Early T cell receptor signals globally modulate ligand:receptor affinities during antigen discrimination

Significance

Antigen discrimination by T cells is based on subtle differences in binding of the T cell receptor (TCR) for its peptide major histocompatibility complex (pMHC) ligand. While such binding characteristics are readily mapped with great precision in reconstituted biochemical systems, it is less clear how these interactions are affected in the live cell environment. Here we utilize single-molecule imaging to individually resolve all of the pMHC:TCR binding events in live T cells. The quantitative measurements reveal an active feedback mechanism that globally modulates the probability of pMHC:TCR binding throughout the cell–cell interface, without affecting the unbinding rate. The result is to increase the efficiency with which TCRs scan for antigen pMHC after the first few molecular encounters have occurred.

Abstract

Antigen discrimination by T cells occurs at the junction between a T cell and an antigen-presenting cell. Juxtacrine binding between numerous adhesion, signaling, and costimulatory molecules defines both the topographical and lateral geometry of this cell–cell interface, within which T cell receptor (TCR) and peptide major histocompatibility complex (pMHC) interact. These physical constraints on receptor and ligand movement have significant potential to modulate their molecular binding properties. Here, we monitor individual ligand:receptor binding and unbinding events in space and time by single-molecule imaging in live primary T cells for a range of different pMHC ligands and surface densities. Direct observations of pMHC:TCR and CD80:CD28 binding events reveal that the in situ affinity of both pMHC and CD80 ligands for their respective receptors is modulated by the steady-state number of agonist pMHC:TCR interactions experienced by the cell. By resolving every single pMHC:TCR interaction it is evident that this cooperativity is accomplished by increasing the kinetic on-rate without altering the off-rate and has a component that is not spatially localized. Furthermore, positive cooperativity is observed under conditions where the T cell activation probability is low. This TCR-mediated feedback is a global effect on the intercellular junction. It is triggered by the first few individual pMHC:TCR binding events and effectively increases the efficiency of TCR scanning for antigen before the T cell is committed to activation.

T cell activation in the adaptive immune system is mediated by ligand:receptor interactions at the interface between a T cell and an antigen-presenting cell (APC). T cells are known to be extremely sensitive (1, 2) and have even been reported to trigger and produce cytokines in response to even a single peptide major histocompatibility complex (pMHC) ligand (3). T cells are also highly selective and can accurately discriminate between different pMHCs based on small differences in molecular binding properties—especially the kinetic dissociation rate (k_off) of the pMHC:T cell receptor (TCR) complex (4–9). The fundamental problem faced by any system that has both single-molecule sensitivity and precise discrimination is that molecular properties such as k_off are intrinsically ensemble averages. At the single-molecule level, each pMHC:TCR interaction consists of a discrete dwell time (τ_off), and k_off is only defined as $1/\langle {\tau }_{\text{off}}\rangle$ for a large number of events. Individual values of τ_off for pMHC:TCR binding events are roughly exponentially distributed and even a weakly binding agonist will occasionally remain bound much longer than average (10–12). Although hypotheses have been put forward, the basic question of how a T cell distinguishes the rare, genuine agonist pMHC from spurious long-lived complexes with the abundant self pMHC is not resolved (13–17).

Here, we monitor the spatial position and temporal duration of all molecular pMHC:TCR binding events during T cell antigen discrimination. Measurements are made using hybrid live T cell–supported membrane junctions, which enable controlled presentation of antigen pMHC in the context of adhesion and costimulatory molecules, all in a fluid environment that mimics many properties of a cell–cell interface (18–20). The binding state of each individual pMHC molecule is resolved based on single-molecule tracking of its lateral mobility, providing a real-time readout of the binding status of every pMHC within the junction (12).

Two key molecular binding parameters are directly revealed by these measurements: the pMHC:TCR dwell time distribution and the fraction of pMHC actually bound to TCR under each cell. The molecular dwell time distribution (related to the in situ kinetic off-rate) was observed to remain constant over a wide range of conditions. However, the fraction of bound pMHC (related to the in situ dissociation constant) exhibited distinct positive feedback at the lowest antigen densities. Progressive accumulation of antigen pMHC:TCR interactions globally increases the affinity of pMHC for TCR throughout the interface. This cooperativity occurs over micrometer distances and is thus not based on physical contact between pMHC:TCR complexes. Cooperativity is instead mediated through changes in the kinetic on-rate (k_on), which result from global changes in the geometry of the cell–cell interface induced by early TCR signaling events. Single-molecule imaging of costimulatory CD80:CD28 binding reveals that these environmental changes also increase the binding efficiency of other ligand:receptor interactions in the interface. Inside-out activation of lymphocyte function associated antigen-1 (LFA-1) on the T cell surface (21) and changes in the cortical actin cytoskeleton (18, 22) are plausible candidates for such feedback. Notably, positive cooperativity is only observed at pMHC densities where T cells are unlikely to be activated (e.g., as measured by NFAT nuclear translocation in these experiments). At progressively higher antigen levels, corresponding to conditions under which central supramolecular activation clusters are clearly visible (23), the observed cooperativity becomes negative.

These observations expose active feedback through the TCR signaling network that modulates pMHC:TCR and other molecular binding affinities in situ. One consequence of this feedback is to increase the efficiency with which TCRs scan for pMHC after the first few agonist pMHC:TCR molecular binding events have occurred, but before the decision to activate is reached. The sensitivity of the feedback mechanism to antigen thus even exceeds the already extreme sensitivity with which T cells activate in response to antigen. Another consequence of such a feedback system is to increase the probability of multiply rebinding the same agonist pMHC (24). Under situations where extremely few or even a single agonist pMHC is available, multiple rebinding events increase the precision with which molecular properties of the agonist, such as the corresponding pMHC:TCR k_off, can be determined. Feedback that favors multiple rebinding events will thus increase the ability of TCR to discriminate among similar pMHCs based on binding kinetics. More generally, these observations underscore how chemical properties, such as binding affinities, can be dynamically manipulated in the living cellular environment.

Results

Single-Molecule pMHC:TCR Binding in Live Cells.

We probe single T cell responses to TCR triggering in hybrid junctions between live primary T cells and supported lipid membranes functionalized with pMHC, the integrin intracellular adhesion molecule-1 (ICAM-1), and the costimulatory ligand CD80 (Fig. 1A). T cells spread rapidly via integrin binding to form essentially planar interfaces with the supported membrane. As reported previously (12), individual pMHC:TCR complexes can be tracked in these hybrid junctions by taking advantage of the dramatic decrease in mobility of pMHC bound to TCR in the T cell plasma membrane relative to pMHC diffusing freely in the supported membrane (Fig. 1B). Using this approach we tracked slow-moving, bound pMHC:TCR complexes and measured the single-molecule pMHC:TCR dwell time distributions for two different TCR mouse model systems (AND and 5c.c7) and a panel of peptide ligands of varying potencies (Fig. 1C). The average molecular dwell times ($\langle {\tau }_{\text{off}}\rangle$) range from <1 s to 68 s, correlate with observed peptide potency in these studies, and are comparable to reported bulk solution binding measurements on isolated proteins made using surface plasmon resonance (SPR) (4, 9, 25). This assay enables measurement of in situ dwell times below the reported detection limit of SPR measurements (9).

