Tapasin edits peptides on MHC class I molecules by accelerating peptide exchange



The endoplasmic reticulum (ER) protein tapasin is essential for the loading of high-affinity peptides onto MHC class I molecules. It mediates peptide editing, i.e. the binding of peptides of successively higher affinity until class I molecules pass ER quality control and exit to the cell surface. The molecular mechanism of action of tapasin is unknown. We describe here the reconstitution of tapasin-mediated peptide editing on class I molecules in the lumen of microsomal membranes. We find that in a competitive situation between high- and low-affinity peptides, tapasin mediates the binding of the high-affinity peptide to class I by accelerating the dissociation of the peptide from an unstable intermediate of the binding reaction.


MHC class I molecules present antigenic peptides on the cell surface for surveillance by cytotoxic T lymphocytes. The peptides (usually 8–10 aa in length) are generated in the cytosol and transported into the endoplasmic reticulum (ER), where they bind to the newly synthesized class I heterodimer, which consists of the heavy chain (HC) and beta-2 microglobulin (β2m). For most class I allotypes, optimal peptide loading (prior to exit toward the cell surface) requires an interaction with the peptide loading complex (PLC), which consists of the peptide transporter TAP, the lectin chaperone calreticulin, the protein disulfide isomerases ERp57 and possibly PDI, and the class I-specific peptide loading factor, tapasin 1, 2. In the PLC, class I binds directly to tapasin 3. This interaction is crucial for the loading of high-affinity peptides onto most (“tapasin-dependent”) class I allotypes, and thus for their expression at the cell surface. Tapasin-mediated peptide binding proceeds in an iterative manner such that low-affinity peptides that are bound initially are successively displaced by higher affinity ones. This optimization process is known as peptide editing 4.

The structure of tapasin has recently been solved 5, but it is still unclear how it induces class I molecules to bind high-affinity peptides in the presence of an excess (estimated to be a 1000-fold, 6) of low-affinity peptides. Its mechanism of action has been addressed in different experimental systems, which have yielded different and sometimes conflicting results. In cell-based assays, tapasin leads to the increased surface expression, thermostability, and persistence of class I-peptide complexes 4, 7–9, but some reports state that it does not influence the composition of the peptide pool bound to class I 10, 11. In assays that use whole cells, the peptide editing function of tapasin may be obscured by ER/Golgi quality control 12, a role of tapasin in the trafficking of class I to or from the Golgi apparatus (Springer et al., unpublished data) 13, 14, class I loading by alternative pathways 2, and the metabolic stabilization of TAP by tapasin 3. In search of the molecular mechanism of peptide editing, several groups have therefore established in vitro assays for tapasin function. Experiments with recombinant tapasin are difficult to perform due to the weak intrinsic affinity of tapasin to class I, and thus in one study, tapasin and class I were modified with leucine zippers 15. In another approach, a part of the PLC was reconstituted with a recombinant tapasin/ERp57 conjugate and cell extracts containing calreticulin and peptide-receptive HC/β2m complexes 16. Both assays detected an increase in the association and dissociation rates of some (but not all) peptides to class I when tapasin was present.

In this report, we describe an alternative approach to exploring tapasin function, namely the reconstitution of tapasin-mediated peptide editing in the natural environment of the reaction, the lumen of intact ER microsomes, where the proteins of the loading complex can interact with their native affinities. Our assay affords a higher degree of resolution to the editing process, and the results suggest that tapasin induces the rapid dissociation of peptides from an unstable intermediate of the binding reaction to promote the binding of high-affinity peptides to class I.


An in vitro assay for peptide binding to class I in microsomal membranes

To study the binding of defined peptides to class I molecules in the context of the entire PLC under natural conditions, we produced the murine class I molecule, H–2Kb, in an in vitro transcription/translation system in the presence of 35S-methionine and microsomal membranes isolated from Raji cells 17, 18. Protease and glycosidase digests showed that the class I molecules inserted cotranslationally into the membranes in the proper orientation, their signal sequences were cleaved, and they became glycosylated (Fig. 1A and B, Supporting Information Fig. 1). When we washed the membranes and added the Kb-specific peptide, SIINFEKL (aa sequence in one-letter code), it was transported into the membranes in an ATP-dependent manner (presumably by TAP as shown previously, 19), and it bound to the Kb molecules, which after detergent lysis of the membranes could be precipitated with the peptide- and β2m-dependent mAb, Y3 (Fig. 1C). Thus, a sizable fraction of Kb translocated in vitro was folded, associated with β2m, and peptide-receptive; where no peptide was added, a low background of binding to microsome-endogenous peptides was visible. To remove this background, we transported biotinylated SIINFEKL peptide into the microsomes and precipitated the peptide-class I complex with streptavidin agarose (Fig. 1D).

Figure 1.

