The essential function of the T cell receptor (TCR) is to translate the engagement of peptides on the major histocompatibility complex (pMHC) into appropriate intracellular signals through the associated cluster of differentiation 3 (CD3) complex. The spatial organization of the TCR–CD3 complex in the membrane is thought to be a key regulatory element of signal transduction, raising the question of how receptor clustering impacts on TCR triggering. How signal transduction at the TCR–CD3 complex encodes the quality and quantity of pMHC molecules is not fully understood. This question can be approached by reconstituting T cell signaling in model and cell membranes and addressed by single-molecule imaging of endogenous proteins in T cells. We highlight such methods and further discuss how TCR clustering could affect pMHC rebinding rates, the local balance between kinase and phosphatase activity and/or the lipid environment to regulate the signal efficiency of the TCR–CD3 complex. We also examine whether clustering could affect the conformation of cytoplasmic CD3 tails through a biophysical mechanism. Taken together, we highlight how the spatial organization of the TCR–CD3 complex – addressed by reconstitution approaches – has emerged as a key regulatory element in signal transduction of this archetypal immune receptor.
An essential step in adaptive immune responses is the recognition and discrimination of peptides that are produced in the organism (self) and foreign, non-self (antigenic or agonistic) material. This is mediated by the T cell receptor on the surface of T lymphocytes, in conjunction with constitutively, but not covalently associated cluster of differentiation 3 (CD3) dimers (TCR–CD3) (Birnbaum et al., 2014; van der Merwe and Dushek, 2011). The TCR–CD3 complex engages antigenic peptide-loaded major histocompatibility complex (pMHC) molecules on the surface of antigen-presenting cells (APCs) and is able to recognize antigenic pMHC from non-antigenic pMHC molecules with high specificity (Feinerman et al., 2008). The TCR–CD3 complex consists of TCRα and TCRβ chains, which enable pMHC binding, and CD3δ–CD3ε and CD3γ–CD3ε and CD3ζ–CD3ζ chain dimers (CD3ζ is also known as CD247, hereafter referred to as ζ chains) (Call and Wucherpfennig, 2005, 2007) (Fig. 1). The CD3δ, γ and ε chains each contain a single conserved immunoreceptor tyrosine-based activation motif (ITAM), whereas the ζ chain contains three ITAMs (Birnbaum et al., 2014). Phosphorylation of the ITAMs on CD3 chains by the Src-family kinase Lck is the initial step that initiates downstream signaling from the receptor (Shah et al., 2016; Smith-garvin et al., 2009; Weiss, 2010). Lck is itself regulated by phosphorylation, with autophosphorylation of the activatory tyrosine 394 (Y394) enhancing the catalytic activity, whereas phosphorylation of the inhibitory Y505 site, mediated by C-terminal Src kinase (Csk), inhibits kinase activity through an allosteric mechanism (Boggon and Eck, 2004). Once ITAMs in the TCR become phosphorylated they become docking sites for the cytosolic tyrosine kinase 70 kDa ζ-chain associated protein (ZAP70) that, once recruited to the membrane, becomes activated and phosphorylates tyrosine residues on the membrane adaptor protein linker for activated T cells (LAT) (Zhang et al., 1998). Multiple cytosolic enzymes and adaptor proteins are recruited to phosphorylated LAT, which ultimately propagate signaling into the cytosol, leading to T cell activation (Balagopalan et al., 2015; Zhang et al., 1998).
The ability of TCR–CD3 complex to recognize the very few antigenic pMHC molecules in a sea of self pMHC molecules (antigenic pMHC numbers are ∼104 lower than self pMHC) is integral to the initiation of signaling through the TCR, and thus, downstream immune responses (Morris and Allen, 2012). Many models have been proposed to account for the apparent high sensitivity and selectivity of the TCR. The highly selective manner by which T cells discriminate between self and antigenic pMHC is attributed to a high sensitivity to the length of interaction between the pMHC and TCR due to a process known as kinetic proofreading (McKeithan, 1995). In this model, after TCR–CD3 engagement, a series of molecular steps, each requiring some length of time, are required before commitment to cellular activation is achieved, but if pMHC–TCR dissociation occurs during this process, the steps are rapidly reversed. Abundant pMHC–TCR self-interactions are effectively filtered out due to their lifetime being too short to initiate activation. Agonist pMHC on the contrary has reduced off-rate (i.e. longer lifetime) and thus persists for long enough to initiate signaling. This explains the observed dependence of T cell activation on pMHC–TCR binding kinetics (Aleksic et al., 2010; Govern et al., 2010; Huang et al., 2010; Lyons et al., 1996), but the model does come with a penalty to sensitivity. To explain the sensitivity of T cells to low antigenic pMHC numbers, induced rebinding by clustering of unligated TCR–CD3 molecules around a bound TCR–CD3 (Dushek and Van der Merwe, 2014) or spreading of triggering from bound TCR to unbound TCR in the same preformed cluster (Schamel and Alarcón, 2013) (Fig. 2B) have been suggested.
