Loss of recognition by cross-reactive T cells and its relation to a C-terminus-induced conformational reorientation of an HLA-B*2705-bound peptide

Abstract

The human major histocompatibility complex class I antigen HLA-B*2705 binds several sequence-related peptides (pVIPR, RRKWRRWHL; pLPM2, RRRWRRLTV; pGR, RRRWHRWRL). Cross-reactivity of cytotoxic T cells (CTL) against these HLA-B*2705:peptide complexes seemed to depend on a particular peptide conformation that is facilitated by the engagement of a crucial residue within the binding groove (Asp116), associated with a noncanonical bulging-in of the middle portion of the bound peptide. We were interested whether a conformational reorientation of the ligand might contribute to the lack of cross-reactivity of these CTL with a peptide derived from voltage-dependent calcium channel α1 subunit (pCAC, SRRWRRWNR), in which the C-terminal peptide residue pArg9 could engage Asp116. Analyses of the HLA-B*2705:pCAC complex by X-ray crystallography at 1.94 Å resolution demonstrated that the peptide had indeed undergone a drastic reorientation, leading it to adopt a canonical binding mode accompanied by the loss of molecular mimicry between pCAC and sequence-related peptides such as pVIPR, pLMP2, and pGR. This was clearly a consequence of interactions of pArg9 with Asp116 and other F-pocket residues. Furthermore, we observed an unprecedented reorientation of several additional residues of the HLA-B*2705 heavy chain near the N-terminal region of the peptide, including also the presence of double conformations of two glutamate residues, Glu63 and Glu163, on opposing sides of the peptide binding groove. Together with the Arg-Ser exchange at peptide position 1, there are thus multiple structural reasons that may explain the observed failure of pVIPR-directed, HLA-B*2705-restricted CTL to cross-react with HLA-B*2705:pCAC complexes.

Introduction

Molecular mimicry between foreign and self-peptides bound to major histocompatibility complex (MHC) class I or II molecules has long been suspected to play a key role in the development of autoimmune diseases, but the evidence for its occurrence is so far limited.^1–6 Peptide binding by MHC class I molecules is a prerequisite for the antigen-specific recognition that controls interactions with CD8⁺-T cells in the immune system. MHC class I antigens are membrane-anchored cell surface glycoproteins that are composed of a polymorphic heavy chain (HC) and noncovalently attached β₂-microglobulin (β₂m). The membrane-distal two domains of the HC form two α-helices and a β-pleated sheet that shape a groove with six pockets (A–F). It allows the molecule to accommodate and present peptides originating from proteasomal degradation of foreign or self-proteins within the cell.⁷

The B*2705-bound peptide repertoire^8–10 is characterized by ligands with an obligatory arginine as primary anchor at peptide position 2 (pArg2). This residue fits ideally into the negatively charged B pocket of the molecule.¹¹ Furthermore, glycine is very frequently observed at p1 as well as, less often, arginine or lysine. Aliphatic or basic residues are preferred at the peptide C-terminus (pΩ).^8–10 Depending on the length of the peptide, there are some deviations from this situation, in particular regarding the frequency by which an amino acid is observed at a given position, but overall the preferences of the pockets are retained.⁹ It has also been observed that pArg5 may act as an additional anchor, because it can contact residue 116, which is polymorphic between certain HLA-B27 subtypes.^5,12 In the B*2705 subtype, this unconventional interaction is promoted by Asp116 and leads to bulging-in of the middle portion of the peptide and a noncanonical conformation which deviates drastically from the canonical [Fig. 1(A)], bulged-out binding mode that characterizes the vast majority of peptides bound to MHC class I molecules.^7,14

Figure. 1.

General structural properties of the B*2705:pCAC complex with the peptide depicted in brown. For the sake of clarity, water molecules are omitted in all representations. (A) Overview of the different peptide binding modes (see text for an explanation of the p4α and p6α conformations) of B*2705 and B*2709 in complex with the pVIPR, pLMP2, pGR, and pVIPR-U5 peptides. (B) The view is from the side of the α2-helix. Final 2F_O-F_C electron density maps (blue mesh) contoured at 1 σ around pCAC shown in stick presentation. The subtype-specific residue Asp116 of the HC is shown in stick presentation as well. It establishes two salt bridges to pArg9 represented by red dashed lines. (C) The view is 90° rotated horizontally towards the viewer in comparison to (B). (D) The pCAC peptide is colored according to isotropic B-factor distribution. Definition of the color scale is given below the peptide. The side chain of pArg5 is particularly flexible. The view is as in (B) (E) Comparison of the binding modes of B*2705:pCAC (brown) and B*2705:pVIPR (violet) (PDB entry 1OGT),¹² with both peptides in the canonical p4α binding mode. The carboxylate function of Asp116 is slightly rotated allowing to establish a salt bridge to pArg9 in B*2705:pCAC. (F) Structural comparison of B*2705:pCAC and B*2705:flu (gray) (PDB entry 2BST).¹³ Whereas the N- and C-terminal peptide residues superimpose nearly indistinguishably, the central part of both peptides shows clear deviation in the conformation of the protein backbone and side chain conformations, respectively. (G) Schematic representation of side chain orientations when viewed from the N- to the C-termini of B*2705 in complexes with pCAC, flu, pVIPR (in p4α and p6α binding modes), pLMP2 (in p6α binding mode), and pGR. The upper panel illustrates the convention for the orientation of the arrows: floor of peptide binding groove indicated by “β-sheet” and binding region for a TCR by “TCR”. The lower panel shows the primary sequence of the peptides, and the approximate orientations of the peptide side chains in the binding pockets are indicated.

