T-cell triggering thresholds are modulated by the number of antigen within individual T-cell receptor clusters

Abstract

T cells react to extremely small numbers of activating agonist peptides. Spatial organization of T-cell receptors (TCR) and their peptide-major histocompatibility complex (pMHC) ligands into microclusters is correlated with T-cell activation. Here we have designed an experimental strategy that enables control over the number of agonist peptides per TCR cluster, without altering the total number engaged by the cell. Supported membranes, partitioned with grids of barriers to lateral mobility, provide an effective way of limiting the total number of pMHC ligands that may be assembled within a single TCR cluster. Observations directly reveal that restriction of pMHC content within individual TCR clusters can decrease T-cell sensitivity for triggering initial calcium flux at fixed total pMHC density. Further analysis suggests that triggering thresholds are determined by the number of activating ligands available to individual TCR clusters, not by the total number encountered by the cell. Results from a series of experiments in which the overall agonist density and the maximum number of agonist per TCR cluster are independently varied in primary T cells indicate that the most probable minimal triggering unit for calcium signaling is at least four pMHC in a single cluster for this system. This threshold is unchanged by inclusion of coagonist pMHC, but costimulation of CD28 by CD80 can modulate the threshold lower.

T cells exhibit an exquisite ability to recognize extremely low densities of agonist peptide antigen displayed in MHC proteins on the surface of antigen presenting cells. Some studies have suggested that T cells may react to even a few individual agonist peptide molecules (1–3). Moreover, T cells maintain this extreme sensitivity without spontaneous triggering in the absence of agonist, which could lead to autoimmune disorders (4). Various mechanisms for how high sensitivity to agonist with apparent immunity to stochastic noise have been proposed. Some include T-cell receptors (TCR) monomer triggering by coreceptor or self-peptide-major histocompatibility complex (pMHC) heterodimerization (5, 6) or force-induced conformational changes (7). Cooperativity among multiple receptors within the TCR-pMHC clusters is also implicated (8), which may result from conformation-induced clustering (9), kinetic segregation (10), or lipid-mediated assembly (11). However, the specific physical mechanisms by which these remarkable capabilities are achieved remain unresolved, due largely to experimental limitations. Key to ultimately determining, and experimentally verifying, the molecular mechanism of TCR triggering is the ability to manipulate TCR cluster assembly in living T cells.

In the following, we present results from experiments that directly probe the consequences of differentially partitioning agonist, coagonist, and null pMHC among TCR clusters in primary T cells. The strategy is based on the hybrid live cell–supported membrane configuration, in which a solid supported lipid bilayer takes the place of the antigen presenting cell (12–14) (Fig. 1). Histidine-tagged variants of MHC class II, intercellular adhesion molecule-1 (ICAM-1), and CD80 for costimulation experiments are linked to the membrane through Ni²⁺-chelating lipid groups (14, 15). The lipids and proteins diffuse freely and as monomers on the supported membrane (16) (Fig. S1). Peptide composition and pMHC density are under direct experimental control. TCR clustering is controlled through grid patterns of metal lines, which have been prefabricated onto the underlying solid substrate. These structures, known as diffusion or mobility barriers (12, 16), block lateral transport of lipids and supported membrane associated proteins. Molecules diffuse freely within each corral, but are unable to hop between separate corrals. As the TCR engage antigen pMHC at high antigen densities, clustering ensues until all pMHC within a single corral of the grid-partitioned supported membrane are coalesced into a single cluster with their cognate TCR receptors (12) (Fig. 1D and Fig. S2). The number of pMHC within each supported membrane corral determines the maximum pMHC content of the corresponding TCR cluster that may assemble on the T cell. Thus adjusting the grid size at constant pMHC density titrates the maximum number of pMHC per TCR cluster without changing the number of antigens engaged by the T cell. The total number of TCR and other signaling molecules within clusters is not limited by the substrate partitions. We refer to this physical manipulation of molecular organization within living cells as a spatial mutation (12, 17, 18). In the present application, T cells differing only in the peptide agonist distribution among TCR clusters are generated and compared side-by-side.

Fig. 1.

