Structure of an MHC I–tapasin–ERp57 editing complex defines chaperone promiscuity

Adaptive immunity depends on cell surface presentation of antigenic peptides by major histocompatibility complex class I (MHC I) molecules and on stringent ER quality control in the secretory pathway. The chaperone tapasin in conjunction with the oxidoreductase ERp57 is crucial for MHC I assembly and for shaping the epitope repertoire for high immunogenicity. However, how the tapasin–ERp57 complex engages MHC I clients has not yet been determined at atomic detail. Here, we present the 2.7-Å crystal structure of a tapasin–ERp57 heterodimer in complex with peptide-receptive MHC I. Our study unveils molecular details of client recognition by the multichaperone complex and highlights elements indispensable for peptide proofreading. The structure of this transient ER quality control complex provides the mechanistic basis for the selector function of tapasin and showcases how the numerous MHC I allomorphs are chaperoned during peptide loading and editing.

Adaptive immunity depends on cell surface presentation of antigenic peptides by major histocompatibility complex class I (MHC I) molecules and on stringent ER quality control in the secretory pathway. The chaperone tapasin in conjunction with the oxidoreductase ERp57 is crucial for MHC I assembly and for shaping the epitope repertoire for high immunogenicity. However, how the tapasin-ERp57 complex engages MHC I clients has not yet been determined at atomic detail. Here, we present the 2.7-Å crystal structure of a tapasin-ERp57 heterodimer in complex with peptide-receptive MHC I. Our study unveils molecular details of client recognition by the multichaperone complex and highlights elements indispensable for peptide proofreading. The structure of this transient ER quality control complex provides the mechanistic basis for the selector function of tapasin and showcases how the numerous MHC I allomorphs are chaperoned during peptide loading and editing.
The presentation of antigenic peptides by major histocompatibility complex class I (MHC I) molecules on the cell surface ensures the detection and elimination of infected or cancerous cells by the adaptive immune system. Any misinterpretation can lead to inflammatory and autoimmune diseases. Cytotoxic T lymphocytes, which are licensed to kill, search for non-self peptides presented by MHC I 1,2 . MHC I complexes are composed of a heavy chain (hc) and the light chain β 2 -microglobulin (β 2 m). Encoded by the most polymorphic gene cluster in humans with more than 24,000 known classical human leukocyte antigen (HLA)-A, -B, and -C alleles, all heavy chains invariantly share their association with β 2 m. Prior to presenting peptides on the cell surface, MHC I molecules undergo a stringent quality control during their maturation process in the endoplasmic reticulum (ER) by transiently interacting with specialized proteins, including glycan sensors, disulfide isomerases, and specific MHC I chaperones 3 . These ER chaperone systems play a pivotal role in adaptive immunity not only by stabilizing folding intermediates but also by shaping the peptide repertoire displayed on MHC I molecules.
The two MHC I-specific chaperones, tapasin and TAP-binding protein related (TAPBPR), were shown to share similar binding interfaces on MHC I and were thus suggested to have common catalytic principles [4][5][6][7][8] . However, the two chaperones differ substantially in their subcellular localization and their molecular liaisons. TAPBPR acts autonomously on a subset of MHC I clients in the peptide-depleted environments of the cis-Golgi and ER-Golgi intermediate compartment (ERGIC) 9 . As the major MHC I editor, tapasin fulfils its function predominantly as part of the peptide-loading complex (PLC) in conjunction with the oxidoreductase ERp57 and the lectin-like chaperone calreticulin in the ER 3,8 . The interaction between tapasin and ERp57 relies on a mixed disulfide between Cys 95 of tapasin and Cys 33 of ERp57 (Cys 57 in immature ERp57) 4 . In the oxidizing environment of the ER, Cys 33 of ERp57 toggles between an intermolecular, mixed disulfide with Cys 95 of tapasin and an intramolecular disulfide with Cys 36 (Cys 60 in immature ERp57) 10 . This toggling, known as the escape pathway, has been described for protein-disulfide isomerase-assisted protein folding and prevents the enzyme from becoming trapped in covalent complexes with its substrates 11 .
Peptide-deficient MHC I molecules are recruited to the PLC 12,13 , undergo peptide loading and editing 14,15 , dissociate from the PLC as stable peptide-MHC I (pMHC I) complexes, and traffic to the cell surface for antigen presentation 3 . During peptide loading and editing, tapasin stabilizes peptide-receptive MHC I heterodimers and catalyzes peptide exchange by accelerating peptide association and dissociation [15][16][17] . Deletion of tapasin and ERp57 results in an impaired immune response and altered peptide repertoire 18,19 . The various MHC I allomorphs exhibit distinct levels of plasticity, differ in peptide-binding preferences, and depend to different degrees on tapasin for optimal peptide loading 20,21 . Conformational plasticity of MHC I scales with the ability to select high-affinity peptides in the absence of tapasin [22][23][24][25] . Insights into the molecular basis of tapasin promiscuity for different MHC I clients as well as structural elements necessary for peptide exchange catalysis at atomic resolution are lacking due to the intrinsically transient nature of the editing complex.
Here, we present the 2.7-Å crystal structure of an MHC I-tapasin-ERp57 editing complex assembled by a photo-triggered approach. The structure of this transient assembly reveals molecular rearrangements within the chaperone complex upon client engagement. Movement of the C-terminal immunoglobulin domain of tapasin extends the interaction interface to MHC I hc and β 2 m. We show by functional assays that structure-guided ablation of interface interactions leads to reduced cell surface presentation of pMHC I complexes. The editing loop of tapasin rigidifies upon client recognition and contributes to a widened MHC I F-pocket. Since the loop structure might also stabilize a peptide-bound MHC I pocket, our data suggest a dual function of the editing loop in peptide exchange catalysis. Our structural and functional data reveal the molecular basis of tapasin promiscuity towards MHC I clients and suggest that the level of plasticity of peptide-receptive MHC I determines chaperone-assisted acquisition of high-affinity peptides.