Conformational changes that are transmitted through the TCR–CD3 complex owing to pMHC binding have also been postulated as a gatekeeper for TCR triggering. The binding of the pMHC to the TCRα–TCRβ chains is thought to cause conformational changes in the CD3 complex (Feng et al., 2018; Minguet et al., 2007; Xu et al., 2008). CD3ε and ζ chain cytoplasmic tails can associate with the plasma membrane, which prevents their association with and phosphorylation by Lck and other kinases, thus keeping them in an inactive state (Aivazian and Stern, 2000; Xu et al., 2008; Zhang et al., 2011). Upon antigen binding, the cytosolic tails of the ζ chain and CD3ε are released from the plasma membrane, thus becoming available for phosphorylation by Lck and allowing TCR triggering. This mechanism of TCR triggering is known as the safety-on model (Kuhns and Davis, 2008; Xu et al., 2008) (Fig. 2A). The permissive geometry model, another leading conformational change model, posits that the TCR exchanges between active and inactive conformations. The model suggests reversible cholesterol binding to the transmembrane domain of the TCRβ subunit stabilizes the complex in the signaling-inactive form, whereas binding of pMHC stabilizes the signaling-active conformation (Swamy et al., 2016). Only in the signaling-active conformation are the ITAMs of the CD3 and ζ chains available for phosphorylation (Swamy et al., 2016) and the proline-rich region of the CD3ε tail is available for binding of the adaptor protein Nck (also known as Nck1) (Blanco et al., 2014; Gil et al., 2002; Martínez-Martín et al., 2009). In addition, the application of shear force during TCR–CD3 binding events is thought to allow for mechanosensing by the TCR–CD3 complex (Feng et al., 2018). This is particularly relevant in the context of APC contacts where T-cells are crawling across and scanning the APC surface for antigenic material. The tension experienced during binding is thought to alter the conformation of the TCR–CD3 complex, possibly by leading to dissociation of CD3 cytoplasmic tails with the membrane (Brazin et al., 2018), to allow triggering (Fig. 2A).
Owing to the dimensions of the TCR and pMHC molecules, bonds between these two protein complexes must occur at tight contact junctions (10–15 nm width) between the APC and T cell membrane (Choudhuri et al., 2005). This has been proposed as a means to exclude membrane proteins with a bulky and highly glycosylated extracellular domain, similar to other signaling interfaces, such as seen in integrin (Paszek et al., 2014) and Fc receptor signaling (Freeman et al., 2016). One such large bulky molecule is the glycosylated phosphatase CD45 (also known as PTPRC), which is segregated from regions of close contact between two opposing membranes (Razvag et al., 2018). CD45 is a highly abundant, constitutively active phosphatase on T cell surfaces and plays a dual role in T cell activation; it has an important role in dephosphorylating Lck on the inhibitory Y505 site, thereby maintaining a pool of active enzyme (Rhee and Veillette, 2012), but also directly dephosphorylates the TCR (Hui and Vale, 2014; Hui et al., 2017; James and Vale, 2012) and is also thought to be a negative regulator of TCR phosphorylation (Burroughs et al., 2006; Cordoba et al., 2013; Davis et al., 2006; Dushek et al., 2012). The kinetic segregation model of TCR triggering proposes that Lck, which lacks a transmembrane or extracellular domain, can freely enter the close contact where CD45 has been locally depleted and thus the balance between phosphorylation and dephosphorylation in this region is shifted toward phosphorylation (Burroughs et al., 2006; Davis et al., 2006) (Fig. 2C). This model highlights the importance of a shift in the kinase–phosphatase balance around the pMHC–TCR–CD3 complex, which is proposed to be crucial for allowing robust, long-lived phosphorylation of the CD3 and ζ chains (Burroughs et al., 2006; Davis et al., 2006). As predicted by the kinetic segregation model, elongation of pMHC–TCR complex (Choudhuri et al., 2005) or truncation of the extracellular dimensions of CD45 (Cordoba et al., 2013) both abrogate signaling.
