Effects of a local auxiliary protein on the two-dimensional affinity of a TCR–peptide MHC interaction

The interaction between T-cell receptors (TCRs) and different peptide-presenting major histocompatibility complex molecules (MHCs) is vital for the response of T lymphocytes in the adaptive immune system. It both decides whether the T lymphocyte should undergo positive and negative selection in the thymus as well as being the initial step in triggering a T-cell response against foreign pathogen (van der Merwe and Dushek, 2011). The affinity of the TCR–MHC interaction has been considered to be an important factor in how the T cell responds in these settings, and has been observed to correlate with T-cell signaling potency (Huang et al., 2010; Huppa et al., 2010). However, it is, to date, not understood how the TCR can distinguish between activating and non-activating ligands (Chakraborty and Weiss, 2014; Dustin et al., 2001; Morris and Allen, 2012). To understand this further, various techniques have been developed to study the relatively weak interactions of TCRs with ligand-presenting MHCs. Surface plasmon resonance was one of the first techniques used to measure the interaction of soluble TCRs to immobilized MHCs, generating three-dimensional (3D) dissociation constants (K_d) (Garcia et al., 1997; Matsui et al., 1994). A vast diversity of TCRs and MHCs have since been investigated ranging from agonistic (1–200 µM) to antagonistic interactions (>5 µM) (Alam et al., 1999; Rosette et al., 2001; Rossjohn et al., 2015; Stone et al., 2009; Sumen et al., 2004). However, 3D K_ds can differ substantially from binding affinities in a cell–cell contact, where both receptors and ligands are laterally confined in two dimensions (2D) and are affected by cellular dynamics and protein flexibility (Hu et al., 2013; Huang et al., 2010; Huppa et al., 2010; Xu et al., 2015). Unfortunately, measuring 2D K_ds is considerably less common, and more difficult, than performing 3D affinity measurements. The number of measured 2D K_ds of TCR–MHC interactions is therefore limited (Dustin et al., 1997; Grakoui et al., 1999; Huang et al., 2010; Huppa et al., 2010; Zhu et al., 2007, 2013), hampering the development of new theories to explain T-cell signaling and antigen discrimination. Furthermore, there are multiple different types of molecules binding across the cell–cell contacts in vivo that can influence the binding behavior in the contact. Auxiliary binding molecules have previously been used in 2D K_d measurements to ensure contact formation and to promote bond formation between weaker-binding ligands and their receptors (Grakoui et al., 1999; Jönsson et al., 2016). However, it is not well known how or whether this affects the 2D K_d of the studied system (Grakoui et al., 1999; Jönsson et al., 2016; O'Donoghue et al., 2013).

The aim of the present study was to investigate how the 2D K_d of a prototypical TRBV9⁺ TCR (L3-12) binding to its agonist MHC class II molecule, the human leukocyte antigen protein DQ8 presenting the α-1 gliadin peptide (HLA-DQ8-glia-α1), depends on cellular context and, in particular, how this interaction is influenced by auxiliary binding molecules in the cell contact. HLA-DQ8-glia-α1 is known to be associated with celiac disease, a T-cell-mediated autoimmune-like disorder characterized by repeated patterns of TRBV9⁺ TCR usage against an immunodominant α-1 gliadin determinant (Broughton et al., 2012). The L3-12 TCR has a higher solution affinity (7 µM) (Broughton et al., 2012) than typically observed for microbial and non-self-peptide-presenting MHC class II molecules (∼30 µM) (Cole et al., 2007) and binds considerably more strongly than autoreactive TCR–MHC complexes (100–200 µM) (Bulek et al., 2012; Deng and Mariuzza, 2007). However, a 2D K_d value for this interaction has not been determined.

Rat CD2 (rCD2) binding the rat CD48 mutant T92A (rCD48_T92A) was used as the auxiliary binding pair in this study. The rCD2–rCD48_T92A interaction has a 3D K_d of 11 µM, which is comparable to that of the human adhesion pair CD2–CD58 (Evans et al., 2006). Both HLA-DQ8-glia-α1 and rCD48_T92A were expressed in Jurkat T cells, which do not naturally express these receptors, and the cells were allowed to bind to supported lipid bilayers (SLBs) with different amounts of fluorescently labelled L3-12 TCR and rCD2 ligands. By using this model system, three sources of confounding effects were avoided: (1) the effects of T-cell activation on receptor reorganization and binding (Limozin et al., 2019; Zhu et al., 2006), allowing us to focus on the conditions prior to T-cell signaling; (2) the influence co-receptor binding can have on the TCR-MHC 2D K_d, which could distort the measurement; and (3) the budding off of TCR-enriched microvesicles into the contact (Choudhuri et al., 2014), which would result in an underestimation of the true 2D K_d value by the Zhu–Golan method. By removing these effects from the experiments, we could focus on the direct influence that auxiliary molecules and cellular context have on the TCR-MHC 2D K_d.