Fig. 1.

Monitoring single-molecule ligand:receptor binding events in living T cells. (A) Single-cell responses of T cells to TCR triggering are probed in hybrid interfaces between live T cells and supported membranes functionalized with receptor ligands (pMHC and CD80) and adhesion molecules [the integrin ligand ICAM-1, Protein Data Bank (PDB) ID code 1IAM and 1P53]. The kinetics and stoichiometry of pMHC:TCR (PDB ID code 3QIU) and CD80:CD28 (PDB ID code 1DR9 and 1YJD) are monitored using total internal reflectance (TIRF) microscopy, where bright organic fluorophores covalently coupled to the ligand (either pMHC or CD80) are used as high-contrast imaging agents. ICAM-YFP, which binds its receptor LFA-1 (PDB ID codes 2K9J and 3K72), is imaged using TIRF to observe adhesion organization. NFAT-GFP nuclear localization and ZAP70-EGFP membrane localization report on the signaling status of living T cells in response to pMHC stimulation. (B) A representative time lapse of single-molecule MCC/MHC:AND trajectories. Each color represents a different pMHC:TCR single-molecule trajectory lasting tens to hundreds of seconds. Trajectories beginning at later time points (red arrow) represent de novo ligand:receptor binding events. (C) The distribution of single-molecule pMHC:TCR dwell times (${\tau }_{\text{off}}$) is measured for different pMHC:TCR combinations. Hundreds of molecular traces from ≥10 cells are used to populate each distribution, with fits shown. In general, higher-potency pMHC ligands have longer $\langle {\tau }_{\text{off}}\rangle$. pMHC:TCR $\langle {\tau }_{\text{off}}\rangle$ were measured with a 500-ms camera exposure time and time-lapse intervals varying between 1 s and 10 s.

Direct readout of the binding status of single pMHC molecules also enables measurement of the fraction of bound pMHC within each individual T cell–supported membrane junction. In these experiments, every component of the reaction quotient ([TCR_Total], [pMHC:TCR], and [pMHC_Total]) is measured independently to calculate a parameter we here refer to as ^cellK_D, which is defined as ^cellK_D = [TCR_free][pMHC_free]/[pMHC:TCR] and is essentially the in situ dissociation constant, K_D, within the interface (Fig. 2 A and B). Although the coreceptor CD4 is not explicitly mentioned here it is present in all measurements and is expected to weakly interact with MHC to form a CD4:pMHC:TCR ternary complex (26). All observations and calculations of ^cellK_D thus intrinsically include the effects of CD4 binding.

Fig. 2.

Direct, single-molecule calculation of pMHC:TCR dissociation quotients. (A) Sequential 500-ms and 40-ms acquisitions, respectively, record the number of bound pMHC (from the 500-ms acquisition) and the total number of pMHC ligand molecules (from the 40-ms acquisition) per individual T cell. Cell outline is determined using reflection interference contrast microscopy (RICM). Representative images are from an AND T cell interacting with 0.09 molecules per micrometer MCC/MHC. Subsequent measurement of the total number of TCR in the live cell-supported membrane interface via TIRF enables calculation of pMHC:TCR K_D on a single cell basis, ^cellK_D. (Scale bar, 5 μm.) (B) ^cellK_D is measured at the single-molecule, single-cell level for the MCC/MHC:AND, T102S/MHC:AND, and MCC/MHC:5c.c7 pMHC:TCR combinations. Each circle represents a cell. Higher-potency ligands correspond with higher-affinity pMHC:TCR interactions. The pMHC density for these data sets are ∼50–300 pMHC per micrometer, ∼50–300 pMHC per micrometer, and 125 and 340 pMHC per micrometer for the MCC:AND, T102S:AND, and MCC:5c.c7 combinations, respectively. (C) Single-cell kinetic traces of fraction bound pMHC at two different MCC densities (0.09 and 0.6 μm⁻²) using AND T cells. All traces begin after 5 min of initial cell landing. Each colored line represents one cell. The black line indicates the steady-state mean. pMHCs are labeled 1:1 with Atto488 for all experiments.

Although ^cellK_D is not strictly an equilibrium parameter, typical kinetic rates of binding and dissociation (k_on and k_off) are fast compared with the timescale of experiment. Under these conditions, the fraction of bound pMHC per T cell maintains at steady state during the course of observation (Fig. 2C and Fig. S1A and Movie S1). Population average values of ^cellK_D calculated directly from single-cell measurements are comparable to equilibrium K_D measurements obtained from parametric fits to bulk measurements of pMHC:TCR binding in supported membranes for all three pMHC:TCR combinations (Fig. 2B and Fig. S1B). There is, however, large variation in the individual values of ^cellK_D measured for each cell, which is not the result of measurement error or stochastic noise (Fig. 2B). We speculated that this variation is not random but rather reflects systematic modulation of pMHC:TCR binding characteristics by cellular activity.

Fig. S1.

pMHC:TCR steady-state kinetics, bulk pMHC:TCR binding curves, and binding assay validation. (A) The density of fraction bound pMHC is the fraction bound pMHC normalized by the cell–bilayer interface area at the same time point. The cell area is determined by RICM. Each line represents one cell, and black lines show the steady state mean. (B) Bulk pMHC:TCR binding curves. The number of pMHC:TCR binding events is averaged over a population of 50–100 cells at each pMHC density measured. The insets display the same data on condensed axes. The data fit a modified Langmuir binding model, which yields the average K_D and the pMHC:TCR binding saturation level for the three pMHC:TCR combinations examined. The fit parameters are used to calculate the average number of bound pMHC per cell at a given overall pMHC density and are consistent with measured values at the lowest recorded pMHC densities. The pMHC:TCR binding saturation level correlates with pMHC:TCR binding dwell times; longer pMHC:TCR dwell times correlate with higher pMHC:TCR binding thresholds. Note that live cell pMHC:TCR binding data appear to be characterized by a single K_D when only one parameter is measured (# pMHC:TCR binding events per cell), as opposed to the varying reaction quotient observed when K_D is calculated from independent measurements of pMHC density, TCR density, and pMHC:TCR density at a given time point, as in Figs. 2 and 3. (C) Varying the ratio of unlabeled:labeled pMHC results in a proportional change in the number of pMHC:TCR binding events per cell detected using the single-molecule assay. (D) Similarly, varying the ratio of labeled:unlabeled pMHC while keeping the overall pMHC density constant results in similar numbers of overall pMHC:TCR binding events at different labeling ratios. This is true across approximately three orders of magnitude of overall pMHC density and approximately two to three orders of magnitude of labeling ratio. MCC:Atto647N in a 1:1 stoichiometry was used as a probe in B and MCC:Atto488 in a 1:1 stoichiometry with AND CD4+ T cells were used in C.