Structure of the MHC I-tapasin-ERp57 editing complex
We captured a human tapasin-ERp57 heterodimer with bound MHC I hc and β 2 m in flagrante by a photo-triggered approach 7 and determined the crystal structure of the MHC I-chaperone complex to 2.7 Å resolution (Fig. 1). To facilitate crystallization, we utilized the ectodomains of tapasin and MHC I hc lacking the transmembrane and cytosolic regions. After photo-cleavage of loaded peptides, we first screened different MHC I allomorphs for complex formation with the tapasin-ERp57 heterodimer by size exclusion chromatography (SEC), resulting in murine H2-D b as the most suited MHC I hc candidate (Fig. 1a, b, and Supplementary Fig. 1a-d). Furthermore, we substituted Cys 36 of ERp57 with Ala (C36A), formerly described as C60A in immature ERp57 (ref. 4), to retain the mixed disulfide between tapasin and ERp57 (ref. 10).
Since a homogeneous population of a kinetically stable chaperone-client complex between disulfide-linked tapasin-ERp57 and refolded MHC I hc/β 2 m could not be isolated (Fig. 1a, b), we directly used a photoactivated mix of refolded photo-peptide/H2-D b / β 2 m and tapasin-ERp57 (1.3: 1 molar ratio) for crystallization. Size exclusion chromatography-coupled mass spectrometry (SEC-MS) revealed an almost complete photo-induced cleavage (>96%) of the peptide bound to MHC I ( Supplementary Fig. 2). Initial crystals of the chaperone-client complex belonging to space group P2 1 2 1 2 1 diffracted X-rays to 4.0 Å resolution with two heterotetrameric editing complexes in the asymmetric unit. Utilizing microseed matrix screening 26 , we obtained crystals in space group P22 1 2 1 with an improved resolution of 2.7 Å and only one complex in the asymmetric unit, enabling us to build an atomic model with good stereochemistry ( Fig. 1c and Supplementary Table 1).
As in the client-free structure of tapasin-ERp57 (ref. 4), the four thioredoxin-like domains of ERp57 (a-b-b′-a′) are arranged in their characteristic twisted U form, whereas tapasin forms its typical L-shaped structure 4,8 (Fig. 1c and Supplementary Movie 1). The a and a′ domains of ERp57 interact with the N-terminal region of tapasin composed of a fusion between a seven-stranded β barrel and an immunoglobin (Ig)-like fold 4 . In contrast to the structure of client-free tapasin-ERp57 (ref. 4), we could not model the mixed disulfide between Cys 95 of tapasin and Cys 33 of ERp57 present in the purified tapasin-ERp57 heterodimer ( Supplementary Fig. 1). Since disulfide bond reduction did not result from UV illumination performed for photocleavage of the peptide (Supplementary Fig. 1), we speculate that reduction is either facilitated by X-ray irradiation 27 or resulted from the MHC I editing process itself.