Jurkat T cells expressing HLA-DQ8-glia-α1 were allowed to attach to an SLB presenting fluorescently labeled L3-12 TCR (Fig. 1A). The L3-12 TCR bound to HLA-DQ8-glia-α1 on the cells and accumulated beneath the cells visibly as an increase in fluorescence intensity (Fig. 1B). This fluorescence intensity can be converted into the total sum of ligands in the contact (bound and free, B+F). The intensity outside the contact gives the free ligand density in the contact (F), after multiplication with a correction factor to take into account the exclusion of free ligands in the contact (see Materials and Methods for details). The subsequent addition of more TCR to the same SLB resulted in a shift in F and B, corresponding to higher free ligand densities and more bound receptors. The changes in B, F and p were plotted in a Zhu–Golan plot as defined by Eqn 1 (Fig. 1C). It was confirmed that using SLBs with multiple additions of ligands did not influence the profile of the Zhu–Golan curve compared to when different SLBs were used for the studied concentrations (Fig. S1). The ratio of bound versus free ligands decreased with an increasing number of added ligands. However, at high ligand densities, typically when the contact-to-surface area ratio p was above 0.2, the Zhu–Golan plot started leveling off with only minor changes in B/F. This effect will be discussed below. For Fig. 1, only ligand density data points below B×p=300 molecules/μm² are included where these effects are non-significant. The Zhu–Golan data points here follow Eqn 1 and fitting the data gave a 2D K_d of 14±5 molecules/µm². This value is in the range of previously measured 2D K_d values of TCR-agonist MHCs (10–110 molecules/µm²) (Grakoui et al., 1999; Huppa et al., 2010; Pielak et al., 2017).

The total amount of mobile HLA-DQ8-glia-α1 molecules on the cells was N_tot×f=405,000±108,000 (mean±s.d.), which was obtained from the x-intersect in the Zhu–Golan plot using an average S_cell of 700 µm² assessed from the cell diameter observed using widefield microscopy. With the mobile fraction of the receptors being f=0.9±0.1 (Movies 1 and 2), measured using fluorescence recovery after photobleaching (FRAP), this corresponds to a total amount of, mobile and immobile, HLA-DQ8-glia-α1 molecules per cell of N_tot=450,000±120,000 molecules. This value is within the experimental error identical to the value measured by flow cytometry (Table 1). Previously, it has been shown that the total receptor number on the cells, N_tot, obtained from the Zhu–Golan curves can be higher than the total number of receptors measured using antibody binding (Dustin et al., 1997; Zhu et al., 2007). This effect was accounted for by a greater tendency of cells expressing larger numbers of receptors to bind to the SLB, biasing the measured total amount of receptors using the Zhu–Golan analysis towards larger N_tot values (Dustin et al., 1997; Zhu et al., 2007). Since our measured N_tot is the same as that obtained from flow cytometry, we conclude that under the experimental conditions used here there was no significant bias towards cells expressing larger numbers of receptors binding to the SLB.

Measurements of rCD2 binding to rCD48_T92A expressed on Jurkat T cells were also performed. First, Jurkat T cells expressing rCD48_T92A were added to a rCD2-functionalized SLB resulting in the accumulation of rCD2 underneath the cell. More rCD2 was added and the obtained data fitted to the Zhu–Golan expression in Eqn 1 (Fig. 1D). The 2D K_d was subsequently found to be 6±1 molecules/µm² (mean±s.d.), with the total amount of rCD48_T92A molecules on the cell N_tot=124,000±9000 molecules, and f=0.6±0.2 measured separately by FRAP (Movies 3 and 4). The total amount of rCD48_T92A molecules on the cell was, as for HLA-DQ8-glia-α1, within the accuracy of the experiment, identical to the number measured by flow cytometry (Table 1). The 2D K_d value is ∼7-fold lower than that for rCD2 binding wild-type rCD48, measured under otherwise similar conditions to those used in this work (Jönsson et al., 2016). This is a larger relative difference compared to the corresponding change in 3D affinity between the T92A mutant and the wild-type CD48 (3–4-fold) (Evans et al., 2006). The 2D affinity for rCD2 binding rCD48_T92A is furthermore similar in magnitude to that measured previously for human CD2 binding human CD58 (Tolentino et al., 2008; Zhu et al., 2007).

