P. A. van der Merwe, Sir William Dunn School of Pathology, University of Oxford, OX1 3RE, Oxford, UK. Email: email@example.com Senior author: P. A. van der Merwe
Results from live-cell microscopy suggest that the behaviour of isolated components of the T-cell activation machinery in vitro does not represent the reality inside cells. Understanding the cellular-scale dynamics of microcluster migration can only be accomplished by in situ observation. Developments in ‘super-resolution’ microscopy have permitted investigators to move beyond tracking the movements of individual molecules, allowing the recognition of protein islands and nanodomains present in quiescent and active T cells. Many high-resolution techniques have their own susceptibilities to artefacts, so it is important to take a multifaceted approach to confirm results. A major challenge for the future will be to integrate all the new information into a coherent model of antigen recognition and T-cell activation.
To become activated, T cells must sample antigen displayed as peptide–MHC (pMHC) complexes on the surface of professional antigen-presenting cells (APCs). For years it was thought that once the T-cell receptor (TCR) encounters pMHC to which it has specificity, binding is prolonged to facilitate cytokine exchange across a relatively homogeneous interface called the immune synapse.1 Only in the late 1990s did researchers discover that the T-cell immune synapse in fact consists of an ordered, yet highly dynamic, arrangement of signalling molecules into microclusters.2,3 Minutes after the formation of a stable cell–cell junction in response to a cognate TCR–pMHC interaction, the TCR microclusters begin to migrate to the centre of the junction in an actin-dependent fashion4 to form the central supramolecular activation cluster (c-SMAC). This is in contrast to other microclusters, such as those containing lymphocyte function-associated antigen 1(LFA-1) [complexed to intercellular adhesion molecule 1 (ICAM-1)], which accumulate around the c-SMAC to form the peripheral supramolecular activation complex (p-SMAC) giving an overall shape reminiscent of a ‘bullseye’2,3 – for a diagrammatical representation see reference 5. Questions about T-cell triggering include the mechanism of microcluster segregation into the SMACs, the role of lipid rafts in the signalling process, and the stoichiometry and kinetics of the central component, the TCR.
The discovery of microcluster segregation by Monks et al.3 underlined the importance of using live cells to study dynamics. T-cell triggering is in fact a superb trial of live-cell imaging techniques because events occur over a range of time scales, from rapid single molecule exchange between microclusters to the overall formation of the supramolecular complex ‘bullseye’ shape. Therefore, high temporal and spatial resolution is required. Furthermore, molecular components of the T-cell activation machinery can be highly sensitive to their local environment.5 Their study demands techniques that allow them to be observed in a setting that replicates as closely as possible the in vivo reality.
This review explores new techniques that are beginning to meet these criteria. The text is written to inform a biologist’s perspective, so the methods are presented in the context of biological questions that they can help answer; however, it is beyond the scope of a short review to do justice to all the published data on models of T-cell activation. With this in mind, the discussion begins with the still fundamental tools for studying the immune synapse: total internal reflection microscopy (TIRFM) and supported lipid bilayers.6 Next, techniques that facilitate the study of nanoplatforms and protein nanodomains are described. The conflict surrounding the valency of the TCR complex is considered in Box 3, as this highlights the pitfalls of different methods.7,8 Finally, a recent study of TCR-pMHC binding kinetics at the single molecule level5 is examined. Two promising super-resolution techniques that have been used fruitfully in live cells but not yet applied to T-cell triggering are looked at in Boxes 1 and 2.