Prototypical for peptides adopting a noncanonical conformation in HLA-B27 subtypes are pVIPR (RRKWRRWHL, derived from vasoactive intestinal peptide Type 1 receptor, residues 400-408)^12,15 and pLMP2 (RRRWRRLTV, derived from latent membrane protein 2 of Epstein-Barr virus, residues 236-244).^5,16 We have also found a sequence-related peptide, pGR (RRRWHRWRL, derived from glucagon receptor, residues 412-420), which is bound in a noncanonical binding mode as well, apparently because the central pHis5 residue contacts Asp116 indirectly, through water-mediated hydrogen bonds.⁶ Cytotoxic T lymphocytes (CTL) directed against the self-antigen pVIPR have been detected in healthy individuals with the HLA-B*2705 (in short, B*2705) subtype which is associated with ankylosing spondylitis (AS), a seronegative spondyloarthropathy,^17,18 and a proportion of these T cells cross-react with the viral pLMP2 peptide and the self-peptide pGR. It appeared thus that the pTrp4-pArg/His5-pArg6 motif, when presented by B*2705 in a noncanonical fashion and possibly in combination with the presence of pArg1, would favor CTL cross-reactivity. This could be explained most easily by molecular mimicry, defined as similarity in overall shape as well as charge distribution for an interaction surface,⁴ between these three peptides.

As no reactivity with pVIPR-directed CTL was found when B*2705 molecules were complexed with a sequence-related peptide derived from human voltage-dependent calcium channel α1 subunit (pCAC, SRRWRRWNR, residues 513-521),⁶ we were interested whether a peptide with the Trp4-Arg5-Arg6 motif, but with a basic pΩ residue (pArg9) in addition would also bind in the noncanonical conformation. We approached this problem by X-ray crystallography and show here that the presence of arginine at pΩ leads to a conformational rearrangement of the pCAC peptide which affects nearly all binding groove pockets of the molecule.

Results and Discussion

Crystal structure of HLA-B*2705:pCAC

As previous work had suggested that the unusual noncanonical conformation of certain B*2705-bound peptides and its connection with the pTrp4-pArg/His5-pArg6 motif might be instrumental in leading to CTL-cross-reactivity and allele-dependent molecular mimicry,^5,6,14,19,20 the nona-peptide pCAC was chosen for the present study, because we were primarily interested whether the presence of pArg9 in this peptide would preclude the interaction of pArg5 with Asp116 seen in pVIPR or pLMP2.

Purified B*2705 was refolded in the presence of the pCAC peptide and successfully crystallized applying a streak seeding protocol. The obtained crystals were suitable for X-ray diffraction data collection beyond 1.94 Å in the space group P2₁. X-ray data collection statistics are summarized in Table I. Since crystals of B*2705:pCAC are isomorphous to B*2705 in complex with the sequence-related peptides pVIPR,¹² pLMP2,⁵ and pGR,⁶ a direct comparison of all complexes is possible. The structure of B*2705:pCAC was solved by molecular replacement with one HLA-B*2705:pCAC complex per asymmetric unit. Refinement converged at R_work of 17.6% and R_free of 22.4% with excellent geometry (Table I), revealing the architecture typical for MHC class I molecules.⁷

Table I.

Data Collection and Refinement Statistics

	HLA-B*2705:pCAC
Data collection
Wavelength (Å)	0.933
Space group	P21
Unit cell (a,b,c [Å]; β [°])	56.3, 54.6, 69.8; 99.3
Solvent content (%)	48.5
Matthews coefficient (Å3 Da−1)	2.4
Resolution (Å)a	20.00–1.94 (1.97–1.94)
Unique reflectionsa	18,393 (1529)
Completeness (%)a	99.8 (99.5)
<I>/<σ(I)>a	16.2 (4.8)
Rsyma,b	0.113 (0.36)
Refinement
Nonhydrogen atoms	3631
Rworka,c	0.176 (0.191)
Rfreea,d	0.224 (0.254)
HC (no. of atoms/ average B factor [Å2])	2335/24.6
β2m (no. of atoms/ average B factor [Å2])	860/32.6
Peptide (no. of atoms/ average B factor [Å2])	106/20.5
Water (no. of molecules/ average B factor [Å2])	312/32.1
Glycerol (no. of atoms/ average B factor [Å2])	18/37.5
Rmsd e from ideal geometry (bond length [Å])	0.011
Bond angles [°]	1.259

^aValues in parentheses refer to the highest resolution shell.

^bR_sym = ∑_h ∑_i |I_h,i - <I_h>|/∑_h∑_iI_h,i.

^cR_work = ∑_h |F_o – F_c|/∑ F_o (working set, no σ cut-off applied)

^dR_free is the same as R_cryst, but calculated on 5% of the data excluded from refinement.

^eRoot-mean-square deviation (Rmsd) from target geometries.