Systematic Variation of pMHC:TCR Binding Affinity.

The dependence of ^cellK_D on pMHC density was characterized by precision titrations ranging from very low pMHC densities (∼0.05 molecules per micrometer) to very high pMHC densities (∼300 molecules per micrometer). For a given pMHC density, ^cellK_D values for at least 50 cells were averaged to calculate a well-defined population average, $\langle {}^{\mathrm{cell}}K_{\mathit{D}}\rangle$ (Fig. 3A and Fig. S2A). This assay is accurate over at least four orders of magnitude of pMHC density (Fig. S1 C and D). Several distinct observations regarding the nature of ligand:receptor binding in T cell intermembrane junctions are revealed by these results.

Fig. 3.

^cellK_D varies regularly with ligand density. (A, Top) $\langle {\tau }_{\text{off}}\rangle$ remains constant during a pMHC density titration over almost four orders of magnitude in the supported membrane. (A, Bottom) Over the same pMHC density range, $\langle {}^{\mathrm{cell}}K_{\mathit{D}}\rangle$, measured using the single-molecule, single-cell assay demonstrated in Fig. 2, varies regularly, reaching an optimum at low pMHC densities. pMHC:TCR binding exhibits positive cooperativity at low ligand densities (corresponding to as few as five steady-state pMHC:TCR binding events) until reaching the optimal pMHC density. At values higher than optimal pMHC density, pMHC:TCR binding exhibits negative cooperativity. (B) NFAT translocation thresholds correspond with pMHC:TCR binding optimums. All data in A and B were recorded using the MCC/MHC:5c.c7 pMHC:TCR combination. Each circle indicates a population average, and error bars show SEM. n ≥ 50 for ^cellK_D and NFAT measurements at each condition, and kinetic measurements were performed as in Fig. 1. All data are representative of at least three biological replicates. (C) NFAT translocation is measured on a cell-by-cell basis by calculating the degree of NFAT-GFP nuclear localization within a 500-nm-thick optical section (obtained using spinning disk confocal microscopy) from the middle of the T cell. If the ratio of nuclear to cytosolic NFAT-GFP intensity (I_nuc/I_cyt) is greater than 1, then that T cell is defined as activated. Numbers in upper right of the two images at right indicate the I_nuc/I_cyt value for the cells shown. (D, Left) The distribution of bound pMHC nearest neighbors was calculated for the MCC:5c.c7 interaction at 0.1 MCC/μm². * and ** correspond to the pMHC densities depicted in B. (D, Right) Schematic of the nearest-neighbor calculation. The MCC outlined in green is the nearest neighbor of the MCC outlined in white. An image of TCR (labeled using H57 Fab conjugated to Alexa647) from the same cell demonstrates that TCR density is not locally increased at sites of pMHC binding. (All scale bars, 5 μm.)

Fig. S2.

Full titration curves for several pMHC:TCR combinations. (A) Full (bound pMHC)/(total pMHC) per cell, ^cellK_D, NFAT, and ρ(TCR_Total) titration curves for MCC:AND, T102S:AND, and MCC:5c.c7 pMHC:TCR binding. Data were collected as in Fig. 3. Each plot is representative of at least three biological replicates, except the T102S:5c.c7 ^cellK_D measurements, which are representative of one replicate. Red: MCC:AND; blue: T102S:AND; green: MCC:5c.c7; purple: T102S:5c.c7. (B) Table of kinetic constants measured in this report. The parameters optimum <^cellK_D> and optimum pMHC density were estimated using a Bayesian estimation method. Feedback strength was derived from fits to Eq. S2 in SI Materials and Methods. The NFAT thresholds are derived from fits to Eq. S1 in SI Materials and Methods. Optimum k_on is calculated from optimum <^cellK_D> and <τ_off>, where k_off = 1/<τ_off> and k_on = k_off/<^cellK_D>. Error in k_on is estimated by propagating error in <^cellK_D> and SEM in k_off. The <bound pMHC> per cell at the NFAT threshold and at the lowest pMHC density are calculated as follows: Measured values correspond to the measured number of bound pMHCs at the measured density closest to the optimum pMHC density (0.15 MCC per micrometer for the MCC:AND combination, 1.78 MCC per micrometer for the MCC:5c.c7 combination, 3.2 T102S per micrometer for the T102S:AND combination, and 8.88 T102S per micrometer for the T102S:5c.c7 combination) and at the lowest pMHC density (0.07 MCC per micrometer for the MCC:AND combination, 0.09 MCC per micrometer for the MCC:5c.c7 combination, 3.2 T102S per micrometer for the T102S:AND combination, and 0.12 T102S per micrometer for the T102S:5c.c7 combination). Predicted values are calculated according to the equation and fit parameters from Fig. S1B. ≤ and ≥ symbols for the T102S:5c.c7 combination reflect the fact that only two densities were measured due to the technical difficulty of observing this pMHC:TCR combination. Error bars in columns eight and nine reflect SD. n.d., no data. (C) Representative images showing binding events and TCR localization for 5c.c7 T cells exposed to 0.1–3.8 MCC per micrometer. TCR is labeled with Alexa647-conjugated H57 Fab at saturating conditions and TCR images are displayed with identical contrast. (Scale bar, 5 μm.)

First, pMHC:TCR binding at the lowest densities, corresponding to as few as four to five individually resolved pMHC:TCR molecular binding events per cell (Figs. S1B and S2B), exhibits distinct positive feedback. This is revealed by the systematically decreasing values of $\langle {}^{\mathrm{cell}}K_{\mathit{D}}\rangle$ observed at the lowest pMHC densities (Fig. 3A and Fig. S2A). The mean pMHC:TCR dwell time,$\langle {\tau }_{\text{off}}\rangle$, remains constant over the same pMHC density range, indicating that the pMHC:TCR k_on is modulated (Fig. 3A and Fig. S2B).

Second, pMHC:TCR binding exhibits positive cooperativity without physical contact between pMHC:TCR complexes. Every pMHC:TCR molecular complex is directly resolved and is observed to be spaced microns apart at the lowest pMHC densities tested (Figs. 1B, 2A, and 3D and Fig. S2C).

Third, regardless of the stimulating pMHC ligand or the T cell clone, $\langle {}^{\mathrm{cell}}K_{\mathit{D}}\rangle$ reaches maximum affinity at the lower pMHC density ranges tested (1.35–5.75 molecules per micrometer) (Fig. S2 A and B). The density at which $\langle {}^{\mathrm{cell}}K_{\mathit{D}}\rangle$ reaches its maximum is here referred to as the optimal pMHC density. At progressively higher pMHC densities pMHC:TCR binding exhibits negative cooperativity (Fig. 3A and Fig. S2A). Thus, under conditions at which TCR microclusters are readily observed (22, 23, 27) (Fig. S2C), pMHC:TCR binding is anticooperative—contrary to common assumption.