To investigate whether having two different binding ligands influences the affinity of each other, SLBs containing both rCD2 and L3-12 TCR were next studied. This was undertaken at either low or high concentrations of the additional ligand. In all cases the proteins rCD2 and L3-12 TCR bound to the Jurkat T cells and distributed homogeneously within the cell contact area (Fig. 2A,B). Having rCD2 at densities below B_rCD2=200 molecules/μm² (low densities) had no significant effect on the 2D K_d of the L3-12 TCR binding HLA-DQ8-glia-α1 interaction (Fig. 2C; Fig. S2). However, increasing the densities of bound rCD2 to above 300 molecules/μm² (high densities) increased the effective 2D K_d by a factor of 1.9 to 26±1 molecules/µm² (Fig. 2D). The limits for the low and high densities of rCD2 were chosen empirically and are different to when L3-12 TCR and HLA-DQ8-glia-α1 were considered as the auxiliary binding pair (see below). Additionally, the maximum amount of bound L3-12 TCR was reduced by 37% compared to L3-12 TCR binding in the absence of rCD2 (Fig. S2). A similar effect was observed when adding the L3-12 TCR as an auxiliary ligand to rCD2-functionalized SLBs. At densities of L3-12 TCR below B_TCR=700 molecules/µm² the rCD2 binding rCD48_T92A interaction was only moderately affected (Fig. 2E), whereas at densities above 900 molecules/µm² of bound L3-12 TCR the Zhu–Golan plot for rCD2 binding rCD48_T92A shifted downward, resulting in a 1.5-fold larger 2D K_d of 9±1 molecules/µm² (Fig. 2F). Moreover, in the latter case, the maximum amount of bound rCD2 in the contact decreased by 43% (Fig. S2). The Zhu–Golan plots leveled off at high ligand concentrations for both protein pairs, which occurred at B×p>300 molecules/μm² for L3-12 TCR binding HLA-DQ8-glia-α1 and at B×p>100 molecules/μm² for rCD2 binding rCD48_T92A. These data points were excluded from Fig. 2, and will be discussed below. A contributing factor to the increase in the 2D K_d values at high densities of auxiliary molecules could be that the auxiliary molecules aid in excluding free ligands and receptors from the cell contact, thereby lowering the effective affinity. Another possible explanation for the increased 2D K_d could be competition between the two ligands in forming receptor–ligand contacts, resulting in a lower on-rate and/or lifetime of the individual interaction thereby lowering the affinity. Both these effects could arise if there is a height mismatch between the two ligand–receptor pairs (Milstein et al., 2008; Weikl et al., 2019). Another influencing factor could be crowding due to the addition of a second ligand; however, from previous measurements of the lateral interactions between rCD2 molecules in an SLB (Junghans et al., 2018), it is unlikely that this would be a significant effect at the concentrations used here. Of note, the observed decrease in the amounts of bound ligands when adding auxiliary molecules also depended on the cell–SLB contact size. Contact area was in general larger when having auxiliary molecules (Fig. 3), thus reducing the receptor density in the contact, resulting in smaller amounts of bound molecules.

Contact formation occurred at ligand densities above F=20 molecules/µm² for both L3-12 TCR and for rCD2. The bound cells at these ligand densities appeared spherical in the widefield images (Fig. 3A). Increasing the ligand densities led to cell spreading on the bilayer, and a contact size that increased approximately linearly with the number of bound receptors (B×p) (Fig. 3). A similar linear trend has previously been noted by Shao et al. (Shao et al., 2005). Cell deformations have been observed for T cells interacting with B cells, which was suggested to play a crucial role in T-cell activation (Negulescu et al., 1996). In the presence of both the L3-12 TCR and rCD2, the contact area was generally larger as compared to the cell contact when there was only one protein species on the SLB (Fig. 3B,C). This effect was especially noticeable at higher ligand densities (see values above p_lam=0.2 in Fig. 3B,C). However, the average contact size in these experiments did not grow larger than 40% of the total cell surface area (Table 1). Cell contacts dominated by the L3-12 TCR were unaffected by low levels of rCD2 but increased in the presence of higher levels of rCD2 when B_rCD2>300 molecules/µm² (Fig. 3B). The measured p-values for rCD2 without L3-12 TCR at the SLB were similar to previously measured contact sizes, ranging from 0.05<p<0.1, for contacts containing rCD2 and various forms of rat CD48 (Fig. 3C) (Dustin et al., 1997; Jönsson et al., 2016; Zhu et al., 2007). However, the contact size increased considerably in the presence of the L3-12 TCR (Fig. 3C) and approximately doubled in size at B_TCR>900 molecules/µm² (Fig. 3C). The increase in contact area was dependent on the amount of bound ligands, which in this study was dominated by the L3-12 TCR binding to HLA-DQ8-glia-α1, since the Jurkat T cells had approximately four times more HLA-DQ8-glia-α1 than rCD48_T92A (Table 1).

