Peptide-independent stabilization of MHC class I molecules breaches cellular quality control

The correct assembly of major histocompatibility complex class I (MHC-I) molecules that are bound to peptide ligands with high affinity (hereafter, high-affinity peptides) is supported by chaperone proteins and monitored at several stations of the secretory pathway to warrant efficient cellular immune responses against pathogens and tumors.

The first quality control step takes place in the endoplasmic reticulum (ER). After their cotranslational insertion into the ER membrane, the lumenal domains of MHC-I heavy chains are first bound by the lectin chaperone calnexin and the disulfide isomerase ERp57 (also known as PDIA3). Then, they bind to the light chain, β2-microglobulin (β₂m). Heterodimers between MHC-I heavy chain and β₂m associate with the lectin chaperone calreticulin and a pre-existing complex of the transporter associated with antigen processing (TAP), the TAP-associated protein tapasin and the ERp57, forming together the peptide loading complex (PLC). Prior to loading with high-affinity peptides, suboptimally loaded MHC-I molecules are usually bound to the PLC and are retained in the ER owing to the KKxx sequence that is present on the cytosolic tail of tapasin (Paulsson et al., 2006). As soon as they are loaded with peptides at sufficient affinity, the trimers of MHC-I heavy chain, β₂m and peptide dissociate from the PLC and exit the ER.

The second quality control step occurs at the border between the cis- and medial Golgi. MHC-I molecules are again monitored for high-affinity peptide occupancy by a system that is related to the standard glycoprotein quality control (Ellgaard and Helenius, 2001). It recognizes suboptimally loaded MHC-I, most likely through UDP-glucose:glycoprotein glucosyltransferase (UGT1), retains them in the ER-Golgi intermediate compartment (ERGIC) and the cis-Golgi, prevents them from becoming endoglycosidase H resistant, and eventually returns them to the ER with the help of calreticulin (Hsu et al., 1991; Paulsson et al., 2006; Garstka et al., 2007; Purcell and Elliott, 2008; Wearsch and Cresswell, 2008; Howe et al., 2009; Zhang et al., 2011). This second system is responsible for the intracellular retention of tapasin-dependent MHC-I allotypes in tapasin-deficient cells (Peh et al., 1998; Garstka et al., 2011).

Third, after some time at the cell surface, MHC-I lose their peptide ligand and β₂m, and the resulting free heavy chains are transported to lysosomes for proteolytic destruction (Mahmutefendić et al., 2011). The molecular mechanisms of this surface quality control process and its exact location in the endocytic branch of the secretory pathway are not well understood.

The molecular properties of suboptimally loaded MHC-I that are recognized by these three quality control systems are poorly characterized because crystal or nuclear magnetic resonance structures of empty MHC-I molecules are still not available. Suboptimally loaded MHC-I have a decreased affinity to β₂m, lower thermal stability, a larger hydrodynamic radius and greater conformational mobility of individual residues compared to high-affinity peptide-bound MHC-I (Elliott et al., 1991; Fahnestock et al., 1992; Springer et al., 1998; Mage et al., 2012; Kurimoto et al., 2013); thus, suboptimally loaded MHC-I molecules have been compared to molten globules (Bouvier and Wiley, 1998; Saini et al., 2013). In support of this view, comparative molecular dynamics simulations on empty and peptide-bound murine and human MHC-I have indicated a substantially increased mobility of peptide-empty MHC-I, especially in those parts of the α₁ and α₂ helices that delineate the F-pocket (Sieker et al., 2007; Sieker et al., 2008). Intriguingly, mobility of the polypeptide backbone and related features, such as exposed hydrophobic patches, are recognized by UGT1 (Ruddock and Molinari, 2006). Based on these data, we have proposed that the increased mobility of empty and suboptimally loaded MHC-I, as well as other derived features, are recognized by the cellular quality control mechanisms (Wright et al., 2004; Van Hateren et al., 2010).

Here, we have tested this hypothesis by designing a novel variant of the murine MHC-I allotype H-2K^b, in which the α₁ and α₂ helices are connected by a disulfide bond close to the F-pocket, restricting their mobility. The C84–C139 disulfide bond allows normal PLC interaction and antigen presentation but renders MHC-I surface expression TAP- and tapasin-independent, accelerates anterograde transport, and greatly decreases the rate of MHC-I endocytosis. We demonstrate that these phenotypes can be mostly explained by the increased affinity of the variant K^b protein for β₂m. Our results reveal (1) that the immobilization of the MHC-I polypeptide backbone by the peptide ligand allows MHC-I to pass all three known cellular quality control steps, (2) that peptide binding and β₂m binding to MHC-I are coupled to each other through the molecular dynamics of MHC-I, and (3) that the quality control processes in the ER and Golgi act on several different forms of MHC-I that are linked by dynamic equilibria.

