Selection of peptides that bind to the HLA-A2.1 molecule by molecular modelling

Molrcular Immunology, Vol. 33, No. 2, pp. 221-230, 1996 Copyright af 1996 Elsevier Science Ltd. All rights reserved Printed in Great Britain 0161-589Oi96 $15.00+0.00

Pergamon 0161~5890(95)00065-8

SELECTION OF PEPTIDES THAT BIND TO THE HLA-A2.1 MOLECULE BY MOLECULAR MODELLING JONG-SEOK *Genetic

LIM,* SEUNGMOAK KIM,* HEE GU LEE,* KI-YOUNG TAE-JONG KWONt and KILHYOUN KIM*‘;5

Engineering

Research Institute,

Korea Institute of Science and Technology,

LEE,*

Taejon, Korea;

tDepartment of Microbial Engineering, Kon-Kuk University, Seoul, Korea; iDepartment of Pharmacy, Ewha Womans University, 11-l Daehyundong. Seoul, Korea (Receivrd

14 December 1994; accepted irz ,evisrd,fotw

18 April 1995)

Abstract-Cytotoxic T lymphocytes recognize antigenic peptides in association with major histocompatibility complex class I proteins. Although a large set of class I binding peptides has been described, it is not yet easy to search for potentially antigenic peptides without synthesis of a panel of peptides, and subsequent binding assays,In order to predict HLA-A2. l-restricted antigenic epitopes, a computer model of the HLA-AZ.1 molecule was established using X-ray crystallography data. In this model nonameric peptide sequences were aligned. In a molecular dynamics (MD) simulation with two sets of peptides known to be presented by HLA-A2. I, it was important to know the anchor amino acid residue preference and the distance between the anchor residues. We show here that the peptides bound to the HLA-AZ.1 model structure possessa side chain of C-terminal

anchor residue oriented into the binding groove with different distances between the two anchor residues from 15 to 21A. We also synthesized a set of nonamer peptides containing amino acid sequencesof Hepatitis B virus protein that were selected on the basis of previously described HLAA2.1 specific motifs. When results obtained from the MD simulation were compared with functional binding assays using the TAP-deficient improves prediction of the HLA-A2.1

cell line T2, it was evident that the MD simulation method binding epitope sequence. These results suggest that this

approach can provide a way to predict peptide epitopes and search for antigenic regions in sequences in a variety of antigens without Key words: MHC binding

screening a large number

peptide. epitope prediction,

INTRODUCTION

Class I major histocompatibility complex (MHC) encoded antigens are expressed on the cell surface as a heterodimeric complex consisting of a membrane-bound 46kDa heavy chain, and a 12-kDa non-MHC-encoded, non-membrane bound light chain, fl2-microglobulin (barn). The heterodimer is bound by an 8-10 amino acid residue long antigenic peptide originating from the endogenous pool of the cell to be presented to T cells. The mechanism by which individual class I MHC molecules bind a broad array of peptides has recently become clearer with accumulation of knowledge on the crystal structures of class I molecules, and the sequences of peptides of endogenous origin bound by the class I molecules. §Author to whom correspondence should be addressed: Dr Kilhyoun

Kim,

Department

of Pharmacy,

Ewha Womans

University, I 1-I Daehyundong, Seoul 120-750, Korea. Abbreviations:

fl2m,

/I2-microglobulin;

fsec, femtosecond;

GAMIg-FITC, fluorescein-conjugated goat anti-mouse IgG; HBV, Hepatitis B virus; mAb, monoclonal antibody; MD, molecular dynamics; MFI, mean fluorescence intensity; MHC, major histocompatibility complex; PBS, phosphatebuffered saline: psec, picosecond; TcR, T cell receptor. 221

of synthetic peptides.

computer

modeling.

The crystal structures of human MHC class I molecules HLA-A2.1 (Bjorkman et al., 1987a; Saper et al., 1991), HLA-Aw68 (Garrett et al., 1989), HLA-B27 (Madden et al., 1991, 1992) and of a murine H-2K” (Fremont et al., 1992; Zhang et al., 1992) have been reported, and a variety of the sequences of peptides bound by the class I molecules has also been reported (Riitzschke et al., 1990; Van Bleek and Nathenson, 1990; Falk et al., 1991a; Jardetzky et al., 1991; Hunt et al., 1992). As data regarding the structural requirements for peptides binding to class I molecules accumulate, it becomes more likely that information on the structure of peptide-MHC complexes will provide a method of selection of class I binding peptides from the amino acid sequences of a variety of antigens. In fact, it has recently been possible to assess the effect of the presence of different side chains, not only in the main anchor positions, but also in other positions of the binding peptides (Chen and Parham, 1989; Corr et al., 1992). The characterization of sequence motifs bound to a certain type of class I molecule has been approached in several ways. A pool of peptides isolated by acid elution and reverse-phase HPLC separation of MHC proteins has revealed preferred amino acid residues at particular

222

J.-S. LIM et ul.

