Mechanistic Basis for Epitope Proofreading in the Peptide-Loading Complex

This article is featured in In This Issue, p.4041

Cytotoxic T lymphocytes recognize virally or malignantly transformed cells via antigenic peptide epitopes presented on MHC class I (MHC I) molecules. Selection of MHC I loaded with immunodominant epitopes requires a sophisticated interplay of various factors including the transporter associated with Ag processing (TAP) and tapasin (Tsn) as key players, as well as auxiliary chaperones such as calreticulin and the thiol-dependent oxidoreductase ERp57 that is disulfide-linked to Tsn. Together, they comprise the endoplasmic reticulum (ER)-resident peptide-loading complex (PLC) centered on the TAP complex (1, 2). The molecular events during MHC I peptide loading are still not well defined. In particular, our understanding of MHC I proofreading through direct contact between Tsn and MHC I still needs further development. X-ray structures of a range of MHC I molecules and the Tsn-ERp57 conjugate have been described previously (3, 4); the architecture of the Tsn/MHC I complex within the PLC, however, remains to be determined.

The crucial function of Tsn in MHC I loading was first described in studies using the Tsn-deficient human cell line 721.220 (5–7). Its critical role in Ag processing was confirmed in Tsn^−/− mice, where a drastic reduction in MHC I cell-surface expression was observed when compared with Tsn-proficient cells (8, 9). Notably, soluble Tsn, which lacks the transmembrane domain and cytosolic tail, can partly complement MHC I loading and cell-surface expression (10). Tsn has been shown to play a key role in catalyzing peptide loading of MHC I (11) in a process called peptide optimization (12). The Tsn-ERp57 conjugate acts as a scaffold for other PLC components, especially in recruiting and stabilizing peptide-receptive MHC I molecules (13). In addition, Tsn binds to TAP, thus bridging peptide donor and acceptor (14, 15). The function of ERp57 is mainly a structural one, promoting the Tsn–MHC I interaction (3, 13). The Tsn–MHC interaction is thought to be mediated by two conserved ER-lumenal interfaces (3, 16–19).

MHC I alleles differ in their dependence on Tsn with respect to the acquisition of peptides (20–22). Notably, the HLA alleles B*44:02 and B*44:05 differ only by a single residue at position 116, yet they diverge markedly in their dependence on Tsn: B*44:02 carries an aspartate and is Tsn dependent, whereas B*44:05 contains a tyrosine and is Tsn independent. It has been suggested that Tsn widens the peptide-binding pocket and generates an energy barrier, which allows only the binding of high-affinity peptides, thereby disengaging Tsn (23). By analyzing peptide loading onto the MHC I allele H-2K^b using isolated microsomes, Tsn was shown to increase dissociation rates of peptides and thus to reduce the concentration of unstable MHC I complexes with low-affinity peptides (24).

Because an atomic-level picture of the structure and dynamics of the complex between MHC I and Tsn was lacking so far, the mechanistic principle of peptide optimization has remained enigmatic (25). We followed peptide editing in real time in the presence and absence of Tsn and determined the kinetics as well as thermodynamics of the interaction between Tsn and MHC I. The process of peptide optimization was synchronized by a photoreaction, converting a high-affinity epitope into a low-affinity cargo. We show that Tsn accelerates the dissociation rate of low- and medium-affinity (suboptimal) peptide epitopes. Moreover, the differential binding of Tsn to peptide-loaded and peptide-deficient MHC I was observed, which is crucial for peptide editing and selection of immunodominant epitopes. To unravel the underlying mechanism at the atomic level, we obtained the structure of the Tsn/MHC I complex from multimicrosecond all-atom molecular dynamics (MD) simulations. Tsn and the Ag peptide compete for opening/closing the MHC I binding groove, thereby modulating the affinity of the Tsn/MHC I complex.

The DNA sequence coding for HLA-B*44:02 (aa 25-298) was PCR-amplified from pCSB53 using the forward primer (5′-CCTTAATTAACGGCTCCCACTCCATGAGGTATTT-3′) and the reverse primer (5′-CTGCACCGGTCCATCTCAGGGTGAGGGGCTTC-3′). The resulting construct was cloned into a plasmid containing a GFP gene after an internal ribosomal entry site for selection, kindly provided by A. Townsend (Oxford University). The complete method is described elsewhere (26). The final plasmid pCSM71 codes for β₂-microglobulin (β₂m; bearing its native signal peptide), a linker sequence (GS)₆LIN, and the ER-lumenal domain of the HLA-B*44:02 H chain, followed by a myc-tag, a biotinylation recognition sequence, and a His₆-tag. For the generation of the retrovirus, the packaging cell line GP2-293 (BD Biosciences) was cotransfected with pVSV-G (27) and pCSM71 using the Effectine transfection kit (Qiagen). Cell culture supernatants were collected after 2–3 d, centrifuged at 500 × g for 5 min, and used directly for transduction of the production cell line or stored at 4°C. HeLa cells were infected with the recombinant retrovirus in the presence of polybrene. The supernatant was replaced with fresh medium after 24 h, and cells were subjected to one additional cycle of retroviral infection after another 24 h. Transduced cells were sorted via their intrinsic GFP using FACS. The cell line was named HeLa-B4402 SPM148 and yielded high expression levels of HLA-B*44:02.

