A random peptide repertoire was synthesized with a distribution of 19 amino acids other than cysteine at an equal molar ratio in every position, as previously reported (13, 29, 30). In brief, random peptide libraries of different lengths were synthesized separately according to this method and mixed in an equimolar ratio according to the average molecular mass before use.

Peptides with defined sequences used in this study, such as RVEDVTNTAEYW, were synthesized and purified to 99% by reverse-phase high performance liquid chromatography (HPLC) and mass spectrometry (SciLight Biotechnology).

All random peptide repertoires and peptides were stored in lyophilized aliquots at -80°C after synthesis and dissolved in dimethyl sulfoxide (DMSO) before use.

The DNA fragments encoding extracellular domains of SLA-1*13:01 (GenBank accession No. AB847437.1) and sβ2-M (GenBank accession No. BAG32341) were cloned into pET-21a (+) vectors (Novagen) and expressed as inclusion bodies in E. coli BL21 (DE3). The mutants at positions 99 of SLA-1*13:01 were cloned by overlap PCR and expressed as inclusion bodies in E. coli BL21 (DE3). The recombinant proteins were purified as described previously and dissolved in 6 M Gua-HCl (34).

The proteins were refolded by stepwise dilution using buffer containing 100 mM Tris-HCl (pH 8.0), 2 mM EDTA, 400 mM L-Arg HCl, 5 mM GSH, and 0.5 mM GSSH, and treated at 277 K for 12 hours. Before renaturation, a random peptide library containing 5.12 mg octapeptide, 5.76 mg nonapeptide, 6.4 mg decapeptide, 7.04 mg undecapeptide, 7.68 mg dodecapeptide, 8.32 mg thirteen peptide, 8.96 mg tetradecepeptide and 9.6 mg pentadecapeptide, or 10 mg of fixed peptide was dissolved and added to 1 L of refolding solution. Then, the solutions dissolved SLA-1 and sβ2-M inclusion bodies were individually added to refolding buffer at a 1:1 molar ratio. The refolded complexes were concentrated and purified with a Superdex 200 16/60 column in Tris buffer (20 mM Tris [pH 8] and 50 mM NaCl), followed by Resource Q anion-exchange chromatography (GE Healthcare) in Tris buffers (Buffer A: 20 mM Tris [pH 8] and 5 mM NaCl; Buffer B: 20 mM Tris [pH 8] and 500 mM NaCl).

The peptide-containing fraction in the complex was eluted according to a previous method (13, 19). In brief, the complex was treated with 10% acetic acid at 65°C for 15 minutes and the peptides were collected through a 3-kDa filter. Then, the peptide-containing fraction was desalted using a C18 tip and separated on the EasyNano LC 1000 system (San Jose, California, Thermo Fisher Scientific). Methods as follows: the peptide component was loaded into a trap column (5-μm pore size, 150-μm i.d. ×3-cm length, 100 Å) and separated by a custom C18 column (3-μm pore size, 75-μm i.d. 315-cm length, 100 Å), with a flow rate of 450 μL/minute. The 60-minute linear gradient was operated as follows: 3% B (0.1% formic acid in acetonitrile [v/v])/97% A (0.1% formic acid in H₂O [v/v]) to 6% B in 8 min, 6% B to 22% B in 37 min, 22% B to 35% B in 8 min, 35% B to 100% B in 2 min, and 100% B for 5 min. The MS data were acquired by a Q Exactive HF (Bremen, Thermo Fisher Scientific) in data-dependent acquisition mode. The top 20 precursors by intensity in the mass range m/z 300 to 1800 were sequentially fragmented with higher-energy collisional dissociation and a normalized collision energy of 27. The dynamic exclusion time was 20 s. The automatic gain control for MS1 and MS2 was set to 3e6 and 1e, and the resolution for MS1 and MS2 was set to 120 and 30K.

Based on MS/MS information, peptides were resolved from each spectrum (FDR=1%) by de novo sequencing in Peaks Studio software. The parameters were set as follows: the enzyme was set to nonspecific, the variable modifications were adjusted oxidation (M)/deamidated (N,Q), the peptide mass tolerance was approximately ± 10 ppm, and the fragment mass tolerance was set to 0.02 Da. The identified peptides adjusted by the detection threshold (score ≥ 50) are listed in Table S1 .

To determine the RXEDVTNTAEYW (X represents any of twenty amino acids) library, MaxQuant1.6 software was used to search peptides based on MS/MS information. Posttranslational modifications (PTMs) were set to false, and the MS/MS match tolerance was set to 20 ppm. No-specificity was selected in the digestion option. The protein FDR was set to 0.01.