Molecular rearrangements within the chaperone complex upon client engagement
The interface between tapasin and MHC I covers a total surface area of 4381 Å 2 (hc: 74%; β 2 m: 26%) ( Supplementary Fig. 3) of which the N-terminal domain of tapasin contributes 2657 Å 2 . The concave binding site of the N-terminal domain embeds the α2-1 helix region of MHC I hc like a clamp (Fig. 1c). On one side of the clamp, a loop comprising residues Glu 11 -Lys 20 , referred to as editing loop in the following, sits on top of the F-pocket of the MHC I peptide-binding groove, whereas a β hairpin element contacts the floor of the peptide-binding groove from below ( Fig. 2a-c). When superimposed with atomic models of clientfree tapasin-ERp57 (ref. 4) and peptide-bound MHC I (ref. 28), our structure reveals that client binding leads to important rearrangements of the α1 and α2-1 helices in MHC I, resulting in a widened F-pocket (Fig. 2a). This widening is stabilized by a hydrogen-bond network between His 70 and Glu 72 of tapasin, which attracts the side chain of Tyr 84 of the MHC I hc (Fig. 3a, b). The side chain of Lys 20 of tapasin thereby facilitates positioning of Tyr 84 by hydrogen bonding with the main-chain oxygen (Fig. 3b). Notably, Glu 72 of tapasin structurally corresponds to Glu 105 of TAPBPR, which was found to contact Tyr 84 of MHC I hc in a similar manner 6,7 . The β-sheet floor (β-strands β5-7) of the peptide-binding groove near the F-pocket is distorted upon binding, presumably by engagement with the β hairpin of tapasin (β strands β9-10) (Fig. 2b). Upon client binding, the β hairpin of tapasin is shifted downwards to accommodate the α2-1 region of MHC I (Fig. 2c) in a similar way as revealed in TAPBPR-MHC I structures 6,7 . Furthermore, the C-terminal Ig-like domain of tapasin swings towards the α3 domain of the MHC I hc and β 2 m (Fig. 2d), allowing for an additional interaction interface with a buried surface area of 1724 Å 2 (hc: 55%; β 2 m: 45%; Supplementary Fig. 3a, b).

The editing loop of tapasin contributes to a widened MHC I F-pocket
Upon client binding, the intrinsically disordered editing loop of the client-free state of tapasin 4,29 rigidifies as evidenced by a clear electron density at the 1.5 σ level for most loop residues ( Fig. 3a-d  loop is stabilized by the side chain of Asn 80 of MHC I, which forms a hydrogen bond with the backbone-carbonyl oxygen of Lys 16 of tapasin (Fig. 3d). The side chain of Asp 12 of tapasin is in salt-bridge distance to the ε-amino group of Lys 146 of the MHC I hc that presumably contributes to the widening of the peptide-binding groove and the rigidification of the editing loop of tapasin (Fig. 3d). Furthermore, the side chain of Leu 18 of the editing loop, which was shown to be relevant for MHC I surface expression 30,31 , compensates in the peptide-free MHC I F-pocket for the lack of a C-terminal hydrophobic side chain of the antigenic peptide and presumably assists in groove widening. Even though sitting directly above the F-pocket, Leu 18 does not mimic the C terminus of cargo peptides. Thus, Leu 18 is unlikely to act as a peptide surrogate per se 30 . In line with this groove-widening function, substitution of Leu 18 with a bulky hydrophobic side chain (tryptophan) did not negatively impact MHC I surface presentation in our flow cytometry-based cellular assays when compared to wildtype (WT) ( Fig. 3e and Supplementary Fig. 3c). Conversely, substitution of Leu 18 with glycine has been shown to nearly abolish peptide exchange activity of tapasin 31 . Surprisingly, abrogation of the hydrogen bond between Tyr 84 of MHC I and Glu 72 of tapasin did not affect MHC I surface presentation (Fig. 3e), as demonstrated by an in vitro peptide-loading assay 4 . Introducing a charge repulsion by mutating tapasin Lys 20 , which helps in positioning the α1 helix of MHC I via hydrogen bonding to the backbone-carbonyl oxygen of Tyr 84 (Fig. 3a, b), did not substantially decrease the surface MHC I level (72 ± 10% of WT; Fig. 3e), suggesting that the editing loop of tapasin mainly stabilizes the empty F-pocket by a hydrophobic plug. At the same time, MHC I F-pocket residues Asn 80 , Tyr 84 , and Lys 146 , which coordinate the terminal carboxyl group of bound peptides, are engaged in interactions with tapasin and its editing loop (Fig. 3a, b and Supplementary Fig. 4a). Our data suggest that tapasin acts in a dual manner, opening the peptide-binding groove and stabilizing the empty F-pocket of the MHC I hc.