positions for class 1 binding peptides. In a collection of peptides that are bound by a type of class I molecule, amino acid positions that are primarily occupied by one amino acid residue, or by very closely related residues are known as anchor residues. These anchor residues bind in well-defined pockets within the binding groove of class I molecules, and it appears that the peptide specificity of different types of class I molecules is primarily determined by the amino acid composition of their anchor residues. In the case of HLA-A2.1, these anchors are leucine (Leu) at position 2, and Leu or valine (Val) at P9, the Cterminus (Falk et al., I99 1b). X-ray crystallographic studies have outlined the molecular structure of six pockets (termed A-F) in the peptide binding groove, and have shown that the two main pockets (B and F) engage the two main anchors located at position 2 and at the Cterminus of the peptides, respectively (Latron et al., 1992). It has been assumed that these allele-specific peptide anchors interact specifically with dominant pockets in the class I molecules, and that the other, shallower pockets have less stringent requirements and accumulate many different side chains (Matsumura et al., 1992). A better understanding of the conformational differences of potential antigenic peptides which bind to class I molecules would help to define the structural requirements in peptide selection, and help in design of an optimum antigenic peptide. In this work attempts were made to select HLA-A2.1binding peptide epitopes from hepatitis B virus (HBV) proteins, and to describe the likely structures of the peptide-MHC complexes using molecular dynamics (MD) simulation. The construction and analysis of a threedimensional model of peptides bound by HLA-A2.1 molecules is presented and characterization of the binding peptide motifs determined by MD simulation is also described. This approach may be useful for prediction of antigenic epitopes in a variety of antigens. This would greatly reduce the number of peptides that have to be screened in binding assays.

MATERIALS

AND METHODS

SJwthetic peptides Peptides were purchased from Cambridge Research Biochemicals (Cheshire, U.K.). All peptides used in this study were subjected to HPLC on a Cl8 reverse-phase column and eluted as a single major peak ( > 90%). They were solubilized in dimethylsulphoxide and diluted with phosphate-buffered saline (PBS). Cell line and monoclonal antibodies The human cell line T2, which was a generous gift from Dr Y. Young at The Scripps Research Institute (La Jolla, California, U.S.A.), is class I assembly deficient and, accordingly, expresses reduced amounts of HLA-A2.1, and no HLA-B5 on the cell surface. The cell line was maintained in RPM1 supplemented with 10% FCS, penicillin (100 U/ml), streptomycin (100 pg/ml), and L-glutamine (300 pg/ml). Monoclonal antibodies (mAbs),

BB7.2 recognizing assembled HLA-A2.1 molecules, and W6/32 recognizing assembled HLA-A, -B, and -C locus products were purchased from American Type Culture Collection (Rockville, Maryland, U.S.A.). Fluoresceinconjugated goat anti-mouse IgG (GAMIg-FITC) was obtained from Tago Immunologicals (Biosource, California, U.S.A.).

Flow cytometric analysis qf‘ HLA-A2.1 T2 cell surface

molecules on the

Cells (I x I 06) were washed twice with PBS and incubated with 10 pg/ml of HBV peptides at 37°C for 3 hr. After washing, 100 ~1 of BB7.2 hybridoma culture supernatant was added and incubated on ice for 40 min. After washing three times with PBS, cells were incubated with 100 ~1of a l/l00 dilution of GAMIg-FITC on ice for another 40 min, washed and analysed on a flow cytometer (Model FACScan, Becton Dickinson, San Jose, California, U.S.A.). Data were acquired with no gate setting and represented the mean fluorescence intensity (MFI) of 10,000 events.

Model building ,fbr HLA-A2.1 complex

and the nonamer peptide

Models of HLA-A2.1 and binding nonamer peptides were built from the crystal structures of the Brookhaven Protein Data Bank: 3HLA for HLA-A2.1 (Saper et al., 1991) and 3HSA for the nonamer peptide (Madden et al., 1992). The HLA-A2.1 model was simplified by using only c(1 and x2 domains (residues 1-I 82) and the 18 water molecules bound on the ~1 and a2 domains, since ct3 and p2m are not directly involved in peptide binding (Madden et al., 1993). For developing models of nonamer peptides, initial coordinates of peptides were fitted to available backbone atoms and C,j coordinates, and other atomic coordinates including hydrogen atoms, were selected by standard atomic configuration since it was not possible to obtain accurate coordinates for nonamer peptides. Calculations were performed by the Discover 2.7 program (Biosym Technologies, San Diego. U.S.A.) on an IRIS340/GTX computer. The first 0 calculation step was energy minimization with a 9.0 A cutoff distance to relax bad contacts between the HLA-A2.1 and the nonamer peptide arising from arbitrary selection of atomic coordinates. Initial coordinates of C, in the peptide were fixed, and an energy minimization consisting of only 100 steps was carried out by the steepest descent method. In a second step, the calculation of molecular dynamics (MD) was performed for a total of 20.0 picoseconds (psec) for each of 20,000 1.O femtosecond (fsec) steps at 300 K. In this HLA-A2.1 model, 18 water molecules and a nonamer peptide had no constraints, and thus could move freely. The analysis was done with a final 1000 steps (1 .Opsec) and the average MD structure calculation was performed using the program Insight II (version 2.2.0., Biosym Technologies, San Diego, U.S.A.).