Above a contact size of approximately p=0.2 the bound cells formed lamellipodia, with larger lamellipodia forming at higher B×p values (Fig. 3). In the Zhu–Golan plots these points were typically shifted to the right with only a minor decrease in B/F with increasing B×p (Fig. 4). For rCD2 it was only in combination with L3-12 TCR in the SLB that the Zhu–Golan plot showed this behavior (Fig. 4D). rCD2 alone did not lead to contact formation larger than 15% of the total cell surface area (Table 1) and no clear lamellipodia, suggesting a possible correlation between lamellipodia formation and the increased apparent binding (Fig. 4C,D). Another possible contributing factor for the increased binding is that cooperative interactions at the higher ligand densities occur, which leads to an elevated affinity (Steinkühler et al., 2018). However, for the Zhu–Golan points at the highest B×p values the total amount of bound receptors exceeds the total receptor number (N_tot in Table 1), indicating that cooperative binding alone cannot explain the increased binding. If the receptor–ligand affinity is assumed to be unchanged, the number of cell surface receptors increases by up to 50% for the highest ligand densities in Fig. 4. To investigate whether the increased apparent binding was caused by trapped ligands under the cells, a mobility study was conducted using FRAP (Fig. S3). Both the L3-12 TCR and the rCD2 in the contact recovered almost completely within 40 s with 89±15% and 90±13% (mean±s.d.) recovery for the L3-12 TCR and rCD2, respectively. This indicates that the increased binding seen under the lamellipodia conditions are not due to trapped ligands under the cells but could instead originate from an increased number of receptors.

To further investigate the role of lamellipodia formation, the cells were treated with latrunculin A, an actin polymerization inhibitor. Since the largest contact areas, and thus most pronounced binding levels, were observed on mixtures containing both L3-12 TCR and rCD2 (Fig. 3), latrunculin A-treated cells were added to SLBs containing both proteins. The treated cells changed shape and formed larger cell–SLB contacts without observable lamellipodia (Fig. S4). Moreover, the contact areas were of similar size with an average p-value of 0.24±0.07, independent of the ligand density at the SLB. The maximum number of bound ligands in the contacts decreased for both rCD2 and L3-12 TCR by 40% (Fig. S4). This was also the case for SLBs functionalized with only one type of ligand (data not shown). Furthermore, no increased ligand binding above X, with X defining the point on the Zhu–Golan curve where B/F=0 corresponding to all receptors being bound, was observed for the latrunculin A-treated cells at the concentrations of L3-12 TCR and rCD2 that, for untreated cells, gave rise to this effect (Fig. S4). This is in agreement with the formation of lamellipodia being a key event in producing increased ligand binding.

Measuring 2D K_d values is considerably less common than measuring their 3D counterparts. However, it has become increasingly clear that there is not always a one-to-one relationship between 2D and 3D affinities (Huang et al., 2010; Huppa et al., 2010), and in order to be able to accurately model the behavior of immune cell signaling we need to know the former. Obtaining 2D K_d values has typically been undertaken using either a mechanical approach based on micropipettes to periodically bring cells presenting the two proteins into contact (Huang et al., 2010) or using fluorescence-based methods.

Here, we have chosen to use the fluorescence-based Zhu–Golan method (Fig. 5A), which allows weak interactions to be measured as well as the effects of auxiliary ligands to be studied. In this method, the density of bound receptors in the flat cell–SLB contact, which is given experimentally by the density of bound ligands, is compared to the free receptor density on the cell surface. The latter is independent of the actual cell surface area in the contact (Zhu et al., 2007), which can be larger than the cell–SLB contact area due to cell surface roughness (Mege et al., 1986). Therefore, only free receptors in the actual cell–SLB contact are counted. In this way, we measured the 2D K_d of L3-12 TCR binding to the celiac disease-related HLA-DQ8-glia-α1 as 14±5 molecules/µm² (mean±s.d.). It should be pointed out that this value corresponds to the 2D K_d for the TCR–MHC interaction in our controlled system. The influence of, among others, cell signaling, cytoskeletal interactions, co-receptors, and interactions with the CD3 complex are here omitted, and how these would influence the 2D K_d value can only be a matter of speculation. The obtained 2D K_d value is nonetheless comparable in magnitude to previously measured 2D K_d values for TCRs binding agonist MHCs (Grakoui et al., 1999; Huang et al., 2010; Pielak et al., 2017). Notably, the L3-12 TCR binding HLA-DQ8-glia-α1 2D affinity was 2.3-fold lower than that of rCD2 binding the high-affinity rCD48_T92A mutant, whereas the corresponding 3D affinity was 1.6-fold higher. Thus, it should again be stressed that there is no direct relationship between 2D and 3D affinities. Different theoretical studies (Hu et al., 2013; Weikl et al., 2019), as well as experiments on cell-derived giant vesicles (Steinkühler et al., 2018), have indicated that membrane fluctuations can influence the 2D K_d value such that it varies with molecular concentration. We did not observe this effect here and instead a single 2D K_d value fitted the data points well, up until the point at which lamellipodia formed. This indicates that topographical effects due to membrane fluctuations do not have a significant influence on the affinity of our system at the molecular concentrations used in this study.