The 2F_O-F_C and F_O-F_C electron density maps show that the pCAC peptide main chain adopts a single conformation [Fig. 1(B,C)], with only the side chain of pArg6 exhibiting a double conformation [Fig. 1(B,C)], contrasting the observations made for the dual binding mode of pVIPR or pGR when bound to B*2705.^6,14 pCAC exhibits the canonical conformation as found for various other B*2705 peptide complexes [Fig. 1(A)]: amino acids p2 and pΩ are buried in the peptide-binding groove of B*2705, whereas the central region of the peptide (p4 to p7) bulges out from the cleft.^14,21–23

Overall peptide conformation

This canonical conformation is characterized by a prominent kink which forces the central region of the peptide from pTrp4 to pTrp7 of pCAC to protrude from the binding groove. As a result, this region of the peptide is solvent-exposed and potentially available for recognition by a T cell receptor on CTL. Unlike the situation in the structures of B*2705 with pVIPR and pLMP2, the central pArg5 exhibits ill-defined and weak electron density of its side chain [Fig. 1(B)]. This is also reflected by high B-factors for this side chain [Fig. 1(D)]. The guanidinium group of pArg5 is stabilized by hydrogen bonds with Gln155^OE1 (Table II) and there is a water-mediated contact to the indole moiety of pTrp7. Whereas one conformation of the side chain of pArg6(A) is solvent-exposed and not in direct contact to any other amino acid, the second conformation pArg6(B) points towards the α1-helix and engages in hydrogen bonds to backbone carbonyl atoms of Ile66, pArg3 and pTrp4 [Fig. 1(B,C); Table II].

Table II.

Peptide Coordination in B*2705:pCAC

Peptide position	Peptide residue	Contact residue	Distance in [Å]	Interaction
p1	pSer1N	Tyr7OH	2.9	HB
	pSer1N	Tyr171OH	2.7	HB
	pSer1OG	Arg62NH1	3.1	HB
	pSer1OG	Glu163(A)OE2	2.9	HB
	pSer1OG	*Arg145NH2	2.8	HB
	pSer1O	Tyr159OH	2.6	HB
	pSer1	Trp167	∼3.4	vdW
p2	pArg2NE	Glu45OE1	2.8	SB
	pArg2NH1	Thr24OG1	2.9	HB
	pArg2NH2	Thr24OG1	3.1	HB
	pArg2NH2	Glu45OE1	2.9	SB
	pArg2O	Glu163(A)OE2	3.3	HB
	pArg2O	Arg62NH1	2.9	HB
	pArg2NE	Glu45OE1	2.9	SB
	pArg2NH2	Glu45OE1	2.9	SB
	pArg2N	Glu63(A)OE1	3.1	HB
	pArg2N	Glu63(B)OE2	2.9	HB
	pArg2N	Glu63(B)OE1	3.3	HB
p3	pArg3N	Tyr99OH	3.0	HB
	pArg3NE	pArg5O	2.7	HB
	pArg3NH2	pArg5O	2.8	HB
	pArg3	Tyr99, Leu156, Tyr159	3.6-4.0	vdW
p4	pTrp4NE1	*pGlu253OE2	2.8	HB
	pTrp4O	pArg6NH2	2.8	HB
	pTrp4	Arg62, Gln65, Ile66	3.6-4.0	vdW
p5	pArg5NE	Gln155OE1	3.1	HB
	pArg5NH1	Gln155OE1	3.0	HB
	pArg5O	pArg3NH2	2.8	HB
	pArg5O	pArg3NE	2.7	HB
p6	pArg6(B)NH2	pArg3O	2.8	HB
	pArg6(B)NH1	Ile66O	3.3	HB
p7	pTrp7	His114, Trp147, Val152, Gln155, Leu156	∼3.5	vdW
p8	pAsn8ND2	Glu76OE1	2.9	HB
	pAsn8ND2	Asp77OD1	3.0	HB
	pAsn8O	Trp147NE1	2.9	HB
p9	pArg9N	Asp77OD1	3.0	HB
	pArg9NE	Asp116OD2	2.8	SB
	pArg9NH1	Asp77OD2	2.8	SB
	pArg9NH1	Asp74OD1	2.9	SB
	pArg9NH2	Asp116OD2	2.5	SB
	pArg9O	Lys146NZ	3.4	HB
	pArg9OXT	Tyr84OH	3.0	HB
	pArg9OXT	Thr143OG1	2.6	HB

Only direct contacts are included, and water-mediated contacts are omitted. Distance cut-off for hydrogen bonds (HB) and salt bridges (SB) is 3.3°Å. Letters in parenthesis indicate the conformations of the amino acid side chains. Interactions between peptide residues are listed according to their first occurrence in the sequence. Interactions of amino acids to a symmetry-related B*2705:pCAC complex are indicated by an asterisk.