There is active debate in the literature about the accuracy and/or primacy of these (and many other) models. For instance, we have not elaborated on models that place critical importance on the regulation of Lck activity. This clearly plays a central role in T cell activation since acute inhibition of the negative regulator Csk leads to phosphorylation of the TCR and downstream signaling (Tan et al., 2014), whereas deletion of CD45 leads to hyperphosphorylation of Y505 on Lck and an inability to activate the TCR (Rhee and Veillette, 2012). Neither have we elaborated on the role of the co-receptors CD8 and CD4, which bind to constant regions of MHC class I and II, respectively, and help to stabilize the pMHC–TCR complex (Hong et al., 2018; Rudolph et al., 2006). This also enhances the efficiency of signaling by bringing Lck close to the CD3 and ζ chains (Artyomov et al., 2010; Stepanek et al., 2014). Co-receptors seem to play an accessory role, since high-affinity pMHC–TCR interactions (Kerry et al., 2003) and chimeric antigen receptors (CARs), which utilize signaling motifs from ζ chains (Fig. 1), are able to generate signaling independent of co-receptors (Sadelain, 2016). We follow the logic of van der Merwe and Dushek in grouping these models as special cases of a shift in the kinase–phosphatase balance (van der Merwe and Dushek, 2011) – the difference being that extra emphasis is placed on the enhancement of kinase (i.e. Lck) activity around the TCR rather than a decrease in phosphatase activity, as with the kinetic segregation model. For an in-depth discussion of TCR triggering models we recommend the review of van der Merwe and Dushek (2011).
In the following sections, we review the literature on the spatial organization of the TCR–CD3 complex in the plasma membrane of T cells, how this organization may affect pMHC–TCR interactions and its potential impact on TCR triggering. Where relevant we pay particular attention to reductionist systems in which reconstitution of components of the signaling cascade has advanced our understanding of the underlying processes (Fig. 3).
Spatial organization of the TCR–CD3 complex
The spatial membrane organization of TCR is an important factor in T cell signaling. Clustering of TCRs is thought to play a key role in the sensitivity of pMHC–TCR interactions (Dushek and Van der Merwe, 2014; Kumar et al., 2011; Pageon et al., 2016), and an inhomogeneous distribution of TCR at the plasma membrane has been observed by many microscopy methods (Kumar et al., 2011; Lillemeier et al., 2006; Pageon et al., 2016; Roh et al., 2015; Schamel et al., 2005). TCR has been found in detergent-insoluble mouse thymocytes membrane fractions, suggesting a preference for cholesterol-rich membranes, with receptor triggering increasing their abundance in these domains (Montixi et al., 1998; Xavier et al., 1998) (Fig. 2D). These fractions also had a higher abundance of phosphorylated ζ chain due to increased Lck activity (Montixi et al., 1998; Xavier et al., 1998). Observation of the TCR complexes within these fractions revealed that there is potential for TCR to be present as higher-order oligomers (for definitions, see Box 1), as well as being monomeric (Schamel et al., 2005). Exposure of cells to low doses of antigen led to triggering only in the fractions with higher-order oligomers, with triggering of monomeric TCR only apparent at higher doses of antigen (Schamel et al., 2005). This work was also coupled with the observation of TCR (specifically TCRβ and CD3ε) distribution in the membrane by freeze-fracture transmission electron microscopy (TEM) (Schamel et al., 2005). Here, TCR molecules indeed presented as a mixture of monomers and higher-order oligomers, with greater than 55% of TCR–CD3 in organized oligomeric structures, which could be disrupted by cholesterol extraction (Schamel et al., 2005). Further investigations by both biochemical and TEM methods revealed that oligomers were present to a much lesser extent in resting T cells (Kumar et al., 2011). Again, stimulation with antigen gave rise to an increased oligomerization over a period of days (Kumar et al., 2011). Interestingly, isolation of primary naïve and memory cells from mice also revealed striking differences in TCR oligomerization between cell types, with memory T cells exhibiting many more higher-order oligomers that the naïve cells, suggesting antigen-experienced cells retain clustering of TCR for effective antigen detection (Kumar et al., 2011).
In this Review, the terms cluster, nanocluster and oligomerization will be used frequently and thus some clarification of what is meant by these terms is warranted. Clustering, or often nanoclustering, has typically been used to describe proteins in super-resolution microscopy data that are grouped together on a sub-diffraction-limit scale (<250 nm) to a greater extent than would be expected for a random distribution (Pageon et al., 2016; Roh et al., 2015). This is in contrast to microclusters of TCRs, which form soon after activation of T cells and are evident in diffraction-limited imaging (Saito and Yokosuka, 2006; Seminario and Bunnell, 2008; Varma et al., 2006). The term TCR oligomerization is used usually in the context of biochemical studies in which extraction and separation of the complex using methods such as Blue Native polyacrylamide gel electrophoresis indicates that a proportion of TCRs have a propensity to self-associate to form dimers or other oligomers (Schamel and Alarcón, 2013). It is likely that TCR ‘nanoclusters’ and ‘oligomers’ are one and the same, and thus we often exchange the terms depending on the context of the results we are discussing.