Selection of class I binding peptides RESULTS Model peptides,for comparison of binding activities To verify the relationship between binding affinity and MD simulation, seven antigenic nonamer peptides of known binding affinity to HLA-A2 molecules were selected (Parker et al., 1992). Peptides used are listed in Table 1. They all contained Leu at position 2 (P2), and Val or Leu at position 9 (P9). Among seven peptides, six peptides had a notable binding affinity to HLA-A2.1, but peptides HIV2, HIV3 and Mel-l had lower affinities and accordingly had relatively short half-lives as assessed by measuring the rate of dissociation of fl2m. The structures of the class I molecules complexed with these peptides were analysed by using MD simulation method. When the bound peptides were viewed from the side of the binding cleft, six out of seven peptide-HLA-A2.1 complexes showed that the peptide backbones adjust their conformation such that they are accommodated in the center of the binding cleft (Fig. 1). The simulation results revealed that carbonyl groups on the C-termini of the peptides were bound more tightly to the binding pocket by hydrophobic interaction with hydrophobic side chains of the binding pocket, than the amino groups on N-termini. The N-termini of the peptides therefore move more easily than the C-termini, which can be seen in the configuration of N- and C-termini of the peptides in Fig. 1. The backbones of the peptides were located deep inside the cleft near the N-termini and bulged with different configuration as a result of a kink between residues P3 and P6. They were gradually descending to the floor of the binding cleft near the C-termini of the peptides. The overall structure of the peptide backbones complexed with HLA-A2.1 molecules revealed that the backbone conformations of the first three and last four residues in each peptide were similar to one another, with the exception of HIV4. The backbone conformation of peptide HIV4 obtained from MD simulation showed a unique feature of a kinked region between residues P5 and PS (Fig. 1). This kinked region rising up toward the

223

surface of the complex constituted the antigenic surface of the complex so that the structural differences of this region of the peptide were directly visible to the T cell receptor (TcR) (Bjorkman et al., 1987b). The peptide backbone structure in the centre of the cleft points up toward the TcR, but the two peptides HIV4 and HIV5 had P4 and P5 pointing down into the binding site. To determine whether there is a correlation between the actual binding affinity of peptides to HLA-A2.1, and the characteristics of the peptide configurations obtained by MD simulation, more defined structural features of individual peptides were examined. Although MD simulation time was very short (20.0 psec of each 1.0 fsec integration step) relative to the model of HLA-A2.1, the results showed a strong tendency. The HIV4 peptide, among the seven peptides tested (Table l), has been reported to be a non-binding peptide and the simulation showed that the peptide was released from the peptide-binding cleft in 10.0 psec. Other peptides with positive binding affinities stayed in the cleft and the anchor side chains of the P2 and P9 residues interacted with the hydrophobic binding pockets. A comparison of the side chains of respective P2 anchor residues clearly showed that all of the side chains point toward the binding pockets. The side chains at Pl protruded toward the solvent, except for HIV3 which had the side chains at Pl with a slighty different orientation toward the binding groove. This Pl side chain orientation in HIV3 is not consistent with the argument that the Pl side chain usually points up toward the solvent so that various amino acid residues can be accommodated at this position (Fremont et al., 1992; Matsumura rt al., 1992). The side chains at P9 anchor residues seemed to be essentially superimposable, whether they are Val or Leu, yet the Val side chains of HIV4 exhibited an opposite orientation, i.e. toward the solvent-accessible surface. Thus, it appears the orientation of an individual peptide side chain depends on the context of the peptide sequence in which it is located. Furthermore, considering that HIV4 does not bind to HLA-A2.1, the orientation of P9 side chains

Table I. Characteristics of the peptides bound to the modelled HLA-AZ. 1

Peptide (origin) HIV1 (HIVpol968-976) HIV2 (HIVenv 747-755) HIV3 (HIVgag U-159) HIV4 (HIVgag 74-82) HIV5 (HIVgag 77-85) RT (HIVpol510-518) Mel-l (Protein MZ2-E)

Sequence LLWKGEGAV RLVNGSLAL TLNAWVKVV ELRSLYNTV SLYNTVATL ILKEPVHGV ILESLFRAV

A2-binding” 6 1.4 1.7

n.b.” 3 3


Orientation Distance (A)” of side (P2-P9) chain(p9)’ 17.85

1

18.32 15.18 20.80 16.50 16.88 13.27

i I 1 1

” Half-life of dissociation of /2m in I hr at 37°C (adopted from the data of Parker et al., 1992). * The distance between P2 and P9 and the orientation of the side chain at P9 were determined by the MD simulation method as described in Fig. 1. ’ Arrows indicate orientation of the P9 residueswith respect to the floor of the binding groove of the class I. ” n.b.. No binding.

224

J.-S. LIM et al.

seems to be important for constituting the potential binding affinity of the peptide to the class I binding groove. The fact that HIV4 has a unique glycosylation site at the P7 (Asn) residue which points up toward the solvent and orients to the vicinity of the P9 residue suggests that the P7 residue may affect the configuration of the side chains of P9 (Val). Recent reports also suggest that the P3 residue of the binding peptides is especially important, and basic amino acid residues like Arg as in this peptide may reduce the binding affinity (Parker et al., 1992; Ruppert et al., 1993).