HeLa-B4402 SPM148 cells were grown at 37°C, 5% (v/v) CO₂, in DMEM, 10% (v/v) FCS supplemented with penicillin/streptomycin in a four-tray cell factory (Nunc) until 90% confluence, washed three times with PBS buffer, pH 7.3, and incubated in DMEM with penicillin/streptomycin. The medium was replaced twice every 3 d. The supernatant was collected, centrifuged for 10 min at 5000 × g, and stored at 4°C until use. Ni-NTA agarose (Qiagen) was equilibrated with HEPES buffer and incubated with supernatant containing single-chain HLA-B*44:02 overnight at 4°C in an overhead shaker. Subsequently, Ni-NTA agarose was collected using an Econo-Pac column (Bio-Rad) and washed. Protein was then eluted with 200 mM histidine in HEPES buffer and finally purified via gel filtration using a Superdex 200 (GE Healthcare) in HEPES-E buffer (20 mM HEPES pH 7.4, 150 mM NaCl, 3 mM EDTA). Purified proteins were concentrated using Amicon Ultra-15 devices (Millipore).

The cDNA encoding for Tsn and ERp57 was a kind gift of Peter Cresswell (Yale University). The ERp57^C60A mutant was used throughout the study to preserve the disulfide bond with Tsn (13). The DNA was amplified via PCR using as forward primer (5′-TATGGATCCATGAAGTCCCTGTCT CTGCTCC-3′) and reverse primer (5′-TAGAAG CTTTTAGTGATGGTGATGGTGATGGTGATGGTGATGAGACTGGAAGTACAGGTTCTCCCACTCAATCTTCTGGGCCTCGAAAATGTCGTTCAGACCGTCCTCAAGGGAGGGCC-3′), resulting in Tsn bearing the native signal sequence and the ER-lumenal domain (1–392), followed by a biotinylation recognition sequence, a TEV-cleavage site, and a His₁₀-tag. A recombinant baculovirus (pFastBac Dual) harboring Tsn and ERp57^C60A was generated according to the manufacturer’s protocol. Sf9 insect cells were cultured in SF900 II medium complemented with 10% FCS (v/v) (Invitrogen) and infected with baculovirus (10% of the total expression volume). Seventy-two hours postinfection, cells were harvested and lysed at 4°C for 1 h in HEPES buffer (20 mM HEPES pH 7.4, 150 mM NaCl) supplemented with 1% (v/v) Triton X-100, 1.0 mM benzamidine, and 2.5 mM PMSF. The cell lysate was centrifuged at 120,000 × g for 30 min at 4°C. The Tsn-ERp57 conjugate was purified via Ni-NTA agarose (Qiagen) and eluted with 200 mM histidine in HEPES buffer. Subsequently, the conjugate was purified via HiTrap Q HP (GE Healthcare) with a NaCl gradient (0–500 mM) in 20 mM HEPES, pH 8.0, and finally polished by gel filtration using a Superdex 200 (GE Healthcare) in HEPES-E buffer. The Tsn-ERp57 conjugate was concentrated using Amicon Ultra-15 devices (Millipore).

Biotin ligase A (BirA) (pET21-BirA; Addgene) was expressed in E. coli and purified as described previously (28). B*44:02 (30 μM) and Tsn-ERp57 (30 μM) were incubated with BirA (1 μM) in BirA reaction buffer (50 mM bicine, pH 8.3, 10 μM biotin, 1 mM MgATP). After 2 h at 30°C, free biotin was removed by rapid gel filtration (MicroSpin G25; Bio-Rad). The biotinylation of the two components was compared by immunoblotting using streptactin-conjugated HRP. Biotinylated B*44:02 (10 μM) was first titrated to equimolar amounts of dimeric NeutrAvidin (dNA, Thermo Scientific). Subsequently, twice the amount of Tsn-ERp57 (20 μM, biotinylation efficiency was lower in comparison with B*44:02) was added and incubated at 4°C for 30 min in HEPES-E buffer to yield a stoichiometric B*44:02/Tsn-ERp57/dNA complex. The monodispersity of the complexes was analyzed in HEPES-E buffer pH 7.0 using size-exclusion chromatography multiangle laser light scattering (SEC-MALLS).

