The structural basis of chicken, swine and bovine CD8αα dimers provides insight into the co-evolution with MHC I in endotherm species

It is unclear how the pivotal molecules of the adaptive immune system (AIS) maintain their inherent characteristics and relationships with their co-receptors over the course of co-evolution. CD8α, a fundamental but simple AIS component with only one immunoglobulin variable (IgV) domain, is a good example with which to explore this question because it can fold correctly to form homodimers (CD8αα) and interact with peptide-MHC I (p/MHC I) with low sequence identities between different species. Hereby, we resolved the crystal structures of chicken, swine and bovine CD8αα. They are typical homodimers consisting of two symmetric IgV domains with distinct species specificities. The CD8αα structures indicated that a few highly conserved residues are important in CD8 dimerization and in interacting with p/MHC I. The dimerization of CD8αα mainly depends on the pivotal residues on the dimer interface; in particular, four aromatic residues provide many intermolecular forces and contact areas. Three residues on the surface of CD8α connecting cavities that formed most of the hydrogen bonds with p/MHC I were also completely conserved. Our data propose that a few key conserved residues are able to ensure the CD8α own structural characteristics despite the great sequence variation that occurs during evolution in endotherms.


Results
The canonical CD8αα homodimers with distinct species-specific features. Three extracellular Ig-like domains of chicken, swine and bovine CD8α (cCD8α , sCD8α and bCD8α ) were crystallized and diffracted to 2.0 Å, 1.8 Å and 1.8 Å, respectively. All of them are CD8α α homodimers, and each CD8α exhibited a typical IgV architecture consisting of two anti-parallel sheets ( Fig. 1A-C). The sheets of CD8α are composed of 10 β strands, but there are 11 β strands in sCD8α and bCD8α because their A and G strands are divided into two separate parts. The inner sheets of these CD8α α homodimers contain C, C' , C", G and F strands, and the outer sheets consist of A, B, D and E strands. The detailed AA compositions of each strand in these different CD8α molecules are shown in Fig. 1D. Although the AA sequence identities of the resolved CD8α IgV domains are quite low (Fig. 1E), especially between the mammals and the chicken (non-mammal) (< 30%), there are 17 conserved residues in all the CD8α molecules (blue). Among these residues, two C (in the B and F strands, respectively) form the critical disulphide bond of the Ig superfamily domains, and highly conserved residues -G (in the AB or Scientific RepoRts | 6:24788 | DOI: 10.1038/srep24788 A'B loop), L (in the B strand), Y (in the C strand), L (in the E strand), G and Y (in the F strand) -that compose the common core of the IgV domains can also be found in these CD8α α structures (Fig. 1D) 44,45 .
By comparing all the resolved CD8α α structures, the root-mean-square deviations (RMSDs) of the mammal CD8α α molecules are below 1.9 Å, which is lower than the RMSDs of the mammal CD8α α s and cCD8α α (Fig. 1E). Among these three structures, cCD8α α is a special one that exhibits unique characteristics. For example, cCD8α α has the longest C and C' strands and a unique alpha helix between C" and D strands (Fig. 1A,D). These indicate there is an obvious gap between non-mammal and mammal CD8α molecules. The two artiodactyl CD8α α dimers also have some distinct structural characteristics, and the most notable feature is that they have an additional A' strand ( Fig. 1B-D). The two artiodactyl sCD8α α can be discriminated from each other by certain characteristics, such as the longer A' strand and shorter complementarity determining region 2 (CDR2) loop in sCD8α α .
Conserved interfacial aromatic residues are critical to CD8αα dimerization. The chicken cCD8α α structure further confirmed that the dimerization of CD8α is beyond that of mammals and was retained quite well during the evolution of endotherms. The buried surface areas (BSAs) of cCD8α α , sCD8α α , bCD8α α , mouse CD8α α (mCD8α α ), rhesus macaque CD8α α (rCD8α α ) and human CD8α α (hCD8α α ) are 828.3 Å 2 , 979.6 Å 2 , 928.7 Å 2 , 1048.7 Å 2 , 1105.5 Å 2 and 1026.3 Å 2 , respectively, indicating that they are tightly binding dimers. However, there are only a few hydrogen bonds formed by non-conserved residues between two monomers of these different species (Fig. 2). There are only 2~6 hydrogen bonds between the two monomers of these CD8α α dimers. The BSAs of sCD8α α and bCD8α α are similar, but there are six hydrogen bonds in sCD8α α and only two hydrogen bonds in bCD8α α , indicating that CD8α dimerization does not mainly rely on hydrogen bonds. The Q residue on the C strand is involved in the formation of hydrogen bonds in five CD8α α dimers, but it was substituted by H in bCD8α , and no hydrogen bond is formed at this location in bCD8α α (Figs 1D and 2).