Immunological synapse on partition patterns. (A) Schematic of experimental strategy. Primary mouse T cells interact with MHC and ICAM-1 functionalized patterned supported lipid bilayer membrane. MHC density is constant. MHC is preloaded with an appropriate ratio of null (T102E) and agonist (MCC) peptide. Chromium lines prefabricated on the glass are diffusion barriers to all lipids and lipid-tethered proteins. (B) En face view of TCR distribution in the immunological synapse. TIRF images of anti-TCR Fab on 0 MCC–MHC μm^-2 1 min after contact (diffuse TCR), 1 MCC–MHC μm^-2 10 min after contact (small spread TCR clusters), and 100 MCC–MHC μm^-2 10 min after contact (large central TCR cluster). Image intensity is scaled to demonstrate clearly the TCR distribution in each case. (Scale bar: 5 μm.) (C) Percent of cells with the TCR distribution phenotypes in B. Ten minutes after contact with bilayer from a representative experiment; n = 140–150 cells per concentration. Cells with both a large central cluster and peripheral small clusters are scored in the central large cluster category, because some TCR transport to the center has occurred. (D) Bright field, reflection interference contrast microscopy (RICM), and TCR distribution of cells off grid, and on 3, 1, and 0.5 μm side partition grids after 5–10 min of contact with the surface. TCRs cluster until there is one TCR cluster per box; 100 ± 5% SD MCC–MHC μm^-2. (Scale bar: 5 μm.)

The primary goal of the present study is to determine how agonist distribution among TCR clusters governs T-cell activation. We perform a two parameter titration experiment in which the overall antigen peptide surface density as well as its partitioning among TCR clusters are independently controlled in living T cells. The results indicate that the threshold antigen densities for triggering Ca²⁺ flux (in terms of the number of antigens per cell) are dependent on agonist partitioning among TCR clusters. Most significantly, when antigen dose-response functions obtained on different grid partition sizes are analyzed in terms of the maximum number of antigens per TCR cluster (determined by pMHC content within individual membrane corrals), they collapse onto a single curve.

We observe that T cells trigger at an average agonist density of approximately two per TCR cluster, irrespective of the total number of agonist engaged by the cell. The term triggering threshold is used here to describe the average density at which half-maximum response is achieved. Observed triggering thresholds were not changed by inclusion of the coagonist null peptide ER60. Interpretation of this invariant activation response in terms of the Poisson distribution of agonist pMHC among TCR clusters indicates that the most likely activation threshold is at least four agonist peptides within a single TCR cluster in this system. However, T cells do not have a universal threshold for triggering; it implies only that under controlled circumstances reproducible thresholds are observed. For example, inclusion in the supported membrane of CD80, which activates the costimulatory receptor CD28 on the T cells, lowers triggering thresholds. The implications of these measurements with respect to the mechanism of TCR signaling are discussed.

Results

Physical Partitioning of T-Cell Receptor Clusters.

T-cell activation is investigated in a hybrid immunological synapse consisting of primary mouse T cells interacting with functionalized supported lipid bilayer membranes (Fig. 1A). This configuration is widely used and has been instrumental in characterizing the immunological synapse (8, 12, 13). Glass substrates organize spontaneous self-assembly of supported membranes from small unilamellar vesicles. A characteristic feature of supported membranes, which is critical for these experiments, is the free lateral mobility of lipids and membrane associated proteins (12). For hybrid immunological synapse reconstitution, the supported membrane is functionalized with ICAM-1 and MHC loaded with various combinations of agonist [moth cytochrome c (MCC)], null (T102E), and coagonist null (ER60) peptides (8, 12, 13). For CD28 costimulation studies, CD80 is included in the supported membrane as well. The cytoplasmic domains of these proteins are expressed as histidine-tag constructs (His10 for ICAM-1 and CD80 and 2XHis6 for MHC) and stably linked to the supported membrane by multivalent interactions with Ni²⁺-chelating lipids mixed in the membrane. We employ kinetic control parameters to achieve desired densities of bound protein (14, 15).

AND TCR effector T cells form hybrid junctions with these supported membranes irrespective of the peptide displayed by the MHC, due to the presence of ICAM-1. When all of the MHC is loaded with the null T102E peptide, T cells spread on the surface and exhibit a relatively homogeneous distribution of TCR (Fig. 1B). At low concentrations of MCC–MHC, TCR organize into clusters throughout the contact zone. At much higher MCC–MHC, these TCR clusters are actively driven toward the center of the interface by an actin polymerization-dependent process, forming the classical immunological synapse pattern (8) (Fig. 1 B and C). TCR cluster formation is coincident with and necessary for activation of downstream calcium flux, but TCR clusters can continue to form after activation (8).