The circular dichroism (CD) spectra of the peptide-SLA-1 (pSLA-1) complexes were measured on a CD instrument (Chirascan; Applied Photophysic, Ltd.) using a Jasco J-810 spectrometer equipped with a water-circulating cell holder. As the temperature rose from 20°C to 80°C at a rate of 1°C/minute. A 1-mm optical path length cell was used for monitoring at 218 nm. The instrument used a thermistor to detect the temperature of the solution. Finally, the ratio of unfolded protein to the mean residue ellipticity (θ) was calculated. The unfolded score is shown as (θ - θ_N)/(θ_U - θ_N ), where θ_N and θ_U are the mean residual ellipticity values in fully folded and fully unfolded states, respectively. The midpoint transition temperature (Tm) is calculated from the denaturation curve data in the Origin 9.1 program (OriginLab).

The refolded and purified protein complex was concentrated and performed using the sitting-drop and hanging-drop vapor diffusion method at 277 K and 291K. Complex crystals of SLA-1*13:01 with peptide RVEDVTNTAEYW (pSLA-1*13:01_RW12) were obtained under PEG/Ion solution No.3 (0.2 M ammonium fluoride, 20% w/v polyethylene glycol 3,350). Prior to X-ray diffraction, the crystals were immersed in a stock solution containing 17% glycerol as a cryoprotectant, and then flash-cooled in a 100 K gaseous nitrogen stream. Then, the diffraction data of pSLA-1*13:01_RW12 were collected and the R-AXIS IV⁺⁺ imaging plate detector was used to resolve the resolution of 2.5 Å on Beamline BL19U1 (wavelength, 0.97892 Å) of the Shanghai Synchrotron Radiation Facility (Shanghai, China). The HKL-3000 software package (HKL Research) was used to index, integrate, scale and merge data (35). The crystallographic statistics for the complexes are listed in Table 1 .

Using SLA-1*13:01 (Protein Data Bank code 6KWO) as a search model, the structure of the pSLA-1*13:01_RW12 complex was determined by molecular replacement using the Phaser program (36). The model was rebuilt using COOT (37) and refined by REFMACS5 (38). Refinement rounds were implemented using the refinement program in the PHENIX package with isotropic atomic displacement parameter (ADP) refinement and bulk solvent modeling (39). Finally, the stereochemical quality of the final model was assessed using the PROCHECK program (40). Structural illustrations and electron density-related figures were drawn using PyMOL software. Multiple-sequence alignment was performed with Clustal Omega (http://www.ebi.ac.uk/Tools/msa/clustalo/). The accessible surface area (ASA) and buried surface area (BSA) were calculated with PDB in Europe Proteins, Interfaces, Structures and Assemblies (PDBePISA, http://www.ebi.ac.uk/msd-srv/prot_int/pistart.html), and the B factor was calculated with CCP4.

SLA-1*13:01 differs from SLA-1*04:01 by only eight residues, and they are all distributed on the peptide binding platform composed of α1 and α2 domains ( Figures 1A, B ). Our previous research showed that SLA-1*04:01 binds much more nonapeptides than SLA-1*13:01 (19). However, the in vitro refolding results of SLA-1*04:01 and SLA-1*13:01 with a mixed random peptide library of 7-15 amino acids showed that the peak of pSLA-1*13:01 is higher than the peak of pSLA-1*04:01, indicating SLA-1*13:01 can bind more random peptides of mixed length than SLA-1*04:01 ( Figure 1C ). RPLD-MS showed that compared with SLA-1*04:01 (13), SLA-1*13:01 could bind fewer short peptides of 8-10 amino acids, but more long peptides of 11-15 amino acids ( Figure 1D and Table S1 ). It should be pointed out that among the peptides bound by SLA-1*13:01 and SLA-1*04:01, nonapeptide and decapeptide are the main peptides, which agrees with the consensus on the length distribution of MHC-I binding peptides. Compared with SLA-1*04:01, SLA-1*13:01 has a lower amount and proportion of binding nonapeptides ( Figures 1D, E ), which consists of the previous data (19).