Gel filtration (TSK-GEL G3000 SW_XL column; Tosoh) was performed with an in-line laser light scattering detector (TREOS), refractometer (OptilabrEX; Wyatt Technology), and UV detector (Jasco). The system was equilibrated with HEPES-E buffer pH 7.0 filtered through a 0.1-μm pore size VVLP filter (Millipore), followed by a recirculation through the system for at least 24 h at 0.1 ml/min to improve the baseline by removing air bubbles, as well as particles, via degasser and preinjection filter (0.1 μm), respectively. A total of 250 μg B*44:02/Tsn-ERp57/dNA or B*44:02/dNA in 200 μl was injected and analyzed at a flow rate of 0.5 ml/min. The obtained signals were processed with the ASTRA software (Wyatt Technology) to calculate the molecular mass.

Single-chain HLA-B*44:02 was purified as described earlier. A total of 147 μl B*44:02 (17 μM stock) was diluted into 1 ml 8 M urea, 20 mM HEPES pH 8.0 for denaturation (29). The protein was concentrated to 50 μl using Amicon Ultra-3 devices (Millipore), thereby removing endogenously bound peptides. Refolding and subsequent complex formation were induced by dilution of the denatured protein (final concentration, 1 μM) into 2.5 ml ice-cold refolding buffer (20 mM HEPES pH 8.0, 400 mM l-arginine/HCL, 2 mM EDTA, 5 mM reduced glutathione, 0.5 mM oxidized glutathione, 0.5 mM PMSF), and 20 μM EEFGRA(Anp)SF. Refolding mixture was kept in the dark and mixed using an overhead rotor for 72 h at 4°C. The protein was concentrated to 200 μl using Amicon Ultra-3 devices (Millipore) and subjected to gel filtration using a Superdex 200 (GE Healthcare) in HEPES-E buffer.

The dissociation and exchange kinetics of low- and medium-affinity peptides labeled with Alexa Fluor 633 [AF633; ADIA(C^AF633)VAKY and AAIA(C^AF633)VAKY, respectively] were measured at λ_em/ex 632/647 nm. Association and exchange kinetics were analyzed using EEFG(C^FL488)AFSF at λ_em/ex 480/520 nm. Peptides were synthesized using standard Fmoc solid-phase chemistry and purified by C₁₈ reverse-phase HPLC. The identity of the peptides was verified by electrospray ionization–mass spectrometry. Data were collected by the FluorEssence software and processed with GraphPad Prism.

Peptide-deficient B*44:02 (170 nM) was loaded with a 10-fold molar excess of the medium-affinity peptide ADIA(C^AF633)VAKY or the low-affinity peptide AAIA(C^AF633)VAKY for 48 h at 16°C in 65 μl HEPES-E buffer. Free peptides were removed within 1 min by rapid gel filtration (MicroSpin G25; Bio-Rad). Dissociation kinetics were monitored after adding a 1000-fold excess of unlabeled peptide at 25°C. The dissociation was followed in real time (λ_ex/em 632/647 nm) in the absence and presence of Tsn-ERp57. Data were fitted to a monoexponential dissociation kinetics using GraphPad Prism.

A pH shock was used to dissociate endogenously bound peptides from B*44:02 (30). In brief, purified B*44:02 was titrated with 0.1 M citric acid to a pH of 2.5. The reaction was kept on ice for 2 min and subsequently neutralized with 1 M Tris/HCl, pH 9.0. Released peptides were removed via rapid gel filtration (MicroSpin G25; Bio-Rad). Peptide exchange was initiated by the addition of 170 nM high-affinity peptide EEFG(C^FL488)AFSF to B*44:02/dNA (170 nM) or B*44:02/Tsn-ERp57/dNA (170 nM). Association of EEFG(C^FL488)AFSF and dissociation of ADIA(C^AF633)VAKY or AAIA(C^AF633)VAKY were monitored at λ_em/ex 480/520 (fluorescein) and 632/647 nm (AF633), then fitted by monoexponential association or dissociation kinetics using GraphPad Prism. The error of the rate constants (±) is an SD.

Peptide-deficient B*44:02 molecules (170 nM) were loaded with a 10-fold excess of the photocleavable, high-affinity peptide EEFGRA(Anp)SF [Anp, 3-amino-3-(2-nitro) phenyl-propionic acid] for 48 h at 16°C in 65 μl HEPES-E buffer. Free peptides were removed within 1 min by rapid gel filtration (MicroSpin G25; Bio-Rad). HLA-B*44:02 preloaded with EEFGRA(Anp)SF (170 nM) was incubated with the high-affinity peptide EEFGC^(FL488)AFSF (170 nM). The anisotropy was monitored at λ_em/ex 480/520 nm and 25°C for 60 min. Photocleavage was performed for 10 min on ice using a 366-nm UV lamp (185 mW/cm²; ThorLabs). After photocleavage, the association kinetics of EEFGC^(FL488)AFSF to B*4402 was monitored for 60 min at 25°C in HEPES-E buffer in the absence and presence of Tsn-ERp57.