To understand how cells differentiate between MHC-I molecules that are bound to high-affinity peptides and suboptimally loaded MHC-I molecules (i.e. MHC-I with low-affinity peptides or without any peptide), we first wished to identify potential molecular differences between these forms by computer simulation. Previously published molecular dynamics simulations of the α₁ and α₂ domain or of entire MHC-I have suggested one eminent difference between empty and peptide-bound MHC-I: in the absence of peptide, the helical sections that flank the F-pocket region (residues 74–85 and 138–149 in the α₁ and α₂ helices, respectively) were significantly more mobile (Zacharias and Springer, 2004; Garstka et al., 2011; Narzi et al., 2012). Indeed, the corresponding region of peptide-empty HLA-B*07:01 is highly susceptible to proteases (Bouvier and Wiley, 1998), suggesting partial unfolding on a longer time scale. We therefore hypothesized that bound peptides restrict the mobility of this region, and that a similar conformational restriction might be achieved by linking the α₁ and α₂ helices with a disulfide bond. Molecular modeling indicated that the distance between residues 139 and 84 was suitable for the formation of a disulfide bond between two cysteine residues (Fig. 1A). In agreement with our hypothesis, molecular dynamics simulations of the peptide-binding domain of the resulting protein, H-2K^b(A139C/Y84C) (hereafter called K^b-Y84C), indicated a strong restraint of the distance between the alpha carbon atoms of C84 and C139 compared to wild-type K^b (visible as a small variance, i.e. fluctuation, of the distance in Fig. 1B, left). This restraint affected the entire F-pocket region [seen as blue (low-mobility pairwise distances) in Fig. 1C and supplementary material Fig. S1B], but weakened after three helical turns, between residues 73 and 150 (Fig. 1B, right). Interestingly, the absolute mobility of the backbone of empty K^b-Y84C was similar to that of K^b in complex with a high-affinity peptide (Fig. 1D, color-coded root mean square fluctuations of the individual residues over the course of the simulation). Taken together, the molecular dynamics simulation data suggest that the effect of the C84–C139 disulfide bond in K^b-Y84C mimics the effect of the bound peptide on the conformational dynamics of wild-type MHC-I.

Next, we tested whether K^b-Y84C was able to bind peptides. The ER-lumenal domains of K^b-Y84C and wild-type K^b were produced in E. coli and folded in vitro with human β₂m and the peptides FAPGNYPAL (Sendai virus nucleoprotein residues 324–332; single-letter amino acid code) or SIINFEKL (ovalbumin residues 257–264). These complexes were then subjected to thermal denaturation monitored by tryptophan fluorescence (TDTF) (Springer et al., 1998; Saini et al., 2013) to assess their structural stability (Morgan et al., 1997) (Fig. 1E). The denaturation temperature (corresponding to the midpoint of transition, T_m) of K^b-Y84C was only slightly lower than that of wild-type K^b, suggesting that peptide-bound K^b-Y84C was nearly as stable as wild-type K^b. In contrast, peptide-free K^b-Y84C (Saini et al., 2013) was significantly more thermostable than wild-type K^b (Fig. 1E, lower panel). The same was true for K^b-Y84C in complex with the C-terminally truncated peptide variants SIINFEK and FAPGNYP-NH₂ (with its C terminus amidated). The thermal denaturation of K^b and K^b-Y84C in complex with AVYNFATM [the K^b-restricted lymphocytic choriomeningitis virus (LCMV)-derived epitope gp34 (Achour et al., 2002)] was also monitored by circular dichroism, revealing that K^b-Y84C–gp34 was less thermostable than K^b–gp34, whereas K^b-Y84C–/AVYNFA was more stable than K^b–AVYNFA (supplementary material Fig. S1D). Thus, in the empty or suboptimally loaded state, K^b-Y84C is more resistant to thermal denaturation than wild-type K^b.

Next, to compare the rates of association of a high-affinity peptide to K^b-Y84C and wild-type K^b, both heavy chains were folded in the absence of peptides (Saini et al., 2013). The peptide SIINFEK_TAMRAL, which was labeled with the fluorescent dye carboxytetramethylrhodamine (TAMRA) on the side chain of the lysine residue, so as not to interfere with binding to K^b, was added, and binding kinetics were measured by fluorescence anisotropy (Fig. 1F). The binding to K^b-Y84C was considerably faster than to wild-type K^b, suggesting that peptide-empty K^b-Y84C is more amenable to peptide binding. Dissociation of SIINFEK_TAMRAL from K^b-Y84C was slightly faster than K^b (Fig. 1G), in accordance with the slightly lower thermostability of K^b-Y84C shown in Fig. 1E.