The pMHC densities at which maximum pMHC:TCR affinity is measured coincide with T cell activation thresholds for all three pMHC:TCR combinations (Fig. 3B and Fig. S2A). NFAT translocates to the nucleus in response to sustained calcium release, and confocal microscopy imaging of NFAT subcellular localization provides a visual, binary readout of individual T cell activation (Fig. 3C). For each peptide, the activation threshold is the density at which half of the maximum fraction of cells translocate NFAT after 30 min of pMHC stimulation (28) (Fig. 3B and Fig. S2 A and B). At pMHC densities below the activation threshold, where T cells are scanning for and engaging antigen but have a low probability of activation (NFAT translocation), we measure positive cooperativity in pMHC:TCR binding. At higher pMHC densities, where T cells are likely to be activating, we measure negative cooperative binding between pMHC and TCR (Fig. 3 A and B and Fig. S2A).

Finally, the feedback strength, which is reflected in the slope of change in $\langle {}^{\mathrm{cell}}K_{\mathit{D}}\rangle$ with pMHC density, depends on $\langle {\tau }_{\text{off}}\rangle$ of the stimulating ligand (Fig. S2 A and B). Thus, there is kinetic discrimination in this effect, which suggests TCR triggering is involved. Here, we refer to these TCR-mediated effects that are initiated by as few as five pMHC:TCR binding events and are insufficient to induce NFAT activation as “early TCR signals” (Fig. S2). This TCR-mediated feedback depends on pMHC:TCR strength (Fig. S2B), but the precise signaling pathways remain to be mapped.

TCR-Mediated Feedback Modulates CD80:CD28 Binding Affinity.

Modulation of pMHC:TCR affinity occurs through changes in k_on of the pMHC:TCR interaction. k_on is a contextual parameter that is intrinsically affected by the intermembrane environment (29–31). We therefore hypothesized that the first few pMHC:TCR binding interactions could trigger changes in the T cell:APC interface, which in turn increased k_on of subsequent pMHC:TCR binding. Such a general morphological mechanism should also produce similar effects on other juxtacrine ligand:receptor interactions at the interface. Both CD80:CD28 and pMHC:TCR complexes have intermembrane distances of ∼13 nm and bind with comparable solution affinities in the low micromolar range (25, 32–34) (Fig. 4A). Therefore, if a general change in membrane morphology is enhancing pMHC:TCR binding, this effect should also be experienced by other, similarly sized intermembrane ligand:receptor complexes, such as CD80:CD28.

Fig. 4.

pMHC:TCR feedback enhances CD80:CD28 binding. (A) Experimental schematic. pMHC:TCR binding enhances CD80:CD28 affinity, and this effect is not reciprocal. (B) CD80:CD28 binding efficiency increases when AND T cell clones are stimulated by MCC/MHC at optimal density (0.6–0.7 molecules per micrometer) relative to stimulation at low density (0.05–0.10 molecules per micrometer). A CD80-SNAP fusion is labeled 1:1 with the Atto488 fluorophore via the SNAP tag and is kept at a constant density of 0.13–0.24 CD80-SNAP per micrometer on the bilayer. MCC/MHC is unlabeled. Data from ≥100 cells from three separate mice were used to populate the histogram for each condition. (C) The distribution of bound CD80 nearest neighbors was calculated at 0.18 CD80-SNAP per micrometer. (D) Addition of unlabeled CD80 to the supported membrane (at ∼200 molecules per micrometer) has a negligible effect on MCC/MHC:AND binding affinity over a range of pMHC densities. (E) Addition of unlabeled CD80 to the supported membrane lowers the NFAT threshold density for the MCC/MHC:AND combination. Each circle indicates a population average, and error bars show SEM. n ≥ 50 for ^cellK_D and NFAT measurements at each condition.

We varied the unlabeled pMHC density and monitored single-molecule binding kinetics of the CD80:CD28 costimulatory interaction using the same imaging strategy applied to pMHC:TCR (Fig. 4A). Observations of individual CD80:CD28 binding events reveal a similar increase in binding efficiency at the same pMHC densities that maximized pMHC:TCR affinity for both MCC and T102S peptides (Fig. 4B and Fig. S3 A and B). Histograms for each condition are populated from AND CD4+ T cell clones from three separate mice. Well-resolved CD80:CD28 binding events are spaced microns apart (Fig. 4C and Fig. S3C) and their intensity distribution remains constant when pMHC density is varied (Fig. S3C), which demonstrates that the increase in CD80 affinity is not due to enhanced CD28:CD80 dimerization (35). Notably, this cross-talk effect of pMHC:TCR binding on CD80:CD28 affinity is not reciprocal. Addition of (unlabeled) CD80 does not appreciably shift pMHC:TCR $\langle {}^{\mathrm{cell}}K_{\mathit{D}}\rangle$ [the difference in $\langle {}^{\mathrm{cell}}K_{\mathit{D}}\rangle$ minima with and without CD80 (∼0.15) is within the SE in the pMHC titration measurement (0.11–0.15)], indicating that CD80:CD28 binding does not contribute to the cooperative effect (Fig. 4D). Addition of CD80 lowers the NFAT translocation threshold density for the agonist MCC ligand (36), but this effect is less prominent for the weaker T102S ligand (Fig. 4E and Fig. S3D). CD80:CD28 complexes travel along linear trajectories, confirming effective engagement with the T cell cytoskeleton (Fig. S3E). Similar to pMHC:TCR, the CD80:CD28 $\langle {\tau }_{\text{off}}\rangle$ is constant as a function of pMHC density (Fig. S3F).

Fig. S3.

Costimulatory signals modulate T cell NFAT thresholds and pMHC:TCR binding. (A) T102S/MHC ligand modulates CD80 binding efficiency in AND T cells. The distribution of fraction bound CD80 per cell at optimal T102S/MHC density (3.2 T102S per micrometer) shifts higher compared with that at low density (0.1 T102S per micrometer). The same CD80-SNAP fusion as in Fig. 4 was used, and T102S/MHC is unlabeled. n = 50 for each histogram. Data are from one experiment. (B) Fraction of bound CD80 as a function of MCC-MHC density. n = 30 for each data point and representative of at least two biological replicates. (C, Top) Representative images showing total and bound CD80, TCR localization, and cell–bilayer interface for AND T cells exposed to 0.1–0.6 MCC per micrometer. TCR is labeled with Alexa647-conjugated H57 Fab at saturating conditions and TCR images are displayed with identical contrast. (Scale bar, 5 μm.) (C, Bottom) Intensity distribution of bound CD80-SNAP-Atto 488 puncta at 0.1 and 0.6 MCC per micrometer. (D) Addition of CD80 to the supported membrane at a density of 200 molecules per micrometer shifts the NFAT translocation threshold for MCC/MHC:AND, but not T102S/MHC:AND interactions. Error bar shows SEM. n = 50 for each data point and are representative of at least two biological replicates. (E) Kymographs of CD80-SNAP-Atto488 molecules follow linear trajectories, indicating engagement with the T cell cytoskeleton. (Scale bar, 1 μm.) (F) CD80:CD28 dwell times are constant with respect to pMHC density. CD80-SNAP-Atto488 ligand was imaged at a 3-s time lapse and analyzed as in Fig. 1. Each data point is calculated from hundreds of single-molecule trajectories from ∼10 cells from one experiment.