Structural analysis of SLA-1*13:01 and SLA-1*04:01 complexed with long peptides is helpful to clarify their different preferences for peptide length. We have previously identified a dodecapeptide (RW12) that can bind to SLA-1*04:01 and solved the structure of their complex pSLA-1*04:01_RW12 (13). Here, an in vitro refolding experiment showed that RW12 can also combine with SLA-1*13:01 to form the stable complex pSLA-1*13:01_RW12 ( Figure 2A ). After purification and crystal screening of pSLA-1*13:01_RW12, we successfully obtained the crystal and finally solved its structure at 2.5 Å, the same resolution of pSLA-1*04:01_RW12 ( Table 1 ). Structural analysis confirms that bound peptide conformation is not affected by crystal-crystal packing. The structural comparison between pSLA-1*04:01_RW12 and pSLA-1*13:01_RW12 showed that their overall structural similarity was high (RMSD = 0.9561), but the conformations of the presented RW12 peptide were obviously different (RMSD = 2.089) ( Figure 2B ).

In the structures of pSLA-1*04:01_RW12 and pSLA-1*13:01_RW12, the electronic density maps of the RW12 peptide are clear and can be used for credible comparative analysis ( Figure 3A ). Our previous study showed that the binding mode of RW12 is different from the classical MHC-I peptide binding mode. The first residue Arg (P-1-R) at the N terminus of RW12 extends out of the peptide binding groove of SLA-1*04:01, the second residue Val (P1-V) is turned over and its side chain is accommodated by the A pocket (13). The same is true of the interaction between the N terminus of RW12 and SLA-1*13:01 ( Figure 3B ). Except for the N terminus, the conformations of the rest of RW12 are significantly different in SLA-1*13:01 and SLA-1*04:01. RW12 obviously protrudes from the peptide binding groove of SLA-1*13:01 but is deeply embedded in the groove of SLA-1*04:01 ( Figure 3B ), and the side chain orientation of each residue is also different ( Figure 3C ).

The hydrogen bonds between RW12 and SLA-1*13:01 are significantly less than that between RW12 and SLA-1*04:01, including the hydrogen bonds mediated by water molecules, especially in the middle region of the peptide binding groove ( Figure 3D and Table 2 ). Compared with pSLA-1*04:01_RW12, the residues from P6 to P10 of RW12 in pSLA-1*13:01_RW12 have higher B factors, which indicates that these residues protruding from the peptide binding groove are less constrained ( Figure 3A ). The results of circular dichroism showed that the thermal stability of pSLA-1*04:01_RW12 is slightly higher than that of pSLA-1*13:01_RW12, consisting with the numbers of hydrogen bonds with RW12, although both of them can keep stable and crystallize ( Figure 3E ).

The conformational difference between RW12 presented by SLA-1*13:01 and SLA-1*04:01 is caused by the different amino acids between these two SLA-I alleles. Structural analysis showed that only five of the eight differential amino acids directly interact with RW12, and the variable residues at positions 57, 151 and 152 do not contact RW12. 62E, 66K and 69D in SLA-1*13:01 can form hydrogen bonds with P-1-R, P2-E and P5-T of RW12 ( Figure 4A ), and 70T and 99Y in SLA-1*04:01 can form hydrogen bonds with P3-D and P5-T of RW12 ( Figure 4B ).

62E/R and 66K/N at the N terminus of the peptide binding groove did significantly alter the conformation of the P-1 to P2 residues of RW12 ( Figures 4C, D ). The conformational difference of RW12 starts from P3-D, the P3-D position in pSLA-1*13:01_RW12 is relatively higher ( Figure 4C ), the positions of the subsequent residues are more elevated, and their side chain orientations are more variable ( Figure 3C ). These results indicated that the difference between SLA-1*04:01 and SLA-1*13:01 in presenting RW12 peptide may start from the interaction with P3-D and be consolidated later, while 69D/E, 70N/T and 99F/Y are just located in this region ( Figures 4C, D ).

Our previous study showed that 99Y/F is the key factor that leads to a significant difference between SLA-1*04:01 and SLA-1*13:01 when binding nonapeptides (19). 99Y in SLA-1*04:01 can form a hydrogen bond with the main chain of the P3 residue, but in SLA-1*13:01, 99F cannot form a similar hydrogen bond and leads to a decrease in its ability to bind nonapeptides. If the 99Y/F of SLA-1*04:01 and SLA-1*13:01 are replaced by mutations, the ability to bind nonapeptides between them will also be exchanged. Similar to the binding of nonapeptides, the main chain of P3-D of RW12 forms a hydrogen bond with 99Y of SLA-1*04:01, but not with 99F of SLA-1*13:01. The conformational difference of RW12 also starts from P3-D, 99Y of SLA-1*04:01 makes P3-D of RW12 tied to the bottom of the groove but 99F of SLA-1*13:01 cannot ( Figure 4 ). To further analyze the role of 99Y/F, we compared the other solved structures of SLA-1*04:01, SLA-1*13:01 and its mutant SLA-1*13:01(F99Y). 99Y can form a conserved hydrogen bond with the P3 main chain in all structures ( Figure 5A ), so we believe that 99Y/F has an important influence on peptide binding, regardless of peptide length.