Total internal reflection microscopy together with supported lipid bilayers (Fig. 3A,B) have been important tools for observing not only TCR clustering, but also how clusters affect TCR triggering through observations of pMHC–TCR interactions at the single-molecule level. It has long been known that upon activation, the TCR forms clusters that are evident in diffraction-limited imaging (termed microclusters), and that these are sites of active and sustained signaling (Yokosuka et al., 2005). Early work revealed the inhomogeneous distribution of TCR and associated proteins within the plasma membrane (Bunnell et al., 2002; Douglass and Vale, 2005). Fluorescence recovery after photobleaching (FRAP) imaging demonstrated that TCR clusters remained static (i.e. did not recover after bleaching), but exchanged signaling proteins, including Lck. This indicated that the inhomogeneous distribution of TCR could transiently trap signaling proteins (Douglass and Vale, 2005). TCR domains at initial contact sites of the immunological synapse have also been shown to have prolonged association with other downstream kinases, such as SH2 domain-containing leukocyte protein of 76 kDa (SLP76, also known as LCP2) and ZAP70 (Yokosuka et al., 2005). This observation was further strengthened by trapping TCR clusters at the periphery of cells using nanolithographic patterns that retarded cluster movement (Mossman et al., 2005). Trapping the clusters, and thus preventing trafficking to the central region of the synapse where internalization and downregulation of signaling occur (Fooksman et al., 2010), allowed longer association of downstream kinases, thus leading to a higher abundance of TCR–CD3-associated phosphorylation (Mossman et al., 2005) (Fig. 3C). More recently, Taylor et al. used a synthetic T cell signaling system that is based on CAR technology (Fig. 1): an extracellular SNAP tag that could be labeled with a short DNA hybridizing strand was fused with the cytoplasmic region of the ζ chain or fused on the N-terminus of the TCRβ so that it could be incorporated into the TCR complex (Taylor et al., 2017). The ‘DNA antigen’ ligand for these constructs was a membrane-anchored CLIP tag, on which a complementary DNA strand was linked. When using both constructs, it was shown by single-molecule imaging and tracking that not only were cells sensitive to single-base alterations in the DNA-antigen, but that signaling was only initiated when the bound receptors clustered (Taylor et al., 2017).
With the advent of super-resolution microscopes, it has become possible to observe TCRs at the level of single molecules at the plasma membrane (Lillemeier et al., 2010; Pageon et al., 2016; Sherman et al., 2013). Photoactivatable localization microscopy (PALM) imaging has revealed a variety of information about TCR distribution in conjunction with its signaling partners at a nanoscale resolution (Lillemeier et al., 2010; Pageon et al., 2016; Sherman et al., 2013). Lillemeier et al. utilized PALM imaging to observe the distribution of both TCR and the adaptor protein LAT clusters in T cell membrane, revealing that TCR and LAT clusters remained apart in unstimulated cells; however, these islands grew and concatenated upon antigen binding of the TCR (Lillemeier et al., 2010). This was corroborated by TEM imaging of TCR–LAT clusters, showing that the two proteins mixed in oligomeric domains when the cell was stimulated (Lillemeier et al., 2010). PALM imaging of ζ chain labeled with the photoswitchable fluorescent protein Dronpa revealed that ZAP70 was only captured by activated clusters, and that the distribution of ZAP70 within these clusters was not uniform (Neve-Oz et al., 2015). These data were used as a basis for modeling the dynamics of TCR triggering through Monte-Carlo simulations; of the six models tested, a model of cooperative conformation changes between activated TCR–CD3 molecules within a cluster (Fig. 2B) and a model of co-clustering of activated Lck with TCR could recapitulate the observed organization of activated TCR–CD3 (Neve-Oz et al., 2015). A combined approach of PALM and direct stochastic optical reconstruction microscopy (dSTORM) from our group also revealed that increased TCR density leads to an increased probability of TCR triggering, as measured by phosphorylation of the ζ chain. It was observed that TCR clusters increased in density upon engagement, and that the ratio of TCR phosphorylation correlated with the molecular density of the clusters, suggesting that the ability of the antigen to cluster TCR is a key property for triggering (Pageon et al., 2016). Implementation of dSTORM with a light-sheet microscope has allowed the imaging of naïve cells in situ within the lymph nodes (Hu et al., 2016). This revealed that within the lymph, TCR clusters were present on the surface of the naïve cells, and these clusters became aggregated in the stimulated condition (Hu et al., 2016). More recently, dSTORM coupled with membrane topology mapping has revealed that TCR clustering could be a mechanism by which the TCR concentrates at the tips of microvilli, as a means of supporting efficient scanning of APCs by T lymphocytes (Jung et al., 2016). These observations have been further tested by advances in 3D imaging, where lattice light-sheet microscopy was employed to probe the cell-scanning behavior of T cells (Cai et al., 2017). Here, they also observed concentration of TCR at the tips of the microvilli; only contacts that bore accumulated TCR led to sustained contact with the ligand, therefore suggesting that inhomogeneity of TCR distributions is a means for generating productive contacts between T cells and APCs (Cai et al., 2017).
