Antigen presentation subverted: Structure of the human cytomegalovirus protein US2 bound to the class I molecule HLA-A2

Abstract

Many persistent viruses have evolved the ability to subvert MHC class I antigen presentation. Indeed, human cytomegalovirus (HCMV) encodes at least four proteins that down-regulate cell-surface expression of class I. The HCMV unique short (US)2 glycoprotein binds newly synthesized class I molecules within the endoplasmic reticulum (ER) and subsequently targets them for proteasomal degradation. We report the crystal structure of US2 bound to the HLA-A2/Tax peptide complex. US2 associates with HLA-A2 at the junction of the peptide-binding region and the α3 domain, a novel binding surface on class I that allows US2 to bind independently of peptide sequence. Mutation of class I heavy chains confirms the importance of this binding site in vivo. Available data on class I-ER chaperone interactions indicate that chaperones would not impede US2 binding. Unexpectedly, the US2 ER-luminal domain forms an Ig-like fold. A US2 structure-based sequence alignment reveals that seven HCMV proteins, at least three of which function in immune evasion, share the same fold as US2. The structure allows design of further experiments to determine how US2 targets class I molecules for degradation.

Presentation of virus-derived antigenic peptides by major histocompatibility (MHC) class I molecules directs CD8⁺ T cell lysis of infected cells, which plays a key role in clearance of many viruses (1). The human herpesviruses nonetheless have evolved the ability to establish lifelong infection. More than 50% of the world's population carry the β-herpesvirus human cytomegalovirus (HCMV), which has become an important opportunistic pathogen (2). Persistent infection is remarkable, as HCMV possesses one of the largest virus genomes. The more than 200 HCMV gene products each serve as potential sources of peptides for MHC class I antigen presentation. Further, HCMV reactivates from latent infection of myeloid lineage cells years after acute infection, even in a fully primed immunocompetent host (3). To enable persistent infection, the CMVs have evolved numerous mechanisms to subvert the immune response, including disruption of the MHC class I antigen presentation pathway (4).

A cluster of immunomodulatory genes are located in the unique short (US) region of the HCMV genome, comprising the US2, US6, and US11 families (5). These encode a series of eight glycoproteins, at least four of which interfere with cell surface expression of MHC class I molecules. Each possess an endoplasmic reticulum (ER)-luminal domain, a single predicted transmembrane domain, and a short cytoplasmic tail. US3 retains class I molecules in the ER by blocking transport of folded class I complexes to the Golgi during the immediate early period of virus infection (6–8). US2 and US11 independently target ER-to-cytosol dislocation of class I during early periods of virus infection (9, 10). US6 blocks the TAP peptide transporter from the ER-luminal side during late periods of virus infection (11–13). The concerted action of this series of HCMV immuno-evasive proteins is believed to diminish CD8⁺ T cell recognition of HCMV peptide antigens. Indeed, it has been difficult to identify cytotoxic T lymphocytes specific for most HCMV protein epitopes (14).

Adenovirus, herpes simplex viruses, HIV and Kaposi's sarcoma-associated herpesvirus encode factors that specifically down-modulate human class I expression (4, 15). Although a growing number of such proteins have been described, little is known about the structural basis for interaction with their target proteins. Here we report the structure of the soluble complex between US2 and HLA-A2, the first structure of a viral protein that down-regulates class I. We find that US2 contains an Ig-like fold, and using structure-based sequence alignments, we predict that seven other HCMV US proteins are folded similarly. This structure also has implications for the mechanism by which US2 targets class I for proteasome degradation.

Materials and Methods

Protein Refolding and Purification.

US2_15–140 was refolded as described (16) in the presence of 7.5 mg of DE-52 purified HLA-A2/Tax, concentrated by pressurized stirred cell and Centricon 10 (Amicon) and purified by Superdex 75 gel filtration (Amersham Pharmacia) in 10 mM Tris (pH 8.0), 150 mM NaCl. Purified US2/HLA-A2/Tax was concentrated by Centricon 10.

Crystallization and Data Collection.