Physical structures on the underlying supported membrane substrate can be used to control TCR clustering in the living cell (Fig. 1D). Barriers to lateral mobility of supported membrane components are created by metal lines (5-nm high, 100-nm wide) that are prefabricated onto the glass substrate by electron beam lithography (Figs. S1 and S3). Grid patterns of barriers trap fixed numbers of pMHC, along with any other supported membrane proteins, within each confined grid corral without otherwise interfering with their mobility. As TCR engage these mobile but trapped pMHC, they also become subject to the substrate-imposed constraint. Generally, we observe TCR to accumulate into one microcluster within a single corral. Thus the number of pMHC per corral determines the maximal number of pMHC per TCR cluster, whereas the number of TCR per cluster is not limited. TCR outnumber MHC in most clusters (Fig. S4) and especially at low pMHC densities. We note that this observation does not indicate that individual TCR-pMHC interactions are long-lived or stable (19); it indicates only that the TCR cluster is stable at high antigen density. Importantly our experiments restrict pMHC, therefore ensuring that no TCR cluster could ever have more than that contained within a single corral. Thus our observations are independent of resting TCR distribution (e.g., as monomers or multimers) (20). When executed properly, this experimental system allows TCR clusters and bilayer proteins to move freely along the barriers, without crossing them or nonspecifically adsorbing to them (21, 22). Other molecules on the T cell not interacting directly with cognate ligands in the supported bilayer move freely throughout the entire interface. As examples, actin and CD45 are successfully transported within the immunological synapse irrespective of constraint patterns (Fig. S5).

Partitioning of Agonists Among TCR Clusters Alters the Threshold Density to Trigger Intracellular Calcium Flux.

The experimental ability to independently control pMHC distribution among TCR clusters enables characterization of TCR cluster signaling with unprecedented specificity. A typical experiment is depicted in Fig. 2. A population of T cells is allowed to interact with a supported membrane that is partitioned by substrate patterns in one region and unpatterned elsewhere. T-cell triggering is monitored by intracellular Ca²⁺ flux using fluorescence ratio imaging of fura-2 dye (13, 23) (Fig. 2 A and B, and Fig. S6). Although calcium flux is a quick cellular response (24), population calcium flux is integrated over 20 min to include cells responding at different times and provide better signal to noise. The choice of integration time does not alter overall results (Fig. S7). The supported membrane is of uniform composition over the entire substrate. Thus differences in T-cell triggering on and off patterned regions result exclusively from differences in TCR-pMHC cluster size. Lymphocyte function-associated antigen 1 and ICAM-1 clustering is not altered, because each corral contains dozens of ICAM-1 molecules. These experiments are performed at overall MHC density of 100 μm^-2 (± 5% SD). If all the peptides are the MCC agonist at these densities, robust intracellular Ca²⁺ flux is observed at all cell grid sizes (12). With increasing dilution of MCC to T102E, threshold densities for triggering can be identified.

Fig. 2.

T-cell activation on patterned partition grids. (A) Bright field of T cells and grid (Left) and false colored calcium levels (fura-2 ratio) (Right). (Upper) Off and on 0.5-μm square grid. (Lower) Off and on 0.5-μm control array of isolated solid squares. Small schematics of patterns are overlaid in bright field image. The white line in the calcium images indicates the border of the grid pattern. Calcium image 8.30 min after cell injection, bright field image approximately 30 min after cell injection. (Scale bar: 40 μm.) (B) Population calcium flux on partition grids. Bright field of representative cells on and off patterns (Left), schematic of possible agonist (MCC) and null (T102E) distribution in grid boxes (Center), and false-colored calcium flux (fura-2 ratio) traces of > 100 cells per region (Right). (Scale bar: 5 μm.) Cell traces (each horizontal line is one cell) are ranked by their integrated calcium flux in descending order. Calcium is imaged every 15 s. (C) Population mean (±SEM) of integrated calcium flux of all cells over 20 min on different size partition grids in A and B. (A–C) All data in this figure are from one representative experiment in one chamber with different patterns spaced approximately 0.3-mm apart; 8.6 ± 5% SD MCC–MHC μm^-2.