The crystal structure of the K^b-Y84C–gp34 complex was determined to 2.1 Å resolution (Fig. 2; supplementary material Tables S1, S2), revealing a very similar overall fold to the previously determined crystal structure of wild-type K^b with the same peptide (Velloso et al., 2004). Importantly, the conformation of the gp34 peptide is very similar in both complexes (Fig. 2B,D). The disulfide bond between residues C84 and C139 of K^b-Y84C–gp34 is clearly indicated by the electron density (Fig. 2C, right panel). The side chain of residue Y84, which plays a key role in the tethering of the C terminus of the peptide in wild-type K^b, is replaced in K^b-Y84C by two water molecules that link the C-terminal carboxylate of the peptide to residues Y123 and K146 of the heavy chain through a network of hydrogen bonds (Fig. 2C, left panel). In conclusion, our data demonstrate that K^b-Y84C binds peptides, and that the three-dimensional structure of K^b-Y84C–gp34 is very similar to K^b–gp34.

We next asked whether K^b-Y84C functions in antigen presentation similarly to the wild-type K^b. We introduced GFP fusions of the full-length K^b-Y84C-encoding gene into BL/6 and 3T3 mouse fibroblasts as well as into HeLa cells. Indeed, the K^b-Y84C protein formed the additional C84–C139 disulfide bond because it migrated faster than wild-type K^b in non-reducing denaturing SDS-PAGE of immunoprecipitates from cell lysates, indicating a more compact state (supplementary material Fig. S1E). The surface levels of K^b-Y84C–GFP (detected by their N-terminal HA tag and analyzed in a dosage-dependent manner that is explained in supplementary material Fig. S1F,G) were comparable to K^b–GFP in murine cells, but significantly higher in HeLa (Fig. 3A, top row). In microscopy, both K^b–GFP and K^b-Y84C–GFP in BL/6 fibroblasts displayed strong ER and surface stain (supplementary material Fig. S2A).

To investigate whether K^b-Y84C can bind endogenous peptides and present them at the cell surface, we transfected 3T3 fibroblasts and HeLa cells with either wild-type K^b–GFP or K^b-Y84C–GFP and with plasmids encoding the peptide MSIINFEKL (pCyto-SIINFEKL). The K^b–SIINFEKL complexes were monitored at the cell surface using the monoclonal antibody (mAb) 25-D1.16, which specifically recognizes the K^b–SIINFEKL complex (Porgador et al., 1997; Mareeva et al., 2008). K^b-Y84C–SIINFEKL complexes were detected at the cell surface in amounts that were slightly lower than those of wild-type K^b–SIINFEKL (Fig. 3A, center row). We therefore considered the possibility that K^b-Y84C might be less efficient in presenting the SIINFEKL peptide, and we plotted the ratio of the 25-D1.16 over HA signals against the heavy chain expression level (Fig. 3A, bottom row). Intriguingly, the 25-D1.16:HA ratio was higher for wild-type K^b than for K^b-Y84C, which suggests that K^b-Y84C is indeed less efficient in presenting the SIINFEKL peptide. Given that productive interaction with tapasin enables K^b to select SIINFEKL from a large excess of low-affinity peptide (Praveen et al., 2010), it appears that K^b-Y84C, although more stable in its empty form, is slightly compromised in binding to tapasin and/or functionally interacting with it.

We next asked whether K^b-Y84C can present bound peptides to T cells, and therefore we introduced the GFP fusion genes into HeLa cells by transfection together with plasmids encoding the peptide SIINFEKL targeted either to the cytosol (pCyto-SIINFEKL) or to the ER (pER-SIINFEKL, with a preceding signal sequence). We then incubated the cells with the B3Z hybridoma cells, which specifically recognize the K^b–SIINFEKL complex (Karttunen et al., 1992), and monitored the activation of B3Z cells by measuring β-galactosidase activity. K^b-Y84C and wild-type K^b were equal in their ability to present endogenous SIINFEKL to B3Z cells (Fig. 3B). Intriguingly, when murine Ltk⁻ cells expressing K^b-Y84C were incubated with increasing amounts of exogenous SIINFEKL peptide before exposure to B3Z cells, B3Z cell activation was much stronger for K^b-Y84C than for wild-type K^b (supplementary material Fig. S2B).