US2_15–140/HLA-A2/Tax crystals were grown by vapor diffusion at 18°C from hanging drops containing 1 μl of protein solution [10 mg/ml in 10 mM Tris (pH 8.0), 150 mM NaCl] and 1 μl of precipitant solution [12–14% polyethylene glycol (PEG) 6000, 7% 2-methyl-2,4-pentanediol (MPD), 100 mM Tris (pH 7.75)]. Pyramidal crystals (0.1 × 0.1 × 0.06 mm³; space group P1; a = 95.48 Å, b = 96.70 Å, c = 99.54 Å, α = 109.02°, β = 109.46°, γ = 107.93°; four US2_15–140/HLA-A2/Tax complexes per asymmetric unit) grew in 1–3 days. Protein samples recovered from crystals migrate identically to inclusion body material on reducing SDS/PAGE. Crystals were soaked briefly in a cryo-protectant solution [18% PEG 6000, 11% glycerol, 8% MPD, 100 mM Tris (pH 7.75)] and flash-cooled in liquid nitrogen. X-ray data were collected to 2.2 Å with a Quantum-4 charge-coupled device detector (ADSC) at BioCARS beamline D of the Advanced Photon Source, Argonne National Laboratory, Argonne, IL. Data were processed (Table 1) by using denzo and scalepack (HKL Research, Charlottesville, VA).

Table 1.

Data collection and refinement statistics

Data processing
Resolution (Å)	40.0–2.20 (2.28–2.20)
Number of reflections	139,486 (13,849)
Redundancy	2.1 (2.1)
I/σI	14.2 (3.3)
Completeness (%)	98.0 (97.1)
Rsym (%)*	4.5 (28.7)
Refinement
Data range (Å)	40.0–2.20 (2.34–2.20)
Reflections in working set	131,678 (20,366)
Reflections in Rfree set	2713 (419)
Rcrys (%)†	20.5 (30.0)
Rfree (%)‡	24.0 (34.8)
Number of protein atoms	15,716
Number of solvent atoms	1,446
rms deviation bond length (Å)	0.194
rms deviation bond angles (°)	1.9
Mean B value (Å2)	41.1

^*R_sym = Σ_h′〈|I_h − I_h′|〉/Σ_h′I_h′, where 〈|I_h − I_h′|〉 is the average of the absolute deviation of a reflection I_h′ from the average I_h of its symmetry and Friedel equivalents. 

^† R_crys = Σ∥F_o| − |F_c∥/Σ|F_o|, where F_c is the calculated structure factor. 

^‡ R_free is as for R_crys but calculated for 2% of randomly chosen reflections that were omitted from the refinement. 

Structure Determination and Refinement.

The structure was determined by molecular replacement in cns (17) using free HLA-A2/Tax as a search model (18). Cross-rotation and translation functions identified four different solutions, one for each complex in the asymmetric unit. Refinement was carried out by using cns version 0.9. Model building was performed in o (19) by using a solvent-flattened 4-fold averaged map. Strict noncrystallographic symmetry (NCS) constraints between the four US2_15–140/HLA-A2/Tax complexes were used for the first three rounds of refinement. For the final four rounds of refinement, NCS restraints of weight 300 were maintained. The refined atomic model consists of four copies each of HLA-A2 residues Gly-1–Glu-275, β₂-microglobulin residues Met-0–Met-99, Tax peptide residues 1–9, US2 residues Pro-43–Leu-137; and 1,446 water molecules. Buried molecular surface area and interatom contacts were calculated by using areaimol and contact, respectively (CCP4, Daresbury Laboratory, Warrington, U.K.). Figures were generated by using ribbons (20).

Mutagenesis of Class I Heavy Chains.

Amino-terminal hemagglutinin epitope-tagged class I heavy chains were constructed by PCR amplification of HLA-A2 residues 25–366 from cDNA, using a 5′ primer containing the hemagglutinin tag. The PCR products were digested and ligated into pCDNA 3.1+ expression vector (Invitrogen) containing the murine K^b signal sequence. Point mutations that convert arginines 131, 157, or 181 to glutamate were introduced by PCR and verified by DNA sequencing. All three arginine mutants maintained the ability to form properly folded class I complexes, as judged by their reactivity with conformation-sensitive mAb W6/32 (21) (data not shown).

Pulse–Chase Analysis.

U373-MG astrocytoma cells stably transfected with US2 were maintained and used for pulse–chase experiments as described (10). Procedures for Nonidet P-40 lysis and immunoprecipitation have been described (9, 10).