A total of 25 μg/ml NeutrAvidin was coupled to the surface via standard amine coupling to a density of ∼4000 response units. Biotinylated Tsn-ERp57 was immobilized on the functionalized surface at a flow rate of 5 μl/min to an average of ∼700 response units. B*44:02 binding to and dissociation from Tsn-ERp57 was monitored for 900 and 5000 s, respectively. Regeneration of the chip was performed with 1 M NaCl for 180 s at 30 μl/min. Using a reference flow cell, we corrected sensorgrams for bulk refractive index changes and unspecific binding. All data were double-referenced using responses from blank injections with running buffer. Data were processed using the BIAevaluation software (GE Healthcare) and fitted to a standard 1:1 interaction model using Eq. 1.

R_eq represents the response units at equilibrium, R_max is the maximum analyte binding capacity of the surface, K_D is the equilibrium dissociation constant, and C is the analyte concentration. The binding isotherms were normalized by the R_max values.

The Abs used were 7F6 against Tsn (31), HC10 for HLA B*44:02 (32), α-ERp57 (ERp57; Abcam), α-avidin (NeutrAvidin; Abs-online GmbH), streptactin-HRP (biotin; IBA). All secondary Abs (anti-rat, anti-mouse, anti-rabbit) were purchased from Sigma-Aldrich.

MD simulations were performed with GROMACS 4.6.5 (33), using the Amber99SB-ILDN protein force field (34, 35) and TIP3P water model (36). The SETTLE (37) and LINCS (38) constraint algorithms were applied to water and other molecules, respectively; LINCS expansion order was 2 with three iterations. In combination with virtual hydrogens (39), this allowed for a 4-fs integration time step. Short-range nonbonded Coulomb and Lennard-Jones 6-12 interactions were treated with a Verlet buffered pair list (40) with potentials smoothly shifted to zero at a 10 Å cutoff. Long-range Coulomb interactions were treated with the PME method (41) with a grid spacing of 1.2 Å and cubic spline interpolation. Analytical dispersion corrections were applied for energy and pressure to compensate for the truncation of the Lennard-Jones interactions. A periodic rhombic dodecahedron cell was used. The thermodynamic ensemble was NpT. Temperature was kept constant at 300 K by a velocity-rescaling thermostat (42) with coupling time constant 0.1 ps. For constant 1.0 bar pressure, an isotropic Berendsen barostat (43) was used with coupling time constant 0.5 ps and compressibility of 4.5 × 10⁻⁵ bar^–1.

B*44:02^pep (with EEFGRAFSF peptide) and Tsn coordinates were taken from crystal structures [PDB ID: 1M6O (44) and 3F8U (3), respectively] to generate starting structures. Missing Tsn loops were filled in with MODELER 9.12 (45). The two components were first aligned along their longitudinal axes (as enforced by anchoring in the membrane, which was not included in our present study) and oriented such that B*44:02 T134 contacts Tsn R187 [because these residues are known to be in proximity in the complex (3)], then translated away from each other by 10 Å. The system was solvated, and randomly picked water molecules were replaced by Na⁺ and Cl⁻ ions to yield a concentration of 0.15 M and a neutral overall charge. The final system contained ∼150,000 atoms. Initial velocities were generated at 65 K, and the system was progressively heated to 300 K over 1.0 ns. The final coordinates from the complex-forming simulation after 1 μs were used as the starting point for the comparison of peptide-loaded B*44:02^pep and peptide-deficient B*44:02^pd. The peptide-deficient form was prepared by replacing the peptide from the MHC I binding groove by water molecules, followed by 500 steps of steepest-descent energy minimization. Finally, five 1.0-μs trajectories were acquired for both peptide-loaded and peptide-deficient systems.

Buried surface was computed as the difference in solvent-accessible surface between the protein complex and its isolated components, with a probe size of 1.4 Å. The mean buried surfaces (for both the peptide-loaded and -deficient complexes) were averaged over the last 900 ns of the five trajectories started after complex formation and are presented with their SD as an error estimate. Residue-residue contact occupancy was defined as the fraction of total simulation time that any pair of atoms in two residues is within a 3.5-Å cutoff. Occupancy was averaged over the last 900 ns of each simulation; SD between the individual simulations was used as an error estimate. F-pocket width was measured from C_α-C_α distances d1 between I85 (α₁) and T138 (α_2-1), d2 between Y74 (α₁) and A149 (α_2-1). Distances were averaged over the last 900 ns of each simulation; statistical errors were estimated using block averaging.