These results reveal that positive feedback generated by early pMHC:TCR binding events also increases k_on for CD80:CD28, while CD80:CD28 binding exhibits no reciprocal effect on pMHC:TCR. Since CD80:CD28 and pMHC:TCR complexes are physically similar, and are therefore expected to exert similar mechanical perturbations on the intermembrane environment, the lack of reciprocal binding cooperativity suggests a passive physical mechanism of membrane pinning is not responsible for the observed cooperativity. Instead, the mechanism appears to involve morphological changes in the interface triggered by signaling activity of pMHC:TCR.

Cytoskeleton and Integrin Signaling Contribute to pMHC:TCR Affinity Enhancement.

Both pMHC:TCR and CD80:CD28 complexes clearly engage the cytoskeleton at densities below NFAT threshold where affinity enhancement is observed. Additionally, integrins play an important role in establishing the physical geometry of the intercellular interface. Both effects could modulate ligand:receptor binding, and we examine these possibilities here with inhibitor studies. We probe the role of cytoskeleton activity in pMHC:TCR affinity enhancement using the small-molecule inhibitor Latrunculin A (LatA), which disrupts actin polymerization by binding G-actin. Effects of integrin signaling were probed using GGTI-298, a geranylgeranyltransferase I inhibitor that targets GTPase Rap1 (37). Rap1 activates inside-out signaling via interactions with LFA-1 (21).

Dose titrations of each inhibitor were performed at pMHC densities close to and below the maximum affinity for the MCC/MHC:AND interaction (Fig. 5A). Titrating the dose of GGTI-298 resulted in a monotonic decrease in MCC:TCR affinity enhancement, and titrating the dose of LatA leads to a similar, but smaller, decrease in affinity enhancement. At the optimum density, phalloidin staining after T cell fixation revealed the expected enrichment of F-actin at the T cell periphery in the absence of LatA and a relatively even distribution of F-actin in the presence of LatA (Fig. S4A). Inclusion of a recombinant ICAM-YFP fusion in the supported membrane revealed the expected ring-like ICAM distribution at the periphery of control T cells (Fig. S4B). Rap1 inhibition not only disrupted this distribution of ICAM, resulting in a relatively even ICAM-YFP distribution across the T cell, but also decreased the probability of the T cell’s landing on the supported membrane. These cytoskeletal and adhesion effects may be related to density- and $\langle {\tau }_{\text{off}}\rangle$-dependent trends in T cell landing on pMHC-conjugated supported membranes (Fig. S4C). Inhibition of either actin polymerization or Rap1 activity alters the pMHC:TCR affinity modulation and thus indicates a mechanistic role for both cytoskeleton and integrin inside-out signaling in pMHC:TCR affinity enhancement.

Fig. 5.

Cytoskeleton and integrin signaling contribute to pMHC:TCR affinity enhancement. (A) The fold change in the fraction of bound MCC-Atto488/MHC at the optimum MCC density relative to the lowest MCC density decreases in the presence of Rap1 and actin inhibition for AND T cells. n ≥ 15 for each condition. Error bars indicate SEM. (B) ZAP70-EGFP speckles in AND T cells are imaged using TIRF microscopy. Each blue dot represents a ZAP70-EGFP speckle; the positions of every ZAP70-EGFP feature collected over an ∼5-min window are projected onto a single image for each cell. Linear ZAP70-EGFP trajectories are visible in T cells exposed to either MCC/MHC or T102S/MHC. These trajectories are centrosymmetric in the case of MCC/MHC and asymmetric in the case of T102S/MHC. (C, Left) An AND T cell scavenges for MCC/MHC over an ∼120-s period. An edge detection algorithm using ZAP70-EGFP in TIRF as a contrast agent detects the cell outline. Individual MCC-Atto647N/MHC molecules (blue dots) are tracked in TIRF. (C, Top Right) Cell movement corresponds with an increase in pMHC binding. *, **, and *** denote time points as indicated in the images at Left. (C, Bottom Right) The pMHC molecule indicated by the red dot at t = 94 s reveals localized recruitment and correlated movement of ZAP70-EGFP at the site of pMHC:TCR binding. (Scale bar, 5 μm.)

Fig. S4.

pMHC:TCR affinity enhancement has a signaling component and is related to cell adhesion. (A) AND T cells presented with MCC/MHC were treated with LatA or vehicle then were fixed and stained with Phalloidin-Alexa568 (Top). For each of the indicated cells, a normalized intensity profile is shown across the diameter of the cell (Middle). (Scale bar, 10 μm.) The fraction of T cells with ring-like actin morphologies as determined by Phalloidin-Alexa568 were hand-counted and tabulated (Bottom). The inhibitor concentration is indicated along the x axis. Data correspond to Fig. 5A. (B) AND T cells treated with GGTI or vehicle were imaged on supported membranes presenting pMHC and ICAM-YFP. The ICAM-YFP signal is shown (Left). GGTI concentration is indicated (Upper Left). Cells in each condition were masked to measure the interfacial area. Histograms of the contact areas are shown (Right). (Scale bar, 10 μm.) The fraction of T cells with ring-like ICAM morphologies were hand-counted and tabulated (Bottom). Data correspond to plot in Fig. 5A. (C) pMHC density was titrated from 0.1 to 100 molecules per micrometer for the MCC and T102S peptides, and the density of AND CD4+ T cell clones on the bilayer was measured using RICM. Identical numbers of cells were injected into the flow chamber for each condition tested. Even at the highest densities tested, negligible numbers of AND T cell clones land on bilayers functionalized with pMHC loaded with the low potency ER60 peptide ($\langle {\tau }_{\text{off}}\rangle$ ∼1 s). n = ∼50 for each condition, from one experiment. (D) ZAP70-EGFP introduced to AND T cells via retrovirus was imaged using TIRF microscopy. Bulk ZAP70-EGFP intensity in the entire TIRF volume was integrated per cell using an automated algorithm and calibrated using single-molecule ZAP70-EGFP bleaching events. MCC/MHC and T102S/MHC densities were calibrated using the binding curves in Fig. S1B. n = ∼25–50 for each data point. Data are averaged from three separate biological replicates; error bars represent SEM.

TCR-Mediated Morphology, Adhesion Dynamics, and Proximal Signaling.