Taken together, our data demonstrate that K^b-Y84C efficiently acquires and presents intracellular peptides, and that it presents exogenous peptides to T cells more efficiently than wild-type K^b.

For efficient expression at the cell surface, most MHC-I allotypes require peptide supply to the ER (mediated by the TAP transporter) and molecular chaperoning by tapasin (Van Kaer et al., 1992; Grandea et al., 2000). To assess whether the surface expression of K^b-Y84C required TAP and tapasin, we investigated its surface levels by flow cytometry in transfected TAP- and tapasin-deficient mouse fibroblasts. Strikingly, surface expression of K^b-Y84C was far higher than that of wild-type K^b (Fig. 4A). This was true for detection both with the antibody specific to the HA epitope tag, with which the K^b proteins were tagged at their N termini, and with the Y3 antibody, which recognizes a β₂m-dependent conformational epitope on the α₁ and α₂ helices of K^b (Hämmerling et al., 1982). In agreement with the flow cytometry results, strong surface expression of K^b-Y84C, but not of wild-type K^b, was observed in immunofluorescence microscopy analyses of TAP- and tapasin-deficient cells with anti-HA antibodies (Fig. 4B). Therefore, in contrast to wild-type K^b, K^b-Y84C can traffic to the cell surface and maintain a steady-state presence there even if peptide or tapasin are unavailable.

To assess whether the TAP and tapasin independence were due to a general inability of K^b-Y84C to interact with the PLC, we next performed co-immunoprecipitation experiments with K^b-Y84C and wild-type K^b in HeLa cells. Both coprecipitated efficiently with TAP and tapasin (Fig. 5A). To test the functionality of this interaction, we used a previously characterized set of variants of the SIINFEKL peptide (SIINFEKM, SIINFEKV, SIINFEKI and SIINYEKL). Their relative surface presentation efficiencies depend on the editing of the peptide repertoire by tapasin (Howarth et al., 2004). We introduced K^b-Y84C or wild-type K^b into HeLa cells, together with plasmids encoding the SIINFEKL variants, and quantified their surface presentation by flow cytometry with mAb 25-D1.16. The presentation hierarchies for K^b-Y84C and wild-type K^b were similar, which suggests that tapasin acts in a similar fashion on both proteins (Fig. 5B). Taken together, these results show that K^b-Y84C is able to interact with tapasin and the PLC, but is not dependent on these interactions for steady-state surface expression.

The stronger steady-state expression of K^b-Y84C at the cell surface compared to wild-type K^b in the absence of TAP or tapasin could be the result of a more efficient anterograde transport of K^b-Y84C, a decreased rate of endocytic destruction of K^b-Y84C, or both. To investigate the first possibility, we compared the rates of anterograde transport of K^b-Y84C and wild-type K^b in a pulse–chase experiment. [To prevent internalization and degradation of any K^b molecules that reached the cell surface, we added the peptide SIINFEKL to the medium; this treatment does not significantly promote anterograde transport of MHC-I from the cell interior (Townsend et al., 1989)]. In the initial time points of the chase in wild-type cells, K^b-Y84C traveled only slightly faster than wild-type K^b (Fig. 6A, top panel; Fig. 6B, left). In contrast, there was a substantial difference between the K^b proteins in TAP- and in tapasin-deficient cells, where wild-type K^b was mostly restricted to the ER, whereas K^b-Y84C was eventually transported to the cell surface (Fig. 6A, center and bottom panels; Fig. 6B, center and right).

In conclusion, our data suggest that the anterograde transport of K^b-Y84C is faster than that of K^b, especially for the suboptimally loaded proteins in TAP- or tapasin-deficient cells.

Following peptide dissociation, MHC-I molecules lose β₂m at the cell surface and become internalized and transported to lysosomes, where they are destroyed by proteases (Mahmutefendić et al., 2011). Thus, the second possible reason for the high steady-state expression levels of K^b-Y84C on the surface of TAP- and tapasin-deficient cells (Fig. 4) is slower internalization and/or lysosomal targeting of suboptimally loaded K^b-Y84C after the loss of high-affinity peptide.