Domain movements in tapasin extend the interaction interface with MHC I
The interface loop region of tapasin on the lateral side of the MHC I α2-1 helix 4,7 is not involved in specific interactions and therefore only plays a minor role in MHC I surface presentation (Fig. 4a). In contrast, the neighboring Gln 261 of tapasin, which is involved in several hydrogen bonds with the α2-1 helix region, leads to a reduction of MHC I surface presentation to 65 ± 6% when mutated to alanine (Fig. 4b, f and Supplementary Fig. 3d). An even stronger suppression of surface presentation was observed when interactions between the β hairpin of tapasin and the α2 domain of MHC I hc were disrupted (R187E; 55 ± 13% compared to WT) (Fig. 4c, f). Mutations in both the interface loop and the β hairpin synergistically reduced MHC I surface presentation to 46 ± 4% (Fig. 4f). Interestingly, as in free tapasin-ERp57 (ref. 4), we observed an intermolecular contact between side chains of Arg 333 of tapasin and Glu 229 of MHC I hc (Fig. 4d), which, when disturbed by charge repulsion, reduced MHC I surface presentation to 37 ± 3% (Fig. 4f). The importance of this interaction between the CD8recognition loop of MHC I and the IgC domain of tapasin was already proposed by the cryogenic electron microscopy structure of the PLC 8    impact on surface presentation (Fig. 4d, f). The displaced IgC domain of tapasin is further stabilized by contacts to β 2 m, involving backbone interactions between Ile 92 of β 2 m and Leu 329 of tapasin and a side chain hydrogen bond between Ser 88 of β 2 m and Glu 307 of tapasin (Fig. 4e). Disruption of the hydrogen bond between Glu 307 of tapasin and Ser 88 of β 2 m reduced the MHC I surface level to 46 ± 1% (Fig. 4e, f); albeit the expression level of mutant tapasin was also lowered compared to WT (Supplementary Fig. 3e).

Dual function of the editing loop in peptide exchange catalysis
Mechanistic studies suggested that the MHC I-specific chaperones tapasin and TAPBPR hold distinct preferences for MHC I allomorphs [33][34][35] . However, our structure of the MHC I-tapasin-ERp57 editing complex demonstrates that the binding interfaces used for client recognition are similar between the two editors with minor differences for specific residues (Supplementary Figs. 5 and 6). Comparing the impact of the two editors on widening the peptide-binding cleft reveals a slightly stronger displacement of the α2-1 helix of MHC I when associated with TAPBPR ( Supplementary Fig. 4b). In our structure, the editing loop lies on top of the MHC I F-pocket (Fig. 3). In chaperone-free pMHC I, the side chains of residues Asn 80 and Lys 146 point into the F-pocket and coordinate the C terminus of bound peptides (Supplementary Fig. 4a). In an intermediate peptide-/chaperone-bound state the carbonyl group of Gly 15 of the editing loop may present a hydrogen-bond acceptor, which could shield the peptidebound MHC I F-pocket, consistent with a peptide trapping function 29 . We propose that the editing loop fulfills two putatively contrasting tasks. On the one hand, the editing loop contributes to the widening of the F-pocket in the peptide-receptive state of MHC I, whereas, on the other hand, it shields the peptide C terminus during the epitopeloading process. Such a scenario may also be applicable to TAPBPR as NMR studies showed that the longer editing loop of TAPBPR (residues Gly 24 -Arg 36 ) can be observed as a lid on top of the F-pocket 29 , whereas in one of the MHC I/TAPBPR structures, the loop dives into the F-pocket and stabilizes peptide-receptive MHC I by acting as a peptide surrogate (scoop loop) 7,36,37 (Supplementary Fig. 4b).   The gating strategy is displayed in Supplementary Fig. 8. α2-1 interface, brown; βhairpin interface, dark orange; α3 interface, orange; β 2 m interface, light orange. WT wildtype, − mock transfected. Source data for e are provided as a Source Data file. The same color code as used in a is applied in b-e.