Box 1. Prospects for live-cell ‘super-resolution’ fluorescence microscopy: SIM
‘Super-resolution’ fluorescence microscopy techniques include pointillism microscopy (PM), stimulated-emission-depletion (STED) microscopy and structured illumination microscopy (SIM). These techniques are so named because they can resolve details below Abbe’s optic diffraction limit of 200 nm, yielding images approaching the resolution of transmission electron microscopy but with reduced risk of artefacts because of cell fixation and labelling. SIM improves lateral resolution to around 100 nm by illuminating the sample with light patterned into parallel lines, so transferring information which is usually outside of the observable region in Fourier space into the observable region.54 The concept behind SIM can be represented intuitively with Muiré fringes – see accompanying diagram (illustration after Gustafsson54). Alone, the boxes of parallel lines resemble homogeneous blue squares. However, when superposed at an angle, Moiré fringes appear, which are coarser than the details in the original pattern. Furthermore, the sites of aberrant spacing between lines (marked by arrows) are visible at a lower resolution with pattern superposition. In early studies, the patterned light was produced by mechanically moving a filter through three to five different parallel positions in three different rotational orientations. This meant that it took at least a few seconds to collect a complete set of the nine to fifteen frames required to generate one SIM image.55 Given the substantial dynamism of individual molecules in the immune synapse on a timescale of 0·5–10 seconds or less,12,28 SIM in this incarnation was not yet ready to be applied to the immune synapse. On the other hand, recent advances in rapid light patterning have increased the rate of frame capture to ∼35 Hz, with a corresponding SIM image capture rate of ∼3·9 Hz.56 In a small field of view, visualization of microtubule dynamics and kinesin transport has even been performed at a SIM image capture rate of 11 Hz.56 This is still a long way from the > 100 frames per second taken as standard with conventional total internal reflection microscopy,12 but the payoff is a dramatic increase in spatial resolution.
Box 2. Prospects for live-cell ‘super-resolution microscopy: STED’
Stimulated-emission-depletion (STED) was originally proposed in 199457 and like confocal microscopy is a scanning technique, where to reduce background only minute volumes are subject to focused excitation light at any given time. The major advantage of laser scanning techniques over total internal reflection microscopy (TIRFM) is that they are not dependent on a wave evanescing from the coverslip, and can therefore probe > 100 nm into the cytoplasm. The increase in resolution is accomplished by reducing the confocal volume. In conventional confocal microscopy, light at the excitation wavelength is focused into a spot, creating a three-dimensional volume containing excited fluorophores. Because of diffraction, the smallest possible full width at half maximum intensity (FWHM) of the focused spot is larger than 200 nm in the x–y axis and 500–700 nm in the z-axis11 (see (a)). In STED, excitation light is rapidly followed (in the order of nanoseconds) by a second doughnut-shaped beam of light. This doughnut-shaped beam is the STED beam, and is at a wavelength that ‘stimulates emission’ of photons from the fluorophores in the same direction as the incident beam, in effect quenching them (see (b)). Only at the zero point of the STED pulse (i.e. the centre of the x–y cross-section) do all the fluorophores remain excited. In 2000 Klar et al. improved upon the initial idea by stimulating emission on either side of the sample volume along the optical axis, giving an almost spherical fluorescence spot of around 90–110 nm in diameter.58 Lateral resolution below 10 nm (approximately half the diameter of a ribosome) has been reported for STED59 although this was not a biological sample. Because of the miniscule fluorescence spot generated in STED, the scanning process is more time-consuming than in conventional confocal microscopy – an obstacle to video-imaging of live cells. Furthermore, the small spot size and rapid scanning make it difficult to detect enough photons to confidently state whether fluorescence is above background. In live cell microscopy, photobleaching and phototoxicity because of STED are reportedly acceptable.60 STED recently facilitated video microscopy of nanoscopic vesicles in neuronal boutons.61 These are small neuronal regions around 1 μm in diameter. This meant that the overall scanned area was possibly smaller than it would have to be to meaningfully image the immune synapse. Taking a different approach, the use of STED to probe retardations in plasma membrane lipid movement only visible on a nanometric scale62 could be useful given the debatable importance of lipids and rafts in T-cell signalling.29,32–34
Total internal reflection microscopy and lipid bilayers
Fluorescence microscopy is well established9 and useful because it allows the specific labelling of species of interest (although the labelling strategy and potential over-expression of the species may affect its behaviour). TIRFM has reinvigorated the field as it allows the microscopist to view a very narrow slice of sample with no background. This advantage, combined with commercial availability of the equipment, has led to its widespread use.10 TIRFM refers to a microscope geometry that generates a wave of light travelling away from the coverslip at a perpendicular angle. This wave rapidly evanesces (i.e. fades), so it only penetrates around 100 nm into the sample. Therefore, only fluorophores lying in a very narrow plane near the coverslip are excited. TIRFM is ideal for studying molecules in the plasma membranes of cells in contact with the coverslip, although it is unsuitable for probing deeper into the cytoplasm. Generally, TIRFM studies of the immune synapse involve ligation of at least one surface receptor on the T-cell, such as the TCR or an accessory molecule like CD2. Often activating ligands are incorporated into a supported lipid bilayer (SLB) that sits on top of the coverslip.6,11 There are several well-established advantages to using SLBs. Unlike directly coating the coverslip with ligands, the fluidity of SLBs allows lateral movement of incorporated molecules, facilitating study of the spatio-temporal dynamics of their bound partners. Furthermore, the lipid bilayer represents a more physiological surface than a bare glass coverslip.