On the other hand, Trp7 is oriented towards the α2-helix, in very similar position to the same residue within the pVIPR peptide in the canonical conformation [Fig. 1(E)].¹² The indole moiety of pTrp7 is hydrogen-bonded via a water molecule to Gln155 of the HC and to the guanidinium function of pArg5. In addition, the side chain of pTrp7 is stabilized by van der Waals interactions to Val152 and Leu156. The position of pAsn8 in the pCAC peptide is equivalent to the position of pHis8 in both conformations of pVIPR in the B*2705:pVIPR complex [Fig. 1(E)]. This location leads to a movement of the α-helical residue Glu76 towards the carboxyamine head group of pAsn8, allowing the formation of a hydrogen bond. In contrast, in the two binding modes of B*2705:pVIPR, Glu76 exhibits a double conformation and is not involved in direct peptide binding. Furthermore, the carboxyamine head group of pAsn8 in pCAC is hydrogen bonded to Asp77, and the backbone carbonyl forms a hydrogen bond with Trp147 in an identical arrangement as described in the structure of HLA-B*2705:pVIPR.¹² Together with the hydrogen bond between pArg9^N and Asp77^OD1 (Table II), the connections of main chain atoms to the penultimate and the C-terminal peptide residues with both α-helices of the HC are highly conserved features of MHC class I molecules.⁷ Based on isotope-edited infrared spectroscopic measurements of B*2705 and B*2709 complexed with pVIPR as well as with another self-peptide (TIS; RRLPIFSRL), and pLMP2, we have recently suggested that the “suspension” of a peptide through interactions of two main chain atoms (pΩ-1^O and pΩ^N) with HC residues residing on the α1- and α2-helices, respectively, permits HLA-B27 subtype-dependent differential HC dynamics.^24,25 These preserved structural features are also present in the B*2705:pCAC complex and must thus be regarded as independent of the peptide's conformation and the type of side chain (hydrophobic or basic, as in case of pCAC) of the C-terminal residue of the peptide.

It is also of interest to compare these results with those of a peptide (flu) derived from influenza virus nucleoprotein (flu, SRYWAIRTR, residues 383-391)¹³ which shares the N- and C-terminal residues as well as pArg2 and pTrp4 with pCAC. As in case of pCAC, flu exhibits a canonical p4α conformation when bound to B*2705, and pSer1, pArg2, and pArg9 assume identical locations [Fig. 1(F) and interactive Fig. 3, views L,M]. However, the main chains of the two peptides are easily distinguishable from residues p3-p7, and all respective side chains are differently located. A schematic comparison of these distinguishing features is provided in [Fig. 1(G)] which shows also how individual side chains of the peptides pVIPR (canonical p4α conformation only), pLMP2, and pGR are oriented within the binding groove of B*2705. Not surprisingly, this comparison reveals that the side chains of residues p3-p7, and in particular those of p5 and p6, are the most variable in the five structures. This is likely to have direct implications for cross-reactive T cells, as their recognition of related MHC molecules is influenced by similarity in shape and charge of the interaction surfaces.⁴

Figure. 3.

An interactive 3D model depicting the architecture of the F pocket. The view is from the α2-helix (removed for clarity) into the F pocket. (A) B*2705:pCAC is shown in stick representation from pTrp7 to pArg9. The indol of pTrp7 has been omitted for clarity. Water molecules are shown as red spheres. pArg9 of pCAC is involved in numerous hydrogen bonds and salt bridges indicated by red dashed lines. Additional hydrogen bonds are either directly or mediated by water formed between pArg9 and of the HC residues Lys70, Asp74, Asp77, and Asp116. For the sake of clarity the backbone atoms have been removed for the HC residues. (B) pArg9 of the pCAC peptide forms a salt bridge to Asp74, Asp74 and to the polymorphic Asp116. The C-terminus of the peptide is stabilized by salt bridges and hydrogen bonds depicted by red dashed lines. (C) pArg5 of the pVIPR peptide in p6α conformation¹² forms two salt bridges to Asp116. The main chain interactions of the peptide C-terminus are unchanged when compared to the pCAC peptide. (D) pArg9 of the flu peptide¹³ forms very similar contacts with B*2705 as described for the pCAC peptide [Fig. 3(B)]. The carboxylate of Asp116 is slightly rotated compared to the structure of B*2705:pCAC. The 3D functions that are integrated into this figure (best viewed with the latest version of the free Adobe Reader) can be activated through a click on any part of the image in the PDF version of the article (the 3D features can be terminated by right-clicking on the 3D display and selecting the “Disable 3D” function from the contextual menu). Access to individual components of the 3D model is possible through the model tree icon. The displays which the tour presents are briefly designated at the left. Each model can be interactively manipulated using the mouse (rotating, panning and zooming tools can be selected through the toolbar or contextual menu), and readers can access the preset “tour” of the model by clicking on the respective designations (either via the middle section of the model tree when it is displayed, or the drop-down menu in the toolbar, or through the contextual menu). Initially, the tour shows views of the peptide binding groove with B*2705:pCAC superimposed on the flu and pVIPR peptides bound to the B*2705 subtype. Further views show in detail the interaction of the respective peptides with HC residues residing in the F-pocket.

Molecular mimicry

Molecular mimicry is a central postulate of the arthritogenic peptide hypothesis,² but its functional relevance in the context of HLA-B27 has yet to be established. We have previously pointed out that molecular mimicry is not only a peptide sequence-dependent property, but relies equally on the conformation of the ligands involved. The latter is, however, an HLA-B27 subtype-dependent feature.^5,6,14 When considering the pVIPR, pLMP2, and pGR peptides, their structural similarity rests entirely on the presence of the polymorphic HC residue 116 which permits all three peptides to be bound in the unusual noncanonical p6α conformation in which the polymorphic HC residue Asp116 is contacted either directly (pLMP2, pVIPR) or indirectly (pGR in conformation “B”).^5,6,12 In the B*2709 subtype, the presence of His116 precludes structural mimicry, because no extensive similarity with regard to structure or surface charge can be found, since pVIPR and pLMP2 are bound in p4α binding mode contrasting pGR in p6α binding mode.¹⁴