Tapasin promiscuity for MHC I clients
In the tapasin-chaperoned state, the position of β 2 m in relation to the MHC I hc is similar (RMSD 1.3 Å) to the peptide-bound state 28 , whereas binding of TAPBPR leads to a repositioning of β 2 m 6,7 . Strikingly, sequence alignments of classical MHC I allomorphs show that the interaction sites for tapasin are conserved among tapasin-dependent and -independent clients and are shared between tapasin and TAPBPR, except for residue Trp 73 of H2-D b (Supplementary Fig. 6). Analyzing the primary structure of non-classical MHC I molecules HLA-E, -F, and -G demonstrated that not all potential interacting residues are conserved, but most of the alternative residues are still capable to form hydrogen bonds ( Supplementary Fig. 7). In essence, the crystal structure of the MHC I-tapasin-ERp57 editing complex reveals that both tapasindependent and -independent allomorphs have the structural prerequisites to interact with the chaperone, suggesting that their degree of plasticity and hence their ability to accommodate high-affinity peptide cargo determines the association, which is stabilized by key structural elements of tapasin (Fig. 5). It needs further structural and functional characterization to understand how e.g., C-terminally extended peptides that protrude from the MHC I F-pocket are edited by tapasin or TAPBPR. As tapasin acts primarily as part of the PLC multichaperone network, we speculate that association of ERp57 with tapasin is not only necessary for coordinating the glycan-chaperone calreticulin, but that the molecular dynamics of this interaction may help unlock tapasin for client engagement and peptide proofreading catalysis.

Plasmid constructs
The constructs of human β 2 -microglobulin (β 2 m) (UniProtKB P61769; amino acids 21-119) and of the ectodomain of H2-D b (UniProtKB P01899; amino acids 21-300) were previously described 7 . DNA fragments encoding the ER-lumenal domain of human tapasin (UniProtKB O15533; amino acids 21-402; tapasin ΔTMD ) and human ERp57 (UniProtKB P30101; amino acids 25-505) were amplified by polymerase chain reaction (PCR) and cloned into the modified MultiBac vectors pAMI and pMIDK, respectively 38 . The C36A escape pathway mutation (ERp57 C36A ) was introduced in the ERp57-encoding construct by Quik-Change mutagenesis 11 . A C-terminal His 6 -tag and a Tobacco Etch Virus (TEV) protease cleavage site were introduced into the pAMI vector encoding tapasin ΔTMD by sequence-and ligation-independent cloning (SLIC) 39 . The pMIDK_ERp57 C36A donor vector was fused by Cremediated recombination to the pAMI_tapasin ΔTMD acceptor vector, resulting in one transfer plasmid for integration into the EMBacY baculoviral genome via Tn7 transposition 40 . Tapasin-encoding plasmids for flow cytometry analyses were generated by assembling the coding sequence of wildtype tapasin (UniProtKB O15533; amino acids 21-448), an internal ribosomal entry site 2 (IRES2) followed by the enhanced green fluorescent protein (eGFP)-encoding sequence into the pAMI vector. Interaction mutations of tapasin were introduced by SLIC 39 . All vectors were verified by Sanger sequencing.

