An extensive literature on the intracellular signals triggered by pMHC-ligand engagement of the TCR suggests that the response is rapid, sensitive, and highly discriminatory. In this study, we have documented another key feature, namely the digital, highly amplified ERK response that occurs at short timescales (<3 min) but correlates with functional responses at >1 h post-TCR engagement. This finding raised a fundamental issue: how can T cells trigger such an “explosive” response while maintaining the specificity of ligand discrimination? In an attempt to construct a model that accounted simultaneously for all four key characteristics of TCR signaling in response to ligand engagement, we combined two recently reported opposing feedback modules [20
] with a core scheme based on kinetic proofreading [10
]. Using realistic kinetic parameter sets for computer simulation of the signaling cascade downstream of TCR engagement, our model yielded an output that had the striking characteristic of a sharp transition in ligand agonist functionality at TCR-binding lifetimes corresponding to those measured in several different T-cell systems [21
]. We showed that this transition is also consistent with the very large (1 × >104
) shift in potency of pMHC ligands that differ by only a few fold in their binding lifetimes. Our modeling suggests that the sharp threshold for pMHC-receptor lifetimes yielding agonist responses originates from the distinct kinetic characteristics of the phosphatase-mediated negative feedback that suppresses signaling by weak ligands and the ERK-mediated positive feedback that is induced effectively only by more avid ligands of the TCR.
We built upon the observations of Stefanova et al. [27
] in constructing a model in which SHP-1 mediated inhibition begins to function quickly upon TCR engagement, but scales in an analog way with input. In contrast, the ERK response was modeled as delayed but (as newly documented here) digital in nature. This combination allows TCR activity induced by a large number of weak ligands to be constantly repressed by a proportional negative feedback that has enough time to quench upstream signals before they reach the limit necessary to trigger the ERK response. Both modeling and experiment confirm that the ERK response is increasingly delayed in onset as the duration of pMHC–TCR binding decreases. In contrast, more strongly binding ligands, though also inducing an initial SHP-1 inhibitory response, override the limited nature of this negative feedback early after ligand engagement by quickly triggering the highly amplified ERK digital response. The magnitude of this ERK activation then prevents inhibition of those TCR not yet inactivated by the gradually rising pSHP-1 levels, permitting effective downstream signaling through diverse pathways that impinge on genes involved in T-cell differentiation. The latter expectation of a transient recruitment of pSHP-1 to agonist-engaged TCRs and the generation of an abortive proximal tyrosine-phosphorylation response in T cells exposed to an agonist when the ERK cascade is inhibited have both been observed in biochemical studies [27
]. Overall, these observations provide new insight into how control circuits can be organized to suppress noise generated by large numbers of ligands while promoting highly sensitive responses to a few optimal stimuli in the same cellular context.
Our simulations enabled us to make several predictions that were verified by experiment. Most relevant to our understanding of how T cells set the threshold for discriminating between foreign and self-ligands to promote effective responses without fostering autoimmunity, we predicted that modest changes in intracellular enzyme levels would “tune” this agonist threshold during differentiation [51
]. This prediction was confirmed in studies showing that the decreased amount of SHP-1 in T cells a few days after activation of naïve T cells accounts for a gain in response to a pMHC ligand that is incapable of stimulating naïve or resting primed cells expressing the same TCR. This sensitivity of the response-threshold position to modest alterations in the intracellular concentrations of key components of the network, particularly SHP-1, was a somewhat surprising result. Stochastic noise in the production and degradation of signaling components might be expected to produce fluctuations of a similar magnitude in key molecules [54
] and hence to jeopardize accurate self-/non-self-discrimination by T cells in the periphery after the threshold is set during positive and negative selective events in the thymus.
One possible explanation for how this is avoided is that naïve T cells may have a very stable metabolism that enables them to preserve the phenotype selected for in the thymus prior to overt activation by foreign ligand. Alternatively, others have proposed that T cells can respond to tonic exposure to self-ligands by abrogating self-responsiveness while maintaining reactivity to pathogen-derived ligand [51
]. Perhaps this “self-tuning” involves dynamic adjustment of the competition between positive and negative feedbacks. A third possibility is that such fluctuations do result in an occasional T cell producing potential activation signals upon self-recognition; however, in the non-inflamed state, this would lead to tolerance through deletion or anergy [56
]. The danger would be if this occurred during an inflammatory response, but indeed it is just such situations that may be inciting events for autoimmunity [57
In this same regard, the acquisition among activated T cells of overt signaling responses to variant pMHC ligands that do not evoke such responses among naïve or resting primed T cells with the same TCR is an intriguing finding whose physiological relevance is only evident in one circumstance. Hogquist et al. originally identified EIINFEKL as a peptide driving positive selection of OT-1 T cells under organ-culture conditions in which the usual display of self-peptides is limited [18
]. EIINFEKL was also the strongest antagonist of OT-1 T-cell activation by SIINFEKL presented in the context of H-2Kb
]. Hence, EIINFEKL-Kb
was a ligand known to induce some positive signaling in the OT-1 thymocytes and antagonistic negative signaling in peripheral OT-1 lymphocytes. Our model and experiments enable us to hypothesize how this divergent signaling capacity of EIINFEKL-Kb
may correlate with up/down expression of components of the TCR signaling machinery and, specifically, SHP-1 [42
]. Thus, actively keeping SHP-1 levels low during early T-cell differentiation could allow self-ligands to have weak-agonist function and drive the positive selection of the T-cell repertoire, while increased SHP-1 levels would eliminate this response capacity among the mature T cells that populate the periphery [19
]. The “bell-shaped” dose response induced by EIINFEKL-Kb
T cells has been observed in other biological systems [58
]: our model suggests that such a nonmonotonic dose response is in fact a reflection of the activation of excess negative feedback at a high dose of weak ligands.