In addition to TCR–MHC interactions, there are also various auxiliary molecules binding across cell contacts in vivo. To investigate whether auxiliary molecules can influence the affinity of other binding pairs, we studied how different concentrations of rCD2 binding to rCD48_T92A affect the Zhu–Golan plot for the L3-12 TCR binding HLA-DQ8-glia-α1 and vice versa. It was found that the affinity was unaffected at lower densities, whereas it could be reduced by up to a factor of two at high densities of bound auxiliary molecules (Fig. 5B). However, there did not seem to be a universal threshold between ‘low’ and ‘high’, which was more than a factor of two different depending on whether the auxiliary molecule was rCD2 or the L3-12 TCR. The concentration at which the auxiliary ligands lower the affinity might be cell dependent, since we, in a previous study (Jönsson et al., 2016), did not observe any significant change in the binding of human CD4 (hCD4) to MHC class II molecules on Raji B cells, when the densities of the auxiliary molecule human CD2, which binds human CD58, was varied in the range of 200 to 1200 bound CD2 molecules/μm² (data not shown). To investigate this further we replaced the L3-12 TCR with hCD4 at the SLB and studied its binding to the MHC class II molecule HLA-DQ8-glia-α1 on the transfected Jurkat cells in the presence of low and high levels of rCD2. It was found that by increasing the rCD2 concentration from B_rCD2=120±40 molecules/μm² to 540±160 molecules/μm², the average B/F value of hCD4 in the cell–SLB contact decreased by a factor of 1.8 (Fig. S5). This decrease in effective affinity is similar to that observed for L3-12 TCR binding HLA-DQ8-glia-α1. It was also observed that higher rCD2 concentrations caused increased hCD4 exclusion from the cell–SLB contact (from 25% exclusion at low rCD2 densities to 41% exclusion at high rCD2 densities). The increased exclusion will result in a decrease in the effective affinity and is thus one possible explanation for the decrease in binding when having higher densities of rCD2. Another possibility is that the auxiliary molecules create a less favorable environment for the studied ligand–receptor pair, thus influencing the binding on- and/or off-rates for the investigated interaction, thereby lowering the affinity.

It is generally suggested that adhesion molecules, such as CD2, aid in forming cell–cell contacts and by doing so promote binding of lower affinity, or less prevalent, molecules in the cell–cell contact, such as CD4 binding MHC class II molecules (Jönsson et al., 2016) or TCR binding to MHC presenting cognate antigens (Davis and van der Merwe, 2006; James and Vale, 2012; Milstein et al., 2008; Shaw and Dustin, 1997). This is also observed here for hCD4 binding HLA-DQ8-glia-α1, which requires rCD2 in the SLB to form contacts, and would have been expected also for L3-12 TCR binding HLA-DQ8-glia-α1 if the density of the latter had been at physiological levels [100–300 cognate MHC molecules per cell instead of 450,000 MHC molecules (Grakoui et al., 1999; Harding and Unanue, 1990)]. The observation that high densities of rCD2 decreases the effective affinity suggests that there is an optimal density range within which rCD2 promotes contact formation but without decreasing the affinities of other interactions. Whether this actually happens in vivo can at the moment only be considered speculatively; however, it is noteworthy that physiological densities of rCD2 (hCD2) and rCD48 (hCD58) are 50 to 100 molecules/μm² (Dustin et al., 2007; Von Bergwelt-Baildon et al., 2002; Zhu et al., 2006, 2007), which are comparable to the densities of rCD48_T92A on the cells (180 molecules/μm²) and of rCD2 in the SLB (20–100 molecules/μm²) used in this study. Since the expression level in vivo is similar to the threshold between low and high densities found in this study, it is possible that CD2 could potentially both align the T-cell surface with the antigen-presenting cell to support the formation of TCR–MHC bonds, as well as, when enough auxiliary molecules have accumulated in the contact, lower the effective affinity of the TCR–MHC interaction. However, if auxiliary binding molecules influence the affinity in vivo, then this also raises the question: which is the correct affinity to measure? There are multiple ligand–receptor pairs acting simultaneously in cell–cell contacts between immune cells in vivo (Huppa and Davis, 2013). It is thus possible that the affinity of a specific ligand–receptor pair at physiological densities of auxiliary molecules deviates from the affinity measured for a system containing only the studied ligand–receptor pair. In addition, if the in vivo affinity depends on the density of auxiliary molecules, it is possible that the cell could modulate the affinity of a specific ligand–receptor interaction by varying the local density of auxiliary binding pairs.