The residues on the dimer interface in the known CD8α α structures are shown in Fig. 3, and their contributions to dimerization are listed in Table S1. The numbers of residues in the interface of CD8α α dimers are approximately the same (26~27 residues). Among them, there are eight conserved residues, and their locations are invariable in the different CD8α α structures. Although the total van der Waals force (VDW) and BSA are variable among the different evolutionary CD8α , contributions by conserved residues are approximately the same. These interactions should be the basic insurance of CD8α α homodimerization. Among the conserved residues, the contributions of four aromatic AAs (F or Y) definitely account for absolute proportions. Three F in C' , as well as F and G strands and one Y in the C' strand provide more than 100 VDW and 200 Å 2 BSA. Especially for chicken, the four conserved aromatic residues contribute 161 VDW and 262.21 Å 2 BSA, which account for 42% and 34% of the total VDW and BSA, respectively (Table S1). These four aromatic residues can interact with each other; for example, F48 contacts F104 in sCD8α α as well as in other CD8α α . Their inner interactions further The cCD8α α homodimer is coloured yellow-orange, and its unique helix is coloured light magenta; its extremely short CDR2-like loop, which only consists of 2 residues, is coloured red. (B) The sCD8α α homodimer is coloured green, and the composed residues of its additional A' strand are displayed. (C) The bCD8α α homodimer is coloured cyan, and its A' strand was also determined. (D) The AA alignment of CD8α molecules based on their crystal structures is shown. Each strand consisting of residues is labelled by a coloured box: yellow-orange for chicken, grey for mouse (PDB ID: 3DMM), green for swine, cyan for bovine, orange for monkey (PDB ID: 2Q3A) and salmon for human (PDB ID: 1CD8). The hollow arrow on the regions of boxes represents the corresponding strand, and the only helix in cCD8α α is also labelled by a light magenta box. The highly conserved residues are coloured blue. (E) The values of AA identities and RMSDs of the CD8α molecules whose structures have been resolved are shown. ensure that dimerization can occur in a conserved manner during CD8α evolution. Therefore, the conserved aromatic AAs in the interface must be critical to preserving CD8α α homodimer formation during evolution.
The binding of MHC I and CD8αα are anchored by three conserved residues in CD8α. It was generally believed that the CDR loops are the core functional domains in CD8α molecules because the human and mouse CD8α α -p/MHC I complex structures show that CD8α α uses three CDR loops to bind p/MHC I. The manner in which the two CD8α α -p/MHC I complexes bind is similar to that of antibody binding [25][26][27] . However, the structural alignment shows a great variation of the CDR1 and CDR2 loops in the known CD8α α structures (Fig. 4). The two loops showed diverse conformations and changed greatly in length, especially the CDR2 loops. The CDR2 loop of chicken cCD8α α only consists of two residues and has shifted dramatically compared with other CDR2 loops of mammalian CD8α α structures. In contrast, the CDR3 loops of the known structures showed great similarities in conformation and length, and there is only one conserved residue in the CDR3 loops (N in CDR3). This indicates that if the binding manner of CD8α α and p/MHC I is conserved during evolution, only the CDR3 loop, not CDR1 and CDR2, should be the principal part relating to binding MHC I molecules.
The cavity formed by CD8α α CDR loops accommodates an MHC I α 3 domain CD loop, and it has been shown that they are the most important interacting parts 25,27 . The cavities of the six known CD8α α structures are shown in Fig. 5. The cavities are formed by identical residues separated symmetrically in the two monomers of each homodimer. The numbers of residues composing the cCD8α α , sCD8α α and bCD8α α cavities are 7, 7 and 9, and their volumes are 202.0 Å 3 , 322.9 Å 3 and 384.6 Å 3 , respectively. Four conserved residues (S, F, Y from one monomer, and N from the other monomer) are involved in the cavities of CD8α α dimers; three of them (S, Y and N) are on the surface, and the other one (F) forms one side wall of the cavity. These conserved residues have been shown to interact MHC I by hydrogen bonds and are crucial for the binding.