Supported lipid bilayers were instrumental in recent attempts to elucidate the relationship between actin dynamics and microcluster migration. By incorporating CD58 molecules (the CD2 binding partner found on APCs) into SLBs, Kaizuka et al. determined that CD2 microclusters follow a centripetal trajectory towards the c-SMAC, in a similar fashion to their previous report on TCR microclusters.12,13 However, as the observed rate of CD2 translocation was substantially lower than for TCR, they argued that CD2 is more weakly coupled to centripetally travelling actin than TCR. One inventive method for elucidating the linkage between surface receptors and actin is the use of ‘spatial mutation’: physical partitioning of the SLB with extremely fine strips of chromium (typically 0·1–1 μm in width, 5 nm in height and 3–4 μm in pitch). When deposited onto the supporting glass coverslip, these chromium strips limit the lateral movement of individual lipids and proteins (Fig. 1b).14–16 If T cells are applied to CD3ε-containing SLBs that have been partitioned in this manner, TCR–CD3ε complexes tend to form clusters against the chromium strips. Using confocal microscopy to look further into the cell than TIRFM allows, Yu et al.15 observed that actin speckles lose centripetal velocity as they pass beneath these physically restrained TCR clusters, and regain velocity once they move clear of the clusters. Taken together these results provide substantial corroborating evidence for a model of microcluster segregation involving differential slippage in receptor–cytoskeleton coupling.12,13 Further work by Hartman et al.17 used cross-linking antibodies to demonstrate that in contrast to non-clustered LFA-1–ICAM, clustered LFA-1–ICAM can proceed to the c-SMAC, possibly because clustered molecules lead to a more concentrated actin-coupling force. This complements a model for segregation where TCR microclusters, which are stable in the absence of actin, can proceed to the actin-poor c-SMAC, whereas microclusters that break up in the absence of actin, such as ICAM–LFA-1 microclusters, are only observable in the actin-rich p-SMAC.12 Another kind of ‘spatial mutation’ involves depositing activation ligands in a pre-defined pattern onto glass coverslips.47 When cells settle on the micro-patterned cover slips, it is possible to observe the dynamics of membrane species as they migrate towards the activating region. Williamson et al.47 made use of this form of micro-patterning to observe bunching of actin at the boundary between the activating and non-activating surfaces, suggesting that the mechanism of linker of activated T-cells (LAT) recruitment to the activating site is not dependent on laterally moving actin.
Supported lipid bilayers are imperfect mimics of the dynamic surface of a live cell, and it is worth considering their shortcomings. The most obvious disadvantage is that unlike real APCs, SLBs are inert. Second, replicating the enormous diversity and functionality of the lipids and proteins found in the plasma membrane is impossible to practically achieve in an artificial setting. Third, unlike SLBs, the plasma membranes of APCs are not planar. The influence of membrane components on the curvature of the bilayer,19 and conversely the influence of curvature on membrane domain organization,18,20 are becoming increasingly documented. Immunologists should be particularly attentive to topography because steric effects are postulated to have a central role in the kinetic segregation model of TCR activation.21,22 Fourth, friction between the lower leaflet and the support can dramatically alter the diffusion kinetics of bilayer lipids and their attached proteins.23,24 Furthermore, the close proximity between the lipids and the support makes incorporation of transmembrane proteins unfeasible. This has led to a dependence upon His-NTA, GPI- or biotinyl- linkage of proteins to the head groups of lipids, which is usually distinctly unphysiological (Fig. 1a).25 The use of lipid bilayers on polymer-based supports allows the generation of SLBs containing full transmembrane proteins (as opposed to lipid-anchored proteins). Such polymer-supported bilayers were recently used to study the v-SNARE requirements for vesicle fusion (Fig. 1c).26 However, to be used with live cells these polymer-supported SLBs would have to demonstrate compatibility with a background reducing platform such as TIRFM (Fig. 1c).