Results and Discussion

Structure Determination.

US2 is a 199-residue protein whose ER-luminal domain is predicted to comprise residues 1–164. A US2 spanning residues 15–140 was refolded from Escherichia coli inclusion bodies. US2_15–140 maintains the ability to associate with HLA-A locus products in vitro (16). We crystallized US2_15–140 refolded in the presence of recombinant of the HLA-A2/human T lymphotropic virus type I Tax peptide complex. The crystal structure of the US2/HLA-A2/Tax-peptide complex was determined to 2.2-Å resolution (see Materials and Methods). Table 1 shows the data collection and refinement statistics. The current model includes all residues of HLA-A2/Tax and residues 43–137 of US2 (Fig. 1A). Terminal US2 residues 15–42 and 138–140 are likely flexible or unstructured, because no electron density was observed for these residues, although US2_15–140 recovered from crystals remains intact (see Materials and Methods). However, replacement of US2 residues 1–40 with an exogenous ER targeting sequence does not alter US2 function in vivo (data not shown). The C termini of US2 and the HLA-A2 heavy chain are positioned on the same face of the US2/HLA-A2 complex (Fig. 1A), and the 25-residue segment (140) that links the US2 Ig-like domain to its predicted transmembrane segment could readily reach the ER membrane.

Figure 1.

Structure of the US2/HLA-A2/Tax complex and topology of the US2 Ig-like domain. (A) The US2/HLA-A2/Tax complex: US2 (magenta), HLA-A2 heavy chain (yellow), β₂-microglobulin (gray); HTLV-1 Tax peptide (LLFGYPVYV, green); disulfide bonds (cyan). (B) Ribbon diagram of US2 showing the two β-sheets (red, C′CFG; blue, ABED; yellow, disulfide bond) and the glycosylation site at asparagine 68 (circled). No glycan is observed in this bacterially expressed domain. (C) Amino acid sequence alignment of the US2 Ig domain with US2, US6, and US11 family members. β-strands (red or blue arrows; letter-labeled as in A and B) and US2 residues that contact HLA-A2 (*) are indicated. Conserved hydrophobic residues are blue, predicted N-linked glycosylation sites are orange, and cysteines predicted to form the A to G strand disulfide bond are magenta. US2 core residues (>90% buried, determined by using spdbv (48) are indicated by ●.

The US2 ER-Luminal Domain Contains an Ig-Like Fold.

Although unanticipated, the ER-luminal domain of US2 contains an Ig-like fold composed of seven β-strands (Fig. 1B). The US2 Ig-like domain belongs to the H subtype of the Ig-fold superfamily (IgSF), where the C′/D strand is split between the two β-sheets (22). The US2 Ig-like fold was not apparent from sequence comparisons, as this region of US2 displays at best 11% sequence identity with its nearest structural neighbors, AP-2, tenascin and fibronectin (23), each members of the IgSF (24). Furthermore, US2 possesses a disulfide bond whose position differs from that found in most Ig-like folds. Whereas Ig-like folds commonly possess a single disulfide bond buried in the core of the molecule connecting β-strands B and F, the US2 intercysteine distance of 82 residues is accommodated by a disulfide linkage between the A and G β-strands of the β-sandwich (Fig. 1B). Of the structurally characterized Ig-like folds, only a few have a similar disulfide bond, including CD2 domain 2 (25) and the first fibronectin III repeat of neuroglian (26).

The US2/HLA-A2 Interaction Surface.

US2 and HLA-A2 form a 1:1 complex in solution (16) and are present in a 1:1 ratio in the crystal. However, packing in the crystal lattice causes each US2 molecule to make significant contacts with two HLA-A2/Tax complexes at distinct locations (Fig. 2A), burying a molecular surface area of 1,320 Ų (binding site 1; Fig. 2B) or 697 Ų (binding site 2; Fig. 2C). The N-linked glycans normally attached to US2 (position 68) and HLA-A2 (position 86) in human cells would not interfere with association at either interaction surface. To identify the physiological binding site, we constructed epitope-tagged HLA-A2 heavy chains with point mutations that convert arginine to glutamate at residue 181 of binding site 1 (R181E), or at position 131 (R131E) or 157 (R157E) of binding site 2. None of these mutations interfere with HLA-A2 maturation in the absence of US2 (data not shown). Upon transfection into US2 expressing cells, the dislocation of these HLA-A2 mutants was examined biochemically. Whereas the R131E and R157E heavy chains are destabilized in the presence of US2, the R181E mutation nearly completely abolishes dislocation of class I heavy chains by US2 (Fig. 2D). Thus, in vivo, US2 binds class I molecules at the larger binding site 1, and this interaction is required for subsequent dislocation of class I heavy chains.