CD8 and MHC I are considered to be a pair of evolutionary molecules that maintain their relationship during co-evolution, and their co-evolutionary manner in endotherms was assessed using both the sequences and structures of six species ( Fig. 6 and Figs S1 and S2). Regarding their sequences, the co-evolutionary relationships might not be obvious (Fig. S1). In human and mouse crystal structures of CD8α α -p/MHC I, three conserved residues of CD8α form hydrogen binds to MHC I α 3 domain CD loop and D strand (Fig. 6A,B). Residues S and Y connect the side chains of D and Q in α 3 CD loop with hydrogen bonds which are vital for the p/MHC I-CD8α α binding proved by mutation analysis. Residue N form hydrogen bonds with the main chain of L in D strand. The residues D and Q in α 3 CD loop are highly conserved in MHC I molecular evolution. Residue L in α 3 D strand is conserved in mammal MHC I molecules, although it changed into S in chicken MHC I, it can also form the hydrogen bonds with residues N in CD8α through its main chain atoms. By combination with chicken, swine, bovine and monkey MHC I and CD8α α , according to human and mouse CD8α α -p/MHC I binding way, we found their main chains are matched well and side chains are without steric hindrance (Fig. 6C). In addition, the structures of CD8α β heterodimers of the rest five species were modelled based on the structure of mouse CD8α β (Fig. S2). The interacting residues of CD8 and MHC I were further investigated, according to human and mouse CD8α α -p/MHC I and CD8α β -p/MHC I crystal structures, and the interacting residues of MHC I were found to be more conserved than those of their partners, CD8α α and CD8α β . Interestingly, only at the key interacting sites between CD8α and MHC I were the residues on their interface highly conserved (Fig. S2). In both CD8α α -p/MHC I and CD8α β -p/MHC I complexes, these conserved residues on the CD8α surface form hydrogen bonds with the CD loop and D strand of the MHC I α 3 domain. These results suggested that the binding manner of p/MHC I and CD8 was conserved and anchored by the residues conserved in them.

Conserved residues in the outer and inner surfaces of CD8α molecules. The conserved residues
and their distribution in CD8α structures were coloured differently (using cCD8α as a model, Fig. 7A). Among all 17 conserved residues, only 4 residues are in loops, and the rest are in strands. From another perspective, the inner surface is more conserved than the outer surface of CD8α . There are only 5 conserved residues on the outer surface, and most of them (L and C in the B strand, L in the E strand and G in the AB loop) are common core components of IgV domains that are not special to CD8α 44,45 . Even in mammal CD8α molecules, the conservation of the outer surface was still less than that of the inner surface. However, there are 12 conserved residues in the inner surface, and 9 of them are special to CD8α . The conserved residues in the CDR3 loop are also located in the inner surface. The data suggested that conserved residues in the outer and inner surfaces are essential for the formation of CD8α α dimers and play critical roles in the core functions of CD8α conserved during evolution.
Because of the low AA identities of CD8α molecules from different species, there are only 17 conserved residues, and 7 of them are commonly conserved in IgV domains; only approximately 10 conserved residues were unique to CD8α . The results showed that the few uniquely conserved CD8α residues were critical to allow CD8α to form homodimers and interact with MHC I. Five conserved aromatic residues on the interface provide considerable intermolecular forces to ensure dimerization, and four conserved residues in the binding cavity form vital hydrogen bonds with p/MHC I. Residue Y in the C' strand can play two different roles simultaneously, indicating it is essential for CD8α . The grouped of conserved residues is shown in Fig. 7B. The superposition of the conserved residues in all resolved CD8α α structures indicated that they are important for the maintenance of the proper form of CD8α α during evolution.

Discussion
In this study, we first resolved the crystal structures of chicken, swine and bovine CD8α α and analysed all the known CD8α α structures to determine how CD8α could form a homodimer and bind p/MHC I.