There are several reasons for titrating agonist:null peptide ratio rather than total MHC. First, under physiological conditions, T cells generally encounter relatively high MHC density and must search for rare agonist peptides. Second, molecules are randomly distributed (Poisson) among the supported membrane corrals and the standard deviation of these distributions is , where N is the average number of molecules per corral. A more uniform MHC distribution is thus achieved at higher MHC densities. Specific effects of the stochastic distribution of dilute MCC agonist peptide must still be considered for interpretation of TCR signaling and are discussed further below. Third, at high MHC density, typical dilutions of agonist to null of approximately 50–100-fold ensure that, in the rare events of aggregation of homogeneous MHC, the probability of directly interacting MHCs loaded with agonist is only minimal.

Quantitative results from a representative TCR cluster size titration experiment are illustrated in Fig. 2 B and C. This experiment is performed with an overall MCC–MHC density of 8.6 μm^-2. At this density, similarly high fractions of cells triggered on 1-μm grids, as did off (Fig. 2B). The intrinsic heterogeneity of the T cells, even in TCR transgenic populations, underlies the range of responses—from high to no calcium flux, displayed in the population heat maps (25). By contrast to 1-μm grid, 0.5-μm grid partitioning effectively blocked activation. Population responses are further compared by the average integrated calcium response over 20 min, which reports on early TCR signaling (2) (Fig. 2C and Fig. S6). Note that the barrier line width is not negligible and must be accurately measured by scanning electron microscopy and quantitative fluorescence (Fig. S3) to calculate the correct membrane area in different grid sizes. Control patterns consist of arrays of solid squares, with similar area coverage as the grids. They do not restrict MHC mobility, hence do not partition TCR clusters, and yield similar activation ratios as observed off patterns. This control experiment provides critical confirmation that the presence of metal on the substrate does not appreciably alter T-cell response.

By partitioning TCR clusters more finely with the 0.5-μm grids, triggering was essentially prevented without changing the total number of agonist pMHC engaged by the cell. Thus, the minimum agonist density recognized by T cells depends on how agonists are distributed among TCR clusters. From this observation, we may also conclude that, at the very least, a single agonist pMHC per TCR cluster is insufficient to trigger.

Minimal Number of Agonist Per Signaling TCR Cluster.

One is not enough, but what is the minimum number of agonist pMHC per TCR cluster sufficient for triggering? We explore this question through a series of experiments in which agonist pMHC density is titrated off and on two different grid sizes, 1 and 0.5 μm. As seen in Fig. 3A, as grid size and TCR-pMHC cluster size get smaller, the activation density for triggering Ca²⁺ flux becomes progressively higher. At sufficiently high concentration of MCC–MHC, cells can activate on both grid sizes. If these independent titration curves are instead plotted in terms of the number of agonist pMHC per cluster, then the 1- and 0.5-μm activation curves collapse to a single dose-response function (Fig. 3B). At average of one MCC–MHC per box, T cells do not activate, but at two they do; this result is irrespective of the total number of agonist pMHC engaged by the cell or the dimensions of the grid box. The fact that there are fourfold more TCR clusters in the more finely partitioned 0.5-μm grid cells does not affect this threshold. Additional increases in Ca²⁺ triggering as agonist pMHC density increases further are much more gradual. The onset of triggering occurs abruptly once a critical number of agonist pMHC per TCR cluster is reached. This fundamental observation has been repeated no less than nine times with populations of T cells from at least four different mice and observed with GPI-tethered proteins as well (SI Text and Figs. S8 and S9).

Fig. 3.

Partition titration of agonists per TCR cluster. (A) T-cell activation with respect to titration of overall MCC–MHC surface density off and on 1- and 0.5-μm partition grids. The minimal density at which T cells activate is shifted higher from off grid (●) to 1-μm grid (□), and even further to 0.5-μm grid (⊠). Lines are guides to the eye. Mean calcium flux (± SEM) of 150–300 cells, pooled from 2–3 independent experiments per concentration. (B) T-cell activation with respect to number of MCC–MHCs per box at 1-μm (□) and 0.5-μm (⊠) partition grids. The transition to activation on 1- and 0.5-μm grids is at the same number of MCC–MHC per box (between the two gray lines). (C) Theoretical Poisson distribution of the probability of discrete number of MCC–MHC per box at different average concentrations per box. At nonactivating density (left gray line), most boxes have zero or one MCC–MHC. At activating densities (right gray line), an appreciable number of boxes have more than three or four MCC–MHCs. (D) Direct comparison of Poisson probability in C of n (MCC–MHC per grid box) at nonactivating concentration of 〈MCC-MHC〉/box = 1 (□) to activating concentration of 2 (▪). Fold increase at each n is marked with an arrow and a numerical value. Probability of n = 4–6 goes from negligible to appreciable and rises sharply between non- and activating concentrations. At activating concentrations, the probability of four agonists is 9%; of 5 is 3.6%; and of 6 is 1.2%, yielding several boxes of each per cell at either grid size (assuming T-cell coverage of at least 100 μm²).