To obtain suboptimally loaded K^b proteins on the cell surface, we used HeLa-ICP47 cells [i.e. HeLa cells that were made functionally TAP-deficient by stable transduction with the herpesvirus protein ICP47 (Früh et al., 1995)], which were transfected with K^b-Y84C or wild-type K^b constructs and incubated overnight at 26°C to block endocytosis (Ljunggren et al., 1990; Day et al., 1995). We then performed a brefeldin A (BFA) decay assay in which we shifted the cells to 37°C to induce endocytosis of suboptimally loaded K^b proteins after adding BFA to prevent any surface transport of newly synthesized MHC-I. Surface levels of K^b and K^b-Y84C were measured at different times by mAb Y3 staining and flow cytometry (Fig. 7A). As expected, suboptimally loaded wild-type K^b was endocytosed with a half-life of ∼4 h and could be stabilized by addition of the peptide SIINFEKL. In striking contrast, K^b-Y84C was stable at the cell surface even without any added peptide, suggesting that it is resistant to endocytosis even when suboptimally loaded. Our data show that the increased steady-state levels of K^b-Y84C at the surface of TAP- and tapasin-deficient cells are due to a combination of two factors: its more efficient anterograde transport, and the resistance of its suboptimally loaded form to endocytosis.

Given that dissociation of β₂m likely precedes the endocytic destruction of MHC-I (Zagorac et al., 2012), we hypothesized that suboptimally loaded K^b-Y84C is stable at the cell surface because of its higher affinity to β₂m. To test this hypothesis, we immunoprecipitated K^b–GFP complexes from transfected HeLa-ICP47 cells with anti-GFP antibody and followed the dissociation of β₂m from the heavy chain over time at 37°C (Fig. 7B). Indeed, β₂m dissociated much faster from wild-type K^b than from K^b-Y84C, suggesting that it binds more tightly to the K^b-Y84C heavy chain. Likewise, decreasing levels of β₂m bound to the heavy chain over time were seen for wild-type K^b but not for K^b-Y84C in pulse–chase experiments in TAP-deficient cells (supplementary material Fig. S2C,D).

Given that K^b-Y84C has a higher affinity to β₂m, it is expected to associate more efficiently with β₂m under conditions where β₂m is limiting. To test whether this phenomenon explains the faster surface transport of K^b-Y84C that we observed in TAP- and tapasin-deficient cells, we performed a pulse–chase analysis similar to Fig. 6, but this time we simultaneously overexpressed β₂m. Interestingly, the rate of acquisition of resistance to EndoF1 was essentially identical for wild-type K^b and K^b-Y84C and much faster than in the absence of β₂m overexpression (Fig. 7C,D). Thus, we conclude that the higher affinity of K^b-Y84C to β₂m is sufficient to explain its faster anterograde transport in tapasin-deficient cells.

We next investigated whether the accelerated transport of wild-type K^b upon β₂m overexpression also decreases the efficiency with which wild-type K^b presents SIINFEKL, in analogy to the lower presentation efficiency of K^b-Y84C (see Fig. 3A). The experiment showed this to be the case (supplementary material Fig. S2E), suggesting that the antigen presentation phenotypes of K^b-Y84C can also be explained by its increased β₂m affinity.

To analyze the trafficking of the K^b proteins in the presence of excess β₂m in more detail, we performed fluorescence microscopy on TAP-deficient cells expressing the C-terminal GFP fusion variants of K^b heavy chains (K^b–GFP). Wild-type K^b–GFP was retained in the ER and showed limited co-localization with the ERGIC marker p58 (also known as LMAN1) (Fig. 7E, top left), whereas K^b-Y84C–GFP was clearly visible at the cell surface and did not appreciably colocalize with p58 (Fig. 7E, top right). When we co-expressed β₂m with the K^b heavy chains, we observed that wild-type K^b now exited the ER but accumulated in the p58-positive compartment (Fig. 7E, bottom left, arrows), indicating that it was retained there by the ERGIC and cis-Golgi quality control (Garstka et al., 2007). Small amounts were detected at the cell surface. K^b-Y84C–GFP displayed increased surface levels upon β₂m overexpression, and again no colocalization with the ERGIC marker was observed (Fig. 7E, bottom right). The much weaker surface expression of wild-type K^b upon β₂m overexpression compared to K^b-Y84C–GFP probably reflects the much higher endocytosis rate of wild-type K^b (see Fig. 7A) because, here, we used no peptide in the medium to trap surface class I molecules. Taken together, these results demonstrate that K^b-Y84C is transported to the cell surface rather efficiently, whereas wild-type K^b is at least partially retained in the ERGIC or cis-Golgi, even at increased β₂m levels. This suggests, in accordance to the pulse–chase experiments in Fig. 6, that suboptimally loaded K^b-Y84C can at least partially bypass the Golgi-based quality control that returns suboptimally loaded wild-type K^b to the ER.

The novel MHC-I variant described in this study, K^b-Y84C, bypasses all three cellular quality control steps that retain suboptimally peptide-loaded MHC-I molecules in the ER (step 1 as discussed in the Introduction), return them from the cis-Golgi to the ER (step 2), or endocytose them from the cell surface for destruction (step 3). We show that this bypass effect is mainly explained by an increased affinity of K^b-Y84C for the light chain, β₂m, and possibly by other effects of the conformational restriction of the F-pocket region by the disulfide bond.