Figure 2.

The US2/HLA-A2/Tax complex and interaction surfaces. (A) Top view of HLA-A2 contacted by US2 at two sites in the crystals. Ribbon diagrams of HLA-A2 heavy chain (yellow), β₂-microglobulin (gray), US2 at binding site 1 (magenta), US2 at binding site 2 (cyan), and ball-and-stick model of the Tax peptide (green). Contact regions (boxes) are depicted in B and C in the same orientation. Although the two US2 molecules nearest a given class I molecule are shown, the stoichiometry in the crystal is 1:1. (B and C) Stereographic views of binding site 1 (B) and 2 (C). Side chains and/or backbone atoms are shown for intersubunit contacts of ≤ 3.5-Å distance. Hydrogen bonds (thin black lines); carbon (yellow); oxygen (red); nitrogen (blue). (D) In US2+ cells, HLA-A2 heavy chains are rapidly targeted for dislocation from the ER to the cytosol, where they are deglycosylated before proteasomal degradation. In the presence of a proteasome inhibitor, a deglycosylated breakdown intermediate accumulates (HC⩵CHO) (10). Pulse–chase analysis of US2+ cells treated with inhibitor carboxylbenzyl-leucyl-leucyl-leucyl vinylsulfone (ZL₃VS) (10) reveals continued dislocation of R131E and R157E heavy chains, evidenced by accumulation of the deglycosylated intermediate (*), similar to that of wild type HA-tagged heavy chains (data not shown). In contrast, the R181E mutation nearly completely abolishes dislocation of heavy chains. CHO designates the N-linked glycan. (E) Alignment of HLA-A2 with consensus heavy chain sequences of class I loci in the three regions contacted by US2. Boxes surround HLA-A2 residues that contact US2. Capital letters denote residues conserved in all class I consensus sequences. Consensus sequences were obtained from the IMGT/HLA Sequence Database (www.ebi.ac.uk/imgt/hla/align.html).

The face of one β-sheet of US2 (C′CFG) makes extensive contacts with HLA-A2 in binding site 1 (Figs. 2B and 3). The buried surface area of the US2/HLA-A2 complex (1,320 Ų) is similar to that of the natural killer cell inhibitory receptor (KIR)/HLA-Cw3 complex (1,386 Ų) (27). The conformation of HLA-A2 does not change significantly upon US2 association, aside from local perturbations at the interface. The Cα rms deviation for HLA-A2/Tax in the absence (18) and presence of US2 is ≈0.7 Å, similar to the value of 0.6 Å between the two free HLA-A2/Tax complexes in an asymmetric unit (18). Indeed, a T cell receptor can still interact with the US2/HLA-A2 complex (16).

Figure 3.

US2 binds remotely from peptide-loading proteins. The positions of mutations (blue) that alter interactions between class I molecules and the peptide-loading machinery are shown relative to the US2 binding site on HLA-A2 (magenta). The HLA-A2 N-linked glycan attachment site, asparagine 86, is also shown in blue. US2 residues that contact HLA-A2 are yellow.

US2 targets MHC class I locus products for degradation in vivo in a locus-specific manner. Whereas US2 destabilizes HLA-A and B locus products, alleles of the HLA-C, HLA-E, and HLA-G complexes do not coimmunoprecipitate with US2 molecules and escape degradation (28, 29). Such locus specificity may allow HCMV-infected cells to avoid lysis by natural killer cells, which eliminate cells that have reduced surface expression of HLA-C, -E, or -G (28). Whereas incubation of soluble US2 with several recombinant HLA-A locus products produces a gel shift in native electrophoresis, several HLA-B complexes, HLA-Cw4 and HLA-E, do not shift in the presence of US2 (16). Sequence alignments of consensus sequences for class I locus products reveal that US2 binds to a relatively conserved region of class I heavy chains (Fig. 2E). Polymorphisms in class I residues corresponding to HLA-A2 175–185 may contribute to the locus specificity of US2 association, because the consensus sequences of class I molecules diverge most significantly at this portion of the US2 binding site.