Next, the question of whether the structure determined by the MD simulation method has a definite peptide backbone length was examined. Simulation results showed that the distances between Cr of the P2 residue and Cn of the P9 residue for six positive binding peptides appeared to be 15 p\ or longer. Of particular interest is an unusual feature of the HIV4 peptide in which the peptide backbone is kinked in a slightly different pattern. The Mel-l peptide, which has a distance of 13.27 A between the two anchor residues, possesses at least a lofold lower binding affinity than other binding peptides (Table 1). This suggests that a short distance reflect a restriction for a peptide in binding to a class I molecule. Taken together, requirements for the peptides binding to HLA-A2 molecules as shown on a molecular modelling are: (1) the side chain of C-termini (P9) be oriented toward the binding pockets of the HLA-A2 molecules; and (2) the distance between Cccof the P2 residue and Ccc of the P9 residue be longer than 14 A. Prediction of’ hepatitis B virus antigenic epitopes The MD simulation method was applied to seven nonamer peptides containing amino acid sequences of HBV proteins which were pre-selected on the basis of the previously described characteristics of HLA-A2. l-specific motifs. In this MD simulation, where the candidate peptides were subjected to binding to the modeled HLAA2. I molecule, only nonamer peptides were examined, since the variable lengths of peptides could not be fitted to available atomic coordinates of the nonamer peptide. For all peptides tested the side chain orientations at Pl were nearly identical, and the P2 residues seemed to be located near the B pocket in the groove of the %l/a2 domains (Fig. 2). Similarly, the P9 residues had an orientation pointing down toward the fi sheet platform, with the exception of HBs 131-139 in which the side chain of the P9 residue was facing the solvent. When the distance between the P2 and P9 anchor residues was calculated six peptides showed distances ranging from 16 to 21 A (Table 2). In the HBs 272-280 peptide the backbone appeared to have the unusually short distance of 13.86 A between the P2 and P9 residues. To investigate potential peptide-associated conformational changes in class 1 molecules induced by peptide binding, the model structures obtained by MD simulation were analysed. Small but significant topological differences in ctl/x2 domains were observed after fitting the peptides into the binding grooves (Fig. 3). In

particular, the binding of the HBs 13 1-l 39 peptide caused a loosening in the c(helical conformation of the ~12domain that spans amino acids 151-163. When the peptide HBs 272-280 was forced to fit into the binding groove of c(1/ct2 domains, a distortion in the a2 domain (between amino acids 15 1 and 163) was observed. Similar changes after the binding of peptide HBs 272-280 also occurred in the helical region of the al domain. A conformational change in the al domain after the binding of other peptides was not observed, although the possibility that several side chains in the groove had small torsional differences and changed their positions cannot be excluded. This main chain variation might result in a slight alteration of topology of the upper surface of the binding groove. Binding GQinity determined by measuring the increased expression of’ T2 cell swface MHC cluss I molecules To more directly assess the relationship between the characteristics of modelled peptide structures and their actual binding to the MHC molecule, T2 cells which show substantially decreased expression of HLA-AZ.1 due to defects in the TAP genes involved in peptide transport (Hosken and Bevan, 1990; Cerundolo et al., 1990) were used. In the presence of exogenous presentable peptides, HLA-A2.1 expression significantly increases as empty class I molecules or unstably assembled class I MHC molecules are stabilized through binding of the peptide. The binding affinities of synthetic peptides were determined by measuring the elevation of HLA-A2.1 expression on the T2 cell surface using anti-HLA-A2 antibody BB7.2 after a 3 hr incubation with each peptide (Fig. 4). Results were reproducible over several independent experiments, and also using another conformation-dependent anti-class I antibody, W6/32, which reacts with HLA-A,B,C (data not shown). Of seven synthetic HBV peptides tested, the HBs 270-278, HBs 346G 354 and HBs 382-390 peptides showed the highest affinities to HLA-A2. These three peptides contain a P9 anchor residue pointing down toward the binding pockets and had a distance of 16&21 A between the P2 and P9 anchor residues when examined in MD simulation. Note that induction of peptide-specific and HLA-A2. I -restricted CTLs by synthetic peptide HBs 346 -354 has recently been described, indicating that the peptide is supposed to bind to the HLA-A2.1 molecule with adequate affinity (Nayersina et al., 1993). As expected from the model structure, little increase of HLA-A2.1 expression on T2 was observed in the cases of the HBs 272-280 and HBs 131-139 peptides. These results suggest that the orientation of the P9 anchor residue and the distance between the P2 and P9 residues in the modelled peptide structures function as potential parameters in selecting candidate peptides that specifically bind to HLA-A2.1 molecules. The other two peptides, HBc 100&108 and HBc 107-m 115, did not appear to bind to the HLA-A2.1 molecule on the T2 cell surface. The orientations of the side chains of the P9 anchor residues, however, were similar in that they pointed down toward the binding pockets while the

Selection of class I binding

peptides

Fig. 1. Configuration of the HLA-A2.1 binding peptides determined by molecular dynamics simulation. Expl erimental details are described in Materials and Methods. Peptide configuration was obtained from the !side view, when bound to MHC molecule during the simulation. The amino acid sequences of the peptides used are listed in Table 1. Backbone structure of the bound peptide is represented in blue, and side chains at P2 in red, at P9 in yellow and at other residues in green. In particular, the backbone of HIV4 peptide is coloured in orange. (+ + + , good binder; + + weak binder; i, non-binder)

Fig. 2. Configuration of the HLA-A2.1 binding peptides of hepatitis B virus antigens determined by molf xular dynamics simulation. Configuration of the peptides was obtained by the procedures described in Materials and Methods. The backbones and side chains are coloured in the same way as in Fig. 1.