Mechanistic analyses of peptide–MHC I association and dissociation have been largely hampered by the dissociation of β₂m from the H chain, which resulted in irreversible inactivation/aggregation of the MHC I molecules. Hence, we linked β₂m covalently to HLA-B*44:02 (residues 25–298). This MHC I allomorph displays the strongest Tsn dependence known so far and is therefore ideal to study the mechanistic basis of peptide editing by Tsn (20, 46). Single-chain B*44:02 with a C-terminal biotinylation tag, purified from stably transduced HeLa cells, eluted as a monodisperse peak at 50 kDa during SEC (Supplemental Fig. 1). The Tsn construct, also containing a C-terminal biotinylation tag, comprises the ER-lumenal domain of Tsn (residues 21–412). Tsn and ERp57, coexpressed in Sf9 cells, were purified as monodisperse Tsn-ERp57 conjugate of 102 kDa (Supplemental Fig. 1). To mimic the lateral organization and interaction within the PLC, we tethered BirA-biotinylated MHC I and Tsn-ERp57 to dNA (Fig. 1A). MHC I/Tsn-ERp57 complexes were assembled by titrating equimolar amounts of B*44:02 to dNA, followed by gel filtration and incubation with biotinylated Tsn-ERp57 (Fig. 1B). Using SEC with in-line MALLS, we detected a stoichiometric complex with a molecular mass of 184 kDa (Fig. 1C).

We first characterized a set of peptides with respect to their affinity as well as association and dissociation rate constants for HLA-B*44:02. The well-characterized B*44:02 epitope EEFGRAFSF derived from HLA-DPα*0201 (47) was labeled with fluorescein via a cysteine positioned at the center of the peptide. EEFG(C^FL488)AFSF has a K_D of 6 nM. ADIA(C^AF633)VAKY and AAIA(C^AF633)VAKY labeled with AF633 display a K_D of 160 and 844 nM for B*44:02 (Table I, Supplemental Fig. 2A–C) and are, therefore, referred to as medium- and low-affinity peptides, respectively. The equilibrium dissociation constants determined by fluorescence anisotropy are in agreement with previously reported data (48). It is noteworthy that binding affinity and kinetics of these model peptides are not affected by the attached fluorophores (Supplemental Fig. 2D).

To investigate the impact of Tsn on the peptide dissociation kinetics, we preloaded peptide-deficient B*44:02 with either the low-, medium-, or high-affinity peptide. Free peptides were removed via rapid-spin gel filtration. Peptide–MHC I complexes were immediately used to analyze peptide dissociation kinetics in the absence or presence of Tsn via fluorescence anisotropy. Peptide dissociation was followed in the presence of a 1000-fold excess of unlabeled peptides to prevent any rebinding of fluorescent peptides. Notably, the Tsn-ERp57 conjugate stimulated the dissociation rate of the medium-affinity complex from 0.34 ± 0.04 × 10⁻³ s⁻¹ to 1.58 ± 0.01 × 10⁻³ s⁻¹ (Fig. 1D). In the case of the low-affinity peptide, the off-rate was found to be 1.78 ± 0.10 × 10⁻³ s⁻¹ in the absence and 7.25 ± 0.43 × 10⁻³ s⁻¹ in the presence of Tsn (Fig. 1E, Table II). The dissociation of the high-affinity epitope was extremely slow and difficult to follow over 40 h, changing only gradually from 0.06 ± 0.03 × 10⁻³ s⁻¹ to 0.10 ± 0.02 × 10⁻³ s⁻¹ when Tsn-ERp57 was added. The Tsn variant 6 (Tn6) with mutations at the predicted MHC I interaction interface (E185K, R187E, Q189S, and Q261S), which is unable to restore B*44:02 surface expression (3), did not alter the off-rate of low- and medium-affinity complexes (Table II, Supplemental Fig. 3). These results demonstrate that a direct interaction between Tsn and MHC I accelerates the dissociation of suboptimal peptides.

To follow peptide editing by Tsn in real time, we loaded B*44:02 molecules with suboptimal epitopes as described earlier. Peptide exchange was initiated by adding an equimolar concentration of the high-affinity epitope EEFG(C^FL488)AFSF (170 nM). Association and dissociation kinetics were monitored simultaneously by dual-color fluorescence anisotropy. Notably, in the absence of Tsn, peptide exchange was very slow, leading to an apparent rate of 0.25 ± 0.14 × 10⁻³ s⁻¹ (k_on,app) for the association of the high-affinity peptide with the low-affinity peptide/MHC I complex, whereas an off-rate (k_off,app) of 0.38 ± 0.18 × 10⁻³ s⁻¹ was measured for the low-affinity peptide (Fig. 2A). These data imply that the dissociation of the suboptimal peptide is the rate-limiting step in peptide exchange. Remarkably, in the presence of Tsn, the rate of peptide exchange was significantly increased, yielding a k_on,app of 2.65 ± 0.07 × 10⁻³ s⁻¹ for the high-affinity peptide and an off-rate of 2.81 ± 0.80 × 10⁻³ s⁻¹ for the low-affinity peptide, indicating that Tsn catalyzes peptide exchange by increasing the dissociation rate of low-affinity peptides.