Specific peptide binding in the in vitro system. (A) Translation of H-2Kb is specific. An in vitro transcription/translation reaction in the presence of microsomes isolated from Raji cells was programmed with a plasmid carrying the gene for H–2Kb or no DNA, separated by SDS-PAGE, and proteins were detected by autoradiography. The upper band (processed) represents the Kb molecules that are inserted into the membranes and glycosylated (see below). (B) Kb translated in vitro is correctly inserted into microsomal membranes and glycosylated. Kb was produced in vitro in the presence of microsomes and digested with proteinase K and/or endoglycosidase F1 in the presence or absence of detergent, as indicated, prior to SDS-PAGE and autoradiography. Glycosylation (compare lanes 2 and 3), membrane insertion in the correct sense (which exposes the cytosolic tail to the protease, compare lanes 2 and 4), and signal sequence cleavage (compare 1 and 5) are detectable. The size relationship between the untranslocated and the translocated protein is shown more clearly in Supporting Information Fig. 1. (C) Peptide is transported into the microsomes in an ATP-dependent fashion and binds to Kb. Kb was inserted into microsomes, and SIINFEKL peptide plus either ATP or the ATP depleting enzyme, apyrase, were added for a peptide transport reaction as indicated. Membranes were washed and lysed with detergent, and Kb was precipitated with the conformation-specific (peptide-dependent) Ab Y3, or with concanavalin A agarose (ConA). The right-most lane is also a control for post-lysis peptide binding. (D) Biotinylated peptide binds specifically to Kb. A peptide transport and binding experiment was done analogous to 1C, but with biotinylated peptide (SIINFEKL-bio) followed by precipitation with streptavidin agarose.

To assess whether the class I molecules that had been translocated into the microsomes were associated with the PLC, we lysed the membranes with digitonin and precipitated with the anti-tapasin Ab, PaSta.1. Kb was found associated with tapasin, but it mostly dissociated when SIINFEKL peptide was transported into the membranes (Fig. 2), in agreement with previous observations in cell lysates 20, 21. Since the T134K (threonine 134 to lysine) mutants of HLA-A*0201 and H–2Kb do not interact with tapasin 22–24, we also tested Kb (T134K) and found that it did not coprecipitate with tapasin. We conclude that our in vitro system faithfully reproduces the cellular conditions of peptide binding to MHC class I in the context of the PLC.

Figure 2.

Association of wild-type Kb, but not the T134K mutant, with tapasin. Wild-type (wt) or T134K mutant Kb were translated in vitro (lane 1) and inserted into microsomes (lanes 2–7), and SIINFEKL peptide was transported into the membranes (lane 6) or added after lysis (lane 3, asterisk). Membranes were lysed with digitonin, and proteins were precipitated with ConA, mAb Y3, mAb PaSta.1, or an anti-CD1 Ab as indicated, and separated by SDS-PAGE. The graph shows the quantification of the experiment. Not all class I has dissociated from tapasin upon addition of peptide (lane 6) since the TAP transporter may not have been active in all microsomal membrane fragments of the assay.

Tapasin accelerates the dissociation of low-affinity peptides

In living cells, tapasin mediates peptide editing, i.e. the binding of peptides with successively higher binding affinity to class I molecules 4. To reconstitute peptide editing in microsomal membranes, we monitored the competition between two peptides: the biotinylated high-affinity index peptide (SIINFEKL-bio) and a non-biotinylated competitor peptide. Both peptides were simultaneously added to the microsomes together with ATP and incubated for several minutes, and then the membranes were lysed with detergent, and the SIINFEKL-bio complex was recovered with streptavidin agarose. The signal of the Kb heavy chain was then quantified after SDS-PAGE and autoradiography. Thus, a strong band on the gel signifies binding of the index peptide, while low intensity shows displacement of the index peptide by the competitor. In this assay, non-biotinylated SIINFEKL as a competitor efficiently displaced the index peptide, whereas the scrambled variant, FILKSINE, did not have any effect on its binding; thus, the binding of both the index and the competitor peptides was sequence-specific (Fig. 3A). This experiment also demonstrated that peptide transport into the microsomes did not become saturated, at least not at the concentration of 20 μM of the FILKSINE competitor peptide (which corresponds to 100 times the concentration of the index peptide), and that both peptide species reached the lumen of the microsomes. We next used as competitor peptide a lower affinity variant of SIINFEKL that was previously designed and characterized by Elliott and collaborators 24. SIINYEKL binds with lower affinity than SIINFEKL to both Kb WT and Kb(T134K), as assessed by heat challenge analysis of the complexes (Fig. 3B). When we used SIINYEKL as a competitor, we found that it displaced the index peptide from Kb(T134K) just like SIINFEKL did in the previous experiment. Intriguingly, this was not the case for the WT Kb, where binding of the index peptide to WT Kb was diminished to 60%, but not abolished at a 100-fold excess of SIINYEKL (Fig. 3A). An even more striking result was obtained for the low-affinity peptide FAPGNYPAA (which is based on the Sendai virus nucleoprotein epitope FAPGNYPAL but missing the C-terminal anchor side chain, Fig. 3A), which could not displace the index peptide at all at ten- or 100-fold excess. Similar results were seen with the SIINFEKL derivatives SIINFEKV and SIINFEKM (not shown).

Figure 3.