Importantly, recent work has suggested that the TCR complex is not clustered in resting T cells and that nanoclusters are an artifact of over-counting in super-resolution techniques (Rossboth et al., 2018). Indeed, there are considerable experimental and conceptual challenges in identifying clusters by imaging. For example, a low detection frequency of receptors caused by poor labeling efficiency or low imaging power could easily result in ‘undercounting’, which would result in a more random distribution and fewer ‘clusters’. Some authors have emphasized these caveats, together with data from other techniques such as Förster resonance energy transfer and fluorescence correlation spectroscopy (Brameshuber et al., 2018), to assert that pre-assembled TCR nanoclusters do not exist (Rossboth et al., 2018). The balance of evidence from biochemical methods, electron microscopy and super-resolution microscopy discussed above suggest that, rather than existing universally within nanoclusters, a propensity for TCR self-association results in dynamic, transient nanoclusters in a subset of TCRs at any one time. Finally, whether clusters pre-exist or not, there is currently no dispute that antigen engagement by the TCR complex leads to a spatial reorganization that is intrinsically linked to signal initiation and propagation. In the following sections, we review evidence for mechanisms by which TCR clustering may impact on activation.
TCR clusters can enhance ligand–receptor interactions
Regardless of whether the TCR is pre-clustered (James et al., 2011; Minguet et al., 2007), there is ample evidence that although monomeric pMHC molecules in solution can bind to the TCR, they do not lead to T cell activation, even at high concentration (Abastado et al., 1995; Boniface et al., 1998; Cochran et al., 2000). Instead, pMHC molecules need to be reconstituted into a complex that confers multivalency, such as trimers or tetramers (e.g. by multivalent streptavidin and biotin interactions, Figs 2E and 3B) to trigger the TCR in suspended cells (Boniface et al., 1998; Cochran et al., 2000).
The requirement of multivalent ligands for activation implies that receptor clustering could be a critical requirement for T cell activation. It is likely that the multivalent ligands produce potent T cell activation because they enable rebinding to the same or neighboring receptors and are, therefore, more efficient at inducing receptor clustering (Dushek and Van der Merwe, 2014). Similar to the concept of the kinetic proofreading for TCR triggering (McKeithan, 1995), full signaling downstream of the TCR may require that the receptor–ligand engagement time lasts longer than the time required for receptor clustering. In support of this view, it was reported that a single pMHC presented on an APC was sufficient to trigger T cell activation by inducing a slow formation of TCR clusters (Huang et al., 2013). Similarly, a recent single-ligand sensitivity (as measured by ZAP70 recruitment) CAR study that used a single-strand DNA of varying length as ligand (Fig. 1) nicely demonstrated that weak ligands had a faster disassociation rate than receptor clustering rate and did not result in T cell activation (Taylor et al., 2017).
Unbound receptors have been observed to preferentially engage ligands at the site of receptors that are already engaged, suggesting that clustering might be caused by ‘zipping up’ the APC and T cell membranes through multiple and densely spaced ligand–receptor interactions (Taylor et al., 2017). Whether TCR clustering impacts on the pMHC binding kinetics has so far only be explored in simulations (Dushek and Van der Merwe, 2014), as measurements of association and dissociation rates in 3D solution and on 2D surfaces result in different values (Huang et al., 2010). The simulations propose that the close contact between two membranes restricts the diffusion, so that the disassociated ligand can rapidly rebind the receptor. This strongly suggests that a high local density of TCRs could indeed increase the ligand association rate (Dushek and Van der Merwe, 2014). Furthermore, clustering could overcome a fast 2D disassociation rate (Huang et al., 2010), as TCR clusters would increase the effective surface area and thus the probability of rebinding (Dushek and Van der Merwe, 2014). This means that even at extremely low ligand concentrations, pMHC could activate T cells through continuous rebinding. Thus, if ligand binding enhanced TCR clustering, a memory-like effect might take place as the initial binding event increases the likelihood of subsequent ligand receptor interactions (Liu et al., 2014; Martin-Blanco et al., 2018; Zarnitsyna et al., 2007). Therefore, clustering could support a nonlinear and cooperative mechanism for ligand–receptor interaction that enhances sensitivity and amplifies ligand discrimination.