A promising approach that sidesteps altogether the need for artificial SLBs involves the use of molecular ‘tweezers’ to manipulate the orientation of cell–cell conjugates.27 In this manner, the interface between two cells can be manipulated such that it lies entirely within the focal plane, removing the usual requirement in confocal microscopy to capture multiple ‘slices’ of sample. This technique therefore allows the visualization of true T-cell–APC synapses, while achieving reasonable spatial and temporal resolution (∼ 250 nm lateral resolution, which is typical of confocal microscopy, and an image capture rate > 1 Hz).27
From tracking single molecules to observing nanodomains
To resolve and track the positions of individual fluorophores with conventional TIRFM, it is required that the point spread functions (PSFs) from individual fluorophores do not substantially overlap (Fig. 2b – the PSF can be thought of as the size of the fluorescence dot detected by the microscope from a single fluorophore). There are two labelling strategies commonly used to overcome this. For proteins with cytosolic domains, low-level expression of a recombinant fluorescent protein, sample protein fusion may be used. For proteins that have accessible ectodomains such as TCR, co-stimulatory receptors and adhesion molecules, fluorescently labelled antibodies can be used to tag a small fraction of the total amount of protein of interest.12,28,29 Either way, observing lipid rafts or protein complexes presents a problem: low-level labelling renders identification of clustered molecules impossible, as the very purpose of low-level labelling is to avoid homo-association of labelled molecules; whereas high-level labelling will result in homogeneous fluorescence (Fig. 2a,b). This section looks at three fluorescence techniques that can be used to help understand the nature of complexed components of the T-cell activation machinery. Different techniques have their own susceptibilities to artefacts, and these are discussed in light of the conflicting results from two studies on the resting TCR in Box 3.
Thinning out clusters while conserving stoichiometry of labelling
Thinning out clusters while conserving stoichiometry of labelling (TOCCSL) works analogously to fluorescence recovery after photobleaching (FRAP). It achieves a virtual reduction of the labelled protein density by the use of a pulse of light that bleaches a high proportion of the fluorophores in a given area. Unlike FRAP, however, TOCCSL examines single centres of fluorescence as they diffuse back into the photobleached region (Fig. 2c), with points of high fluorescence representing clusters of labelled molecules.30 An initial application to Jurkat T cells showed a clear reduction of homo-associated lipid raft marker (GPI-linked green fluorescent protein) following either chemical or enzymatic depletion of cholesterol.31 The ability to definitively differentiate nanoscopic lipid rafts or protein islands from a random distribution of molecules could prove timely given the controversy surrounding the relevance of lipid rafts to T-cell activation.29,32–34 One would have to be careful when interpreting TOCCSL results. Less mobile fractions (e.g. large clusters) may move into the photobleached area more slowly, meaning that they are under-represented.
So far, TOCCSL has only been used to study plasma membrane components. This is presumably because to reduce background to interpretable levels it must be used with TIRFM. However, TOCCSL can be used to infer events inside the T-cell cytoplasm, as demonstrated by a recent study into the insertion of cytosolic Lck into the plasma membrane of Jurkat T cells.35 After photobleaching of membranous Lck fused to yellow fluorescent protein (Lck-YFP), Zimmerman et al.35 observed the appearance of rapidly diffusing, randomly appearing fluorescent particles in the photobleached area. These were presumably cytosolic Lck-YFP molecules drifting in and out of the evanescent field generated during TIRFM. Interestingly, between 0 and 75 ms after photobleaching they noticed a gradual swelling of the population of less rapidly diffusing molecules. They postulated that this was a result of the insertion of cytosolic Lck-YFP into the membrane, with a corresponding fall in diffusion constant.