The AA identities of CD8α molecules from different endotherm species were quite low (Fig. 1E); the identities between chicken and other mammalian CD8α sequences were even below 30%. So, all the resolved CD8α α structures showed their own significant specific characteristics. There was a special helix and a short CDR2 loop found in cCD8α α , and the inter-chain hydrogen bonds and the dimer interface area of cCD8α α were minimal. sCD8α α and bCD8α α showed an additional A' strand ( Fig. 1B-D), but the A' strand in sCD8α α was longer. Even so, the CD8α α structures demonstrated that they are all homodimers formed by V-type immunoglobulins and have similar overall architectures. Additionally, we found that seven common conserved residues resulted in CD8α folding in a V-type conformation, and eight conserved residues of CD8 located on the dimer interface offered a large amount of VDW for the formation of homodimers. VDW and BSA provided by conserved residues in cCD8α α account were greater than 42% and 33% (Table S1), respectively, which were higher than those of sCD8α α and bCD8α α ; these results indicated these conserved residues may be the initial key elements of CD8 dimerization. Four conserved aromatic residues offered the most VDW among all the conserved residues, indicating they play the most important role in homodimerization. In the structure of mouse CD8α β (the only currently known heterodimer, PDB ID: 2ATP), there are 7 conserved residues on the interface of CD8α and six conserved residues of CD8β among these six species (Fig. S3). The four conserved aromatic residues we found in CD8α were also involved in the heterodimerization, and their BSA was 190 Å 2 , approximately 20% of the total amount of interface area. In CD8β , the total BSA of the conserved residues was 228 Å 2 , accounting for 25% of the total interacting area. These data suggested the conserved residues of CD8 (both CD8α and CD8β ) are critical in dimerization 5 . Interestingly, there were also four conserved aromatic residues in CD8β , which provided the most BSA (221 Å 2 ), and a sequence alignment showed three of them are identical to the key aromatic residues we found in CD8α . The only variation was in the chicken CD8β sequence, but the substituted residues had similar properties (Fig. S3).
The elucidated crystal structures of the human and mouse CD8α α -p/MHC I complexes suggested that the manners of CD8α α and MHC I interaction are very similar 25,27 . However, it was unclear that CD8α α from other species could bind p/MHC I in the same way. Based on the (modelled) structures from six species, we proposed that the interaction manner of CD8α α and p/MHC I is consistent and preserved by three completely conserved residues of CD8 during the evolution of endotherms. The compositions of CD8α α dimer binding cavities and three residues that play key roles in interacting with MHC I were highly conserved. These three residues are located on the surface of the cavities and can bind with MHC I strongly by hydrogen bonds and VDW. The residues in different MHC I molecules that are connected by the three residues were also highly conserved. Structural alignment showed that both of them are well superposed in the manner of CD8α α -p/MHC I interaction (Fig. 6). The three conserved residues acted as a three-point register, fixing the interaction of CD8α α and p/MHC I in a consistent manner during evolution. In the crystal structures of the CD8α α -TL (PDB ID: 1NEZ) and CD8α β -p/ MHC I complexes, these three residues could also form the same hydrogen bonds as in the CD8α α -MHC I complex 16 . Moreover, in CD8α β -p/MHC I (PDB ID: 3DMM), CD8β occupies a T-cell membrane proximal position and mainly interacts with the CD loop of the MHC I α 3 domain. However, the conservation of residues in this region was not as high as in CD8α (Fig. S2). The structures showed that CD8α is at the same position and binds MHC I in the same way in both the CD8α α -p/MHC I and CD8α β -p/MHC I complexes. These results strongly indicated that both CD8 isoforms maintain the manner of binding MHC I by relying on the conserved residues of CD8α .
CD8α and p/MHC I are considered as a set of co-evolution molecules because their binding is critical to CTL immunity in vertebrate species. The co-evolution relationship is not obvious by their AA sequences alone, and the interacting residues of MHC I were found to be more conserved than those of their CD8 partners (Fig. S2). However, the key residues for complex binding are highly conserved in both molecules according to the elucidated and modelled MHC I-CD8α α structures (Fig. 6). In consideration of the weak binding affinity between CD8α α and MHC I 6 , we posit that a few conserved residues playing critical roles in this interaction are enough to ensure the binding continues during evolution.