The pCAC peptide is presented in a comparable p4α conformation as well [Figs. 1(G) and Fig. 2(A,B)], exhibiting greatest similarity with canonically bound pVIPR. Since pLMP2 and pGR are not displayed by B*2705 in this binding mode, the limited degree of molecular mimicry between pCAC and pVIPR does not extend to these two peptides [Figs. 1(G) and 2(A,C,D)]. In addition, the exchange of pArg1 (pVIPR, pLMP2, pGR) by pSer1 in pCAC, although leading to loss of a positive surface charge, has already been demonstrated to be functionally irrelevant, at least for selected CTL.²⁶ It seems therefore possible that the middle of the peptide, roughly from pTrp4 to pTrp7 (or pLeu7 in case of pLMP2), provides sufficient differences between pCAC and the other peptides to preclude recognition by cross-reactive CTL. The exchanges at p8 (Asn, His, Thr, or Arg in pCAC, pVIPR, pLMP2, or pGR, respectively) could obviously contribute to the observed lack of CTL cross-reactivity between the different complexes as well. This demonstrates that the structural and the electrostatic surface properties of the B*2705:pCAC complex are unquestionably distinct from those of the other three peptides, even when canonically bound pVIPR is considered (Fig. 2).

Figure. 2.

Surface representation of B*2705 complexed with pCAC, pVIPR, pLMP2 and pGR. The view direction is that of an approaching TCR (A) in brown pCAC in p4α conformation, (B) in violet pVIPR in p4α conformation (PDB entry 1OGT)¹² (C) in cyan pLMP2 in p6α conformation (PDB entry 1UXS)⁵ (D) in orange pGR (PDB entry 2A83)⁶ (E–H) Electrostatic surface potential of B*2705 complexed with pCAC, pVIPR, pLMP2, and pGR in the range of ±9kT/e colored red and blue for negative and positive electrostatic potential (see inlet in panel E). Gray areas on surface are uncharged. The yellow line indicates the peptides presented by B*2705.

Therefore, despite the fact that there are a number of TCR-accessible, exposed side chains that exhibit structural equivalence between B*2705-bound pCAC or pVIPR, such as pTrp4, pTrp7 and possibly pArg5 [Fig. 1(E)], the lack of extensive molecular mimicry between the B*2705:pCAC complex and the other three peptides is obvious. It can thus explain the inability of B*2705:pVIPR-directed CTL to lyse cells expressing B*2705:pCAC complexes.⁶

Anchoring of the C-terminal arginine of pCAC within the B*2705 F pocket

Ultimately, the presence of pArg9 within the F pocket of pCAC is likely to be responsible for the observed drastic conformational re-orientation of pCAC when compared to sequence-related peptides such as noncanonically bound pVIPR. The change in binding mode exhibited by pCAC is apparently due to the preclusion of the interaction of pArg5 with Asp116 that is observed in the structures with pVIPR (p6α conformation), pLMP2, and pGR.¹⁴ Instead, Asp116 engages in two of four salt bridges that are formed by the guanidinium moiety of pArg9, the other two with Asp74 and Asp77. Furthermore, pArg9 is involved in several hydrogen bonds with F pocket residues [Fig. 3(A,B), Table II]. Together with the hydrophobic residues that flank the F pocket, these multiple contacts serve to anchor pArg9 firmly within the binding groove.

The interactive 3D images that are embedded within Figure 3 reveal similarities (e.g., between the structures with pCAC and flu), but also pronounced differences (e.g., between the structures with pCAC and pVIPR, both in canonical or noncanonical conformation) of interactions between the peptide C-termini and F pocket residues. Regarding the pCAC and flu peptides, the extensive network of interactions within the F pocket is nearly identical, with one notable exception: the orientation of the carboxylate of Asp116. Its altered orientation influences the length of the bidentate salt bridges between Asp116 and pArg9 in the B*2705:pCAC structure. These interactions are significantly shorter (2.5 and 2.8 Å) compared to the B*2705:flu complex, where lengths of 3.3 and 3.5 Å were reported.¹³ Consequently, the pCAC peptide is likely to be more tightly anchored to Asp116. A combination of spectroscopic techniques and pK_a-calculations²⁷ might be able to assess the predicted effect quantitatively. Other than the vicinity of Asp116, the backbone amide and terminal carboxylate of pArg9 are in hydrogen bonding distance to Asp77, Tyr84, Thr143 and Lys146 [Fig. 3(A,B), Table II]. These backbone interactions are identical as compared to the complexes of B*2705 with pVIPR, pLMP2, or pGR, except for subtle alterations in the rotamer conformations adopted by Lys146.

We also observed that the anchoring of the basic amino acid (pArg9) within the F pocket resulted in an alteration of the water cluster within the binding groove, in comparison to the situation found in the structures of B*2705 with pVIPR, pLMP2, or pGR (interactive Fig. 3). In these cases, the side chain of Lys70 is oriented towards the F pocket and is connected via a water molecule to Asp116 (interactive Fig. 3 views D,E,F). In the B*2705:pCAC structure, the side chain of Lys70 adopts a very similar orientation, but is additionally connected, also via water molecules, to Asp74 and Asp77 (Fig. 3). The distinct anchoring of the pCAC peptide in comparison to the noncanonical binding modes seen in pVIPR, pLMP2, and pGR (interactive Fig. 3, views J,K) leads to a re-arrangement of the water cluster in the F-pocket beneath the pCAC peptide. Instead of the guanidinium moiety of pArg5 (pVIPR, pLMP2) or the imidazole ring of pHis5 (pGR), additional water molecules reside in the F pocket of the B*2705:pCAC complex, serving to bridge Lys70, Asp116 and pArg9 (interactive Fig. 3, views A,B,C).