225

226

J.-S. LIM et al.

Fig. 3. A computerized model depicting association of the peptides of hepatitis B virus antigens with the HLA-A2.1 molecule. The top views of the peptide-MHC complexes are shown. The MHC model was constructed by using crystallographic data of al and ct2 heavy domains only (yellow). Peptides are represented as a backbone structure (red). The carboxy terminus of peptide is located on the left-hand side. The upper region of the MHC molecule show t12helix. The amino and carboxy termini of the MHC are shown on the right-hand side.

Selection of class I binding Table 2. Characteristic of the peptides of hepatitis gen bound to the modelled HLA-AZ.1

Sequence

Peptide HBs131-139 HBs 27G-278 HBs2722280 HBs 346354 HBs382-390 HBc 100-108 HBc 107-115

Distance (A)’ (P2-P9)

B virus anti-

Orientation of side chain(P9)h

LLDPRVRGL

16.23

T

LDYQGMLPV VLLDYQGML

13.86 16.14

i

ILSPFLPLL WLSLLVPFV LLWFHISCL CLTFGRETV

20.71 18.12 17.95 18.11

i i

ClThe distance between P2 and P9 and the orientation of side chain at P9 were determined by the MD simulation method as described in Fig. 2. ’ Arrows indicate orientation of the P9 residues with respect to the floor of the binding groove of the class 1.

distance between the P2 and P9 anchor residues appeared to be 17.95 A for HBc 100-108 and 18.11 A for HBc 107115 (Table 2). Moreover, no significant conformational changes in the al and a2 domains induced by the fitting of the peptides was observed, compared to the conformation of class I molecules resulting from the binding of other high affinity peptides. The reason for this inconsistency in peptide binding assay is not certain, but it should be considered that the anchor residues would not always be responsible for specific binding to HLA-A2.1 molecules (Jameson and Bevan, 1992; Parker et al., 1992). In the case of HBc 100-108, the three side chains of the peptide which contain bulky aromatic or imidazole side chains, namely Trp, Phe and His at P3, P4 and P5, respectively, might be responsible for poor binding to HLA600

,x '3 P 5 8 8 3 e! $i E

500

-

HBs131-139 Hf3s270-276 HBs272-260 H&346-354

-

H&362-390 HEclOO-106 HBc107-115

400

300

t 1

1

Peptide Concentration

10

1IO0

( pg /ml)

Fig. 4. Binding of peptides to the HLA-A2.1 molecule: elevation of HLA-A2.1 expression on T2 cells after incubation with peptides. Cells were washed and incubated with peptides of different concentrations at 37°C for 3 hr, respectively. After washing with PBS, they were incubated with 100 ~1 of diluted GAMIgFITC for 40 min. Surface expression of HLA-AZ.1 was determined by a flow cytometer using anti-HLA-AZ.1 antibody, BB7.2 as described in Materials and Methods. Fluorescence intensity of the cells without peptide treatment was 340.

221

peptides

A2.1. Similarly, it has been reported that the peptide HBc 107-l 15 did not bind to HLA-A2.1 molecules (Ruppert et al., 1993). Nevertheless, these results suggest that peptide structures obtained from MD simulation are, in general, comparable to actual peptide conformations of resulting from interactions between class I MHC and antigenic peptides. DISCUSSION As the crystal structures of human and murine class I molecules, and the sequences of endogenous peptides bound by these molecules have become available, efforts have been made in the last few years to predict antigenic epitopes of endogenous origin in several ways. A large amount of knowledge about prominent anchor residues was obtained from the results of sequencing of peptides eluted from purified class I molecules (Falk et al., 199 1b; Hunt et al., 1992). Similarly, the determination of binding affinity by class I MHC reconstitution assays was useful to formulate an algorithm which can be used to predict potential epitope peptides from the primary structure of antigenic