The decreasing trend in ^cellK_D before NFAT activation indicates that T cells become more responsive to their surroundings after a few initial binding events. This enhanced sensitivity is also reflected in changes in cell morphology, adhesion, and proximal signaling at pMHC densities where NFAT activation is not observed. At T102S/MHC densities well below NFAT threshold, T cells adopt an asymmetric, crawling morphology and exhibit low levels of Zeta-chain-associated protein kinase 70 (ZAP70) recruitment to the plasma membrane (Fig. 5B and Fig. S4D). Identical T cells exposed to MCC/MHC densities above NFAT translocation threshold exhibit a stationary, centrosymmetric cell morphology and enhanced ZAP70 recruitment. T cell crawling leads to binding of fresh pMHC and correlated ZAP70-EGFP recruitment in the newly engaged region of the supported membrane (Fig. 5C). The transition between these modes of behavior is mediated by TCR triggering and correlates with ^cellK_D changes reported here.

Discussion and Conclusion

Even at pMHC densities well below NFAT translocation thresholds we observe T cells to exhibit a global response to antigen that modulates the binding affinities of ligand:receptor complexes within the interface. We have previously reported that individual pMHC:TCR binding events have been observed to elicit macroscopic changes in cytoskeleton behavior (12). This effect, also readily observed in the data presented here (Fig. 1B and Fig. 5), illustrates how different signaling pathways triggered by the TCR can have different set points for activation. We find that, depending on the pMHC:TCR interaction and the composition of the T cell clone, at the lowest densities measured (corresponding to as few as approximately five steady-state pMHC molecular binding events per cell), T cells can activate retrograde transport of pMHC:TCR complexes and induce an increase in affinity for pMHC. Although the effects we report here are not spatially localized, this positive feedback may be enhanced by mechanisms dependent on close physical proximity of pMHC:TCR complexes (38) and by interplay between local Ca²⁺ release and actin polymerization at the supported membrane:T cell interface (39). By contrast, at densities that result in half-maximal NFAT nuclear translocation in both AND and 5c.c7 systems, T cells have at least 30 simultaneous pMHC:TCR engagement events per cell (Fig. S2B). Addition of the costimulatory molecule CD80 shifts NFAT triggering thresholds to lower pMHC densities (Fig. 4E).

The ability of T cells to respond to a small number of pMHC (here observed by modulating k_on) without fully activating may provide a mechanism by which individual T cells reduce their error rate during antigen detection. This would enable T cells to transiently slow down and check for more antigen on a particular APC, or to rebind the same antigen for multiple measurements of its dwell time. The cell may keep crawling if the initial binding events do not lead to sufficient additional pMHC:TCR binding and subsequent T cell activation (Fig. 5C). Such a mechanism would favor efficient scanning for antigen while allowing the cell to slow for a closer look if even just a few binding events above a certain temporal threshold are detected. T cells also express over 30 costimulatory ligands, depending on the differentiation state and physical context of the T cell (40). The TCR-gated affinity enhancement observed here is likely to affect other costimulatory receptors as well.

Alternatively, it is possible that the affinity modulation we observe is a reflection of the physical mechanism of TCR signaling, without a specific physiological function of its own. The molecular interactions between TCR and its pMHC ligand are intrinsically coupled to mechanical aspects of the cell, through the cytoskeleton and the intermembrane interface. The experimental observations made here provide highly quantitative information on the in situ molecular binding events between pMHC and TCR and thus capture reflections of other features of the signaling activity. This general line of study has the ability to expose aspects of TCR signaling and T cell ligand discrimination not readily resolved by more classical experiments.

Materials and Methods

Peptides (MCC, T102S, K3, and ER60), MHC, CD80, ICAM-1, ICAM-YFP, supported membranes, and T cells were prepared as described previously (12, 36, 41). CD80 fused to decahistidine-tagged SNAP_f (GenScript) was expressed in ES-sf9 cells and purified by Ni²⁺-nitrilotriacetic acid agarose affinity (Qiagen) with 1 mM cysteine. T cells were transduced as described previously with Zap70-EGFP (12) or NFAT-GFP (42). All mouse work was approved by Lawrence Berkeley National Laboratory Animal Welfare and Research Committee under the Animal Use Protocol 17702. T cells were pretreated in suspension with Latrunculin A (Sigma) or GGTI-298 (Sigma) for 15 min or 1 h, respectively, before exposure to the lipid bilayer. Single-molecule and confocal imaging experiments were performed on separate motorized inverted microscopes as described previously (12, 43). Single-molecule data were analyzed using a custom-written particle analysis suite developed in MATLAB (The MathWorks) and NFAT data were analyzed using ImageJ (44). More detailed methods are provided in SI Materials and Methods.

SI Materials and Methods

DNA, Protein, and T Cell Preparation.

Bihexahistidine-tagged MHC class II I-Ek protein and decahistidine-tagged ICAM-1 were produced and purified as previously described. A decahistidine-tagged CD80 protein (a gift of M. Davis, Stanford University, Palo Alto, CA) was expressed in High Five cells and purified on a Ni²⁺-nitrilotriacetic acid (NTA) agarose (Qiagen) affinity column, eluted with an imidazole gradient, dialyzed, and stored in Tris buffer containing 10% glycerol. A plasmid containing CD80 fused to decahistidine-tagged SNAP_f (GenScript) was subcloned into a pFastBac1 vector (Life Technologies). The protein was expressed in ES-sf9 cells and purified by Ni²⁺-NTA agarose affinity (Qiagen) with 1 mM cysteine. The fusion protein was labeled with a fourfold molar excess of SNAP-Surface 488 (New England BioLabs). Excess dye was removed using a Vivaspin centrifugal concentrator (GE Healthcare) and the labeling efficiency was determined to be 60–65% via UV-visible (Nanodrop 2000 Spectrophotometer; Thermo Fisher Scientific). AND CD4+ T cells and 5 c.c7 CD4+ T cells were purchased from Jackson Lab and Taconic Farms, respectively. All T cells were harvested and cultured essentially as previously described (12).

A plasmid containing EGFP fused to CD3 zeta-chain-associated protein of 70 kDa (Zap70-EGFP) was a gift of Takashi Saito, RIKEN Research Center for Allergy and Immunology, Yokohama, Japan (27). The Zap70-EGFP gene was amplified by PCR and subcloned into a murine stem cell virus parent vector (pMSCV). The pMSCV-NFAT1_(1–460)-GFP plasmid contains a fusion of GFP with a truncated murine NFAT1 (NFATc2) gene that lacks most of the DNA-binding domain but incorporates the regulatory domain, which governs nucleocytoplasmic shuttling in NFAT. The plasmid was a gift of Francesco Marangoni, Harvard Medical School, Boston (28). T cells were transduced with Zap70-EGFP as described previously (12) or NFAT-GFP using retroviruses harvested from the Platinum Eco packaging cell line (42). Retroviral transduction with the undiluted pMSCV-NFAT-GFP supernatants was performed by spinfection on days 3 and 4 after the initial splenocyte harvest. ZAP70-EGFP– or NFAT-GFP–positive cells were sorted using FACS according to viability and GFP/EGFP expression. The population of ZAP70-EGFP transduced cells that was used expressed EGFP at no more than 50% of the highest EGFP level in the overall EGFP-positive population. The entire NFAT-GFP–positive population was used for imaging.