Peptide editing in the in vitro system. (A) Interaction with tapasin enables Kb to preferentially bind high-affinity peptides. H-2Kb was inserted into microsomes, and different amounts of competitor peptides (SIINYEKL and FAPGNYPAA, low affinity; FILKSINE, non-binding; SIINFEKL, high affinity) were added to peptide transport reactions, mixed with 200 nM SIINFEKL-bio index peptide. Following lysis, complexes with the index peptide were precipitated with streptavidin agarose and visualized by SDS-PAGE and autoradiography. One representative experiment is shown. The charts show the averages and SEM of two or three separate experiments for each peptide; the binding signal without competitor peptide is taken as 100%. Black circles, Kb WT; white circles, Kb (T134K). In the SIINYEKL and FAPGNYPAA experiment images, the lane for 200 μM competitor peptide in the T134K mutant was separated from the remainder of the gel for display reasons. (B) SIINFEKL and SIINYEKL peptides bind with the same affinities to Kb WT and Kb(T134K). Following binding reactions as in Fig. 1C, membranes were washed and heated to 37°C or left at 4°C, and class I-peptide complexes were precipitated with Y3 and quantified by autoradiography. The chart shows the ratios of band intensities between the 37°C and the 4°C samples for SIINFEKL (black bars) and SIINYEKL (white bars) of three experiments±SEM.

These experiments suggest that with the help of tapasin, H–2Kb can select, and stably bind, a high-affinity peptide (e.g. SIINFEKL-bio) in the presence of a 100-fold excess of a low-affinity peptide (e.g. FAPGNYPAA). Assuming random collisions between the reaction partners, it must therefore sample, on average, 100 low-affinity peptides for each high-affinity peptide that is finally bound. Since the complexes of peptides like SIINYEKL and FAPGNYPAA are still quite stable in kinetic terms 24 (despite being shorter lived than those of optimal peptides, see below), the simplest explanation of our results is that tapasin accelerates the dissociation of the low-affinity peptides during the competitive binding reaction, allowing class I to sample more peptides in a given time, and to thus arrive at a greater proportion of stable complexes with high-affinity peptide.

We next tried to attain independent evidence of this rapid dissociation of peptide during the binding reaction (Fig. 4A). We reasoned that if rapid dissociation of SIINYEKL did occur, a low concentration of SIINFEKL-bio should still be able to bind to Kb even when added well after saturation with SIINYEKL, since it would bind to the empty binding sites that became available by the dissociation of SIINYEKL. We therefore took microsomal membranes that contained Kb WT or Kb(T134K) and either added to them a mixture of 20 μM SIINYEKL and 200 nM SIINFEKL-bio, followed by a 2-min incubation (mix→2 in the figure), or we added (with the same peptide concentrations) SIINYEKL first and, after a pre-binding (saturation) time of 1 min, SIINFEKL-bio for another minute (1→1). We lysed the membranes, split the samples, and precipitated one half with streptavidin agarose (to measure the bound index peptide) and the other half with Y3 (to measure total bound peptide). We used the ratio of streptavidin agarose to Y3 counts (SA/Y3) as a measure of the extent of SIINFEKL-bio binding. Indeed, for WT Kb, both mixture and pre-binding samples showed high SA/Y3 values, suggesting that even after saturation of Kb molecules with low-affinity peptide, binding of high-affinity peptide was still possible within a very short time frame. Strikingly, however, no such SIINFEKL binding was observed for the T134K mutant. This observation supports the hypothesis that tapasin is responsible for a fast dissociation of low-affinity peptide that takes place during the peptide editing process.

Figure 4.

Tapasin accelerates dissociation from the encounter complex. (A) Rapid dissociation of SIINYEKL from the encounter complex. In a binding reaction as in Fig. 3A, at 30°C, 20 μM of SIINYEKL were added to the membranes, and 200 nM of the index peptide SIINFEKL-bio were added simultaneously (“mix,” gel not shown) or after a 1-min (1→1) or 10-min incubation (10→1), and incubated for another minute. The membranes were lysed, the samples split in two, and class I-peptide complexes were precipitated by Y3 or by streptavidin agarose (SA). The diagram shows the ratio of the SA to the Y3 signals (black bars, Kb WT; white bars, Kb(T134K)) as averages of three experiments. Error bars show the SEM. (B) Rapid dissociation of peptide during competitive binding occurs in a tapasin-dependent manner. SIINYEKL-bio (200 nM) was added to the binding reaction of WT (top and center row) or T134K mutant H-2Kb (bottom row). After 1 min at 15°C (top and bottom rows) or 10 min at 30°C (center row), 20 μM SIINFEKL was added to displace the index peptide, and samples were taken at the indicated time points and lysed. Complexes of index peptide and Kb were precipitated with streptavidin agarose. The graph on the right shows the quantification; for this, integrated band densities were divided by the total radioactivity in each sample (to compensate for variations in the amount of membranes) and normalized to the t=0 value as 100%. Averages and SEM of three or four independent experiments are shown. The numerical values are listed in Supporting Information Table 1. (C) Dissociation of high-affinity peptide from the encounter complex. The dissociation of the complexes of WT Kb and Kb(T134K) with SIINFEKL-bio was measured as in B. The numerical values are listed in Supporting Information Table 1. (D) Peptide editing still takes place at low temperatures. SIINFEKL-bio and SIINYEKL were together added to a binding reaction and incubated for 1 min at 30°C. Following lysis, class I-peptide complexes were precipitated by Y3 or by streptavidin agarose (SA). The diagram shows the ratio of the SA to the Y3 signals (black bars, Kb WT 30°C; white bars, Kb(T134K) 30°C; grey bars, Kb WT 15°C; mean (n=3)±SEM). The complete experiment, with peptides added in different order, is shown in Supporting Information Fig. 3. (E) Numerical simulation. A reaction scheme that assumes competition between a high-affinity and a low-affinity peptide for binding to class I was converted into a set of differential equations and modeled using values for the rate constants from our experiments and from the literature (see Materials and methods). The diagrams show simulations of an experiment in Fig. 3A (FAPGNYPAA/SIINFEKL), with tapasin (black circles, WT situation) and without tapasin (white circles, T134K situation) with the following assumptions: top, tapasin accelerates k−1 (CP*→C+P) by a factor of 100; center, tapasin accelerates both k-1 and k1 (C+P→CP*); bottom, tapasin slows k2 (CP*→CP) but does not influence the other constants.