Protein production and purification
H2-D b and β 2 m were expressed as inclusion bodies in Escherichia coli Rosetta(DE3)pLysS as described 41 . In brief, Rosetta(DE3)pLysS cells were transformed with plasmids encoding β 2 m or H2-D b . Overnight cultures of 20-30 ml in TYP media (16 g tryptone, 16 g yeast extract, 5 g NaCl, 1.25 g K 2 HPO 4 per 1 l) supplemented with ampicillin and chloramphenicol were used to inoculate 2 l of TYP main culture. The main culture was incubated at 37°C and 160 rpm until an optical density at 600 nm of 0.8 was reached. Temperature was reduced to 20°C and protein production was induced by addition of 1 mM isopropyl-β-Dthiogalactopyranoside (IPTG). After 20 h, cells were harvested by centrifugation at 4°C and 4500 × g for 15 min. Cells were flash-frozen in liquid nitrogen and stored at −80°C.
The tapasin-ERp57 heterodimer was overexpressed in Spodoptera frugiperda (Sf) 21 cells essentially as previously described 40,42 . In brief, Sf21 cells were grown in Sf900 II SFM medium (Thermo Fisher Scientific) at 28°C and transfected with modified EMBacY bacterial artificial chromosomes (BACs) using X-tremeGENE DNA transfection reagent (Roche). After incubation for 72 h at 28°C, recombinant baculovirus V 0 was harvested and utilized for production of amplified baculovirus V 1 . The tapasin-ERp57 heterodimer was expressed in 800 ml suspension culture at a cell density of 10 6 cells ml −1 by infection with 0.5-1.0% (v/v) baculovirus V 1 . Cells were harvested 72 h post cell proliferation arrest by centrifugation at 4°C and 1300 × g for 10 min. Cell pellets were flash-frozen in liquid nitrogen and stored at −80°C.

Refolding and assembly of H2-D b /β 2 m complexes
Inclusion bodies of β 2 m and H2-D b were isolated and refolded by dialysis according to established protocols 41 . First, cell pellets were resuspended in TE buffer (20 mM Tris-HCl, pH 8.0, 100 mM NaCl, 10 mM DTT) containing 1% (v/v) Triton X-100, followed by centrifugation at 10,000 × g for 20 min at 4°C. Inclusion bodies were washed five times followed by two washing steps with TE buffer without Triton X-100. Pure inclusion bodies were resuspended in TE buffer, flash-frozen in liquid nitrogen, and stored at −80°C in aliquots.
Purified inclusion bodies from a 2 l expression culture of β 2 m were dissolved in 50 ml denaturation buffer (8 M urea, 100 mM Tris-HCl, pH 8.0) and incubated at RT for 1 h. After centrifugation at 16,000 × g for 5 min at 4°C, supernatant was dialyzed twice against 2 l of 10 mM Tris-HCl, pH 8.0, for at least 9 h. Refolded β 2 m was polished by size exclusion chromatography (SEC) using a Superdex75 16/60 column (GE Healthcare) equilibrated with 50 mM HEPES-NaOH, pH 7.4, 150 mM NaCl. Elution fractions corresponding to β 2 m were concentrated by ultrafiltration (Amicon Ultra 3 kDa, MWCO, Merck), flashfrozen in liquid nitrogen, and stored at −80°C.
Inclusion bodies of H2-D b were isolated and refolded by rapid dialysis according to established protocols 7

Data collection, structure determination, and refinement
The synchrotron data set was collected at beamline P13 operated by EMBL Hamburg at the PETRA III storage ring (DESY, Hamburg, Germany) 43 . Data were indexed, integrated, and scaled using the XDS package 44 . The structure of the editor complex was solved by molecular replacement using the coordinates of the tapasin-ERp57/MHC I editing module of the peptide-loading complex (PDB ID 6ENY) 8 as search model with the program Phaser within the Phenix software package 45 . The atomic model of the tapasin-ERp57/H2-D b /β 2 m complex was manually built in COOT 46 and refined in Phenix applying translation-liberation-screw (TLS) parameters. The atomic model was further improved using the PDB-REDO server 47 followed by manual adjustments in COOT 46 and refinement in Phenix 45 . The residual electron density in the MHC I binding pocket did not allow modeling of the photo-P18-I10 peptide used for refolding of MHC I or photo-induced fragments thereof with high occupancy. This result is consistent with SEC-MS analyses of H2-D b /β 2 m/photo-P18-I10 complexes after phototriggered peptide cleavage that showed residual amounts of uncleaved peptide (4%) or of the 6-mer peptide fragment (2%) still bound to MHC I ( Supplementary Fig. 2). Numberings of modeled amino acids correspond to the mature polypeptides lacking the respective signal sequences. Molecular graphics images were prepared using PyMOL (Schrödinger).

