HLA-DM Captures Partially Empty HLA-DR Molecules for Catalyzed Peptide Removal

Peptide binding destabilizes the complex of DM and empty DR

DM-catalyzed peptide exchange is a multi-step process that has been difficult to study mechanistically because most assays do not isolate individual steps but rather represent a mixture of multiple rates. We developed a real-time surface plasmon resonance (SPR, Biacore) assay to directly study the individual steps of the DM-DR interaction. DM was oriented on streptavidin chips through C-terminal site-specific biotinylation so that the large lateral surface required for DR binding was fully accessible. Soluble DR-CLIP complexes were generated with a cleavable CLIP linker. In SPR experiments, it is critical to employ suitable control proteins to ensure specificity of binding, and we therefore identified DM and DR mutants with loss of activity in a real-time peptide binding assay based on fluorescence polarization (FP) (Fig. 1a,c). Mutation of one DMα arginine residue (R98A; hereafter referred to as DM Mut1) reduced DM activity, while an additional mutation (DMα R98A and R194A; called DM Mut2) resulted in loss of activity (Fig. 1a). The SPR binding data closely correlated with these functional data, showing little DR binding to DM-Mut1 and no binding to DM Mut2 (Fig. 1b). Also, a DR2 (DRB1*1501) mutant (DRα S53D) unresponsive to DM in the FP assay (Fig. 1c) failed to bind to DM by SPR (Fig. 1d; Supplementary Figure 1). This DR2 mutant bound peptide with similar kinetics as wild-type DR2 (Fig. 1c), indicating correct folding.

In these SPR experiments, DR-CLIP complexes were injected over the DM surface (stage 1), followed by injection of buffer (stage 2) (Fig. 1b,d). Surprisingly, the dissociation of DR from DM during stage 2 was very slow, even though soluble DM and DR are thought to interact only weakly³⁶. We reasoned that the catalytic activity of DM removed the DR-bound CLIP peptide which exited with the buffer flow, while peptide remains available for rebinding in typical tube-based experiments. We therefore postulated that peptide injection would destabilize the complex of DM and empty DR. Indeed, injection of a high-affinity peptide from myelin basic protein (amino acids 85–99, MBP_85-99) induced dissociation of DR2 from DM (Fig. 1e). A single amino acid mutant of the MBP_85-99 peptide with very low-affinity for DR2 (MBP P4D, aspartic acid [D] at P4 position) induced minimal dissociation at the same peptide concentration (1 μM) (Fig. 1e). Similar binding profiles were obtained at flow rates of 15 to 50μl/min, indicating that rebinding (mass transport) was negligible (Supplementary Fig. 2).

The rate of peptide-induced DM-DR dissociation (stage 3) correlated closely with the affinity of these peptides for the DR2 binding site, both for sets of CLIP (Fig. 2a-d) and MBP_85-99 (Fig. 2e,f) mutants, regardless of which peptide anchors were mutated. Peptides with high affinities for DR2 induced DM-DR dissociation at faster rates, such as CLIP P4F (Fig. 2c) or MBP_85-99 (Fig. 2f), than lower affinity peptides. Peptides with very low affinity, such as MBP P4D, did not induce dissociation at low concentrations (such as 1 μM) (Fig. 1e) but led to some dissociation at higher concentrations (10 μM) (Fig. 2e).

High-affinity DR-peptide complexes bind weakly to DM

The fact that peptide binding induces dissociation of DM-DR complexes suggests that the DR-bound peptide is likely to also have a major effect on the reverse process, the binding of DR to DM. Initial experiments showed that DR molecules with covalently bound peptides did not bind to DM, both for DR2-CLIP and DR1 (DRB1*0101) with a low-affinity CLIP variant (DR1-CLIP_low,), even though binding was readily detectable upon linker cleavage (Fig. 3a,b). These results could either be explained by steric hindrance caused by the linker attached to the N-terminus of the DRβ chain or by a substantial affinity gain resulting from covalent peptide attachment (rebinding of peptide is likely to be extremely rapid). The linker is flexible and also not in the immediate proximity of the predicted DM interaction site, making the latter explanation more plausible.