The suggestion that HLA-DR and HLA-DM are targeted for degradation by US2 (30) is not readily explained by the structural data shown here, if a homologous site of interaction were involved. Superposition of HLA-DR or HLA-DM complexes with HLA-A2 shows poor conservation of the US2 binding site of HLA-A2 in these class II complexes, both in sequence and structure. The DRβ and DMβ chains most closely resemble the HLA-A2 region bound by US2, although the DRα and DMα chains are the subunits destabilized by US2 (30). In addition, incubation of recombinant US2_15–140 with either soluble HLA-DR or HLA-DM does not produce a gel shift by native electrophoresis (16). Other portions of the US2 and/or DR/DM molecules not contained in the soluble constructs may account for degradation of HLA-DR or HLA-DM.

US2 Binding Site Is Remote from Known Class I/Peptide-Loading Complex Interactions.

US2 recognizes a class I surface that apparently arises late in the biosynthesis and assembly of class I molecules (16). Although US2 targets a population of newly synthesized class I molecules for degradation, recombinant US2 maintains the ability to associate with folded HLA-A2 complexed with a peptide. US2 would probably bind class I molecules in the absence of peptide as well, given the location of its binding site. HCMV US3 retains folded, peptide-loaded class I in the ER during the immediate early period of virus infection (6). Upon the early period of infection, the retained class I complexes are susceptible to attack by US2 and US11 (31). Thus, as revealed by the structure, US2 evolved a class I binding site that allows reactivity with this population of peptide-loaded, folded class I molecules.

Late in class I biogenesis, empty class I molecules associate with a multiprotein assembly, the peptide loading complex. This complex, which includes calreticulin, ERp57, tapasin, and TAP, facilitates delivery of peptide cargo to class I molecules (32). Although there are no structural data for intermediates of class I assembly, mutations that alter peptide loading of class I molecules have suggested probable sites of interaction with the peptide-loading complex (33–36). These mutations are opposite the US2 binding site and map to a face of HLA-A2 that includes the single N-linked glycan (residue 86) implicated in the binding of the lectins calnexin and calreticulin (mutations numbered in Fig. 3). Thus, US2 has evolved a class I binding site that still allows access to class I molecules associated with ER-resident chaperones, and so diminishes the chances that class I molecules escape degradation.

US2, US6, and US11 Family Members Contain an Ig-Like Fold.

The intricate relationship between HCMV and its host is exemplified by a group of eight genes in the HCMV US region, comprising the US2, US6, and US11 gene families (5). Although not essential for replication in vitro, four genes in this cluster alter cell surface expression of class I molecules (4). By aligning the sequences of HCMV US proteins based on the US2 structure (Fig. 1C), each of these viral proteins appears to contain an Ig-like fold similar to US2, suggesting that these three gene families arose by duplication from an ancestral Ig-like fold-encoding gene (Fig. 1C) (37). Alignment with US2 reveals conservation of the noncanonical disulfide bond and many core hydrophobic amino acids, particularly in the central β-strands B and F. Based on the US2 structure, all predicted glycosylation sites would be on outer surfaces. Because many Ig-like folds mediate adhesion to other protein surfaces, it is likely that the US3 and US11 Ig domains dictate association with class I. The sequence conservation is too low to predict which surface will be used for these interactions. US7–10 are likewise predicted to be type I membrane glycoproteins whose ER-luminal domains appear to contain Ig-like folds (Fig. 1C). However, the binding partners of US7–10 are presently unknown.

Human adenovirus encodes a glycoprotein, E3–19K, which retains newly synthesized class I molecules in the ER (38). Sequence similarity searches reveal significant conservation between E3–19K and Ig light chains. However, the putative E3–19K ER-luminal Ig-fold likely differs from that found in the US genes, because the sequence similarity is insignificant and the location of cysteines is not conserved between E3–19K and the US family proteins.