Transfection of HAP1 cells
One day prior to transfection, 40 × 10 3 cells per well of a 6-well plate were seeded. On the day of transfection, cells were washed using 1× Dulbecco's phosphate buffered saline, pH 7.4 (DPBS, Gibco), and 1 ml of fresh IMDM medium per well was added. For the transfection mixture, 2 µg DNA and 6 µl X-tremeGENE HP (1:3 ratio) (Roche) were separately dissolved in 50 µl Opti-MEM (Gibco) and incubated for 5 min at RT. The solutions were mixed, incubated for 15 min at RT and added dropwise to the cells. 1 ml of medium was added after 4-6 h, and cells were harvested 48 h after transfection.

MHC I surface expression
MHC I surface expression was analyzed by flow cytometry. All steps were carried out on ice. The cells were washed in 0.2 ml FACS buffer (1x DPBS, 2% BSA, 2 mM EDTA, 0.02% (w/v) sodium azide) and centrifuged for 5 min at 300 × g, 4°C. The supernatant was discarded, and cells were blocked by 10% FcR Blocking Reagent, human (Miltenyi Biotec) in 50 µl FACS buffer for 15 min. Afterwards, cells were washed with 0.2 ml FACS buffer, centrifuged for 5 min at 300 × g and 4°C, and stained with 2 µl of APC anti-human HLA-A/B/C antibody (64 μg/ml, clone W6/32, BioLegend) in 0.1 ml FACS buffer for 30 min. Subsequently, cells were washed twice with 0.2 ml FACS buffer and resuspended in 0.2 ml FACS buffer for analysis by flow cytometry. Data was recorded on a FACSMelody Cell Sorter (BD Bioscience), processed using FlowJo V10 software and analyzed using Excel (Version 16.7) and GraphPad Prism 8.21 MacOS. The gating strategy is displayed in Supplementary Fig. 8.

EndoH digest
For LC-MS analysis, tapasin-ERp57 was deglycosylated using EndoH. Glycoprotein was mixed with Glycobuffer 3 (New England Biolabs), and 625 U of EndoH (New England Biolabs) per 1 μg of protein were added. The mixture was incubated overnight at RT. Deglycosylated tapasin-ERp57 was purified by SEC.

LC-MS analysis
All LC-MS measurements were performed using a BioAccord System (Waters). Peptides were analyzed on an ACQUITY UPLC Peptide BEH C 18 Column, 130 Å, 1.7 µm, 2.1 mm × 100 mm (Waters), with a linear water/acetonitrile gradient complemented with 0.1% (v/v) formic acid at 60°C, 30 V cone voltage, 0.8 kV capillary voltage, and a desolvation temperature of 550°C in positive polarity at 50-2000 m/z. Intact protein LC-MS data was acquired using a cone voltage of 60 V, 1.5 kV capillary voltage, and a desolvation temperature of 500°C on an ACQUITY UPLC Protein BEH C 4 Column, 300 Å, 1.7 μm, 2.1 mm × 50 mm (Waters) at 80°C running a linear water/acetonitrile gradient, supplemented with 0.1% (v/v) formic acid. Mass spectra were recorded in positive polarity at 2 Hz in full scan mode at 400-7000 m/z. SEC-MS measurements were performed using an ACQUITY UPLC Protein BEH SEC Column, 200 Å, 1.7 μm, 2.1 mm × 150 mm (Waters) in 20 mM ammonium acetate. Mass spectra were recorded in positive mode at 1 Hz in full scan mode at 400-7000 m/z with a cone voltage of 42 V, capillary voltage of 1.5 kV, and a desolvation temperature of 450°C. Masses of peptides and proteins were calculated and confirmed in Unify. Intact mass spectra were deconvoluted in Unify using MaxEnt1 algorithm iterating to convergence. Spectra with high background noise were subjected to automatic baseline correction before deconvolution. Deconvoluted spectra were centroidized based on peak height and used for mass calculations. UV spectra were recorded at 280 nm with a sampling rate of 10 Hz.

Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The structure of the MHC I-tapasin-ERp57 editing complex was deposited to the Protein Data Bank (http://www.rcsb.org) under accession number PDB ID 7QNG. Previously published structural data used in this study are accessible at the PDB (human PLC editing module tapasin-ERp57/calreticulin/MHC I, PDB ID 6ENY; peptidereceptive MHC I, PDB ID 2F74; client-free tapasin-ERp57 heterodimer, PDB ID 3F8U; MHC I-TAPBPR complex, PDB ID 5OPI). Mass spectrometry data are available via Zenodo [https://doi.org/10.5281/ zenodo.5939241]. Source data are provided with this paper.