We next examined the effect of peptide affinity on DM-DR association using DR molecules loaded with single peptides; peptides were synthesized with a dinitrophenol (DNP) tag for affinity isolation of homogenous DR-peptide complexes. There was strong DM binding by the low-affinity DR2-CLIP complex, but substantially slower binding by the high-affinity DR2-MBP_85-99 complex (Fig. 3c,d). This effect was also observed for DR4 preloaded either with low-affinity CLIP or high-affinity influenza hemagglutinin 306-318 peptide (HA_306-318) (Fig. 3g). CLIP binds with high affinity to DR1 (ref.³⁷) and binding of the DR1-CLIP complex to DM was also slow (Fig. 3e). Nevertheless, concentration-dependent DM binding of the high-affinity DR2-MBP_85-99 and DR1-HA_{306- 318} complexes could be demonstrated (Fig. 3d,f), indicating that both low and high-affinity DR-peptide complexes bound to DM in this SPR assay, only with different rates. In a functional peptide binding assay, DM-catalyzed binding of a fluorescently labeled peptide occurred with an 11.8-fold slower rate when DR2-MBP_85-99 rather than DR2-CLIP was used as input protein. In the SPR assay, a similar difference in the rates of DR2-MBP_85-99 and DR2-CLIP binding to DM was observed (8.7-fold), showing overall agreement between the SPR and peptide binding assays (Supplementary Fig. 3). Previous studies with full length DR and DM molecules isolated from cells reported a stronger DM interaction with DR-CLIP than DR-peptide complexes^{21, 38}; also a previous SPR experiment showed stronger binding of immobilized soluble DM to detergent solubilized full-length DR-CLIP than DR-peptide complexes³⁹. In both sets of experiments, a complex mixture of peptides was bound to these cell-derived DR-peptide complexes. In the experiments reported here, differences in binding were much more substantial because purified low and high-affinity DR-peptide complexes were utilized.

Comparison of the DR-CLIP binding data among these three allelic forms (DR1, DR2 and DR4) showed large differences in the DM binding kinetics. DR2-CLIP bound with the fastest and DR1-CLIP with the slowest kinetics (Fig. 3). Large differences in peptide induced dissociation were also observed among these allelic forms, with DR2 being more sensitive to peptide-induced dissociation than DR1 and DR4 (Fig. 3).

DM binding requires partial peptide dissociation

The results with covalently linked peptides (Fig. 3a,b) raised the question whether DR molecules with fully bound peptides do not bind to DM and whether partial release of the bound peptide is actually required to generate a DM-sensitive conformer. To investigate this hypothesis, we generated DR1 molecules to which peptides could be covalently attached in one of the pockets so that peptide termini could be truncated without peptide loss. Mutant HA peptides were either covalently linked in the P6 pocket (HA₆ peptides) or the P3 site (HA₃ peptides) by mutation of DRα Val65 or Gly58 to cysteine; both residues are located on the DRα chain helix. These complexes were made using a DR1 molecule with a very low-affinity peptide as input protein; redox conditions were chosen that enabled efficient formation of a disulfide bond between the peptide and the DRα helix. A large set of such complexes was generated (Table 1) which were affinity purified using DNP affinity tags on the peptides to ensure the absence of empty or other peptide-loaded DR species (Supplementary Fig. 4). Full occupancy was confirmed by monitoring the ratio of 280 nm (DR) and 350 nm (DNP group on peptide) measurements. Also, DR1 molecules with linked HA₆ P₂-P₁₁ peptide could be immunoprecipitated by conformation-sensitive antibodies to DRα and DRβ (Supplementary Fig. 5).

Covalent attachment of the high-affinity HA_306-318 peptide in the P6 pocket (HA₆) (Fig. 4a) yielded a complex that failed to bind to DM by SPR (Fig. 4a,d). The full length HA peptide, which spanned from the P-2 to the P11 residue, was then systematically truncated from the N-terminus, which resulted in loss of conserved hydrogen bonds between the peptide backbone and the DR helices. Removal of the P-2 residue (peptide HA₆ P_-1–P₁₁) resulted in loss of hydrogen bonds to Phe α51 and Ser α53 and further truncation of P-1 (peptide HA₆ P₁–P₁₁) also eliminated a hydrogen bond to His β81 (Fig. 4a). Deletion of these two N-terminal peptide residues was not sufficient for DM binding (Fig. 4d). However, strong DM binding was observed with further truncation of the P1 tyrosine (peptide HA₆ P₂–P₁₁), a critical anchor residue for DR1 binding (Fig. 4b-d)⁴⁰. In contrast, truncation of five C-terminal residues (HA₆ P_-2–P₆) resulted in little DM binding; the small amount of apparent binding was possibly due to an overall increase in protein mobility due to loss of five peptide residues (Fig. 4e).