proteins

(Parker

et al., 1992). It has recently

been described that the HLA-A2.1 binding capacity of peptides could be tested on the basis of the binding inhibition of a radiolabelled standard peptide to detergentsolubilized MHC molecules (Ruppert ef al., 1993). Apart from a direct binding assay, a computer-aided scoring program deduced from the HLA-A2. l-binding motif, or the pattern search method, has also been used to predict the epitopes that are included in longer antigenic peptides (Hobohm and Meyerhans, 1993; Nijman et al., 1993). Given that the most common strategy to find class I-restricted peptide motifs used to be the mapping of antigenic regions with synthetic peptides and the substitution of individual amino acid residues, it would be useful to make a reliable tool for prediction of the binding affinity of peptides. In this study a computer model of the HLA-A2.1 molecule, using X-ray crystallography data, was constructed and tested as to whether the analysis of the relationship between the model structure of the peptide from molecular dynamics (MD) simulation, and the actual binding affinity provides a reliable prediction for peptide binding. A set of peptides of known binding affinity to HLA-A2.1, which were nine amino acid residues long were used, since the variable length of the peptide did not fit available atomic coordinates of the nonamer peptide. The results suggest that those peptides which bound most tightly to HLA-A2.1 preferentially had anchor residues at P2 and P9, and the distance between the anchor residues appeared to be 15 A or longer. Additionally, the side chain orientation of anchor residues seemed to play an important role in serving as primary contact points between the class I binding site and the peptide. In nonor weak-binding peptides the side chain orientation of the P9 residue or distances between the P2 and P9 anchor residues appeared to be distinct, compared with those of peptides of high binding affinities. They had side chains of P9 residues pointing toward the solvent. not toward

228

J.-S. LIM et ul.

the binding pocket, and they had an unusually short distance between the P2 and P9 residues. The structural natures of peptides bound by a particular allele of class I molecules obtained from co-crystallized class 1 molecules with bound peptides, were reported in the case of the vesicular stomatitis virus peptide and the Sendai virus peptide complexed with H-2Kh (Fremont et al., 1992; Matsumura et ccl., 1992). Another report described that differences in the structure and antigenicity caused by changes in the sequences of class Ibound peptides could be analysed (Madden et al., 1993). They prepared several different crystals of HLA-AZ.1 molecules complexed with a decameric or nonameric viral peptide, and compared the structures of the peptideHLA-A2.1 complexes. The structure of a peptide RTl of H JVpol, bound to HLA-AZ. 1, appeared to have a similar configuration to the structure obtained from the MD simulation method in this report. The heavy atom root mean square (rms) deviation of PI-P9 distance between the crystal structure (Madden et al., 1993) and the modelled peptide (this report) of RTI appeared to be only 0.97 A, and the backbone rms deviation of PI-P9 distance was 2.1 I A. The distance between P2 and P9 of the crystal structure of RTI peptide was 1.26 A longer than that of the model, but the orientation of the side chains of the two structures were similar. Notable difference between the crystal and the modelled structures was the configuration of P5 (Pro); P5 of the crystal structure was of cis-configuration whereas that of the model appeared to be of tuans-configuration. These results suggest that the modelled structure represents the actual structure of the peptide bound to HLA-A2.1 molecule. With respect to the side chain preferences of bound peptides, it was suggested that interactions favoring tight peptide binding might also require a particular sequencedependent peptide main chain conformation. In addition, it was reported that the predictability of HLA-A2.1 -binding motifs could be improved by considering the sequence dependence of secondary anchors. In fact, it has been shown that consideration of secondary anchor preferences increases the prediction of HLA-A2.1 binding epitopes by approximately 70% (Ruppert et al., 1993). In nonameric peptides which are able to bind HLA-A2.1, amino acids with aromatic side chains at PI, P3 or P5 can function as HLA-A2 secondary anchor residues. By analysing a large collection of peptides it was confirmed that charged residues at these positions are responsible for most negative effects on binding, while hydrophobic or aromatic residues appeared among good binders. However, although the predictability of binding peptides could be improved by considering the secondary anchors, in addition to analysis on the basis of their P2 and P9 anchor motifs, it is very likely that, without taking into account of conformational changes induced by individual peptides, this kind of prediction would remain unsatisfactory. Moreover, the co-presence of both favoured and unfavoured residues would make it difficult to predict the binding affinity of the peptide. Among nonameric peptides tested in this study, five peptides had aromatic side chains at Pl, P3 or P5 as

potential secondary anchors. However, in spite of favoured residues at P3 the two peptides HBs 2722280 and HBc 100-l 08 turned out to be non-binders in the binding assay using T2 cells. This discrepancy between the actual binding affinity and the approximation based on theoretical binding of several peptides might be due to a peptide-induced conformational change in HLA-A2. I, which can be seen in the model structure. This is supported by the antigenic identity of peptide-MHC complexes in that the ability of the side chains to exploit several interactions with the class I binding site depends on the overall sequence of the presented peptides (Madden et al., 1993). Note that two of the HLA-A2.1 binding peptides examined in this work were recognized by CTL prepared from HBV carrier and thus the CTL destroyed the peptidepulsed T2 cells in vitro (Lee et al., manuscript in preparation). Currently available crystallographic data indicate that the overall structures of the binding grooves are similar to one another for all peptide-MHC complexes, and they adjust only slightly to different peptides (Madden et ul., 199 1, 1992; Fremont et al., 1992; Guo et al., 1992; Silver et al., 1992; Zhang et al., 1992). Recently, it has been proposed that MHC binding of antigenic peptides could induce changes in the structure of the binding cleft itself, and the modes of change depend on the sequence of the binding peptide (Bluestone et ul., 1992). When conformational changes in the 2 1/a2 domains of class I molecules were examined after fitting of a peptide through MD simulation, partial distortion was observed in a specific region: for example, in the cc-helical region consisting of amino acids 151-163. These conformational changes in MHC were especially obvious in the MHCpeptide complexes of peptides HBs 13 l- 139 and HBs 2722280, which are not able to bind HLA-A2.1. Furthermore, these peptides appeared to adopt strikingly different contacts with MHC side chains, compared with other peptides located in the central part of the peptide-binding groove. The main chain of the binding cleft was little affected by binding peptides, although changes in individual MHC side chains were observed. Peptide-induced conformational changes in MHC may contribute to the uniqueness of the complex recognized by TcR. Recent studies have demonstrated that changes at inaccessible positions of peptide residues can alter T cell recognition through the introduction of conformational changes in the peptideeMHC complex (Chen PI ul., 1993). The xlhelix has more potential peptide binding sites than the x2-helix. and the ct2-helix has more sites which potentially interact with the TcR than the al-helix (Parham et ul., 1988). Based on these observation, the conformational changes in the x2-helix by peptide binding, if the binding occurs, might disrupt recognition of the MHC--peptide complex by TcR. Moreover, considering that the x2-helix has many more conserved residues than the xl-helix, limited conformational changes after peptide binding might be allowed for TcR recognition. Since the immune response, however, is ultimately guided by the uniqueness of an individual peptide bound to a particular class I molecule, we favour the notion that the structure of the