Peptide Purification and Labeling.

Using the basic sequence of moth cytochrome c (amino acids 88–103) and previously described variants (10), the following peptides were synthesized by David King at the Howard Hughes Medical Institute Mass Spectrometry Laboratory at the University of California, Berkeley and/or commercially (Elim Biopharmaceuticals): MCC (ANERADLIAYLKQATK), MCC(C) (ANERADLIAYLKQATKGGSC), K3(C) (ANERADLIAYPKAATKFGGSC), T102S (ANERADLIAYLKQASK), T102S(C) (ANERADLIAYLKQASKGGSC), and ER60(C) (GFPTIYFSPANKKLGGSC). For fluorophore labeling, cysteine-containing peptides were dissolved in a small amount of phosphate buffer and mixed in a 1:2 molar ratio with Atto 647N resuspended in a small amount of 1-propanol or lyophilized Atto 488 (Atto-Tec GmbH) and labeled using maleimide-thiol chemistry. The peptides were then incubated at room temperature for at least 1 h and purified on a C18 reverse phase column (Grace–Vydac) and H₂O:acetonitrile gradient using ÄKTA explorer 100 FPLC system (Amersham Pharmacia Biotech). Peptide identity was confirmed after purification using mass spectrometry.

Microscopy.

TIRF experiments were performed on a motorized inverted microscope (Nikon Eclipse Ti-E; Technical Instruments) equipped with a motorized Epi/TIRF illuminator, motorized Intensilight mercury lamp (Nikon C-HGFIE), Perfect Focus system, and a motorized stage (Applied Scientific Instrumentation MS-2000). A laser launch with 488- and 640-nm (Coherent OBIS) diode lasers was controlled by an OBIS Scientific Remote (Coherent Inc.) and aligned into a fiber launch custom built by Solamere Technology Group, Inc. Laser powers measured at the sample were 0.8 mW (640 nm) and 0.5 mW (488 nm) for 500-ms exposures and 4.4 mW (640 nm) and 5 mW (488 nm) for 40-ms exposures. A dichroic beamsplitter (z488/647rpc; Chroma Technology Corp.) reflected the laser light through the objective lens and fluorescence images were recorded using an EM-CCD (iXon 897DU; Andor Inc.) after passing through a laser-blocking filter (Z488/647M; Chroma Technology Corp.). Exposure times, multidimensional acquisitions, and time-lapse periods for all experiments were set using Micro-Manager (45). A TTL signal from the appropriate laser triggered the camera exposure.

A 500-ms exposure time was used for all measurements of pMHC:TCR and CD80:CD28 $\langle {\tau }_{\text{off}}\rangle$; exposure times for ^cellK_D measurements are described below. Time lapses for $\langle {\tau }_{\text{off}}\rangle$ measurements varied between 1 s and 10 s, depending on the dwell time of the ligand. Intervals of 10 s were used for the ligand:receptor combinations with the longest dwell times to avoid photobleaching effects; shorter 1-s intervals were used for the shortest dwell time combinations, and intermediate (2, 3, 5, and 7 s) intervals were used for ligand:receptor combinations with intermediate dwell times. In cases where a longer time-lapse interval was used, $\langle {\tau }_{\text{off}}\rangle$ values calculated after photobleaching correction from measurements made with different intervals (e.g., 3 and 7 s) were confirmed to be identical. The relatively long length of the exposure time was corrected in the $\langle {\tau }_{\text{off}}\rangle$ value by including a correction equation for each time-lapse interval. For instance, a time lapse interval of 2 s yields the following relationship between frame number and time: Time (seconds) = 2.25*(no. of frames) − 2. A time-lapse interval of 5 s is corrected using Time (seconds) = 5.25*(no. of frames) − 5.

Spinning disk confocal measurements of NFAT-GFP nuclear localization were performed on a second motorized inverted microscope (Nikon Eclipse Ti-E, Technical Instruments), as described previously (43).

Imaging Chamber and Supported Lipid Bilayer Preparation.

Small unilamellar vesicles (SUVs) were formed by tip sonication of a solution composed of 98 mol % 1,2-dioleoyl-sn-glycero-3-phosphocholine and 2 mol % 1,2 dioleoyl-sn-glycero-3-[(N-(5-amino-1-carboxypentyl) iminodiacetic acid) succinyl] (nickel salt) (Ni2+-NTA-DOGS) (Avanti Polar Lipids) in Mill-Q water (EMD Millipore). Tip sonication was preferred to vesicle extrusion due to the introduction of significant levels of fluorescent impurities into the SUVs during extrusion. Before experiments, number 2 40-mm-diameter round coverslips were ultrasonicated for 30 min in 50:50 isopropyl alcohol:water, rinsed thoroughly in Milli-Q water (EMD Millipore), etched for 5 min in piranha solution (3:1 sulfuric acid:hydrogen peroxide), and again rinsed thoroughly in Milli-Q water. The coverslips were used in the assembly of FCS2 Closed Chamber Systems (flow cells; Bioptechs), which were prefilled with Tris-buffered saline (TBS, 19.98 mM Tris, 136 mM NaCl, pH 7.4; Mediatech Inc.). SUVs were then flowed into the chambers, and bilayers were allowed to form for at least 30 min. The bilayers were rinsed once with TBS, incubated for 5 min with 100 mM NiCl₂ in TBS, rinsed with TBS, and then rinsed with a T cell imaging buffer composed of 1 mM CaCl₂, 2 mM MgCl₂, 20 mM Hepes, 137 mM NaCl, 5 mM KCl, 0.7 mM Na₂HPO₄, 6 mM d-glucose, and 1% wt/vol BSA. Twenty-four hours before experiments, MHC was loaded with peptide at 37 °C in a buffer composed of 1% wt/vol BSA in PBS and brought to pH 4.5 with citric acid. Unbound peptide was separated from pMHC using 10,000 molecular-weight cutoff spin concentrators (Vivaspin 500; GE Healthcare) and then pMHC was diluted in imaging buffer. ICAM1 and pMHC were further diluted with imaging buffer, introduced into the flow cells, and incubated for 35 min followed by a rinse with imaging buffer. T cells were resuspended in imaging buffer and added to the flow cells 35 min after the final rinse and imaged immediately for 30–60 min. To visualize TCR, T cells were incubated in a solution of 4 μL Alexa 647 (Invitrogen Inc.)-labeled H57 anti-TCR Fab and 200 μL imaging buffer for 20 min at 4 °C and rinsed with imaging buffer before the regular imaging buffer resuspension. All other incubations during this protocol were performed at room temperature, and imaging experiments were performed at 37 °C.

NFAT Translocation Assay.