Tapasin acts on an intermediate of the peptide binding process

Such rapid dissociation of peptides from class I during the binding process may be explained mechanistically by a single-state model, in which class I (C) and peptide (P) are in equilibrium with a complex (CP):

equation image(1)

In the simplest concept, tapasin may accelerate the dissociation of CP such that a thermodynamic equilibrium is achieved in the time of the assay, leading to selection of more high-affinity peptides for binding. As an alternative, peptide binding may occur via an intermediate, or encounter complex (CP*), from which the rapid dissociation occurs, and which slowly converts into a stable state that does not dissociate in the time-frame of the assay 25:

equation image(2)

Since such an encounter complex has been suggested for class I in the literature 15 but never conclusively demonstrated, we decided to test whether it was present during peptide binding and whether it is influenced by tapasin. We reasoned that if a slow conversion to a stable state was taking place, then peptide dissociation and exchange should become less pronounced with the increasing duration of a binding reaction. We therefore repeated our experiment of Fig. 4A, this time leaving 10 min between the high-concentration SIINYEKL and the low-concentration SIINFEKL to allow the postulated conversion between CP* and CP to occur (10→1 in the figure). Indeed, for WT Kb, the amount of SIINFEKL binding fell to T134K levels after 10 min of pre-incubation. This demonstrates that after a short reaction time, the complex with SIINYEKL was still able to dissociate to some extent, but as time progressed, it became less able to do so, presumably because it had isomerized to a stable form on a time scale of minutes. The lack of peptide editing after 10 min was not due to the denaturation of any of the components of the PLC, including tapasin, since a 10-min incubation prior to peptide addition did not lead to a loss of editing (not shown).

We next aimed to directly measure the rapid dissociation of peptide, and to this end, we modified our assay (Fig. 4B, and Supporting Information Table 1). We now used as our index peptide the low-affinity SIINYEKL-bio, and we prebound it to Kb in the membranes for 1 min. We then added SIINFEKL in a 100-fold excess (to inhibit the rebinding of dissociated SIINYEKL-bio) and followed the dissociation of the index peptide in a time course on the scale of seconds and minutes. The SIINYEKL/Kb (WT) complex decayed completely and rapidly, with exponential kinetics and a half-life of about 30 s (top panel), demonstrating that during competitive peptide binding, rapid dissociation of the low-affinity peptide indeed takes place. To assess the rate of conversion of the unstable intermediate CP* to the stable form CP, we prebound the SIINYEKL-bio index peptide for 10 min instead of one (middle panel), and we found that some degree of rapid dissociation still occurred but that the majority of Kb/SIINYEKL (63%) now remained undissociated. This suggests that the half-time of the conversion from the intermediate (CP*) to the stable form (CP) is less than 10 min. In contrast to the Kb WT, the SIINYEKL complex with Kb(T134K) reached the same degree of stability much faster. After even 1 min of preincubation with SIINYEKL-bio, it showed a substantial fraction (≈72%) of dissociation-stable complex (bottom panel). Thus, tapasin, directly or indirectly, disables the CP*→CP conversion of the complex with low-affinity peptide, and thus more of the CP* complex remains available for rapid peptide exchange.

Finally, we asked the question whether rapid dissociation of the complex with SIINFEKL would also be visible. We repeated the experiment from Fig. 4B but used SIINFEKL-bio for the preincubation (Fig. 4C, and Supporting Information Table 1). While the complex with WT Kb showed some dissociation, albeit less than that of SIINYEKL-bio, we found that for Kb(T134K), no rapid dissociation of SIINFEKL was measurable. The simplest interpretation of this finding is that the encounter complex of Kb with SIINFEKL converts faster to the stable state than that with SIINYEKL, rendering the high-affinity peptide almost dissociation-resistant during a binding reaction. This conclusion was supported by an independent experiment in which unmodified SIINFEKL and SIINYEKL peptides were bound to Kb and then attempted to displace with SIINFEKL-bio (Supporting Information Fig. 2, explained in the legend).