Our molecular dynamics simulations suggest that the effects of the C84–C139 disulfide bond on the conformational dynamics of the K^b heavy chain are similar to the effects of binding a high-affinity peptide. Given that binding of peptide and β₂m to the heavy chain are cooperative (Townsend et al., 1990; Elliott et al., 1991), it is mechanistically consistent that the C84–C139 disulfide bond also increases the β₂m binding affinity of K^b. As we see no significant differences between wild-type K^b and K^b-Y84C in the crystal structure positions of the residues at the interface of the heavy chain with β₂m (Shields et al., 1999; Achour et al., 2002; Achour et al., 2006; Hee et al., 2012), we propose that the coupling between peptide binding and β₂m binding occurs through the conformational dynamics of the heavy chain (Zacharias and Springer, 2004; Beerbaum et al., 2013; Bailey et al., 2014). Further investigation of K^b-Y84C might elucidate this coupling, which is central to MHC-I stability and to the molecular understanding of cellular peptide selection. To analyze conformational dynamics, novel tools are required; as the investigation of heavy-chain–β₂m interactions by nuclear magnetic resonance spectroscopy is only just emerging (Beerbaum et al., 2013; Kurimoto et al., 2013), we think that molecular dynamics analyses with extended simulation times and more refined analytical methods will remain an important tool.

In the cell, we believe that both major phenotypes of K^b-Y84C, namely faster trafficking (i.e. bypass of quality control steps 1 and 2) and increased stability at the cell surface (i.e. bypass of quality control step 3), can be explained by its increased β₂m affinity. At the cell surface, after losing peptide, MHC-I molecules are quickly endocytosed and destroyed. However, the suboptimally loaded form of K^b-Y84C is endocytosed more slowly, probably owing to the reason that it can hold onto β₂m for a longer time. These data support the notion that it is the free heavy chain species of MHC-I that is recognized for rapid transport to lysosomes and proteolytic destruction (Neefjes et al., 1992). Thus, β₂m affinity is a crucial determinant of MHC-I surface levels.

In the forward direction, the overexpression experiments with β₂m (Fig. 7C,D; supplementary material Fig. S2E) show that the more efficient ER-to-surface transport of suboptimally loaded K^b-Y84C (compared to suboptimally loaded wild-type K^b) can be explained by its tighter binding to β₂m. Remarkably, the greater stability of the empty or suboptimally loaded K^b-Y84C–β₂m complex seems to impair its efficient loading with high-affinity peptide, which suggests that its interaction with tapasin is less efficient, perhaps only slightly, concomitant with its faster transport out of the ER.

Our results extend the finding that β₂m is a conformational chaperone for the folding of MHC-I heavy chains (Wang et al., 1994). For the molecular mechanism of this effect, they do not necessarily imply a new mode of β₂m–heavy-chain interaction or a second binding site for β₂m on the heavy chain. More simply, they suggest that the interaction between β₂m and the K^b heavy chain in the ER is close to equilibrium and that in the absence of peptide, β₂m continuously dissociates from and re-associates to the heavy chain, making a certain fraction of the MHC-I pool in the ER unavailable for peptide binding. This interpretation explains how K^b-Y84C, by virtue of its higher β₂m-binding affinity, might bind peptides faster than K^b.

In addition to its increased β₂m affinity, the conformational restriction of the F-pocket region by the C84–C139 disulfide bond might directly help K^b-Y84C to bypass the cis-Golgi quality control. For this, there are two pieces of preliminary evidence. First, when β₂m is overexpressed, suboptimally loaded wild-type K^b still accumulates in the ERGIC and cis-Golgi. We have shown previously that this accumulation is connected to Golgi-to-ER retrieval. In contrast, suboptimally loaded K^b-Y84C barely shows such accumulation (Fig. 7E), suggesting that it is not recognized by the step 2 quality control mechanism (i.e. by UGT1). Second, tapasin – in addition to its effect on peptide exchange and optimization (Williams et al., 2002; Chen and Bouvier, 2007; Wearsch and Cresswell, 2007) – helps to structure the peptide-binding site of suboptimally loaded (and thus conformationally disordered) MHC-I molecules, such that they can bind peptide (Kienast et al., 2007; Garstka et al., 2011; Kurimoto et al., 2013). The efficient surface transport of K^b-Y84C in tapasin-deficient cells (Figs 4, 6) might be at least partially caused by the ordering of the F-pocket region by the C84–C139 disulfide bond, which thus substitutes for tapasin. In addition, a stabilizing effect of the increased β₂m affinity might also be at work here.