We also examined whether Tsn stimulates the exchange of medium-affinity peptides by a high-affinity epitope (Fig. 2B). In the absence of Tsn, peptide exchange was not observed within the time period of the measurement. In contrast, in the presence of Tsn, the dissociation of the medium-affinity peptide and the association of the high-affinity peptide were dramatically accelerated, yielding apparent association and dissociation rates of 1.01 ± 0.06 × 10⁻³ s⁻¹ and 0.97 ± 0.17 × 10⁻³ s⁻¹, respectively. The real-time analysis of peptide exchange in MHC I/Tsn-ERp57 complexes, reconstituted by isolated components, provides direct proof of the function of Tsn-ERp57 in peptide exchange in favor of high-affinity, immunodominant epitopes.

To follow the immediate response when a high-affinity epitope is converted into a low-affinity cargo, we made use of the photocleavable, high-affinity peptide EEFGRA(Anp)SF. The photolabile amino acid (Anp) was incorporated at a solvent-exposed, nonanchored position, predicted to have no effect on the peptide–MHC I complex (Fig. 3A). Upon illumination at 366 nm for 1 min, this high-affinity epitope is converted into two low-affinity fragments (Fig. 3A). MHC I molecules loaded with the photocleavable, high-affinity peptide were incubated with equimolar amounts of the high-affinity peptide EEFG(C^FL488)AFSF. Before photocleavage, no significant peptide exchange of MHC I loaded with the photocleavable, high-affinity peptide was observed either in the absence or presence of Tsn (Fig. 3B). This is in line with the findings described above, demonstrating that the exchange of high-affinity peptides on B*44:02 is kinetically disfavored even in the presence of Tsn. Upon photoconversion into a suboptimal cargo, the function of Tsn becomes obvious in accelerating the exchange of low-affinity cargo against high-affinity peptide. It is worth mentioning that 90% of the MHC I molecules loaded with the photocleavable peptide are reloaded with the high-affinity peptide EEFG(C^FL488)AFSF after the light-triggered peptide editing (Supplemental Fig. 4). In conclusion, Tsn catalyzes the thermodynamically favored, but kinetically disfavored, exchange of suboptimal cargo against high-affinity epitopes, and therefore preferentially enables kinetically and thermodynamically stable peptide–MHC I complexes to travel to the cell surface.

Based on the previous results, Tsn must control the quality of peptide–MHC I complexes. However, the direct interaction between Tsn and peptide–MHC I has not yet been characterized. Using an oriented site-specific immobilization of Tsn-ERp57, we determined the kinetics and thermodynamics of this transient interaction with MHC I by surface plasmon resonance (SPR). To convert peptide-loaded into peptide-deficient MHC I complexes in situ, we preloaded HLA-B*44:02 with the photocleavable, high-affinity peptide. As expected, MHC I molecules loaded with the high-affinity peptide (B*44:02^pep) displayed a relatively slow association and a fast dissociation from Tsn-ERp57, reflecting a low-affinity, transient interaction (Fig. 4A). Strikingly, after fast photoconversion of the high-affinity epitope into a suboptimal cargo, peptide-deficient B*44:02^pd now revealed a high-affinity interaction, reflected by the fast association and slow dissociation kinetics (Fig. 4B). Reloading B*44:02^pd/Tsn-ERp57 complexes with the high-affinity peptide converts the high-affinity into a transient interaction (Fig. 4C). This proves that the affinity and kinetics of the MHC/Tsn-ERp57 interaction are specifically and reversibly altered by the peptide-loading status of MHC I. The binding profiles at 1.5 μM of B*44:02 were overlaid to illustrate the differences between peptide-loaded and -deficient B*44:02, as well as the impact of peptide loading on the dissociation of B*44:02^pd/Tsn-ERp57 complexes. Equilibrium dissociation constants K_D were derived from the equilibrium binding (R_eq; Fig. 4D). The data are in excellent agreement with a Langmuir-type (1:1) interaction model, reflecting a high-affinity interaction between Tsn and peptide-deficient B*44:02 (K_D,1 = 0.20 ± 0.01 μM) and transient, low-affinity interaction with peptide-loaded B*44:02 (K_D,2 = 4.77 ± 0.29 μM). These affinities correspond to a free energy difference of ΔΔG = −RT ln (K_D,2/K_D,1) = −8 kJ/mol by which Tsn stabilizes B*44:02^pd over B*44:02^pep. Because the peptide-deficient state is traversed during peptide exchange, stabilization of this high-energy intermediate explains the observed accelerated peptide exchange.

We also compared the Tsn-binding characteristics of B*44:02 and B*44:05. The two allomorphs, loaded with endogenous peptide, were probed for interaction with immobilized Tsn-ERp57 (Fig. 5). The affinities (K_D) of the peptide-loaded MHC I molecules to Tsn-ERp57 were determined as 1.8 ± 0.2 μM for B*44:02 and 10.3 ± 0.6 μM for B*44:05. For the peptide-deficient forms B*44:02^pd and B*44:05^pd, K_Ds of 0.31 ± 0.02 and 1.3 ± 0.2 μM, respectively, were determined (Table III). Taken together, these results demonstrate that the Tsn-dependent MHC I allele, B*44:02, has a significantly higher affinity toward Tsn-ERp57 than the Tsn-independent B*44:05, which is in line with the notion that HLA-B*44:05 does not require Tsn for peptide loading, whereas Tsn is indispensable for loading of B*44:02 (49). We also checked by SPR whether the Tn6 mutant is able to discriminate between peptide-loaded and -deficient MHC I. This is not the case; as with immobilized Tn6-ERp57, we were unable to reach saturation for both peptide-loaded and -deficient B*44:02 even at a concentration of 40 μM (Supplemental Fig. 3F).