Tapasin acts in peptide dissociation, not in the conversion step

In the reaction scheme shown in eq. (2), tapasin could achieve editing either by accelerating the dissociation of peptide from CP* (increasing k−1, CP*→C+P) and thus lowering the concentration of CP*, or else by directly inhibiting the CP*→CP conversion (decreasing k2). In an effort to distinguish between these modes of tapasin action, we decided to investigate whether peptide editing was temperature-sensitive. Since protein conformational changes are generally slower at lower temperatures 26, we reasoned that peptide editing should be compromised at lower temperatures if the conformational change of class I played a rate-determining role in it. Thus, we performed at 30°C and 15°C a competition experiment with the index peptide SIINFEKL-bio and a 100-fold excess of the competitor SIINYEKL (Fig. 4D), and we found that editing at 15°C was not significantly altered. In an accompanying order-of-addition experiment with both peptides, 15°C and 30°C incubation were never significantly different (Supporting Information Fig. 3). Lowering the temperature also did not lead to peptide editing in the T134K mutant, suggesting that low temperature cannot substitute for the effect of tapasin (not shown). These results provide evidence that the rate of the CP*→CP conversion does not determine peptide editing and that tapasin may instead regulate the equation image equilibrium.

To see whether our theoretical scheme in Equation (2) can indeed explain peptide editing, we performed a numerical simulation, assuming a competitive reaction between two peptides (one of high and one of low affinity, just like in the experiments in Fig. 3 and 4). We used rate constants from the literature and from the experiments reported here (see Materials and methods and Discussion). When we implemented the hypothesis that tapasin accelerates the CP*→C+P dissociation reaction indiscriminately for both peptides, we detected peptide editing similar to that seen in the experiment for SIINYEKL and FAPGNYPAA peptides (Fig. 4E, top panel). A very similar effect was seen when tapasin was assumed to accelerate both reactions of the equation image equilibrium, leaving the equilibrium constant unaltered (center panel). In stark contrast, when tapasin was assumed to decrease the CP*→CP conversion rate, the editing effect was abolished (bottom panel). Thus, while a stringent proof of mechanism remains out of reach, the simplest explanation for the ability of tapasin to edit peptides on class I molecules is that it accelerates the dissociation of peptides from the unstable intermediate, CP*.

Tapasin destabilizes binding of the carboxy terminus of the peptide

If tapasin forces the dissociation of peptides from class I molecules, it most likely disrupts specific interactions that hold them in the binding groove. We hypothesized that if a low-affinity peptide lacked such a tapasin-labile interaction with class I, it would be displaced by tapasin to a lesser degree, or not at all. Peptides are held in the binding groove of class I molecules by two kinds of interactions: the sequence-dependent binding of anchor residue side chains into specificity pockets at the bottom of the binding groove, and sequence-independent hydrogen bonds between the termini of the peptide and the helices that delineate the binding groove. The latter contribute most to the binding energy 27. We synthesized two variants of SIINFEKL, which lacked either the N-terminal amino group (SIINFEKL-Ndel) or the C-terminal carboxy group (SIINFEKL-Cdel), and assessed the role of the hydrogen bonds at the termini by comparing the ability of these peptides to displace the SIINFEKL index peptide from WT Kb and the T134K mutant, in experiments analogous to those shown in Fig. 3A. Both peptides had a rather low affinity to Kb, but with the Cdel peptide, intriguingly, differences between WT and T134K were insignificant (p=0.055) while for the Ndel peptide, the difference had marginal significance (p=0.037) (Fig. 5). This result implies the C terminus of the peptide in the mode of action of tapasin.

Figure 5.

Tapasin acts on the C terminus of the peptide. In a binding reaction as in Fig. 3A, the two low-affinity peptides SIINFEKL-Ndel and SIINFEKL-Cdel (at the concentrations indicated) competed with SIINFEKL-bio (at 200 nM). The 200* lanes show competition by 200 μM SIINFEKL. Graphs are as in Fig. 3. In the paired Student's t-test with one-tailed distribution, the difference between WT Kb and Kb(T134K) is marginally statistically significant in the case of the Ndel peptide (0.0373<0.05), but insignificant for the Cdel peptide (0.0549).


In the work described here, we have established the first experimental system to study the binding of defined species of peptides to class I molecules in their natural environment, i.e. in the lumen of the ER, with auxiliary proteins such as tapasin present in their natural abundance. In this system, we have reconstituted the tapasin-dependent editing of peptides, i.e. the displacement of low-affinity peptides by higher affinity ones on a cohort of class I molecules over time, which is seen in live cells.

Our first important finding is that during a binding reaction, tapasin mediates the rapid dissociation of peptides from class I molecules. A dissociation enhancement by tapasin has been shown previously 15, 16, but never in the course of a competitive binding reaction. This result explains a long-standing paradox of class I peptide selection. In live cells, peptides of different affinities compete for binding to class I. Since even the low-affinity peptides used in this work, such as SIINYEKL, can form fairly stable complexes with class I at equilibrium (see our Fig. 3B, 25, and 24), it has been a mystery how class I molecules, during their limited trafficking time through the ER (which may be as short as 10 min for H-2Kb, 28), could sample more than one peptide 29. With tapasin inducing low-affinity peptides to rapidly dissociate from class I, it becomes clear how a large number of peptides can bind to and dissociate from a class I molecule in vivo, and thus how class I can optimize its peptide ligands over time.