Fig. 8 shows the simplest model of the cellular assembly of the heavy chain (HC), β₂m, and peptide (P) that is consistent with the data in this paper and elsewhere. It is an extension of the unified qualitative model for peptide binding to MHC-I that we presented previously (Garstka et al., 2011). The model depicts how, in mechanistic terms, the amount of MHC-I available for peptide binding (HC_so_β₂m in the model) depends on the β₂m affinity of the heavy chain (K_β), on the amount of β₂m available, and on tapasin dependence (K_S). In this way, we hope that in the future, the differences in the cellular biochemistry of the different MHC-I heavy chain allotypes, which have different β₂m affinities and tapasin dependences, might be explained.

Taken together, our observations show the central role of β₂m in directing the folding and trafficking of MHC-I molecules. Binding of β₂m to the heavy chain is modulated by the bound peptide through conformational and dynamic effects. Much work remains in order to elucidate these fascinating connections on a molecular level.

Wild-type and tapasin-deficient BL/6 mouse fibroblasts (Grandea et al., 2000), TAP-deficient BL/6 mouse fibroblasts (Van Kaer et al., 1992), B3Z hybridoma cells (Karttunen et al., 1992), STF1 and STF1_TAP2 cells (Zimmer et al., 1999) and HeLa ICP47 cells (Edd James, Southampton, UK) were obtained from colleagues. Cells were grown and transfected with DNA using standard protocols. We used monoclonal antibodies 12CA5 (Niman et al., 1983), 25.D1–16 (Porgador et al., 1997), 1–87 (Roder et al., 2009), Y3 and K10–56 (Hämmerling et al., 1982; Arnold et al., 1984) provided by colleagues. Other antibodies were obtained from commercial sources. Radiolabeling and immunoprecipitation were performed as described previously (Fritzsche and Springer, 2013). Immunofluorescence microscopy was performed as described previously (Garstka et al., 2007).

The A139C and Y84C mutations and the HA epitope tag (YPYDVPDYA) from influenza virus hemagglutinin were sequentially introduced into the wild-type H-2K^b sequence found between the SalI and BamHI sites in pEGFP-N1 (Clontech) by QuikChange site-directed mutagenesis (Agilent). An AgeI site was introduced into the lentiviral vector puc2CL6IPwo (Halenius et al., 2011) by QuikChange. N-terminally HA-tagged H-2K^b expression constructs (wild-type and A139C/Y84C) were then cloned into this vector via the XhoI and AgeI sites. Lentiviruses were produced and used for gene delivery as described previously (Hanenberg et al., 1996; Moritz et al., 1996; Hanenberg et al., 1997). For cytoplasmic SIINFEKL (residues 257–264 of ovalbumin) expression, the MSIINFEKL sequence was inserted into the pEGFP-N1 by site-directed mutagenesis. pCMVbipep-neo expressing SIINFEKL variant minigenes under the control of the CMV promoter were described previously (Howarth et al., 2004). pER-SIINFEKL and pCyto-SIINFEKL, kind gifts from Edd James, were generated as described previously (Shastri and Gonzalez, 1993; Serwold et al., 2001) using pcDNA3 (Invitrogen).

For the T cell activation assay, HeLa wild-type and HeLa ICP47 cells were transfected with the H-2K^b wild-type or Y84C plasmids and pER-SIINFEKL or pCyto-SIINFEKL plasmids. After 2 days, B3Z cells were added and their response quantified as described previously (Sanderson and Shastri, 1994). For Ltk⁻ cells, transfection and culture was as described previously (Tiwari et al., 2007). For the detection of K^b–SIINFEKL complexes at the cell surface, BALB/c 3T3 or HeLa cells were transiently transfected with plasmids coding for K^b–GFP and SIINFEKL constructs, harvested, stained with mAb 25.D1–16, and counted in a CyFlow Space (Partec) or a FACSCanto (Becton Dickinson). Mean antigen dose and mean antibody staining intensity were calculated as in supplementary material Fig. S1G. For the BFA decay, HeLa ICP47 transfected with wild-type K^b–GFP or K^b-Y84C–GFP were incubated overnight at 26°C and then incubated with 20 µM SIINFEKL peptide or with DMSO for 1 h at 26°C. After washing with medium and adding brefeldin A, cells were transferred to 37°C. At time points as described in the figure legends, cells were trypsinized and stained with Y3 or 25.D1. Data were collected on a FACSCanto (Becton Dickinson).