We also found that the cell–SLB contact size increased monotonically with the number of bound ligands. This increase in contact size did not influence the binding behavior of the ligands significantly at p-values less than 0.2; however, at high levels of free ligands, the cells started to form lamellipodia, which coincided with an apparent increase in the number of bound ligands in the SLB, by as much as 50% for the largest cell contacts. At these high ligand densities the initially linear Zhu–Golan curve started to level off (Fig. 5C). One possible cause of the apparent increase of receptors on the cell surface is that the rate of receptor recycling is reduced upon lamellipodia formation, given that it is likely an active site of cytoplasmic rearrangement. It has previously been shown that cell activation can lead to a 1.5-fold increase in the number of cell surface receptors, which were argued to be transiently stored in cytoplasmic vesicles (Thatte et al., 1994; Zhu et al., 2006). Additional cell membrane receptors that are stored within these cytoplasmic vesicles could also be the source of the ‘extra’ receptors on the cell surface. Prominent, but T-cell activation independent, vesicular fusion linked to the formation of the lamellipodia might have an important role, since experiments with the actin-polymerizing drug latrunculin A abolished both lamellipodia formation as well as the high-density binding. The increased amount of bound ligands was, however, not due to trapped ligands because FRAP measurements showed that the majority of ligands in the cell contact were mobile with an effective diffusion coefficient of D=0.3±0.05 µm² s⁻¹ and D=0.6±0.13 µm² s⁻¹ for L3-12 TCR and rCD2, respectively. These FRAP measurements also set an upper limit to the lifetime of the L3-12 TCR binding HLA-DQ8-glia-α1 and rCD2 binding rCD48_T92A interactions, corresponding to 9.5±3 s and 7±2 s (mean±s.d.), respectively. These are in the range of intermediate lifetimes (Axmann et al., 2012; Huppa et al., 2010; Lin et al., 2019), indicating that it is mainly a large on-rate that elevates the affinity of L3-12 TCR binding to HLA-DQ8-glia-α1. It should finally be noted that we do not see a significant influence of topographical effects in the binding studies, which would be expected to show up as a non-linear distortion of the Zhu–Golan plot. However, it is possible that a reduction of membrane fluctuations at high ligand densities contributes to the increased binding we observe for these data points.

In summary, our study shows that 2D affinities not only depend on the protein–protein interaction per se but also on several other factors, such as auxiliary binding molecules, ligand density and the dynamic behavior of the cell. Although measurements of ligand to receptor binding for soluble molecules have provided vital information about how our immune system distinguishes between self, transformed self (for example neo-epitopes) and non-self, it is clear that a better understanding of how different factors affect binding in cell contacts will help with modeling these processes with even higher accuracy. The results obtained here take a step in this direction, helping to understand the effect of cellular context, and in particular auxiliary binding molecules, on the affinities of ligand–receptor interactions at cell–cell contacts.

The human E6.1 Jurkat T-cell line (ATCC^® TIB-152™) was cultured in 1640 RPMI medium (#R8758, Sigma) supplemented with 10% fetal bovine serum (FBS, #F9665, Sigma), 2.05 mM L-glutamine (#G7513, Sigma), 1 mM sodium pyruvate (#S8636, Sigma), 1% HEPES (#H0887, Sigma), 1% of penicillin (stock concentration of 5000 units ml⁻¹) and streptomycin (stock concentration of 5000 µg ml⁻¹) (#P4083, Sigma) at 5% CO₂ and 37°C. These cells were transduced using a lentivirus to express the rCD48 mutant T92A and the human MHC class II molecule HLA-DQ8 (DQA1*0301; DQB1*0302) attached covalently to the α1-gliadin peptide (PSGEGSFQPSQENPQ) and cultured as above. The cells were kept at a cell density between 0.5×10⁶–0.8×10⁶ cells ml⁻¹ between each experiment. Human embryonic kidney 293T (HEK293T; ATCC^® CRL-3216™) cells were cultured in DMEM (#41966029, Thermo Fisher Scientific) supplemented with 10% FBS, 2.05 mM L-glutamine, 1% HEPES, 1% of penicillin and streptomycin (both from 5000 µg ml⁻¹ stocks) at 5% CO₂ and 37°C. HEK293T cells were used for lentivirus production only. All cells had been tested for mycoplasma prior to the experiments and were subcultured every 2–3 days.

4×10⁵ cells ml⁻¹ HEK293T cells were plated in a six-well plate and incubated overnight at 5% CO₂ and 37°C in supplemented DMEM. After 24 h, the HEK293T cells reached a confluency of 80% and were transfected with 0.5 µg of the pHR-SIN lentiviral expression vector containing the DNA of interest (HLA-DQ8-glia-α1 or rCD48_T92A) mixed with 0.5 µg of each of the packaging vectors p8.91 and pMD.G (2nd generation) using GeneJuice^® (#70967, MERCK) according to the manufacturer's instructions. The supernatant containing virus particles was harvested 72 h post transfection at 5% CO₂ and 37°C and centrifuged for 15 min at 2082 g (Eppendorf Centrifuge 5810R, #EP022628188, Sigma) to separate cell debris from the lentiviral-conditioned medium. 1.5 ml of the lentiviral-conditioned medium were subsequently added to 1×10⁶ Jurkat T cells and the expression of the transfected proteins tested 3 days later by using flow cytometry.