Why more mature T cells that have been recently activated should alter SHP-1 levels so as to regain sensitivity to stimulation by weak ligands is not yet clear, but one possibility is that activated cells use this reprogramming of the signaling threshold to take advantage of abundant self-ligands to promote further differentiation once their initial activation has been “validated” by foreign-agonist recognition. Our biological studies and simulations were both conducted in the absence of such potentially active self-pMHCs. However, a very recent study indicates that this synergy can occur in a narrow time window after previous agonist-mediated T-cell activation [36
], consistent with the gain in sensitivity to fast off-rate pMHCs that has been shown here to be due to decreased SHP-1 levels in this time frame.
What are some of the potential limitations of the current model? While it has proved very successful in simulating aspects of T-cell biology that can be verified experimentally and even has correctly predicted some behaviors not previously recognized, we do not know the extent to which the simplifications we have introduced to keep the model tractable have compromised its ability to reflect T-cell physiology. First, this model lacks spatial constraints and treats the T cell as a well-stirred vessel for the first 3 min of TCR signaling. We believe this is justified, based on our quantitative analysis of naïve T cells and their contents, which reemphasized the small cytoplasmic volume of these cells and the resulting high concentrations of signaling components. For this reason, most enzymatic reactions involved in T-cell signaling are not diffusion-limited. Moreover, the spatial reorganization of membrane signaling proteins during T-cell activation that results in a mature immunological synapse [60
] takes place over a substantially longer timescale than the one considered in our model [61
], and initial signaling occurs prior to the large-scale protein clustering involved in the formation of this synaptic structure. This does not mean that local inhomogeneities in protein distribution in the membrane (e.g., “rafts”), or involving scaffolded protein complexes in the cytoplasm, do not influence signaling behavior.
More elaborate modeling tools that preserve spatial information ([62
]; M. Meier-Schellersheim et al., unpublished data) will be needed to expand analyses of T-cell signaling. This may be particularly relevant in understanding how self-ligands can synergize with agonist pMHCs in promoting T-cell activation [36
] and in better modeling the role of signal spreading among engaged and nonengaged TCR in the action of pSHP-1 and ppERK. Second, we have omitted explicit specification of a number of molecules that are well documented in the literature to affect T-cell signaling responses, such as CD45, Csk, and several adapter proteins [63
]. However, the influence of these components on signaling was implicitly incorporated in some of the kinetic parameters (e.g., tonic dephosphorylation), and we feel this is justified by the absence of evidence that any of these components has first-order sensitivity to the quality of the pMHC–TCR ligand interaction. Third, we have introduced modifications to the kinetic parameters of pMHC–TCR interaction measured at room temperature to match the model's output to biological experiments conducted at 37 °C. Whether our approximations in this regard are accurate are not yet clear, because evidence for both linear and nonlinear effects of temperature on pMHC interactions with TCRs have been reported [9
]. Finally, we have considered here only the signaling involved under conditions in which the CD8 coreceptor plays a key role in the response. This is not an absolute necessity for all TCR-mediated activation, but it is a common feature of many physiological T-cell responses including that of the OT-1 cells we used for the biological component of the present study.
Although the primary aim of this work has been to better understand how TCR signaling is regulated and contributes to the proper performance of T cells in the immune system, the results we have obtained showing how simple feedback loops operate to suppress biological noise, amplify responses, and allow flexibility of function are likely to be relevant to many other biological systems. The roles of negative and positive feedback are well documented in gene regulatory networks [66
] and, in particular, in developmental systems, where they can impose irreversible state changes on the system, providing unidirectionality to differentiation events [67
]. It remains to be seen whether the specific features of the opposing feedback pathways modeled here, (rapidly initiating analog negative feedback versus delayed, digital positive feedback) are critical in other biological systems. More generally, the value we document here for quantitative modeling rather than just qualitative cartoon depiction of signaling circuits, and the importance of documenting physiologically relevant signaling dynamics in simulation outputs, indicate that methods and tools for constructing models, conducting simulations, and measuring values will be increasingly critical aspects of experimental biology.