At such low densities, the stochastic distribution of agonist pMHC among the many TCR clusters in each cell must be considered. In our experimental procedure, both agonist and null peptide loaded MHC are exposed to the membrane from solution simultaneously (26). Therefore we expect uncorrelated binding of both peptide–MHC complexes. The peptide off rate is extremely slow and peptide reloading from solution is negligible at neutral pH (27). Thus the original loading configuration is expected to persist throughout the entire experiment, which results in a Poisson distribution for the number of agonist MCC peptides in each grid box. The probability of a given TCR cluster having a particular number of agonist pMHC for different average numbers of agonist pMHC per cluster is plotted in Fig. 3C.

Next, we compare the distribution of agonist among TCR clusters at an average MCC–MHC density of one per cluster, which was not activating, to an MCC–MHC density of two per cluster, which was activating, as traced by gray lines between Fig. 3 B and C. Notice that the probability of two MCC–MHC in a single cluster is relatively unchanged and the probability of two or more changes by less than twofold between the nonactivating and activating situations. In contrast, the number of TCR clusters with three and, to an even greater extent, four or more MCC–MHCs has risen sharply (Fig. 3D). These results indicate that the most probable minimal triggering unit contains at least four agonist pMHC in a single TCR cluster and could require as many as six. It is unlikely that seven or more agonists are required because the probability of these becomes so low that many cells will have no TCR clusters with this many agonists.

In the experiments described above, agonist MCC peptide was titrated into a majority background of the null peptide, T102E. There is another group of endogenous peptides, referred to as coagonists, that do not activate by themselves at any density but enhance activation in the presence of agonist pMHC (5, 28). A peptide from the cysteine protease of the endoplasmic reticulum (ER60) has been reported as a coagonist peptide to effector T cells, bearing the 5C.C7 TCR receptor that also recognizes MCC–MHC (5). MCC titration experiments were also performed by dilution into a background of ER60 (Fig. 4). Triggering thresholds with ER60 are the same as measured for MCC titrations into the absolute null T102E, and exhibit the same universality on a per TCR cluster basis. However, ER60 did yield an apparent increase in the magnitude of Ca²⁺ flux.

Fig. 4.

Partition titration of agonists per TCR cluster with ER60. (A) Comparison of T-cell activation with respect to titration of MCC–MHC surface density off and on 1- and 0.5-μm partition grids, with T102E and ER60 null peptides. The threshold density at which T cells activate off (circles), and on 1-μm grid (diamonds) (2–3 MCC–MHC μm^-2) or 0.5-μm grid (squares) (5–6 MCC–MHC μm^-2) is the same for background null peptide T102E (empty symbols) and ER60 (filled symbols). Lines are guides to the eye. Mean calcium flux (± SEM) of more than 80 cells per concentration, from a representative experiment. T102E data are the same as Fig. 3A.

The costimulatory molecule CD80 can also modulate T-cell signaling by activating CD28 on the T-cell surface (29, 30). Inclusion of CD80 (approximately 300 molecules μm^-2) (14) in the supported membrane during MCC:T102E titration experiments yields two distinct effects on TCR signaling (Fig. 5A). First, the triggering threshold is notably shifted to lower agonist pMHC densities. Additionally, the apparent magnitude of Ca²⁺ flux was increased, in a manner similar to that observed with the coagonist ER60 peptide. This stronger activation with CD80–CD28 costimulation has been observed before (29).

Fig. 5.