Photoactivated Localization Microscopy (PALM)36 and stochastic optical reconstruction microscopy (STORM)37 are ‘pointillism’ microscopy (PM)38 techniques and are part of the ‘super-resolution’ optical microscopy vanguard (see Boxes 1 and 2). Like TOCCSL, PM overcomes the problem of PSF overlap by decreasing the density of active fluorophores. In PM, sample molecules are tagged with photo-activatable or photo-switchable fluorophores. A weak burst of homogeneously applied light at the appropriate wavelength is used to turn on a small, randomly distributed proportion of the fluorophores, whose position can be precisely determined. This is followed by deactivation of the active dyes. The process is cycled, allowing the progressive development of an image that can resolve molecules tens of nanometres apart (Fig. 2d–f). There is a technical expense of requiring two laser wavelengths per dye, with complications for multicolour labelling and a prolonged data acquisition time. The development of multicolour PM using both synthetic dyes and fluorescent proteins39–42 continues to expand PM’s range of applications. Reports of PALM yielding super-resolution information in all three axes in studies of focal adhesions between the cytoskeleton and the extracellular membrane are particularly exciting43–45 given the paucity of three-dimensional information about the T-cell–APC junction.46
Pointillism microscopy has already had an impact on our understanding of signalling downstream of the TCR.47,48 The Gaus group’s analysis of the surface of activated T cells by PALM revealed previously undescribed heterogeneity between clusters of phosphorylated LAT.49 Around the same time, Mark Davis’s group used high-speed PALM (hsPALM) to follow the behaviour of TCR and LAT ‘islands’ during the process of T-cell activation.8 In the membranes of quiescent T cells they observed nanoscopic islands containing TCR or LAT. In activated T cells, these islands appeared to have associated, giving rise to TCR and LAT microclusters, but curiously the individual islands within the microclusters remained intact and intermingling of TCR and LAT molecules did not occur. This puzzling result may be explained in part by a more recent revelation from the Gaus group.47 Using PALM in the TIRFM geometry, over the course of 47 seconds they visualized LAT clusters growing in fluorescence intensity and then fading. This is characteristic of species localized in vesicles – possibly accounting for the failure of individual LAT and TCR molecules to mix. These observations helped to determine that it is vesicular LAT, and not plasma-membrane-bound LAT, which makes the bulk contribution to signalling.48
Two-colour coincidence detection
The observation with hsPALM of nanoscopic islands containing multiple TCRs in resting T cells seemingly contradicts an earlier study by James et al. that used two-colour coincidence detection (TCCD).7 Rather than generating a two-dimensional image of the cell–slide interface, TCCD works by analysing a very small area of the cell surface.7,50,51 Two populations of Fabs were produced that both bound the same TCR β-chain epitope, but were labelled with different coloured fluorophores.7 As all the Fabs had identical specificity, association of more than one Fab with the same TCR complex would only be possible if the complex contained more than one αβ heterodimer. Whether or not a significant proportion of TCR complexes had bound more than one Fab was determined as follows. Two laser beams at the excitation wavelengths of the two fluorophores were overlapped so that they both focused on the same point on the upper surface of settled, resting T cells. One would only expect to see simultaneous emission in both fluorescence channels if the confocal volume contained both fluorophores (and therefore, more than one TCR β-chain) at once. The proportion of dual-channel fluorescence events above background can be quantitatively expressed as Q.50,51 The investigators found that the value of Q for the tagged β-chain TCR was no higher than for CD86, a known monomer, and that both were significantly lower than the Q value for dual-tagged CD28, a known homodimer. They therefore concluded that each TCR complex only contains only one αβ heterodimer, and that the complex is therefore monovalent (see Box 3).
Box 3: Resolving the conflict of T-cell receptor (TCR) islands in resting T cells
There are several ways to reconcile the fact that two-colour coincidence detection (TCCD) observed monovalent TCR in resting cells, whereas high-speed photoactivated localization microscopy (hsPALM) detected islands of TCR7,8 (see body text for an introduction to the techniques). PALM in live cells may have a predisposition to detect immobile species over mobile species, as Lillemeier et al. acknowledge.8 A study using FRAP reported a TCR diffusion constant of ∼ 0·12 μm2/s.63 It is conceivable that this averaged figure masks a highly mobile fraction. If these species move sufficiently during a single acquisition frame then their photon emission in any one pixel might not register above background. Assuming clustered TCR is less mobile than unclustered TCR, hsPALM may selectively detect the clustered molecules (although this charge could possibly be levelled against all fluorescence video-microscopy, as one must always detect a threshold number of photons per frame in a given pixel). The reverse may be true for TCCD, and other fluorescence correlation spectroscopies. For TCCD to detect an event, the fluorophore has to diffuse into the area being excited by the overlapped lasers. Therefore, a less mobile fraction will not be so readily detected, skewing the calculated ratio between the species towards the more mobile fraction. It would be interesting to see if thinning out clusters while conserving stoichiometry of labelling (TOCCSL)30 (see text) is able to detect a mixture of TCR islands, or whether their immobility prevents their detection by this technique also.