Despite the identity of the C-terminal residue, a different arrangement of Lys70 was reported in the B*2705:flu complex, with its side chain pointing away from the F pocket. The authors ascribed this to electrostatic repulsion between Lys70 and pArg9.¹³ However, the present study demonstrates that the situation is more complicated, revealing a conformational interdependence of the side chains of several peptide residues. In B*2705:pCAC, an orientation of Lys70 as reported in the B*2705:flu structure is not feasible, since the guanidinium function of pArg3 and one of the side chain conformations of pArg6 induce electrostatic repulsion between pArg3 and pArg6 (interactive Fig. 3, views G–I). In turn, this makes it impossible for Lys70 to adopt a different rotamer conformation as observed in B*2705:flu, since this would be in steric conflict with the side chains of pArg3 and pArg6 (interactive Fig. 3, views L,M).

Finally, besides arginine, a preferred C-terminal residue of B*2705-bound peptides is lysine.¹⁰ It is therefore interesting to compare the structures of the complexes of B*2705 with pCAC and with the model peptide m9 (ARFAAAIAK).²² In B*2705:m9, a direct salt bridge connects pLys9^NZ and Asp116^OD2, resembling the contact observed in B*2705:pCAC (pArg9^NH2 and pArg9^NE each contact Asp116^OD2). A fundamental difference between the binding modes of these two basic amino acids is, however, the numerous contacts that are only possible in case of arginine (Table II) due to the presence of its guanidinium function.

Thermodynamic, kinetic, and pK_a measurements as well as molecular dynamics simulations have also revealed that pLys9 is very firmly bound within the F pocket of B*2705 molecules, endowing the complex with the m9 peptide with a half-life of about 10 days.^27–30 Since peptides with a basic C-terminus are likely to account for the majority of ligands during positive selection of MHC class I-specific CTL,³¹ the B*2705 subtype belongs to the relatively few human MHC molecules that can bind this subset of peptides with high affinity.³² As suggested by theoretical considerations, this property has interesting implications for positive T cell selection. Firmly binding peptides may in part be responsible for the inappropriate selection of a T cell repertoire predisposing the B*2705 allele for an involvement in diseases with an autoimmune aetiology.³² The firm anchoring of basic C-terminal peptide residues such as observed in pCAC is likely to cause a reduced flexibility of the entire peptide binding groove of B*2705. For the B*2705:m9 complex, an additional stabilization of ∼ 7 to 30 kJ/mol compared with the complex of another subtype (B*2709, which does not bind peptides with a basic C-terminus)⁸ was estimated.²⁷ This is in contrast to the gain in stabilization of ∼ 4 to 7 kJ/mol of B*2705:pVIPR in the noncanonical p6α conformation, in which a direct salt bridge between pArg5 and Asp116 is formed,¹² over the B*2709:pVIPR complex.²⁷ These results underscore the importance of a basic pΩ residue. Since the B*2709 subtype, in contrast to B*2705, is not associated with AS,³³ it appears possible that the inability of the former to bind peptides with a basic residue under in vivo conditions might be connected to the disease association of the latter. The differential association of these subtypes is a long-standing, unresolved problem that has puzzled immunologists for more than a decade.^14,20,33,34

Distinctive design of the A pocket of the B*2705:pCAC complex

The classical hydrogen bonding network in the A-pocket of virtually all peptide-binding MHC class I molecules⁷ is formed by the three highly conserved HC residues Tyr7, Tyr59, and Tyr171, a water molecule, and the N-terminus of the peptide.⁷ In addition, the solvent-directed side chain of p1 may also contribute to the anchoring of the peptide N-terminus. In case of pArg1, its guanidinium function interacts with the carboxylate of Glu163 [Fig. 4(A)], allowing the formation of a unique and tightly packed hydrophobic stacking between the guanidinium group of Arg62 and the aromatic side chain of Trp167.²⁸ The positive charges of the two guanidinium groups of pArg1 and Arg62 are partly compensated by the adjacent Glu63 and Glu163 that, together with Arg62, form a water-mediated clamp over the peptide [Fig. 4(A)], strongly suggesting a considerable contribution to the stability of the entire complex and the prevention of peptide dissociation. Such structural arrangements are found, for example, in the structures of B*2705 or B*2709 in complex with pVIPR,¹² pLMP2,⁵ pVIPR-U5 (RRKW[citrulline]RWHL),¹⁹ as well as pGR.⁶ All four of these peptides have a pArg1 residue in common, in contrast to the pCAC peptide with a pSer1 residue. However, the relatively frequent occurrence of the N-terminal pArg1-pArg2 sequence in peptides bound by HLA-B27 may not exclusively be a consequence of the Arg62-pArg1-Trp167 stacking interaction, as peptides with the pArg1-pArg2 motif appear to have an increased cytosolic stability.³⁵ In support of this contention, López de Castro and co-workers concluded that the preferential ability of HLA-B27 molecules to bind peptides with the dibasic N-terminal motif is probably not connected to the presence of Glu163 and Trp167, since mutations of these residues do not seem to affect the frequency with which pArg1-pArg2-containing peptides are bound.³⁶

Figure. 4.