TCR clustering and the kinase–phosphatase balance
In addition to enhancing ligand–receptor interactions, TCR clusters might also enhance signal transmission (Sigalov, 2010). Since Lck is constitutively active in T cells and a pool of Lck not bound to co-receptors is sufficient for activation, TCR phosphorylation might be limited by the Lck search strategy: whereas finding a single receptor or a cluster of receptors might take the same amount of time, repeated Lck–TCR interactions inside of a TCR cluster could result in amplified phosphorylation rates (Mugler et al., 2012). Lck co-clusters with TCR (Rossy et al., 2013), and phosphorylated ITAMs on CD3 chains and phosphotyrosine on ZAP70, which is recruited to the TCR complex, can bind Lck through its SH2 domain (Lo et al., 2018; Straus et al., 1996) and thus trap Lck in TCR clusters. Indeed, single-molecule tracking has demonstrated the fast and multiple rebinding of SH2-containing proteins to densely packed phosphotyrosine-containing receptors, and that this dense packing increases the dwell time of the SH2 protein at the receptor site (Oh et al., 2012).
TCR clustering could also result in the preferential spatial redistribution of membrane-bound inhibitory and activatory molecules. In resting T cells, highly abundant membrane-bound protein tyrosine phosphatases, particularly CD45, function to maintain the TCR in a dephosphorylated state (Davis et al., 2006). It is thought that a dynamic equilibrium between phosphorylation and dephosphorylation is dominated by phosphatases in resting conditions. Hui and Vale reconstituted a signaling circuit of the ζ chain with Lck and CD45 in liposomes to demonstrate that CD45 dephosphorylates the TCR at a rate more than 20 times faster than the Lck-mediated phosphorylation rate (Hui and Vale, 2014) (Fig. 3F). As discussed above, several lines of evidence indicate that CD45 needs to be depleted from the immediate vicinity of the TCR to maintain phosphorylation of CD3 chains, which is thought to occur through passive segregation due to differences in ectodomain dimensions (Choudhuri et al., 2005; Davis et al., 2006) (Fig. 2C). Dense receptor clusters can also induce this segregation through steric effects since the molecular size of the extracellular domain of CD45 is substantially larger than the lipid-anchored kinase Lck, which lacks a transmembrane and ectodomain. By using a cellular reconstitution system, James and Vale demonstrated lateral steric exclusion of CD45 from dense TCR clusters even when the extracellular domain was replaced with much smaller ectodomains (James and Vale, 2012) (Fig. 3E). Additionally, the intracellular domain of CD45 lacks motifs that bind to TCR ITAMs, whereas Lck contains an SH2 domain that can allow it to interact weakly with phosphorylated tyrosine residues in the TCR complex (Straus et al., 1996). Reduced steric exclusion and weak interactions with rapid rebinding might allow Lck to become enriched in CD45-depleted TCR clusters.
Taken together, these results suggest that upon TCR clustering, the interaction of TCR and Lck is encouraged due to the SH2 and phosphotyrosine interaction, whereas CD45 is excluded due to steric hindrance effects. This shift in the balance of kinases and phosphatases around the TCR leads to increased CD3 chain phosphorylation and ultimate activation of downstream signaling events.
Clustering-induced TCR activation
It has long been clear that cross-linking TCRs with antibodies or with pMHC tetramers can activate T cells (Altman et al., 1996; Wooldridge et al., 2009), but monomeric pMHC or Fab fragments of activating antibodies do not activate them (Wen Chang et al., 1981). Although soluble monomeric pMHC is incapable of activating T cells, soluble dimeric pMHC can activate them, provided that the linker joining the pMHC monomers is less than 50 nm wide (Cebecauer et al., 2005). Monomeric ligands are only able to activate T cells if they are presented on solid surfaces (Huang et al., 2013; Ma et al., 2008), and in this context, it is difficult to exclude the possibility of TCRs being engaged in close proximity. Clustering of pMHC–TCR bonds is likely to occur in this case because the intermembrane distance is optimal for pMHC–TCR binding where existing bonds are, and thus new interactions are funneled to regions of existing bonds (Goyette and Gaus, 2017; Taylor et al., 2017) (Fig. 2C). Several of the mechanisms discussed above could potentially explain the ability of soluble multimeric pMHC to trigger TCR–CD3 activation, which we have summarized in Fig. 2E.