We then investigated the mechanism of binding and tapasin action. Our data suggest that peptides form an unstable encounter complex with class I, CP*, that slowly converts to the stable form, CP (Eq. 2). The rate of this conversion, k2, appears to be about tenfold higher (see below) for high- than for low-affinity peptides. The unstable encounter complex lies on the reaction pathway toward the final stable complex, since otherwise, binding of peptides to class I would simply mirror the stoichiometry of the peptides present in the reaction mixture, and no editing could be observed. The concept of a short-lived encounter complex that slowly converts to a stable form is seen with many receptor–ligand interactions 30 and was proven for MHC class II molecules 31; in MHC class I molecules, it has been postulated but never conclusively demonstrated 15, 32, and the molecular nature of the encounter complex and of the CP*→CP conversion can at this point only be inferred (see below).

Our model in Eq. (2) explains why in the literature, two ranges of association constants for peptide onto class I have been found, depending on the method used to investigate them. With real-time stopped-flow label-free tryptophan fluorescence spectroscopy, which does not separate bound from free ligand, association rates in the range of 2×106 M−1s−1 were observed 33, while immunoprecipitation methods, which test the establishment of a biochemically stable complex with a slow off-rate, found only 104 M−1s−134. Most likely, the former data represent the fast formation of an encounter complex, while the latter show the slow conversion to the stable form.

Which rate constant in the reaction scheme of Eq. (2) is influenced by tapasin? In principle, peptide editing could be brought about by decreasing k2 (the conversion rate) or by increasing k-1 (the dissociation constant). Both changes would make it more likely for the CP* intermediate of the low-affinity peptide to fall apart into class I and peptide (as seen in Fig. 4B), and thus increase the possibility that a high-affinity peptide could be bound (which would then trigger a rapid CP*→CP transition and form a stable complex). While our low-temperature experiment (Fig. 4D) and the numerical simulation do not provide a final proof, they suggest that the CP*→C+P reaction is the critical step that is influenced by tapasin. We propose that tapasin accelerates peptide dissociation from the encounter complex, decreasing the amount of available CP* and thus the net rate of CP*→CP conversion for the low-affinity peptide. The concept of enhanced dissociation is supported by observations in live cells that show faster dissociation rates for class I molecules trapped inside the cell 35, and by two other recent biochemical approaches to tapasin function 15, 16 (see below), as well as a numerical approach that models class I peptide binding and traffic 36. Most importantly, it is corroborated by the observation that the peptide displaces tapasin from class I 20, which suggests that tapasin behaves like an allosteric inhibitor of peptide binding. If, in contrast, tapasin were to bind to a peptide-containing form of class I such as CP* or CP, then one would expect to see cooperativity between tapasin and peptide for binding to class I. Thus, tapasin may bind to the empty class I molecule, C, as suggested by us earlier 1, or it may bind to a hypothetical transition state (C) on the pathway between C and CP* (see below).

Can we determine the values of the rate constants in the reaction scheme? For a low-affinity peptide such as SIINYEKL, some estimates are possible. Since CP* must dissociate 10–100 times in 10 min (the incubation time of the experiments in Fig. 3A), its dissociation rate, k1, must be at least 10/10 min–1, or >0.02 s–1 (i.e. a dissociation half-time of 35 s or less). In agreement with this estimate, the direct measurement of fast peptide dissociation (Fig. 4C) suggests that k−1≈0.05 s−1. The corresponding association rate, k1, should be in the same order of magnitude; if the peptide concentration inside the membranes is identical to that outside, it is at least 5000 M−1s−1, but it could be much larger. We have previously measured 35 000 in detergent lysates, and 2×106 with recombinant protein 33. The rate of conversion of the unstable CP* form to the stable form (CP) is about 0.001 s−1 for the WT Kb molecule in the presence of tapasin, and about 0.01 s−1 for T134K (since in Fig. 4C, after 10 min, about 60% had converted to CP). The test of these values in our numerical model (Fig. 4E) shows that they are realistic, but they may vary considerably between different peptides and class I alleles, which are known to bind peptides with different kinetics and to depend on tapasin to varying degrees 4.

Indeed, one can assume that tapasin-independent class I molecules may have an intrinsically higher k−1 rate constant. This does not necessarily imply a lower equilibrium affinity for peptides, since the association rate, k1, may be influenced to the same extent, which would make tapasin a catalyst for rapid peptide binding and dissociation. Remarkably, our numerical simulation shows very little influence of k1 on peptide editing (compare Fig. 4E top and center panel, and data not shown), making such an increase in both k1 and k−1 equally likely as an influence of tapasin on k1 alone. The concept of tapasin as a catalyst is supported by recent molecular dynamics simulation results, which suggest that the binding site of HLA-B*4405, a tapasin-independent class I molecule, is structurally more stable in the peptide-free state and resembles the peptide-bound state, whereas that of B*4402, which is tapasin-dependent but otherwise closely related, is severely disorganized around the peptide C terminus 37; B*4405 may thus be able to bind to peptide in a more productive fashion with faster kinetics. In this scenario, the role of tapasin would be to stabilize a peptide-free transition state on the path between C and CP*, C, providing structural support for the binding site of B*4402 to enable faster peptide binding and dissociation. This hypothesis, namely that the differing degrees of tapasin dependence can be explained by differing intrinsic k1 rate constants, could be tested by a comparative measurement of the equilibrium peptide binding affinities and association and dissociation rate constants of B*4402 and B*4405. Further work in this direction is required.