Architecture of the A pocket. The α1- and α2-helices (gray) are indicated in cartoon representation and HC residues are drawn as green sticks. For the sake of clarity, the peptide side chains of p3 and p4 are truncated. Water molecules are drawn as red spheres. (A) B*2705:pVIPR (PDB entry 1OGT)¹² with pArg1 of the pVIPR peptide (violet) represents the classical F pocket interaction of HLA-B27 molecules presenting a pArg1-containing peptide. (B) The hydroxyl function of pSer1 of the B*2705:pCAC (brown sticks), with Glu63 and Glu163 each in “A” conformation, is in direct hydrogen bonding distance (red dashed lines) to Glu163. (C) A water molecule mediates a contact between the hydroxyl of pSer1 and Glu163 in the B*2705:pCAC complex, with Glu63 and Glu163 in the “B” conformation. The occurrence of two side chain conformations of Glu63 and Glu163 has further influence on the hydrogen bonding network in the A pocket. (D) B*2705:flu (PDB entry 2BST)¹³ with pSer1 of the flu peptide (grey) reveals a different orientation of Glu163 and a water mediated contact of the hydroxyl of pSer1 to Glu63.

We were primarily interested in two aspects of the anchoring of pCAC within the A pocket: (i) can the pCAC peptide establish the classical hydrogen bonding network as observed in other HLA class I antigen complexes, and (ii) can the truncated side chain function of pSer1 lead to changes in the architecture of the A-pocket by preventing the formation of the “sandwich”-like packing interaction involving Arg62, the side chain of the p1 residue, and Trp167? Both interactions are expected to influence the presentation of a peptide and its recognition by CTL, despite evidence to the contrary in case of pArg1 of the pLMP2 peptide.²⁶ Since a superposition of the N-terminal region of pCAC with that of peptides with pArg1 in complex with B*2705 or B*2709 revealed no differences for the Cα-atom of the pSer1 residue of the pCAC peptide, the classical pentagonal hydrogen bonding network characterizing the A pocket of MHC class I molecules is not affected. Therefore, changes in interaction with pSer1 are expected to be solely restricted to the side chain function, leaving the interaction with main chain atoms unaltered.

The substitution of pArg1 by pSer1 in B*2705:pCAC does not allow the formation of the Arg62–p1–Trp167 clamp. This leads to a slightly different arrangement of the guanidinium function of Arg62 and results in the establishment of two novel contacts: one hydrogen bond to pSer1^OG, and the other to the backbone carbonyl of pArg2 [Fig. 4(B,C); Table II]. The indole moiety of Trp167 in the structure of B*2705:pCAC adopts a conformation similar to that observed in B*2705 complexes with an N-terminal pArg1 (see Ref.28 for further explanations). Beside the interaction with Arg62, the side chain of pSer1 is further stabilized by hydrogen bonding to Glu163(A). Since the side chain of Glu163 occurs in a double conformation, pSer1^OG is either stabilized by a direct hydrogen bond to the carboxylate function of Glu163(A) or indirectly via a water molecule in the other conformation Glu163(B) [Fig. 4(B,C); Table II]. The side chain of Glu63 occurs in a double conformation as well, resulting in two distinct modes of interaction with pSer1. In Glu63(A), the formation of a contact to pSer1 is prevented, but the establishment of a hydrogen bond to the backbone amide of pArg2 is allowed instead [Fig. 4(B)]. In the other side chain conformation of Glu63, an indirect hydrogen bond is formed between the carboxylate function of Glu63(B) via a water molecule to pSer1^OG and a second water mediated contact is established to the HC residue Glu163(B) [Fig. 4(C); Table II]. The occurrence of the double conformations of the Glu63 and Glu163 side chains seems to be a unique feature of the B*2705:pCAC complex, since these residues occur in distinct single conformations in the structures of B*2705 or B*2709 with, for example, pVIPR, pLMP2, pVIPR-U5 or pGR. In these eight structures, Glu163 is in direct contact with the guanidinium function of pArg1 [Fig. 4(A)].^5,6,12,19

Remarkable differences can also be observed when the structures of B*2705 in complex with the peptides pCAC and flu are compared. Superposition of both peptides reveals an r.m.s.d. value of 0.2 Å for all Cα-atoms, with nearly identical side chain conformations for the anchoring residues pSer1 and pArg9 as well as for pArg2 and pThr8 (pCAC) / pAsn8 (flu) [Fig. 1(F,G)]. However, a detailed inspection of the A pocket in both structures reveals significant differences in stabilization of the anchoring residue pSer1. Whereas the hydroxyl of pSer1 can be directly recognized by Arg62 as well as by Glu163(A), and water-mediated, by Glu63(B) in the B*2705:pCAC structure [Fig. 4(B,C)], no direct interactions are observed between these HC residues and pSer1 of the flu-peptide. In the B*2705:flu complex [Fig. 4(D)], pSer1^OG is merely contacted by Arg62 through a water molecule, whereas the side chain of Glu163 points away from the peptide and is directed towards the α2-helix of the HC.¹³ It appears possible that some of the differences between the peptides pCAC and flu might be attributable to the distinct resolutions at which the two structures were solved. Despite these differences, however, the comparison of the structures of distinct peptides in complex with B*2705 permits us to conclude that shorter side chains such as pSer1 of pCAC or flu can be accommodated due to the conformational adaptability of the side chains of HC residues that line the A-pocket.