To determine whether this effect was solely due to removal of the P1 anchor residue, full-length peptides with alanine or glycine substitutions at P1 were tested. Mutation of the P1 anchor to alanine (HA₆ P₁A) or glycine (HA₆ P₁G) resulted in only low amounts of DM binding (Fig. 4f). In contrast, substantial binding was enabled by simultaneous loss of the P1 side chain (P1 glycine) and the two N-terminal residues (P₁G_me–P₁₁) (Fig. 4f). Substantial DM binding therefore requires release of both the P1 anchor and the peptide residues N-terminal to it.

Loss of the P1 anchor could enable dissociation of a longer peptide segment toward the P6 position, which could be required for DM binding. To address this possibility, we moved the covalent attachment site from the P6 pocket to the P3 site by mutation of DRα Gly58 to cysteine (Table 1). Again, removal of P-2 and P-1 did not permit DM binding (HA₃ P_1me–P₁₁) (Fig. 5a) but further removal of the P1 side chain (HA₃ P₁G_me–P₁₁) or the P1 residue (HA₃ P₂–P₁₁) resulted in rapid binding of these two complexes to DM (Fig. 5a). Interestingly, additional truncation of the P2 residue (HA₃ P₃ P₁₁) appeared to slow DM binding (Fig. 5b), indicating that disengagement of the P1 anchor and two N-terminal peptide residues was sufficient for DM binding. DRβ Asn82 forms bidentate hydrogen bonds to the peptide backbone at the P2 position and mutation to alanine increases the intrinsic dissociation rate of the HA peptide >3,000-fold, indicating that it plays a dominant role in stabilizing peptides in the binding groove^{32, 41}. Loss of the P2 peptide residue may thus reduce the activity of such a DR-peptide complex due to loss of these critical hydrogen bonds. These results show that DM only binds to DR-peptide complexes with a disengaged peptide N-terminus, but not to complexes in which the peptide is fully bound.

Important role for DRα Trp43 in DM binding

DRα Trp43 directly interacts with the P1 tyrosine and is partially accessible on the lateral surface of the DRα40-54 segment (Fig. 6a) implicated in DM binding. Conservative mutation to phenylalanine greatly reduced susceptibility to DM in a peptide binding assay (Fig. 6b), and little DM binding by this mutant was observed by SPR (Fig. 6c). This mutation also accelerated spontaneous release of labeled MBP_85-99 peptide from DR2 (Fig. 6d), a property that would actually be expected to increase DM binding based on the data shown above. We also directly compared a panel of four DR mutants for DM-catalyzed peptide exchange and DM binding (Fig. 6e,f). Three mutants had greatly reduced activity (αW43F, αS53D, αS53H) in both assays while one mutant (αG49S) showed enhanced activity. These data show that DR binding to DM as well as DM-catalyzed peptide exchange require the same DR residues and hence involve the same binding site on DR.

High energy barrier for DM binding

Peptide exchange in the presence or absence of DM is strongly dependent on both temperature and pH^{11, 12, 42}. DR2-CLIP and DR1-CLIP_low (Fig. 7a,c, Supplementary Fig. 6) showed a steep temperature dependence in DM binding that is unusual for protein-protein interactions. However, binding of DM to DR1 with a covalent peptide lacking three N-terminal peptide residues (DR1-HA₆ P₂–P₁₁) was substantially less temperature dependent (Fig. 7b,c), demonstrating that dissociation of the peptide N-terminus represents a significant component of the energy barrier. With higher affinity peptides this energy barrier is expected to be even higher. Based on the DM association and dissociation kinetics of the covalent DR1-HA₆ P₂–P₁₁ complex, a K_d of 1 μM was calculated (Fig. 7d), which is consistent with the Michaelis constant (K_M) of ~750 nM for the enzymatic reaction executed by DM³⁹. Interestingly, both on- and off-rates increased to a similar extent with temperature, and the affinity is therefore maintained over this temperature range (Supplementary Fig. 7a-c).