Although the other variant residues do not interact with peptides in the stable way ( Figures 5A, B ), but they also have their own unique role in influencing peptide conformation. 70T of SLA-1*04:01forms a hydrogen bond with the main chain of P5-T, further limiting the subsequent residues of RW12 in the peptide binding groove ( Figures 4B–D ). 69D of SLA-1*13:01 forms a hydrogen bond with P5-T, which fixes the middle of RW12 outside the peptide binding groove, jointly determine the conformation of RW12 protruding out of the SLA-I*13:01 groove ( Figures 4A, C, D ). Overall, RW12 presents a typical ‘bulge’ conformation in pSLA-1*13:01_RW12, while it is embedded in the groove in pSLA-1*04:01_RW12. The maximum position deviation between them can reach 6.0 Å ( Figure 4C ).

By comparing the structures of other resolved MHC-I and undecapeptide complexes, we also found that the 99^th residue cannot independently determine the conformation of the peptide. For example, 99S of the chicken MHC-I molecule (BF2*2101) does not form hydrogen bonds with the P3 residue, similar to SLA-1*13:01 and RW12, but its 69S, 73N and 76N residues are bound to peptide, which limits the whole peptide in the groove, similar to SLA-1*04:01 and RW12. 99Y of the monkey MHC-I molecule (Mamu*B17) forms a hydrogen bond with the P3 residue, similar to SLA-1*04:01 and RW12, but there is no strong restriction on the peptide, similar to SLA-1*13:01 and RW12 ( Figures 5C, D ). Therefore, for SLA-1*04:01 and SLA-1*13:01, the difference of 99Y/F should be critical factor for the binding of peptide, but it needs to cooperate with other different amino acids to jointly determine the conformation of peptide.

The N terminus of RW12 can extend out of the peptide binding groove of SLA-1*04:01 and SLA-1*13:01, while 62E and 66K in SLA-1*13:01 can form hydrogen bonds with P-1-R and P2-E ( Figure 4A ). This may result in SLA-1*13:01 having a stronger ability to present N-terminal extended peptides than SLA-1*04:01, thus affecting their preference for long peptides. Therefore, it is necessary to compare the capabilities of SLA-1*04:01 and SLA-1*13:01 in presenting N-terminal extended peptides.

At present, there are only four pMHC-I complexes with N-terminal extending peptides, pSLA-1*04:01_RW12, pSLA-1*13:01_RW12, pHLA-B*57:01_TSTLQEQIGW (pHLA-B*57:01_TW10) and pHLA-B*58:01_TW10 (9, 12, 13). SLA-I and HLA-I have similar patterns of binding N-terminal extension peptide, but they also clearly show distinguishable species characteristics ( Figure 6A ). The change in the binding mode of the P1 residue of the peptide is the key to the extension of the peptide N terminus. Compared with other peptides in classical presentation mode, the P1 residues of RW12 and TW10 are reversed, and their side chains are inserted into the A pocket, which leads to the elevation of the Cα position of the P1 residue. Then, the main chain N atom and O atom of the P1 residue form hydrogen bonds with 63E on the α1 helix and 159Y on the α2 helix, respectively. The 63^rd residue is polymorphic, and our previous mutation studies show that 63E is necessary for this binding pattern (13). The angle between P-1 and P1 residues in RW12 is obviously larger than that in TW10 ( Figure 6B ) because the species characteristic of 167S in SLA-I alleles makes the N terminus of the peptide binding groove of SLA-I open and is beneficial to the extension of the P-1 residue (13).

Although 62E and 66K in SLA-1*13:01 can form hydrogen bonds with P-1-R and P2-E, there is no strong interaction with the P1 residue ( Figure 4A ), nor does it change the RW12 N-terminal extension ( Figure 6 ). By adding excess residues at the N terminus, we previously proved that SLA-I * 04:01 can allow the extension of up to three residues (13). SLA-1*13:01 could only tolerate the extension of 1-3 residues from the N terminus of RW12 ( Figure 6C ). Therefore, we believe that SLA-1*13:01 does not have a stronger binding ability for the N-terminal extended peptide than SLA-I*04:01, and the N-terminal extended peptide presentation mode has no obvious influence on the difference in peptide length preferences between SLA-1*13:01 and SLA-1*04:01.