In this study, crystal structures of CD8α α and p/MHC I from six different species were analysed. The structures and sequences were significantly different between these six species, especially between chickens and mammals; however, they indicate that a few key conserved residues could ensure the structural basis of CD8α α dimerization and binding with p/MHC I via great sequence variations during endotherm evolution.

Materials and Methods
Preparation of proteins. The genes encoding cCD8α , sCD8α and bCD8α mature peptides (extracellular IgV domains) were chemically synthesized and ligated into a pET21a vector (Novagen) by the Shanghai Generay Biotechnology Company according to the sequences in GeneBank (NM_205235, NM_001001907 and NM_174015). The plasmids were transformed into the Escherichia coli strain BL21 (DE3), and 0.5 mM IPTG was used to induce the expression of these three inclusion bodies 46 . The bacteria were harvested by centrifugation at 6 000 g for 10 min and were then resuspended in cold phosphate-buffered saline (PBS). After sonication, the samples were centrifuged at 16 000 g, and the pellets were washed three times with a solution consisting of 20 mM Tris-HCl pH 8.0, 100 mM NaCl, 1 mM EDTA, 1 mM DTT and 0.5% Triton X-100. Finally, the inclusion bodies were dissolved in guanidinium chloride (Gua-HCl) buffer [6 M Gua-HCl, 50 mM Tris-HCl pH 8.0, 10 mM EDTA, 100 mM NaCl, 10% (v/v) glycerine, 10 mM DTT] to a concentration of 30 mg ml −1 .
Refolding and purification. The dissolved cCD8α , sCD8α and bCD8α inclusion bodies were gradually added into refolding buffer (100 mM Tris-HCl, 2 mM EDTA, 400 mM L-arginine-HCl, 0.5 mM oxidised glutathione, 5 mM reduced glutathione, pH 7.4) until a concentration of 60 mg ml −1 was reached. After incubation for 24 h at 277 K, the soluble portions were concentrated and purified by chromatography on a Superdex 75 10/300 column (GE Healthcare). The eluted peaks were collected by 0.5 ml per tube and tested by SDS-PAGE. Then, the refolded cCD8α , sCD8α and bCD8α were pooled together.
Crystallisation. Crystals of sCD8α were obtained as described previously 42  Data collection and processing. Diffraction data of three different CD8α crystals were collected using the NE3A beamline at the KEK synchrotron facility (Tsukuba, Japan) and an ADSC Q270 imaging-plate detector at a wavelength of 1.0 Angstrom, the BSRF 3W1A beamline and an MAR scanner 345-mm plate at a wavelength of 1.0 Angstrom, and a Rigaku MicroMax-007 HF and Rigaku Raxis IV+ + at a wavelength of 1.54178 Angstrom, respectively. In each case, the crystals were first soaked in reservoir solution containing 15% glycerol as a cryoprotectant for several seconds and then flash-cooled in a stream of gaseous nitrogen at 100 K 47 . The collected intensities were indexed, integrated, corrected for absorption, scaled and merged using HKL2000 48 . The classification of CD8α conserved residues based on their functions. All the conserved residues in the six CD8α α structures are shown in stick. The conserved residues which are critical to dimerization are coloured purple, and the residues taking part in the binding of MHC I according to elucidated human and mouse MHC I-CD8α α structures are coloured yellow. Only one conserved residue that plays a vital role in both dimerization and MHC I interaction is coloured red. The rest nine conserved CD8α residues not involved in these two aspects are coloured blue, and seven of them are common in other IgV molecules. The grouping indicates that only CD8α -specific conserved residues are critical to guarantee its continued function during evolution. Structure determination and refinement. The structures of cCD8α , sCD8α and bCD8α were resolved by molecular replacement using the MOLREP programme with human CD8α (PDB code: 1CD8) as the search model. Extensive model building was performed by hand using COOT 49 , and restrained refinement was performed using REFMAC5. Further rounds of refinement were performed using the phenix.refine programme implemented in the PHENIX package with isotropic ADP refinement and bulk solvent modelling 50 . The stereochemical quality of the final model was assessed with the PROCHECK programme 51 . Data collection and refinement statistics are listed in Table 1.