HeLa wild-type and HeLa ICP47 cells transfected with wild-type K^b–GFP or K^b-Y84C–GFP construct were lysed on ice for 30 min in 1% Triton X-100 in water. K^b molecules were immunoprecipitated using anti-GFP rabbit polyclonal antibody bound to Dynabeads–protein-G (Life Technologies). Complexes bound to beads were incubated at 37°C. At time points as described in the figure legends, samples were placed on the magnet to separate the dissociated β₂m from the fraction bound to the heavy chain. Both fractions were separated by SDS-PAGE and transferred onto Hybond C-Extra membranes. Membranes were probed for K^b heavy chain using anti-GFP-HRP rabbit polyclonal antibody and β₂m using an anti-human β₂m rabbit polyclonal antibody (DAKO). Band intensity was quantified using ImageJ (Schneider et al., 2012).

The K^b-Y84C-AVYNFATM complex was produced as described previously (Achour et al., 1999), and folded complexes were purified on Superdex 20 columns (GE Healthcare) and concentrated to 4–7 mg/ml in 20 mM Tris-HCl pH 7.3. Crystallization assays were performed in 24-well plates at 25°C using the hanging-drop vapor diffusion method. Data collection was performed at beamline ID23-2 at the European Synchrotron Radiation Facility (Grenoble, France). Diffraction data were processed and scaled using the programs MOSFLM 7.0.3 (Leslie, 1992) and SCALA (Evans, 2006). The crystal structure of K^b-Y84C–AVYNFATM was determined by molecular replacement using PHASER (McCoy, 2007) and the crystal structure of wild-type K^b–AVYNFATM (PDB ID: 1S7S) (Velloso et al., 2004), with the peptide omitted, as the search model. A random 5% of the reflections were set aside for monitoring the refinement by R_free cross-validation (Brünger, 1992). Refinement was performed using Refmac5 (Murshudov et al., 1997), Phenix.refine (Adams et al., 2010) and Coot (Emsley et al., 2010). Individual atom B factors were refined isotropically and the position of all water molecules, added to the complexes using Coot, was inspected manually. The stereochemistry of the final models was verified using PROCHECK (Laskowski et al., 1993). All figures were created with PyMol (Schrödinger).

Association rate measurements were performed as described previously (Saini et al., 2013). For the dissociation measurements, K^b–β₂m complexes with SIINFEK_TAMRAL peptide were incubated with 250 µM unlabeled SIINFEKL peptide at room temperature, and dissociation was followed with a Molecular Devices Analyst AD with excitation at 530 nm and emission at 580 nm and using a 561-nm dichroic mirror.

Molecular dynamics simulations were performed as described previously (Garstka et al., 2011) using the sander module of Amber 9.0 (Case et al., 2005) and the parm03 force field (Duan et al., 2003) with TIP3 water (Jorgensen et al., 1983). starting from 1KPV (Protein Data Bank) and using VMD (Humphrey et al., 1996) for visualization.

The complete Materials and Methods are available from the authors upon request.

We would like to thank Malgorzata Garstka (Cell Biology II, Netherlands Cancer Institute, The Netherlands), Mohammed Al-Balushi (Department of Microbiology and Immunology, Al Qaboos University, Oman) and Ute Claus (Molecular Life Science Center, Jacobs University Bremen, Germany) for crucial initial experiments with the Y84C/A139C mutants; Tatyana Sandalova and Gengsi Chen (both Science for Life Laboratory, Karolinska Institutet, Sweden) for support with the crystallization; Uschi Wellbrock (Molecular Life Science Center, Jacobs University Bremen, Germany) and Marilyn Aram (Cancer Sciences Unit, University of Southampton, UK) for excellent technical assistance; Henri de la Salle (Inserm, France), Luc van Kaer (Department of Pathology, Microbiology and Immunology, Vanderbilt University, USA) and Edd James (Cancer Sciences, University of Southampton, United Kingdom) for cell lines; Ron Germain (National Institute of Allergy and Infectious Diseases, National Institute of Health, USA) and Kajsa Paulsson (Clinical Genetics, University of Lund, Sweden) for reagents; Martin Zacharias (Department of Physics, Technical University of Munich, Germany) for crucial advice and support with the molecular dynamics simulations; and Tobias Dick (Division of Redox Regulation, German Cancer Research Center, Germany) for a discussion. Anca Tigan (Department of Pharmacology and Toxicology, Veterinary University of Vienna, Austria, Sinan Zhu (Molecular Medicine and Translational Science, Wake Forest School of Medicine, USA) and Eva Weiss (Department of Biology, University of York, UK) provided important data in earlier stages of the work.