DM has a sharp pH optimum in the acidic range found in late endosomes^11,12. As expected, the binding of DR-CLIP to DM was increased by lowering the pH from 5.5 to 4.6 (Supplementary Fig. 8a). However, peptide binding experiments with DR2 molecules rendered empty by photocleavage of bound peptide showed that peptide binding was faster at higher pH (Supplementary Fig. 8b)²³. Consistent with this result, peptide-induced dissociation of the complex was slower at pH 4.6 (Supplementary Fig. 8a). DR1 with a covalent truncated peptide (DR1-HA₆ P₂–P₁₁) showed 40% of maximal DM binding at pH 7.0, while complexes with a full-length peptide (DR1-CLIP_low and DR2-CLIP) did not bind to DM at this pH (Fig. 7e-g). Furthermore, raising the pH during stage 2, where empty DR dissociates from DM, had little to no effect on dissociation rates (data not shown). The pH-facilitated release of the peptide N-terminus therefore explains, in part, the low pH optimum of DR-peptide binding to DM.

Model of DM-DR interaction

These results significantly change our model of DM action (Supplementary Fig. 9). We show that DR molecules with fully engaged peptides fail to interact with DM until the N-terminal part of the peptide dissociates through constant motion within the DR-peptide complex. DM captures this short-lived transition state and peptide quickly departs following DM binding. The empty DM-DR complex retains the ability to quickly bind a new peptide over extended periods of time. If an interacting peptide has a low affinity, DM may catalyze its removal (editing) while binding of a high-affinity peptide is more likely to induce dissociation of the DM-DR complex. The resulting high-affinity DR-peptide complex has a low likelihood of rebinding DM. This model is consistent with a large body of prior work in the field, including the identification of empty DR-DM complexes in cells and the demonstration of an editing function of DM that drives selection of high-affinity peptides^{14–22, 43}.

DM expression

Soluble DM proteins carried C-terminal BirA (GLNDIFEAQKIEWHE) and protein C tags on DMα and DMβ extracellular domains (EC), respectively. Two N-linked glycosylation sites (DMα N165D, DMβ N92D) were removed to reduce heterogeneity, leaving glycosylation site DMα N15⁴⁷. Sf9 cells were infected with recombinant Baculoviruses at an MOI of >5 and proteins were purified after three days via affinity chromatography using an anti-Protein C matrix (Roche Applied Science); aggregates/impurities were subsequently removed with a Superose 6 gel filtration column (GE Healthcare Bio-Sciences Corp.).

Site-specific biotinylation of the BirA tag was performed at a protein concentration of 2-3mg/ml at a molar ratio of 20:1 DR to BirA. The reaction proceeded for 2 h at 30 °C in the presence of biotin (100 μM), ATP (10 mM) and protease inhibitors. Excess biotin was removed by dialysis and biotinylation was confirmed by gel shift with streptavidin on native polyacrylamide gels.

DR expression

Peptides were synthesized by Peptide 2.0 and JPT Peptide Technologies. Soluble DR–CLIP complexes were produced using stable CHO cell transfectants in hollow fiber bioreactors, as previously described⁴⁸. The CLIP peptide (PVSKMRMATPLLMQA) was covalently attached via a thrombin-cleavable linker at the N-terminus of the respective DRβ chain (DRB1*0101, DRB1*0401, DRB1*1501).

For covalent attachment of peptides, soluble DR1 proteins were generated with cysteines at either DRα 58 (G58C) or 65 (V65C). A low-affinity mutant of CLIP (SKARMATGALAQA, DR1-CLIP_low; mutated anchor residues underlined) was covalently attached to the N-terminus of the DR1β chain (DRB1*0101) to enable efficient exchange after linker cleavage with human rhinovirus 3C protease (provided by J. Fraser, University of Auckland, Auckland, New Zealand). This protease site enabled cleavage under low temperature conditions (4 °C) that minimized dissociation of CLIP_low prior to peptide loading. These DR variants were produced using a Baculovirus system. Soluble DR proteins were purified via affinity chromatography with mAb L243 (American Type Culture Collection); aggregates and impurities were removed using a Superose 6 gel filtration column. Depending on the construct, covalently linked CLIP or low-affinity CLIP were cleaved either with thrombin (Novagen, EMD Chemicals) or 3C protease. Thrombin cleavage was performed using 20 units thrombin per mg of protein at room temperature for 1.5 h. 3C cleavage was done using 50 μg 3C protease per mg of protein under reducing conditions (2 mM GSH) at 4 °C for 2 h. Protein purity and completeness of cleavage were assessed by SDS-PAGE.