One explanation64 that does not rely on microscopic artefacts is that the TCR islands observed by Lillemeier et al. using hsPALM8 had formed in response to contact between the cell surface and the lipids in the supported lipid bilayer, so represent a state of partial T-cell activation.8,64 Regardless, the result from TCCD indicates that at least in some cases, some TCR is monovalent. Previous debate has envisaged the TCR complex existing in a defined stoichiometry. The report of variable amounts of CD3ζ in TCR islands, on the other hand, recalls lipid rafts.8 A model involving lipid rafts is more compatible with the long range diffusion of single molecules observed in an earlier study.28
Studying the binding kinetics of individual pMHC complexes
Traditional kinetics studies reveal only the ensemble behaviour of a large population of molecules. However, the binding kinetics of an individual molecule can alter dramatically depending upon the local conditions. Huppa et al.5 studied the formation of individual TCR–pMHC complexes using Förster resonance energy transfer (FRET). The TCR on T cells was tagged with a donor fluorophore, and pMHC in an SLB was tagged with an acceptor fluorophore. Using TIRFM, the binding of individual TCR complexes to pMHC complexes was therefore observable as FRET signal in the acceptor fluorophore channel. The authors demonstrated that the half-life of the synaptic TCR–pMHC complex in situ was 3-fold to 12-fold shorter than the solution-state half-life. The TCR–pMHC half-life was prolonged upon treatment of the cells with actin-depolymerizing agents, suggesting that actin destabilizes the TCR–pMHC interaction. It would be interesting to probe the relationship between in situ off-rate and the potential for T-cell activation using a panel of peptides restricted to the same MHC II molecule, as has been done for in vitro off-rates using surface plasmon resonance.52 Surprisingly, from cluster to cluster within a given immune synapse the two-dimensional Kd for the TCR interaction with pMHC was found to vary by over two orders of magnitude. However, when interpreting two-dimensional and three-dimensional Kd measurements made in this manner it is important to exercise caution. Two-dimensional Kd would typically be used to describe the degree of association between species constrained to diffuse in two dimensions in the same plane, and could be considered a measurement of the intrinsic binding chemistry. In this case, however, TCR and pMHC are confined to different membranes, and hence to two separate two-dimensional planes. The observed differences in two-dimensional Kd therefore arise from factors intrinsic and extrinsic to the chemistry of binding, most obviously the distance between the two host membranes. This could be influenced by a variety of factors such as the presence of molecules with bulky ectodomains or ruffling of the T-cell surface. Making sense of these observations will therefore require analytical methodologies that take account of diverse biological influences such as proximity of signalling molecules and cytoskeleton dynamics.5,53
Supported lipid bilayers and TIRFM remain indispensible for live-cell studies of immune synapse dynamics. Although still in their infancy, these techniques have begun to highlight gaps in our knowledge whose existence was unsuspected a few years ago. Williamson’s discovery of the importance of vesicular LAT47 demonstrates how high-resolution information can lead to revisions of established signalling models. It also illustrates the dangers of proposing an overarching model of T-cell triggering before the behaviour of all the components is fully understood. The conflicting results surrounding TCR clustering highlight that each method has specific pitfalls and that different techniques are necessary complements to each other. Homing in on the TCR, explaining Huppa et al.’s observation that the kinetics of TCR–pMHC binding varies even on the same contact interface5 will require the accrual of still more detailed information about factors local to each signalling cluster. Looking further ahead, the challenge now is to take the nano-scale information we are gaining and integrate it with our micro-scale knowledge of cluster migration and SMAC formation. The resulting model may not be wholly intuitive.48
We are grateful to Simon Davis, Facundo Batista, Ian Dobbie and Alexander Douglas for constructive feedback on the manuscript.