Selection of class 1 binding

bound peptide is the essential determinant of the antigenie identity of the complex. In the determination of class I binding epitope sequences, a direct binding assay with massive peptide synthesis is cumbersome and time-consuming. The prediction, which depends only on statistical analysis, so far does not seem to be reliable. The determination of binding peptide sequences by crystallographic analysis requires crystallization of MHC molecules containing a single type of peptide and should be limited to a specific area of analysis. From this point of view it may be possible that a structural model by MD simulation provides a convenient screening tool for a large pool of candidate peptides in combination with general prediction schemes. In addition, it should be possible to use MD stimulation to investigate structural changes of the peptide-binding groove resulting from an interaction between the peptide and HLA-A2 molecules. Since this study is limited to HLA-A2.1, accumulation of data from crystallographic analysis would be necessary for further study with other alleles of HLA. Taken together, these MD simulation results suggest that the side chain orientation of P9, the distance between the two anchor residues and conformational changes in the peptide-binding groove induced by peptide-binding function as important parameters to determine whether potentially antigenic peptides will bind to HLA-A2.1 molecules.

REFERENCES Bjorkman P. J., Saper M. A., Samraoui B., Bennett W. S., Strominger J. L. and Wiley D. C. (1987a) Structure of the human class I histocompatibility antigen, HLA-AZ. Nature 329,506-512.

Bjorkman P. J., Saper M. A., Samraoui B., Bennett W. S., Strominger J. L. and Wiley D. C. (1987b) The foreign antigen binding site and T cell recognition regions of class 1 histocompatibility antigens. Na/urr 329, 5 12-5 18. Bluestone J. A., Jameson S., Miller S. and Dick R. 1. (1992) Peptide-induced conformational changes in class I heavy chains alter major histocompatibility compex recognition. J. esp. Med. 176, 1757-1761 Cerundolo V., Alexander J., Anderson K., Lamb C., Cresswell P., McMichael A., Gotch F. and Townsend A. (1990) Presentation of viral antigen controlled by a gene in the major histocompatibility complex. Nuture 345,449.452. Chen B. P. and Parham P. (1989) Direct binding of influenza peptides to class I HLA molecules. Nature 337, 743-745. Chen W., McCluskey J., Rodda S. and Carbone F. R. (1993) Changes at peptide residues buried in the major histocompatibility complex (MHC) class I binding cleft influence T cell recognition: a possible role for indirect conformational alterations in the MHC class I or bound peptide in determining T cell recognition. J. e.~p. Med. 177, 869-873.

Corr M., Boyd L. F., Frankel S. R., Kozlowski S., Padlan E. A. and Margulies D. H. (1992) Endogenous peptides of a soluble major histocompatibility complex class I molecule, H-2Ld: sequence motif, quantitative binding, and molecular modeling of the complex. J. exp. Med. 176, 168 IL1 692. Falk K., Riitzschke O., Deres K., Metzger J., Jung G. and Rammensee H.-G. (1991a) Identification of naturally pro-

peptides

229

cessed viral nonapeptides allows their quantification in infected cells and suggests an allele-specific T cell epitope forecast. J. esp. Med. 174,425-434. Falk K., RGtzschke O., Stefanivic S., Jung G. and Rammensee H.-G. ( I99 I b) Allele-specific motifs revealed by sequencing of self-peptides eluted from MHC molecules. Nature 351, 290-296.