0.5×10⁶ cells were centrifuged at 4°C for 2 min at 270 g (Heraeus™ Multifuge™ X1R, Thermo Fisher Scientific) and washed twice with HEPES-buffered saline (HBS) buffer [10 mM HEPES (#H3375, Sigma) pH 7.4, 150 mM NaCl (#7647-14-5, VWR)] containing 0.05% sodium azide. The cells were labeled with either the isotype phycoerythrin (PE) anti-mouse IgG1 (clone MOPC-21, #400112, BioLegend®, 1:5 and 1:10 dilution), PE-anti-human HLA-DQ (clone HLADQ1, #318106 BioLegend^®, 1:5 dilution) or PE-anti-rat CD48 (clone OX-45, #MA5-17528, Thermo Fisher Scientific, 1:10 dilution) for 45 min on ice, and afterwards washed twice with HBS buffer plus 0.05% sodium azide. An initial antibody titration experiment was conducted to assess the antibody concentrations for optimal antigen binding. For quantification purposes, BD Quantibrite™ PE beads (#340495, BD Biosciences) were diluted in 300 µl HBS buffer plus 0.05% sodium azide and measured alongside with the antibody-stained cells in a BD Accuri™ C6 Flow Cytometer (BD Biosciences). This allowed for the calculation of the total number of antibodies per cell, which, at saturating concentrations of antibodies, was assumed to be equal to the total number of receptors per cell (Poncelet and Carayon, 1985). All data were analyzed using FlowJo™ (v10.5.2, BD Biosciences) and Microsoft Excel (Microsoft).

Vesicle solutions containing 5% wt% [10% wt% for the hCD4 experiments] 1,2-dioleoyl-sn-glycero-3-[(N-(5-amino-1-carboxypentyl)iminodiacetic acid)succinyl] (nickel salt) (DGS-NTA; #790404C, Avanti^® Polar Lipids, Inc) and 95% wt% [90% wt% for the hCD4 measurements] 1-palmitoyl-2-oleoyl-glycero-3-phosphocholine (POPC; #850457C, Avanti^® Polar Lipids, Inc, USA) were prepared at a total lipid concentration of 0.5 mg ml⁻¹. First, the different lipids were mixed in 100 µl chloroform and then dried to remove the chloroform by using a N₂ gas flow for 10 min. The dried lipids were re-suspended and thoroughly mixed in 1 ml of filtered (0.2 µm Minisart^® Syringe filter, Sartorius, #16534K, Sigma) HBS buffer and incubated on ice for 1–2 h. To create small unilamellar vesicles, the vesicle solution was tip sonicated with a CV18 model tip sonicator (Chemical instruments AB) for 15 min with a pulse time of 10 s and an amplitude of 55%. The vesicle solutions were stored at 4°C until used and renewed every month.

For the formation of SLBs, 0.15 mm thick, round glass slides (number one coverslips, 25 mm diameter, #CB00150RA020MNT0, Thermo Fisher Scientific) were used as support. All glass slides were cleaned for 30 min in 80°C heated piranha solution [mixture of 75% sulfuric acid (99.9%, #100731, Merck) and 25% hydrogen peroxide (30%, #H1009, Sigma)]. After the piranha cleaning, the glass slides were rinsed for 1 min with deionized water and dried using N₂ gas. Silicon wells (Silicon isolators, 12×4.5 mm diameter, 1.7 mm depth; cat. no. 665206, Grace Biolabs, USA) were attached to the glass slide by pressing the clean glass onto the silicon well. Next, 30 µl HBS buffer was added to the wells followed by 30 µl of a 1:10 dilution of the vesicles in HBS buffer. After the addition to the prepared well the vesicles were incubated for 1 h at room temperature (RT) to allow for the formation of a fluid and continuous SLB on the glass surface. The SLB was washed five times with 50 µl HBS buffer to rinse away non-ruptured vesicles. The ability of the lipids to form SLBs was confirmed using FRAP as previously described (Junghans et al., 2018).