In an attempt to separate the contribution of pMHC–TCR bond density from overall numbers of pMHC presented, several groups have attempted to pattern TCR ligands on the nanometer scale. Manz et al. used supported lipid bilayers partitioned with grids (Fig. 3C), which allowed for control over the number of engaged TCRs that could form clusters at a constant density of pMHC (Manz et al., 2011). Results from this system suggested that one pMHC per grid box was not enough to activate T cells, but robust activation (measured by Ca2+-flux) could be achieved when there was an average of two pMHC per grid area (Manz et al., 2011). This contrasts with the results of Huang et al. that suggest as low as one pMHC per T cell contact is enough to generate activation (measured by cytokine production) (Huang et al., 2013), which may be explained by differences in the experimental approach taken. When Manz et al. included CD80 (a ligand for the costimulatory receptor CD28) in their partitioned bilayers, they found that threshold for activation was reduced. Since Huang et al. used antigen presenting cells, rather than supported lipid bilayers, it is likely that multiple costimulatory interactions and additional adhesion interactions contributed to an enhanced efficiency of activation. Huang et al. also measured cytokine response hours after initial interaction with APCs finding ∼30% of T cell blasts respond to one pMHC per interface, whereas Manz et al. measured Ca2+ flux, a much more upstream readout of activation, within 20 mins of interaction with the bilayer. Thus it is conceivable that very low pMHC concentrations could result in delayed activation and indeed Huang et al. report a slow accumulation of TCR at the contact point with the single pMHC that seems to plateau after 120 min (Huang et al., 2013).
Taking a sophisticated approach to ligand patterning, Cai et al. used electron lithography to control the lateral and axial spacing of activating CD3 antibodies (UCHT1) on surfaces containing a constant level of intercellular adhesion molecule 1 (ICAM-1) presented on supported lipid bilayers (Cai et al., 2018) (Fig. 3D). When the axial spacing of ligands above the surface is increased by 10 nm, CD45 was only weakly excluded and CD3 chain phosphorylation showed a sharp threshold with respect to lateral spacing of ∼50 nm (Cai et al., 2018). This 50 nm lateral spacing threshold is in line with a previous report (Delcassian et al., 2013) and compares well with the spacing in soluble dimers that supports signaling (Cebecauer et al., 2005). Cai et al. attributed this lateral threshold without CD45 exclusion to the distance ZAP70 can reach when bound to the cytoplasmic tails of CD3 and ζ chains and therefore the distance at which the activities of ZAP70 could synergize to phosphorylate LAT (Cai et al., 2018). Despite being a tour de force of ligand patterning, the sole readout in this report was pan anti-phosphotyrosine staining and thus it is difficult to determine whether the observed differences are due to less efficient TCR phosphorylation or inability of the signal to progress to phosphorylation of downstream effectors like LAT. Future work will hopefully shed light on what aspect of signaling is most sensitive to lateral TCR density, which could give valuable clues as to the nature of the mechanism.
The lipid environment and TCR clustering
Clustering of TCRs could also impact on the local lipid environment, which could in turn affect signaling. Several lines of evidence suggest lipid-ordered domains and protein clustering share a bidirectional relationship – clustering of proteins that favor lipid-order can create large-scale regions of lipid order, which in turn enrich or deplete other proteins (Levental and Veatch, 2016; Raghupathy et al., 2015; Stone et al., 2017). Since Lck and LAT favor lipid-ordered domains and CD45 does not (Stone et al., 2017), this could shift the kinase–phosphatase balance around the TCR and facilitate the phosphorylation of LAT by ITAM-recruited ZAP70 (Fig. 2D). A similar mechanism was recently suggested for the B cell receptor (BCR). In B cells, clustering of the BCR induces a local region of lipid order, which also enriches Lyn, the Src family kinase responsible for phosphorylating the BCR (Stone et al., 2017). Furthermore, depletion of CD45 in the vicinity of BCR clusters occurred, but the authors did not separate the steric effects previously described for CD45 exclusion from TCR clusters (James and Vale, 2012). Interestingly, ordered domains extended beyond the BCR cluster and were diminished by the Src family kinase inhibitor PP2, indicating that recruitment of other signaling molecules could contribute to the growth and stabilization of lipid-ordered domains surrounding BCR clusters (Stone et al., 2017).