Comparison of the US2 Interaction to That of Other MHC Class I Molecule Ligands.

Crystal structures have now been obtained for MHC class I molecules associated with five of its membrane protein ligands (Fig. 4): the αβ T cell receptor (39), CD8αα from T cells (40), KIR natural killer cell receptor (27, 49), lectin-like natural killer receptors (41, 42), and the HCMV US2 immunoevasin. These are a subset of the proteins known to interact with class I (others include tapasin, calnexin, calreticulin, and additional viral ligands), and indicate the centrally important position class I occupies in determining the fate of a cell during the immune response to infection. The US2/HLA-A2 structure is the only example determined for a complex between class I and a protein that associates with it in cis (with their transmembrane domains embedded within the same membrane). In contrast, other class I ligands for which structural data have been obtained interact with class I on the surface of another cell. Each receptor interacts with class I molecules in a distinct fashion, using mostly nonoverlapping binding sites on class I molecules (color-coded in Fig. 4) (43–45). CD8αα and US2 bind more highly conserved surfaces of class I, presumably to maximize the number of alleles with which each receptor can interact. T cell receptors, on the contrary, bind with high specificity to a particular peptide-class I surface, and KIR binds to sites that allow a high degree of allelic specificity. Ly49A binds to a murine class I surface beneath the α1/α2 domain that partially overlaps with the CD8αα binding site. This site was recently identified by using mutational studies (42) of the two Ly49A binding sites observed in the Ly49A/class I complex structure (41). Because this set of five ligands bind different surfaces of class I, they comprise a useful tool for mapping the binding sites of other class I receptors for which no structural data are available.

Figure 4.

Protein ligands for class I molecules interact with different surfaces. Superposition of class I/receptor complexes for which the structure is known reveals that the five class I ligand types bind at distinct locations. Class I interaction surfaces are colored according to each ligand: US2 (magenta), Ly49A (green), B7 TCR (blue), KIR2DL1 (red), CD8αα (cyan). Equivalent Cα atoms for class I heavy chain were used to generate pairwise superpositions. HLA-A2 heavy chain (yellow), β₂-microglobulin (gray), US2 (magenta), Tax peptide (green).

Implications for the ER Dislocation Mechanism.

Ig superfamily domains generally mediate adherence between structures, rather than possessing direct enzymatic function. The US2 residues amino-terminal to the Ig-like fold can be replaced by an exogenous signal sequence (unpublished data), and the US2 Ig-like fold spans residues 43–137. The remaining 62 residues of US2, which are predicted to include a transmembrane domain and a 14-residue cytoplasmic tail, are unlikely to directly mediate dislocation of class I. In addition, US2 binding does not significantly alter the conformation of class I complexes. Thus, US2 may act as an adaptor protein that recruits additional, unidentified components of the ER-dislocation machinery to the US2/class I complex. The US2/HLA-A2 interaction is sufficiently stable to survive consecutive size exclusion and ion exchange chromatography steps, implying a relatively high affinity and/or slow off-rate (16). This immediately raises the question of the status of the US2/HLA-A2 complex in the course of dislocation. Either this complex must be disrupted immediately before dislocation across the ER membrane, or the complex would have to be translocated in its entirety before cytosolic destruction. The first hypothesis would require a disassembly reaction catalyzed by proteins presently unknown; the second model presupposes a translocation pore, the Sec61 complex, of a diameter sufficient to allow passage of the US2/HLA-A2 complex (10). Estimates for the diameter of the Sec61 translocon range from 20 Å as determined by electron cryomicroscopy-based methods (46), to 40–60 Å, based on fluorescence quenching and electrophysiological measurements (47). The upper size estimate of the Sec61 pore might allow dislocation of the entire US2/HLA-A2 (the entire complex could fit in a rectangular box of approximate dimensions 50 × 70 × 80 ų). The structure of the US2/HLA-A2 complex will allow further experiments that test these hypotheses. In addition, knowledge of US2 surface residues will allow us to chemically probe the surface environment of US2 via substitution with residues that react with membrane-permeable bifunctional cross-linkers. Such mutants will be an important tool for identification of factors that participate in ER dislocation.

Abbreviations

HCMV

human cytomegalovirus

ER

endoplasmic reticulum

US

unique short