A histidine tag covalently bound to the proteins allowed for binding of rCD2 and L3-12 TCR to DGS-NTA lipids in the SLB. Recombinant rCD2 and hCD4 were produced as previously described (Chang et al., 2016; Jönsson et al., 2016) and were genetically modified at the C-terminus with a double histidine tag (12xH) or with a single histidine tag (6xH), respectively, while the human L3-12 TCR had one histidine tag (6xH) each on the C-terminus of the α- and β-chain. The L3-12 TCR was produced as described by Broughton et al. (2012). All proteins were fluorescently labeled using an Alexa Fluor^® antibody labeling kit from Thermo Fisher Scientific (#A20181 and #A10475) with a labeling efficiency in the range 1.1 to 1.7 Alexa Fluor^® molecules per protein. The L3-12 TCR and hCD4 were labeled with Alexa Fluor^® 488 and the rCD2 with Alexa Fluor^® 647. For the L3-12 TCR and rCD2 experiments, both proteins were diluted in HBS buffer and added either (1) as single proteins or (2) in combination to the SLB. Both the formation of SLBs and the subsequent binding of proteins was prepared at RT. After 45 min of incubation, the proteins were washed off and the sample moved to a 37°C temperature chamber. To allow for the sample to heat up to 37°C, it was incubated for 5–7 min before the addition of 37°C warm Jurkat T cells expressing HLA-DQ8-glia-α1 and rCD48_T92A. The Jurkat T cells were incubated on the SLB in supplemented RPMI containing 10% FBS to reduce unspecific cell–SLB adhesion (Movie 5). The ligands in the SLB were still mobile after the addition of supplemented RPMI as confirmed by FRAP. Contact growth and protein binding reached steady state after 15 min incubation, and the SLB was washed five times with 50 µl HBS buffer to rinse away unbound cells and traces of cell medium. Increasing amounts of the labeled L3-12 TCR and/or rCD2 were added to the SLB and were incubated for 20 min after the cells had bound. Images were captured after washing off unbound proteins. This was repeated three to four times for each SLB. All images were acquired on a Zeiss observer Z1 TIRF microscope (Plan-Apochromat 100× NA=1.46 objective, Zeiss, Germany) equipped with a Hamamatsu ORCA-Flash4.0 LT Digital CMOS camera (C11440-42U) and Slidebook 5.5 as the operating software. A 3i solid state diode laser stack operating at a wavelength of 488 nm (100 mW) and 638 nm (100 mW) (3il33, Denver, Colorado USA) were used to illuminate the sample together with a 446/523/600/677 nm TIRF quad-band dichroic filter cube. All these measurements were carried out at 37°C. The hCD4 measurements were performed at RT and the bound cells were incubated for 40 min before the images were acquired with a Nikon Eclipse Ti microscope using a 100× magnification oil immersion objective (Plan Apo TIRF, NA=1.45, Nikon Corporation, Japan), a back-illuminated scientific CMOS camera (Prime 95B, Photometrics, USA), 488 nm and 638 nm diode lasers (LBX, Oxxius, France) with a 531/46 nm BrightLine^® single-band bandpass filter, and a 635 nm laser BrightLine^® single-edge laser dichroic beamsplitter (Semrock, USA), respectively.

1×10⁶ transfected Jurkat T cells were incubated in 1 ml supplemented RPMI 1640 medium containing 25 µM latrunculin A (#ab144290, Abcam, Sweden) for 2 h at 37°C. The cells were then added as described above to the SLB. All subsequent washing steps on the SLB were done in HBS buffer containing 25 µM latrunculin A to allow for an effective drug response throughout the complete measurement.

FRAP measurements were performed for both (1) receptors on stained, transfected Jurkat T cells and (2) fluorescently labeled recombinant proteins in a cell–SLB contact. (1) PE α-human HLA-DQ or PE α-rat CD48 labeled Jurkat T cells (for the labeling protocol, see the flow cytometry section) were used to determine the mobile fraction f of the receptors on the T-cell surface, and (2) functionalized SLBs with L3-12 TCR or rCD2 binding to their corresponding receptors on Jurkat T cells were used to determine the mobile fraction of the ligands within the contact area, f_contact, as well as to estimate the diffusion coefficient of ligands in the contact and the lifetime of the ligand–receptor interaction. Images were acquired with an CSU-X1 spinning disc confocal microscope (Plan-Apochromat 100× NA=1.46 objective, Zeiss, Germany), an Evolve EMCCD (512×512) camera, a 3i solid state diode laser stack operating at a wavelength of 488 nm (100 mW) and 638 nm (100 mW) (3il33, Denver, CO) with a GFP confocal emission filter (524/30) and an RFP confocal emission filter (617/73). A series of pre-bleached images was taken before a full laser power bleach of <50 ms was conducted. The recovery was followed for (1) 250 s or (2) 120 s. To calculate the mobile fraction the intensity of an area in the bleached spot, I_bleach, as well as outside the bleached spot along the membrane or within the contact area, I_control, was measured over the course of recovery. To account for fluorophore bleaching during recovery, I_bleach was normalized by I_control. The normalized total intensity over time I_tot(t) reached steady state after 40–50 s and the normalized intensity ratio at steady state gave the mobile fraction. The diffusion coefficient, D, was calculated from the recovery curve by fitting the intensities over time according to Eqn 20 in Jönsson et al. (2008). The time taken for the recovery curve to recover by a factor 1−exp(−1) was used to estimate an upper limit of the lifetime of the receptor–ligand interaction.