Conversely, the lipid environment may also have important impacts on TCR clustering and function. When reconstituted into large unilamellar vesicles (LUVs) containing a mixture of phosphatidylcholine and sphingomyelin, the TCR forms nanoclusters or oligomers, but remains monomeric in LUVs composed only of phosphatidylcholine (Molnár et al., 2012) (Fig. 3G). Removal of cholesterol from T cell membranes using methyl β-cyclodextran reduced the binding of pMHC tetramers, indicating that cholesterol-dependent TCR clusters may enhance pMHC binding (Molnár et al., 2012). In support of a positive role in TCR activation, inhibition of cholesterol esterification enhances responses in CD8+ T cells by increasing plasma membrane cholesterol levels and TCR clustering (Yang et al., 2016). In contrast to these positive roles, another report identified a specific binding site for cholesterol in the transmembrane domain of the TCRβ subunit, where evidence suggested that binding locks the TCR into an inactive conformation that is inaccessible to phosphorylation, regardless of pMHC binding (Swamy et al., 2016).
It is also conceivable that changing the lipid environment could change the association of the CD3ε and ζ chain cytoplasmic tails with the membrane. CD3ε and ζ chains with phosphorylated ITAMs do not associate with the membrane; however, it is unclear whether disruption of membrane association is required for phosphorylation or whether phosphorylation disrupts membrane association (Ma et al., 2017). Since phosphatidylserine species typically have long acyl chains and are found within lipid-order phases (Raghupathy et al., 2015), it is conceivable that changes in lipid order around the TCR could have an impact on CD3ε and ζ chain membrane association, or vice versa.
The precise role of the local lipid environment surrounding the TCR remains elusive, with both positive and negative regulatory functions proposed. One consistent finding appears to be that there is a relationship between cholesterol and nanoscale clustering of the TCR, but a deeper understanding of this relationship is made challenging by the dynamic nature of the nanoscale lipid domains (Levental and Veatch, 2016) and the caveats inherent in methodologies of probing them (Sevcsik and Schütz, 2016). As improvements are made in methodologies, the precise role of the lipid environment will hopefully become clearer.
Conclusions and perspectives
TCR clustering is clearly observed post triggering, but it is currently debated to what degree TCRs exist in nanoscale clusters before triggering. Part of this controversy may be a result of all-or-nothing views that TCRs must all exist either as stable, pre-formed nanoclusters or oligomers, or as randomly dispersed, independent units, whereas the reality likely lies somewhere in between, as was suggested previously (Schamel et al., 2005). Given the close connection between TCR clustering and cholesterol, a deeper understanding of the nanoscale organization of TCR complexes may require further developments in our understanding of lipid nanodomains.
Whereas collective evidence has shown that activation of T cells is more efficient with multimeric ligands and that the TCR forms clusters upon activation, it remains unclear if ligand-induced clustering of TCR alone can truly initialize TCR triggering. Minguet et al. went some way to addressing this by cross-linking the TCR with different length linkers between pMHC and then using co-immunoprecipitation of Nck and the TCR as a readout for conformational changes (Minguet et al., 2007). The results indicate that close engagement of two TCRs by pMHC is necessary to induce both conformational change and activation, but that conformational change alone (induced using anti-CD3ε Fab fragments) is insufficient (Minguet et al., 2007). Since the importance of Nck recruitment to T cell activation has been questioned (Szymczak et al., 2005), future studies dissecting the effects clustering and conformational change on ITAM phosphorylation will be important for clarifying the relationship of these processes with triggering.
Finally, in addition to being an exciting clinical strategy for the treatment of cancer (Sadelain, 2016), CARs have also been highly useful in investigating the molecular mechanisms of TCR activation (Taylor et al., 2017). In some ways it is surprising that, in the context of a CAR fused to various transmembrane and extracellular domains (Sadelain, 2016), the cytoplasmic portion of ζ chain still signals very similarly to when it is in the native environment of the TCR. One could argue that CARs trigger signaling by a completely distinct mechanism from TCRs, but in our view, it seems more likely that they share a single mechanism. From this perspective, CARs could be invaluable tools for investigating which mechanisms are critical for TCR triggering and which may be related to enhanced sensitivity or affinity discrimination.
The authors would like to acknowledge funding from the Australian Research Council (ARC) Centre of Excellence in Advanced Molecular Imaging (CE140100011 to K.G.) and National Health and Medical Research Council of Australia (1059278 and 1037320 to K.G., APP1139003 to Y.M., and APP1163814 to J.G.).
The authors declare no competing or financial interests.