Fremont D. H.. Matsumura M., Stura E. A.. Peterson P. A. and Wilson I. A. (1992) Crystal structures of two viral peptides in complex with murine MHC class I H-2Kh. Science 257, 919~ 921. Garrett T. P. J, Saper M. A., Bjorkman P. J., Strominger J. L. and Wiley D. C. (1989) Specificity pockets for the side chains of peptide antigens in HLA-Aw68. Nature 342,692-696. Guo H.-C., Jardetzky T. S., Garett T. P. J.. Lane W. S.. Strominger J. L. and Wiley D. C. (1992) Different length peptides bind to HLA-Aw68 similarly at their ends but bulge out in the middle. Nuturr 360, 364366. Hobohm U. and Meyerhans A. (1993) A pattern search method for putative anchor residues in T cell epitopes. Eur. J. Immun. 23, 1271~1276. Hosken N, A. and Bevan M. J. (1990) Defective presentation of endogenous antigen by a cell line expressing class I molecules. Science

248,

367-370.

Hunt D. F., Henderson R. A., Shabanowitz J., Sakaguchi K., Michel J.. Sevilir N., Cox A. L., Appella E. and Engelhard V. H. (1992) Characterization of peptides bound to the class I MHC molecule HLA-A2.1 by mass spectrometry. Science 255, 1261.~1263. Jameson S. C. and Bevan M. J. (1992) Dissection of major histocompatibility complex (MHC) and T cell receptor contact residues in a K”-restricted ovalbumin peptide and an assessment of the predictive power of MHC-binding motifs. Eur. J. Irwmn. 22, 2663-2667. Jardetzky T. S.. Lane W. S., Robinson R. A., Madden D. R. and Wiley D. C. (199 I) Identification of self peptides bound to purified HLA-A27. Nature 353, 326.-329. Latron F., Pazmany L.. Morrison J., Moots R., Saper M. A., McMichael A. and Strominger J. L. (1992) A critical role for conserved residues in the cleft of HLA-A2 in the presentation of a nonapeptide to T cells. Science 257, 964-967. Madden D. R.. Garboczi D. N. and Wiley D. C. (1993) The antigenic identity of peptide-MHC complexes: a comparison of the conformations of five viral peptides presented by HLAA2. Cull 75, 693-708. Madden D. R., Gorga J. C., Strominger J. L. and Wiley D. C. (1991) The structure of HLA-B27 reveals nonamer selfpeptides bound in an extended conformation. Nuture 353, 321-325.

Madden D. R., Gorga J. C., Strominger J. L. and Wiley D. C. (1992) The three-dimensional structure of HLA-B27 at 2. I A resolution suggests a general mechanism for tight peptide binding to MHC. Cell 70, 1035-1048. Matsumura M.. Fremont D. H., Peterson P. A. and Wilson I. A. (1992). Emerging principles for the recognition of peptide antigens by MHC class I molecules. Science 257, 927-934. Nayersina R., Fowler P., Guilhot S., Missale G., Cerny A., Schlicht H.-J., Vitiello A., Chesnut R., Person J. L., Redeker A. G. and Chisari F. V. (1993) HLA A2 restricted cytotoxic T lymphocyte responses to multiple hepatitis B surface antigen epitopes during hepatitis B virus infection. J. hnmun. 150, 46594671.

Nijman H. W., Houbiers J. G. A., Vierboom M., van der Burg S. H.. Drijfhout J. W., D’Amaro J., Kenemans P., Melief C. J. M. and Kast W. M. (1993) Identification of peptide

230

J.-S. LIM et al.

sequences that potentially trigger HLA-A2.1-restricted cytotoxic T lymphocytes. Eur. J. Immun. 23, 1215-1219. Parham P., Lomen C. E., Lawlor D. A., Ways J. P., Holmes N., Coppin H. L., Salter R. D., Wan A. M. and Ennis P. D. (1988) Nature of polymorphism in HLA-A, -B, and -C molecules. Proc. natn. Acad. Sci. U.S.A. 85,40054009. Parker K. C., Bednarek M. A., Hull L. K., Utz U., Cunningham B., Zweerink H. J., Biddison W. E. and Coligan J. E. (1992) Sequence motifs important for peptide binding to the human MHC class I molecule, HLA-AI?. J. Zmmun. 149, 358C3587. Rotzschke O., Falk K., Deres K., Schild H., Norda M., Metzger J., Jung G. and Rammensee H.-G. (1990) Isolation and analysis of naturally processed peptides as recognized by cytotoxic T cells. Nature 348,252-254. Ruppert J., Sidney J., Celis E., Kubo R. T., Grey H. M. and Sette A. (1993) Prominent role of secondary anchor residues

in peptide binding to HLA-A2.1 molecules. Cell 74, 929937. Saper M. A., Bjorkman P. J. and Wiley D. C. (1991) Refined structure of the human histocompatibility antigen HLA-A2 at 2.6 A resolution. J. molec. Biol. 219, 277-3 19. Silver M. L., Guo H.-C., Strominger J. L. and Wiley D. C. (1992) Atomic structure of a human MHC molecule presenting an influenza virus peptide. Nature 360, 367-369. Van Bleek G. M. and Nathenson S. G. (1990) Isolation of an immunodominant viral peptide from the class I H-2Kh molecule. Nature 348, 2 13-2 16. Zhang W., Young A. C., Imarai M., Nathenson S. G. and Sacchettini J. C. (1992) Crystal structure of major histocompatibility complex class I H-2Kh molecule containing a single viral peptide: implication for peptide binding and Tcell receptor recognition. Proc. natn. ,4cad. Sci. U.S.A. 89, 8403-8407.