What could the structure of such a transition state look like? The small displacement effect of tapasin on the SIINFEKL-Cdel low-affinity peptide in Fig. 5 (which was also observed in an approach with recombinant proteins, 15) suggests that tapasin destabilizes the binding of the C terminus of the peptide to class I. Indeed, in this region of the binding site, the small α2-1 helix (residues 140–150), in whose proximity tapasin binds, occupies variable positions in crystals with different peptides 38, 39, suggesting that it may be mobile. In molecular dynamics simulations, the α2-1 helix is highly flexible in the absence of peptide or during its dissociation 39, and its flexibility differs greatly between tapasin-dependent and – independent class I molecules 37, 40. Since the α2-1 helix forms three hydrogen bonds with the C-terminal region of the peptide, from Thr-143, Lys-146, and Trp-147, its movement away from the peptide, perhaps mediated by tapasin, would significantly decrease the overall peptide binding affinity of a class I molecule 1, 41 but also open up the groove to allow peptides to bind and dissociate more rapidly 40. Such an opened structure would have the kinetic properties of the transition state C that was postulated above. The remaining effect of tapasin on the Cdel peptide in our experiment might be explained by the fact that only two of the three hydrogen bonds are absent in the Cdel peptide, since the carbonyl group of the last peptide bond still forms the hydrogen bond with the side chain of Trp-147 in the α2-1 helix 41. Even though considerably more work is needed, our experiments are consistent with the simple explanation that tapasin influences the conformation of the α2-1 helix, reduces the peptide binding affinity of class I, and thus accelerates the dissociation of all peptides from the unstable encounter complex, depleting it and thus slowing the formation of the stable complex of low-affinity peptide with class I. The conversion from the encounter complex, CP*, to the stable form, CP, may then represent either a “locking in” of the α2-1 helix, such that it becomes resistant to the action of tapasin, or another movement of the peptide binding groove, perhaps involving the central part of the α1 helix, whose flexibility is known to be essential for efficient peptide binding 42.

Our results expand two previous in vitro approaches, both of which have found increased association and dissociation rates for some peptides under the influence of tapasin. Chen et al. have brought together recombinant tapasin and HLA-B*0801 fused to jun/fos helical domains 15. The dissociation rates that they observe are much slower than those that we postulate for the CP* complex in our work, which suggests that they observe the dissociation of the stable complex, CP, under the non-physiological conditions of tight tapasin-class I tethering. Wearsch and Cresswell 16, in an elegant approach, have used recombinant ERp57-tapasin complexes to demonstrate peptide editing on class I molecules in cell lysates. Even though in the latter experiments, ERp57 was necessary to achieve the editing effect, the Chen et al. data suggest that tapasin is sufficient for the effect, and that the role of ERp57 may be structural, perhaps increasing the affinity of tapasin for class I through cooperative binding via calreticulin 43. In our assay, where peptide binding takes place inside whole membranes, the role of the individual components can be addressed in future work with microsomes from chaperone-mutant cells.

Finally, our work may help to understand the molecular mechanism of peptide selection and thus the prevalence of certain epitopes over others in APC, which often leads to the focusing of the CTL response onto a few epitopes, with a crucial importance for the design of vaccines (immunodominance, 44). It suggests that in vivo, the degree of binding of any peptide to class I (and thus its dominance or subdominance) is determined by the competition with the highest-affinity binders available. Still, if no higher affinity peptide is present, many peptides can become trapped in the binding groove and presented at the surface because their encounter complexes can convert to the stable form. Intriguingly, it was recently shown that only in the presence, but not in the absence, of tapasin, peptides are presented according to their kinetic stability 45. This suggests that the sequence features of the peptide that bring about the CP*→CP conversion are not identical to those that determine high-affinity equilibrium binding. Such sequence features of the peptide, and the structure of the encounter complex, will require further attention.

Materials and methods

Ab and peptides

The Ab Y3 recognizes peptide-bound Kb46. The PaSta.1 antiserum against human tapasin was a kind gift from Peter Cresswell. Peptides were synthesized by our in-house facility and purified by HPLC to ≥99% purity. SIINFEKL-Ndel contains a propionic acid instead of the serine, and SIINFEKL-Cdel has isopentylamine instead of leucine.