To understand at the atomic level the mechanism by which Tsn differentiates between peptide-loaded and -deficient B*44:02, we used multimicrosecond all-atom MD simulations in explicit solvent to characterize the formation of the Tsn/B*44:02 complex. Starting from the two solvent-separated proteins, spontaneous formation of the Tsn/B*44:02 complex was observed in an unbiased 1.0-μs MD simulation (Fig. 6). Two distinct protein–protein interfaces were formed (Fig. 6A), one between the Tsn N-terminal domain and B*44:02-α₂ (Fig. 6B), the other between the Tsn C-terminal domain and B*44:02-α₃ (Fig. 6C). To improve the sampling of these identified Tsn/B*44:02 interfaces, we initiated five additional 1-μs simulations for both peptide-bound and -free B*44:02 from the final complex structure of the first step (Fig. 6D). No significant differences were observed between the peptide-loaded and -deficient forms in terms of size of the interfaces or nature of the individual residue–residue contacts. At the N-terminal interface (Fig. 6B), Tsn cradles the B*44:02 α_2-1 helix, potentially stabilizing B*44:02 in a peptide-receptive conformation and preventing partial unfolding of α_2-1 (50). Tsn contacts B*44:02 α_2-1 with K16, L18, and L79, the α_2-1/2 hinge with W85, and the underside of the β-sheet with R187 and Q189. B*44:02 residues on the Tsn-facing side of α_2-1 (mainly I142, R145) are involved in the interface, as are R151 and β-sheet residues (D129, S132, T134). Our results are consistent with previous studies proposing that Tsn acts on the α_2-1 helix (16–19, 51); furthermore, they support structural data and mutagenesis experiments showing loss of peptide-loading activity in the Tsn mutant Tn6, in which R187 and Q189 are mutated (3). For the C-terminal interface (Fig. 6C), our MD simulations predict interactions between the CD8 binding site of B*44:02 and a cluster of basic Tsn residues (R333, H334, H335, H345). To establish these contacts, the C-terminal domain of Tsn has to rotate compared with its initial orientation in the X-ray crystal structure. The C-terminal Tsn residues observed in our MD simulations have previously been suggested to be involved in the interaction and have been shown to influence assembly and surface expression of MHC I molecules (52–55).

Our observation that the protein–protein interfaces are similar, irrespective of the peptide-loading status of B*44:02, raises the intriguing question of how Tsn distinguishes between B*44:02^pep and B*44:02^pd. Fig. 6E compares the occupancies of the most prevalent contacts between Tsn and B*44:02 residues. At the N-terminal interface, all contacts with >60% occupancy, which are thus likely to play a key role in the complex, are more prevalent in B*44:02^pd than in B*44:02^pep. By contrast, at the C-terminal interface, this occupancy difference is not observed (inset). In line with these tighter contacts between Tsn and B*44:02^pd, the peptide-binding groove F-pocket is widened (Fig. 6B): Upon peptide removal, the I85-T138 (d1) and Y74-A149 (d2) C_α-C_α distances increase from 10.8 ± 0.7 to 11.8 ± 0.4 and from 21.8 ± 0.4 to 23.8 ± 0.8 Å, respectively. This small but significant widening by 1–2 Å suffices to stretch or even break H-bonds and van der Waals contacts between the F-pocket and the peptide, hence lowering its affinity. At the same time, it allows a better match between B*44:02 α_2-1 and Tsn, therefore increasing the affinity, as reflected in the occupancy differences and observed experimentally.

Various studies have tried to identify the mechanism by which Tsn promotes peptide loading of MHC I molecules (12, 13, 20, 23, 24, 56). These studies have led to different interpretations. Unexpectedly, peptides had a lower affinity for HLA-B*08:01 in the presence of Tsn than in its absence (57). A facilitator function for Tsn was therefore proposed. Intriguingly and in contrast with the previous study, the same affinities were found in Tsn-negative cells expressing HLA-B*27:05 or A*02:01 (7, 12). In both systems, the MHC I–peptide complexes expressed at the cell surface were much more stable and the peptide repertoire was evidently altered in the presence of Tsn.