Partition titration of agonists per TCR cluster with CD80. (A) T-cell activation with respect to titration of overall MCC–MHC surface density off and on 1- and 0.5-μm partition grids, with and without CD80. The threshold density at which T cells activate in the presence of CD80 (filled symbols) is shifted to the left for cells off grid (circle), on 1-μm grid (diamonds), and on 0.5-μm grid (squares). Lines are guides to the eye. Mean calcium flux (± SEM) of more than 80 cells per concentration, from a representative experiment. Data without CD80 are the same as Fig. 3A. Null peptide is T102E for all data points. (B) T-cell activation with respect to number of MCC–MHCs per box at 1-μm (diamonds) and 0.5-μm (squares) partition grids, in the presence of CD80. The activation profiles for 1- and 0.5-μm grids is the same in terms of number of MCC–MHC per grid box. (C) Theoretical Poisson distribution of the probability of discrete number of MCC–MHC per box at different average concentrations per box. At nonactivating density (left gray line), most boxes have zero or one MCC–MHC. At activating densities (right gray line), an appreciable number of boxes have more than two MCC–MHCs. (D) Direct comparison of Poisson probability in C of n (MCC–MHC per grid box) at nonactivating concentration of 〈MCC-MHC〉/box = 0.1 (open squares) to activating concentration of 1 (filled squares). Fold increase at each n is marked with an arrow and a numerical value. Probability of n = 2 goes from negligible (0.5%) to appreciable (18%) and rises sharply between non- and activating concentrations, in the presence of CD80.

Renormalization of the MCC:T102E titration responses from both grid sizes in terms of the number of antigen per grid box again yields a singular response curve (Fig. 5B). We observe that triggering thresholds are shifted by CD80–CD28 costimulation, but they are still defined on a per TCR cluster basis, rather than a per cell basis. A stochastic analysis of agonist distribution among clusters as done above reveals the most probable minimum activating unit in the presence of CD80–CD28 costimulation is two agonist pMHC per TCR cluster (Fig. 5 C and D). The data are a bit rougher at these very low triggering thresholds, but it is important to realize that if a single agonist peptide could cause a TCR cluster to trigger, then there would be no partition effect. The data presented here are not consistent with individual agonist activating a TCR cluster.

Discussion

The difference between the measured triggering threshold based on average and stochastic analyses deserves consideration. First, we note that triggering as defined by intracellular Ca²⁺ is essentially digital. The onset of Ca²⁺ flux is abrupt (Fig. S6E). The next question then concerns whether individual TCR clusters trigger with an all-or-nothing or a graded response. Based on the experimental evidence reported here, it is unlikely TCR clusters could yield a graded response at minimal antigen levels. We arrive at this conclusion based on the fact that partitioning TCR clusters at a constant total amount of antigen leads to an abrupt cessation of triggering. If multiple partially triggered TCR clusters were somehow integrated within the cell to yield the equivalent of one fully triggered cluster, then such discrete changes as a function of partitioning alone are not expected.

These results are most consistent with a mechanism in which individual TCR clusters trigger, all-or-nothing, based on achieving some minimum threshold. Such a response necessitates extensive cooperativity among the TCRs within individual clusters (31). It is possible that more TCRs are triggered than the number of agonists within individual clusters, at least to the extent that triggering of an individual TCR can be defined (3). The serial triggering model (3, 32) provides one mechanism for such a phenomenon. There are also reports suggesting some degree of TCR preclustering (20), which would provide a mechanism of recruiting and possibly triggering TCR not directly engaged with antigen pMHC.

With this all-or-nothing response, the probability of a TCR cluster triggering will be a function of the number of agonist in the cluster. However, it is not necessary that a specific threshold number of agonist per cluster, above which the cluster always triggers and below which it never does, even exists. Rather, all we can expect is that the probability of a cluster triggering increases with agonist number and that this increase may be particularly abrupt around a certain critical number of agonists per cluster. Our experimental results reveal abrupt increases in the probability of triggering intracellular Ca²⁺ flux corresponding with the appearance of TCR clusters at least four agonist pMHC in the absence of costimulation.

Observation with CD80–CD28 costimulation reveal that TCR triggering thresholds are not absolute. CD80–CD28 costimulation is one of perhaps multiple ways that triggering thresholds may be tuned in vivo (5, 19, 30, 33, 34). Naïve T cells, which are dependent on this costimulation, may have much higher signaling thresholds. The one truly invariant observation among all of the results presented here is that triggering thresholds are determined on a per TCR cluster basis, and not on a per T-cell basis.