Preparation of microsomal membranes

Raji cells were harvested by centrifugation, resuspended in buffer F (10 mM HEPES-KOH, pH 7.2, 250 mM sorbitol, 10 mM KOAc, 1.5 mM MgOAc), lysed by 2×10 passes through a 22-G needle, and centrifuged twice at 800×g for 5 min. From the supernatant, microsomes were sedimented at 15 000×g for 15 min and resuspended in B88 (20 mM HEPES, pH 7.2, 250 mM sorbitol, 150 mM KOAc, 5 mM MgOAc). Quantification of microsomes was by OD280.

In vitro transcription/translation

The genes for H-2Kb (WT and T134K mutant) were cloned into pTM1 47. DNA was prepared by alkaline lysis according to 48 and treated successively with RNase and with proteinase K, then extracted with phenol/chloroform. In vitro transcription/translation was performed with the TNT reticulocyte lysate system (Promega, Mannheim, Germany) using T7 polymerase in siliconized microcentrifuge tubes. Where indicated, 1 μL of microsome suspension (OD280=10) was added.

Peptide translocation into microsomal membranes

Microsomes were sedimented from an in vitro transcription/translation reaction at 16 000×g for 10 min and resuspended in B88, then peptide (concentration as indicated), apyrase (2 μg; New England Biolabs) and ATP regenerating system (10×: 55 mg ATP, 1.3 g creatine phosphate, 20 mg creatine phosphokinase in 10 mL B88) were added as indicated and incubated at 30°C for 10 min or as indicated in the figure. Microsomes were then harvested from the reactions at 16 000×g for 10 min.

Membrane lysis and precipitation of class I molecules

Microsomes were lysed in NLB (50 mM Tris-Cl, 5 mM EDTA, 150 mM NaCl, 1% Triton X-100) for 1 h, and after centrifugation (16 000×g, 10 min), the supernatant was added to protein A agarose (Calbiochem) with pre-bound Ab, and incubated for 1 h. The beads were washed three times with NLB (the last time without detergent) and heated to 95°C in SDS sample buffer. Samples were separated by SDS-PAGE and analyzed by autoradiography with a Fuji FLA-3000. For precipitation of biotinylated peptides, the lysis supernatant was added to streptavidin agarose (prepared according to standard protocols), incubated for 1 h, and treated as above. For precipitation of glycosylated proteins, the lysis supernatant was added to concanavalin A agarose, incubated for 1 h, and precipitates were washed and resolved by SDS-PAGE as above. Some samples were treated with recombinant endoglycosidase F1 (EndoF1), which is identical in cleavage specificity to endoglycosidase H (EndoH) 49. Coimmunoprecipitations of class I and tapasin were performed as described 3.

Quantification of the MHC class I signals

The radioactive bands on SDS gels all represent H-2Kb molecules, since these were the only radioactive protein species in the samples. Thus, bands obtained after precipitation with streptavidin agarose represent class I molecules that were bound to biotinylated index peptide in the assay mixture. The radioactive bands were quantified using either the manufacturer's software or ImageJ (http://rsbweb.nih.gov/ij/). Background squares from the same lane of the gel were subtracted from the signal. For standardization within one experiment, quantification values of each peptide transport reaction (lane) were divided by the total radioactivity of the reaction determined by scintillation counting. This eliminated the variation in aliquoting the translocation reactions into the individual tubes for peptide transport and binding reactions.

Numerical simulations

The reaction scheme

original image

which assumes competition between a high-affinity (P1) and a low-affinity peptide (P2) for binding to class I, was converted into a set of differential equations, assuming first- and second-order reactions, and implemented in Microsoft Excel. The rate constants were: k−1_P1 and k−1_P2, 0.02 and 0.2 s−1, respectively; k2_P1, and k2_P2, 1 and 0.01 s−1, respectively. Constants k1_P1 and k1_P2 were 2×105 M−1 s−1, 10× lower than reported 33, to keep the number of simulation steps manageable by the software, but this did not influence the conclusions from the simulation (not shown). C was supplied to the system at 10−9 M−1s−1, [P1] was constant at 200 nM, and [P2] was varied between zero and 20 μM as indicated. The step-width of the simulation was 0.003 s, and after 40 000 steps (120 s), [CP1] was plotted against the concentration of competitor peptide, expressed as percentage of the [CP1] value of a simulation where no P2 was present. This allows direct comparison of the graph with the experiments in Fig. 3A. The qualitative results of the simulation were robust toward alterations in any of the constants by factor 10 (not shown).


The authors acknowledge excellent technical support by Ute Claus, crucial discussions with Tim Elliott and Martin Zacharias, and the critical reading of the manuscript by Malgorzata Garstka. One experiment was performed in the laboratory of Peter Cresswell, whose advice and kind donation of reagents is gratefully acknowledged. Our work was financially supported by the DFG (SP583/2-3 to S. S.), the Tönjes Vagt Foundation (P. V. K. P.), the German Academic Exchange Service (R. Y.) and a PhD stipend of Jacobs University (P. V. K. P.). Personal contributions of the authors: P. V. K. P. and R. Y. designed, carried out, and analyzed experiments and collaborated in writing the manuscript; H. K. designed and synthesized the peptides; S. S. conceived the project, worked with P. V. K. P. and R. Y. in designing and evaluating experiments, and wrote the manuscript.

Conflicts of interest: The authors declare no financial or commercial conflict of interest.