Conclusions

Three of our observations deserve specific mention: (i) a basic peptide C-terminus will engage in several favorable contacts with F pocket residues of the B*2705 subtype and will, in particular, contact Asp116 through a salt bridge; (ii) the engagement of Asp116 by pArg9 does not permit pArg5, if present as in case of pCAC, to form a salt bridge with this HC residue, resulting in the prevention of the formation of a noncanonical p6α peptide conformation; (iii) despite the canonical p4α binding mode of pCAC, this peptide forms several unconventional interactions with HC residues which are likely to contribute to the observed loss of molecular mimicry and the evasion of pCAC from recognition by pVIPR-directed cross-reactive CTL.

That Asp116 “prefers” to interact with pArg9 instead of pArg5 is probably not an artefact of the in vitro assembly, since it is likely that the C-terminal peptide residue is the first that finds its place within the MHC class I binding groove in vivo.³⁷ The different proteasomes within mammalian cells produce peptides with distinct C-termini that are not further trimmed inside the endoplasmic reticulum,³⁸ in contrast to the N-termini of peptides which undergo proteolytic modification by the protease ERAP1 in this compartment.^39,40 The multiple and strong contacts between F pocket residues of the B*2705 subtype and a basic peptide C-terminus^22,13 will thus preclude an interaction of further peptide residues with Asp116 of the HC. Consequently, and as exemplified by pCAC, the peptide cannot assume a noncanonical binding mode and is forced to be displayed in a conventional conformation. This alone is probably sufficient to cause lack of recognition by pVIPR-directed cross-reactive CTL^5,6 whose reactivity appears to rely on peptides which are presented in a noncanonical conformation. Therefore, together with the Arg/Ser exchange at p1 and the structural alterations that characterize A pocket residues, molecular mimicry between pVIPR, pLMP2, pGR on one hand and pCAC on the other is effectively abolished when these peptides are displayed by B*2705 molecules.

Materials and Methods

Protein preparation

The purified nona-peptide SRRWRRWNR (pCAC) was purchased from Alta Bioscience (Birmingham, UK). The extracellular region of the B*2705 HC and β₂m were expressed separately as inclusion bodies in Escherichia coli, dissolved in 50% (w/v) urea and HLA-B27:peptide complexes were reconstituted for 14 d at 4 °C as described previously^28,41,42 with slight modifications. Briefly, unfolded B*2705 HC (12 mg), β2m (10 mg), and 4 mg of pCAC was rapidly injected into 400 mL of refolding buffer (400 mM arginine/HCl, 2 mM EDTA, 5 mM reduced glutathione, 0.5 mM oxidized glutathione, and 100 mM Tris/HCl pH 7.5). Refolded B*2705:pCAC complexes were isolated by size exclusion chromatography. For crystallization experiments, the preparation was concentrated to 10–12 mg/mL using AMICON Ultra-15 concentrators.

Crystallization, cryo-cooling, and X-ray data collection

Crystallization trials were setup according to the hanging drop vapor diffusion protocol at 18 °C. The expertise gained from crystallizations of several additional HLA-B27:peptide complexes^5,6,12,28,43 was exploited to design a grid screen varying the concentration (14 - 24% (w/v)) of PEG6000 and PEG8000 against different pH values in the range from 7.0 to 8.0. As initial crystals were not of sufficient size for X-ray diffraction experiments, streak-seeding was applied to grow plate-like crystals with a maximum size of 200x100x20 μm³ from 20% (w/v) PEG8000, 150 mM NaCl, 20 mM Tris-HCl, pH 7.0. For cryo-protection, the concentration of PEG8000 was increased to 24% (w/v), and glycerol was added to a final concentration of 10% (v/v). The X-ray diffraction dataset of B*2705:pCAC was collected at beamline ID14-2 of the European Synchrotron Radiation facility (ESRF, Grenoble, France) at 100 K equipped with an ADSC-Q4 detector (Area Detector Systems Corporation). The X-ray diffraction data were indexed in monoclinic space group P2₁ with the program DENZO and scaled with the program SCALEPACK.⁴⁴

Structure determination, refinement and analysis

The structure was solved by molecular replacement using the program EPMR⁴⁵ with the peptide- and water-stripped B*2705:pGR crystal structure as initial search model (PDB entry 2A83).⁶ Molecular replacement yielded an unambiguous solution with a correlation coefficient of 0.7 and an R-factor of 35% for diffraction data from 30.0 to 3.0 Å resolution with one B*2705:pCAC complex in the asymmetric unit. After rigid body refinement as implemented in REFMAC5,⁴⁶ the initial 2F_O-F_C and F_O-F_C electron density maps clearly revealed the presence of the nona-peptide when inspected with the graphical program O.⁴⁷ After modeling of the peptide in the electron density maps, the model was further subjected to maximum-likelihood restrained refinement (REFMAC5)⁴⁶ followed by iterative model building cycles. These included also the addition of double conformations of amino acid side chains and the modeling of glycerol molecules originating from the cryo-protectant. Water molecules were positioned with ARP/wARP.⁴⁸ After translation, libration and screw-rotation (TLS) refinement⁴⁹ as implemented in REFMAC5 with four TLS-groups, intermediate and final structures were evaluated with PROCHECK⁵⁰ and MOLPROBITY.⁵¹ Figures were prepared with PyMOL.⁵² Three-dimensional (3D) images were created as described previously.⁵³ Electrochemical potentials were calculated with the program APBS.⁵⁴ Atomic coordinates and structure factor amplitudes were deposited in the Protein Data Bank under accession code 3LV3 .