Peptide loading and production of disulfide-linked DR-peptide complexes

The key aspects of covalent linkage of very low-affinity peptides to DR1 were as follows: 1. Use of a very low-affinity DR1-peptide complex as input protein (DR1-CLIP_low); 2. Accelerated dissociation of this low-affinity peptide from DR1 by the small molecule J10-1; 3. Use of redox conditions that enabled efficient formation of a disulfide bond between the DRα helix and incoming peptide. Most peptides with a DNP group were labeled via a Lys side chain at the peptide C-terminus; HA_306-318 C-terminal truncation variants were labeled via the Lys side chain at the P-1 position. DNP-labeled peptides were exchanged onto 5μM cut DR-CLIP complexes at a 10- to 20-fold molar excess of peptide in citrate-phosphate buffer (pH 5.3) in the presence or absence of a small molecule that accelerates peptide loading (J10-1)⁴⁹ overnight at 25 °C, 30 °C or 37 °C, depending on the DR and peptide. Unbound peptide was removed using a Superose 12 gel filtration column (GE Healthcare Bio-Sciences Corp.) and loaded complex was separated from other DR species via affinity chromatography using an anti-DNP-1 column (Biotrend Chemikalien GmbH). DNP and protein absorbance were monitored at 350 nm and 280 nm, respectively⁴⁹.

DNP-labeled HA peptides with a cysteine at the P3 or P6 position were reduced in 500 μM GSH and added to 3C-cleaved DR1-Cys protein; the reaction was then incubated for 2 h in the presence of 50 μM J10-1 and 750μM GSH. The reaction was then shifted for 2 h to oxidizing conditions (950 μM GSSG) to promote disulfide bond formation. Unbound peptide and unloaded complex were removed by gel filtration and DNP affinity chromatography, as described above. Disulfide linkage of HA-Cys peptides to DR1- P₆Cys and DR1-P₃Cys proteins was confirmed based on decreased DRα motility by SDS-PAGE.

Surface Plasmon Resonance

Streptavidin chips (GE Healthcare Bio-Sciences Corp.) were primed and normalized before immobilization of ligand. Biotinylated DM proteins (ligand) were diluted to 0.5μg/ml in HBS-EP buffer (GE Healthcare Bio-Sciences Corp.) and 500 RU of protein (usually DM Mut2 and DM WT) were immobilized in two consecutive flow cells on a SA chip at a flow rate of 2-10 μl/min at 25 °C unless otherwise indicated. SPR experiments were usually performed at 25 °C (range of 20 °C to 37 °C). When an experiment was carried out at a different temperature, the chip was again normalized after change of temperature was completed. DR-peptide complexes were purified by gel filtration chromatography prior to SPR experiments to remove any aggregates. DR-peptide complexes (analyte) were diluted into the running buffer just before injection and injected at the indicated concentrations and conditions (stage 1). This was followed by buffer (stage 2) and injection of peptide (stage 3). Regeneration of the chip was carried out via injection of 50–100 μM of high-affinity peptide for the respective DR allele. Experiments were carried out in degassed 50 mM citrate-phosphate buffer (pH 5.35) with 150 mM NaCl and 0.06% C₁₂E₉ detergent (Calbiochem, EMD Chemicals) unless otherwise indicated. Binding in the reference flow cell (in most cases DM Mut2) was subtracted from binding in the DM WT flow cell.

Peptide binding assay

FP assays were carried out as previously described⁴⁷. Briefly, DR-peptide complexes were incubated with Alexa488-labeled MBP_85-99 or HA_306-318 peptide in a 40 μl volume in black polystyrene 384-well flat bottom plates (Corning) in the indicated conditions and binding of labeled peptide was measured using a Victor³ V plate reader (PerkinElmer). Measurements were made in triplicates. MBP_85-99 was labeled at the P5 position (P5 Lys to Cys mutation) with a maleimide derivative of Alexa-488 (MolecularProbes) ⁴⁷.