmol M9C for 30 h. At 9 h, cells were treated with 1 µM MEK inhibitor (Trametinib, Selleckchem Cat. no. S2673), CsA (Selleckchem Cat. no. 52286), or both, or DMSO (0.1%). Paired-end library preparation and transcriptome sequencing were conducted by Novogene Co., LTD (Sacramento, CA, USA). Alignment to the GRCm38/mm10 reference genome was done with Kallisto (v0.46.1) (93). Differential expression analysis was done with the sleuth package (v0.30.0) (94) in R.

scRNA Sequencing.

Cells were stimulated for 30 h and then stained with one of six hashtag (HTO) antibodies for multiplexing. Single-cell library preparation was done with the 10× Genomics platform (95). Paired-end sequencing was performed on a NextSeq2000 (Illumina). Analysis was done with Monocle3 (v 1.2.9) (96), and gene–gene correlations were quantified using LASSO regression (71) in R.

Image Processing.

Image preprocessing, cell segmentation, and tracking were performed with a custom MATLAB (MathWorks) pipeline (79) (SI Appendix, Fig. S1B). Activity levels were determined by the ratio of nuclear (NFAT) or cytoplasmic (Erk) reporter fluorescence to the total cell fluorescence.

Computational Analysis of Erk and NFAT Signaling.

Bivariate, 14-dimensional time-series vectors were scaled from zero to one and clustered with the dtwclust package (v 5.5.11) (97) in R using the global alignment kernel distance (98), which allows for calculating distances between multivariate single-cell responses. Activity pulses were defined by at least 2 consecutive hours with dAdt > 0 and A_smooth > 95% of the mean A_smooth from all analyzed cells. Early and late integrated activities were computed using the 2 to 4 h and 8 to 15 h activities, respectively. Random forest classification models used a fivefold nested cross-validation training and validation workflow (99).

Mathematical Modeling.

Ordinary differential equation simulations of gene regulation models for Pdcd1 and Il2ra were performed using MATLAB. Parameter values for these models were determined by the fitting gene expression values obtained by modeling to experimental measurements of mean fluorescence intensities (PD-1 and CD25 Late, Fig. 5 C and D) or gene activation percentages (CD25 Early, Fig. 5D). The sum-squared error of this fit was then minimized using the Nelder–Mead simplex method. In these models, cell-to-cell variability in gene expression was modeled using a cell-extrinsic distribution via log-normally distributed differences in maximal transcriptional activity. In our model, activation times for Il2ra were exponentially distributed. For simulations, these random activation times were determined through random number generation, followed by the distribution of likelihoods.

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (XLSX)

Dataset S02 (XLSX)

Movie S1.

Live-cell microscopy of T cell stimulation. Representative time-lapse movies of individual well positions corresponding to each of the five pMHC conditions assayed in this study. Shown to the right of the movies for each condition is the mean +/- 99% confidence interval Erk and NFAT signaling responses for all cells analyzed from the specific well position shown in the movie. Nuclear and cytoplasmic segmentations are shown in dashed and solid lines, respectively. Scale bar = 5 um.

Data, Materials, and Software Availability

RNA sequencing data have been deposited in Gene Expression Omnibus (GSE242495) (100). Analysis and image processing code is available at https://github.com/kuehlab/TcellAntigenPerception_Wither-et-al-2023. The DPR mouse strain generated in this study is available from Jackson Laboratory (Stock # 039008, C57BL/6-Gt(ROSA)26Sor^tm1Hyku/J).