Attempts to elucidate the function of Tsn were hampered by the fact that the direct interaction between MHC and Tsn could not be analyzed so far. Thus, the mechanistic basis of peptide editing and proofreading mediated by Tsn within the PLC could not be fully elucidated. By tethering biotinylated components of the PLC to a NeutrAvidin dimer, we have established a defined platform to study the transient interaction between Tsn and B*44:02, thus mimicking the structural organization of the MHC I/Tsn-ERp57 complex at the ER membrane. By using a set of peptides with affinities ranging from high to medium and low (6–844 nM), we were able to analyze the function of Tsn in peptide editing. In the absence and presence of Tsn, peptide dissociation followed monoexponential kinetics, which is in contrast with other studies, where dissociation kinetics were biphasic (23, 58). In these studies, the authors proposed that β₂m first dissociates from MHC I followed by the peptide (58). Because β₂m was covalently linked to the B*44:02 H chain in this study, its dissociation was prevented.

Tsn increased the dissociation rate of low- and medium-affinity peptides up to 10-fold, whereas the Tn6 mutant, which has a mutated Tsn/MHC I interface and is therefore inactive in peptide loading (3), has no impact on the dissociation rate of suboptimal peptides, proving the accuracy of the interaction analysis (Supplemental Fig. 3). We further demonstrate that Tsn-ERp57 significantly accelerates peptide exchange of suboptimal for optimal, high-affinity peptides. In the absence of Tsn-ERp57, exchange of suboptimal peptides by high-affinity ones is extremely slow, although thermodynamically favored. However, in the presence of Tsn-ERp57, the peptide exchange rate was drastically increased in favor of the high-affinity peptide. To demonstrate peptide editing in real time, we made use of MHC I complexes preloaded with a photocleavable, high-affinity peptide. This epitope bound to B*44:02 cannot be displaced by another high-affinity peptide, neither in the absence nor in the presence of Tsn. Upon photoconversion of the high-affinity into the low-affinity cargo, peptide exchange is synchronized and significantly accelerated by Tsn.

To the best of our knowledge, our study determines for the first time in atomic detail the interaction between MHC I and Tsn and provides direct evidence for a differential binding of Tsn to peptide-loaded and -deficient MHC I, which is key for understanding the observed acceleration of the peptide exchange reaction. Peptide exchange proceeds via the peptide-free form of MHC I as a high-energy intermediate that crucially determines the energy barrier (and thus the rate) of the process (Fig. 7). Tsn monitors the quality of peptide–MHC I complexes and binds peptide-deficient MHC I much stronger than the peptide-loaded complex because of competition of Tsn and peptide for binding to MHC I. Using SPR, we show that Tsn stabilizes peptide-free MHC I relative to the peptide-bound form by ΔΔG = −8 kJ/mol (Fig. 7). This explains the observed accelerated exchange kinetics, which are essential for efficient sampling of high-affinity epitopes. Accordingly, in the presence of Tsn, peptide loading is shifted from kinetic control toward the thermodynamically controlled regime, hence increasing the probability of binding of a high-affinity peptide against a background of many competing low-affinity peptides during the transit time of typical class I molecules through the ER. The mechanism derived from the kinetic and thermodynamic data of this study agrees with phenomenological kinetic network modeling (59); the kinetic network model, however, could not take the quantitative stabilization of the peptide-free intermediate into account.

The mechanism of how Tsn acts on MHC I–peptide binding involves differential binding of Tsn to MHC I depending on its peptide-loading status. Our MD simulations provide detailed atomic-level insights into this intricate mechanism: Tsn interacts with the α_2-1 helix of MHC I and its supporting β-strands (Fig. 6B), which are part of the peptide-binding groove. Tsn and the Ag peptide thus act on the groove as the two players of a molecular tug-of-war mechanism: a high-affinity peptide will succeed in tightly closing the binding groove, whereas the absence of peptide or the binding of a low-affinity peptide widens the groove. This facilitates the release of suboptimal peptides, thus generating peptide-deficient MHC I with high affinity for Tsn (Fig. 7). In contrast, MHC I molecules loaded with high-affinity peptides display a very tight peptide binding groove, thereby interacting with Tsn at low affinity and priming the PLC for dissociation.

In this study, we investigated the peptide loading of B*44:02, which shows the strongest dependency on Tsn. However, care must be taken when extrapolating results to other MHC I molecules, which are less affected by Tsn. Taking published data (50, 60, 61) into account, we propose a model for the working cycle of Tsn (Fig. 7). Tsn monitors the quality of bound peptides (suboptimal versus optimal epitopes) by acting on the MHC I binding groove. Upon engagement of MHC I with Tsn, suboptimal peptides are displaced, and the peptide repertoire of MHC I is edited in favor of high-affinity peptides. Based on this working cycle, only kinetically stable peptide–MHC complexes reach the cell surface, which is crucial for a prolonged CD8⁺ T cell–mediated immune response against tumors and intracellular pathogens.

We thank Tim Elliott (University of Southampton) for the B*44:02 cDNA, Ute Claus for technical support, and Alain Townsend (University of Oxford) for help in generating the cell line expressing the soluble, single-chain HLA-B*4402.

The authors have no financial conflicts of interest.