To measure the nuclear translocation of NFAT-GFP, confocal images of T cells were acquired at 4 μm above the coverslip. For each pMHC density, 40–100 cells were imaged at 10–30 min after cell injection. All NFAT-GFP images are analyzed using ImageJ. For each cell, the cytosol and nucleus were identified and masked off as different regions, and the average intensities of the two regions were measured and the ratio <I_Nucleus>/<I_Cytosol> was calculated. All images with the masked regions were inspected visually. When NFAT is fully cytoplasmic, the ratio is less than 1. When NFAT is fully translocated into the nucleus, the ratio is greater than 1.

NFAT titration curves were fit to the following equation:

$$P\left(x\right)=\frac{P_{max}}{1+{10}^{b\ast \left(x-c\right)}}+d,$$ [S1]

where x is log₁₀(pMHC density). These fit parameters, such as the NFAT threshold density and its associated error, are based off of an ad hoc interpolating function and are not related to physical parameters.

Inhibitor Experiments.

T cells were pretreated with Latrunculin A (Sigma) or GGTI-298 (Sigma) for 15 min or 1 h, respectively, in suspension, before exposure to the lipid bilayer. For phalloidin staining, T cells were fixed in imaging chambers with ice-cold 2% paraformaldehyde for 20 s. Following a wash with 1× PBS, samples were permeabilized with 0.05% Triton X-100 for 20 s. Samples were stained with 1 nM Phalloidin-Alexa568 (Thermo) for 15 min before washing and visualization. Effects of GGTI-298 on T cell contact with the supported membrane were monitored by imaging ICAM-YFP.

Interfacial contact areas were measured from binary masks of single cells in the 488 channel. Cells were selected using the DipImage toolbox for MATLAB and binarized using standard filtering methods.

Single-Cell Binding Assay.

Single-cell binding was measured by taking a series of images, one in each condition: short exposure time (40 ms) in the pMHC channel, long exposure time (500 ms) in pMHC channel, long exposure time (500 ms) in the anti-TCR Fab channel, and 30 ms in the RICM channel. For each pMHC density, 25–50 cells were imaged at 10–30 min after cell injection. All single-cell bindings were analyzed using MATLAB, as described below.

The optimal ^cellK_D and pMHC densities were estimated by fitting the data around the minimum to a second-order polynomial using a Bayesian estimation method (Markov chain Monte Carlo, or MCMC). A standard Metropolis MCMC algorithm was run for 50,000 iterations, after which 1,000 fitted parameter vectors were obtained. The resulting samples were used to obtain posterior means and 95% confidence intervals for both ^cellK_D and pMHC; 95% confidence intervals for ^cellK_D and SEM in k_off were propagated to calculate the error in k_on (46).

All ^cellK_D curves in the manuscript have also been fit to the following empirical equation:

$$f\left(x\right)=\frac{p_1{x}^3+p_2{x}^2+p_{3}x+p_4}{x+q_1},$$ [S2]

where x is pMHC density in molecules per micrometer. Since this effect is mediated through TCR signaling, and is not a passive biophysical mechanism, there is no physical basis to derive an equation to fit these data. All physical parameters are derived from fitting second-order polynomials to the data close to the mimimum, as described above.

Data Analysis.

Single-molecule diffraction-limited spots were detected in raw .tif image stacks of agonist pMHC labeled with MCC-Atto488 and Atto647N molecules by filtering for both size and intensity and linked into tracks using published particle detection and tracking algorithms (47) adapted for MATLAB (The MathWorks) by Daniel Blair and Eric Dufresne (site.physics.georgetown.edu/matlab/; accessed August 16, 2012). Size and intensity thresholds were first determined by eye using a test dataset and then applied uniformly to all data collected with the same exposure time and incident laser intensities. Single-molecule behavior was confirmed by step photobleaching detected in an automated way using a Bayesian change point detection algorithm (48).

Bright ZAP70-EGFP features were detected and tracked using the same algorithm as is used for single-molecule pMHC. Cells were imaged between 0 and 30 min after sample injection. Bulk ZAP70-EGFP intensity per T cell was integrated using MATLAB. ZAP70-EGFP intensity per cell was calibrated to molecules per cell using single-molecule intensities from kinetic traces, as in ref. 12. Cells were imaged between 5 and 30 min after sample injection.

Lifetime distributions are roughly exponential and of the form $f\left({\tau }_{obs}\right)={\langle {\tau }_{obs}\rangle }^{-1}{e}^{-{\tau }_{obs}/\langle {\tau }_{obs}\rangle }$_, where τ_obs is the observed dwell time in our experiments. The individual kinetic transitions were derived assuming the following model:

$$pMH{C}_{f\mathit{a}st-mobility}\underset{k_{off}}{\overset{k_{on}}{\rightleftharpoons }}pMH{C}_{slow-mobility}\stackrel{k_{\text{bl}}}{\to }pMH{C}_{\text{bleached}},$$

where pMHC_{fast-mobility} is the fast-mobility state, pMHC_{slow-mobility} is the slow-mobility state (or the TCR stably bound state), pMHC_bleached is the bleached slow-mobility state, k_off and k_on are the rates of transitions between the slow-mobility state and the fast-mobility state, and k_bl is the rate of transition from the slow-mobility state to the bleached state. We assume that the transitions between the kinetic states represent a Markov memoryless process and use probability theory to derive a probability density function (PDF) for the single-molecule dwell time distribution described by the above model. The corresponding PDF describing observed dwell time distribution is as follows:

$$f\left({\tau }_{obs}\right)=\left({\langle {\tau }_{bl}\rangle }^{-1}+{\langle {\tau }_{off}\rangle }^{-1}\right)e^{-{\tau }_{obs}\left({\langle {\tau }_{bl}\rangle }^{-1}+{\langle {\tau }_{off}\rangle }^{-1}\right)},$$

where ${\left({\langle {\tau }_{bl}\rangle }^{-1}+{\langle {\tau }_{off}\rangle }^{-1}\right)}^{-1}=\langle {\tau }_{obs}\rangle$ is the observed mean dwell time in our experiments. Agonist pMHC labeled with Atto488 and Atto647N SLB bleaching curves were background subtracted and then fit to an exponential decay function of the form $f\left(t\right)=k_{bl}{e}^{-k_{bl}t}$. Fitting was done using MATLAB; reported error is the SEM of the fit.

Single-cell binding affinity is calculated by determining the densities of pMHC and TCR molecules, both bound and unbound, at the interface between cell and bilayer. The boundary of the cell–bilayer interface is determined from the RICM image using the native MATLAB image thresholding algorithm, and the resulting boundary is used to mask out molecules directly under the cell. Single molecules of total pMHC and bound pMHC beneath the cell are detected from the short- and long-exposure images, respectively, using the detection algorithm described above. The number of TCR is measured by summing the anti-TCR Fab intensity within the cell boundary and converting to molecules using the average intensity per anti-TCR Fab, as determined from single-particle tracking and intensity counting.