Previously, direct coupling of two TCRs, by bivalent antibody or agonist dimers, has been shown to trigger T cells (5, 35). These well known facts, along with structural evidence of binding to CD4 (36, 37), naturally lead to the speculation that a TCR dimer may be the fundamental triggering unit. However, the artificial tether between agonists alters the dynamics of agonist–TCR and TCR–TCR association, possibly stabilizing it beyond physiological conditions, and therefore cannot reveal the native threshold for triggering. The experimental design used here does not enforce or require direct dimerization between agonist pMHCs.

Estimates of T-cell sensitivity to pMHC have been made with increasing precision over the past two decades because of advancements in quantitative measurements of peptide number and characterization early T-cell responses (1, 38–41). Single agonist sensitivity has been reported by fluorescence, in a cell–cell interface where multiple ligands, protein complexes, and coagonist endogenous peptides are present (1, 2, 5). There are several factors to consider when interpreting these reports in light of the results we present here. First, our definition of triggering threshold as the average antigen number at which half-maximum response is achieved may differ from earlier interpretations. We chose this definition because it converges to an informative value in the limit of a large number of experiments. Triggering threshold cannot properly be defined as the lowest level of antigen at which any triggering was observed, because there is a nonzero probability of a cell triggering with no antigen due to stochastic fluctuations; this cannot be distinguished from causal triggering on a cell-by-cell basis. Secondly, supported membranes differ from live antigen presenting cell (APC) experiments because costimulatory molecule and peptide antigen distributions are directly controllable. These are intrinsically unknown in live APCs, and may alter triggering thresholds (5, 42). Although the supported membrane system may be less physiological than live APCs, it is far better controlled and therefore offers more quantitative information on the behavior of the TCR signaling system under specific conditions.

In conclusion, the TCR cluster size titration experiments described here arguably constitute the most quantitative analysis of TCR triggering thresholds to date. The results are generally consistent with the extreme, near-single-molecule sensitivity of T cells to antigen long thought to exist. Cluster size titration further reveals that spatial organization into TCR clusters is critical to achieve this sensitivity. The universal observation that Ca²⁺ flux triggering thresholds were determined on a per TCR cluster basis, rather than a per T-cell basis, raises interesting questions concerning large-scale cooperativity in receptor signaling clusters and apparent lack of cooperativity between clusters.

Materials and Methods

Mice and T cells.

AND CD4+ T-cell blasts were prepared by in vitro stimulation of spleen and lymph node cells from F1 cross of ANDx B10.Br mice (Jackson Laboratory) with 1–2 μM MCC peptide. Cells were maintained in IL-2 every 48 hr and used on days 5 and 7.

Patterned Supported Proteolipid Bilayers.

Glass slides were patterned by e-beam lithography at University of California, Berkeley Microfabrication Lab and Molecular Foundry, Lawrence Berkeley National Laboratory. Liposomes with 97.5 mol % 1,2-dioleoyl-sn-glycero-3-phosphocholine, 2 mol % 1,2-dioleoyl-sn-glycero-3-[(N-(5-amino-1-carboxypentyl)iminodiacetic acid)succinyl] (nickel salt) and 0.5 mol% Texas Red® 1,2-dihexadecanoyl-sn-glycero-3-phosphoethanolamine, triethylammonium salt, prepared by sonication or extrusion, were injected over substrate in Biophtech FCS2 flow chamber. His-tagged proteins were expressed in Hi5 cells and purified by Ni-nitrilotriacetic acid affinity (14, 22). Surface density of MCC–MHC was measured with AlexaFluor conjugated MCC peptide and calibrated with supported bilayers (43).

Fluorescence Imaging.

For calcium flux, cells were loaded with 1 μM fura-2-acetoxymethyl and imaged at 340 and 380 nm with emission at 510 nm via 40xS Fluor objective on Quantix 57 or CoolSnap K4 cameras. Fields of view with patterns were monitored every 7 or 15 s for 20 min. Cell motion and fluorescence intensity were tracked and analyzed semiautomatically in Imaris, Metamorph, Matlab, and Excel. TCR, actin, and CD45 were imaged via 100x N.A.1.3 objective on CoolSnapHQ camera. Total internal reflection fluorescence (TIRF) was done via 100x 1.45 N.A. TIRF objective on a Cascade 512B EM CCD camera.