JOURNAL OF MOLECULAR RECOGNITION, VOL. 4,7-15 (1991)
Binding of the Neuroleptic Drug Haloperidol to a Monoclonal Antibody: Refinement of the Binding Site Molecular Model Using Canonical Structures? Jerry M. Anchin, Shankar Subramaniam and D. Scott Linthicum* Department of Veterinary Pathobiology, College of Veterinary Medicine, Texas A and M University, College Station,
TX 77843-4467, USA and Department of Chemistry, Princeton University, Princeton, NJ 08544, USA
The ligand binding site of a monoclonal antibody (185), which binds the neuroleptic drug haloperidol, has been modelled using canonical structures and energy minimization techniques. This refined modelling protocol has allowed us to predict the variable region loop conformations. Three key residues, H:50(W), H:lOOa(D)and L W Y )appear to create the basis of the electrostatic, n-n stacking interactionsand hydrogen bonding required for the high affinity binding site characteristics present in this antibody. The use of computer-aided graphics techniques and appropriate three-dimensional modeUing permits inspection of the predicted molecular recognition features of the ligand binding site.
~~
~
INTRODUCTION Much of our understanding of immunoglobulin (Ig) structure and function comes from numerous chemical, genetic and biophysical studies, but crystallization and x-ray diffraction analyses of Fab fragments derived from monoclonal antibodies (mAb) and myeloma proteins have provided the most detailed molecular picture of the antigen binding site architecture (Poljak et al., 1973; Poljak, 1975; Davies and Metzger, 1983; Padlan and Davies, 1975; Kabat et al., 1977; Alzari et al., 1988). Antibody molecules are heterodimeric proteins composed of two heavy (H) and two light (L) polypeptide chains. Assembly of the whole Ig molecule involves both covalent (disulfide) and non-covalent (hydrogen bonds and hydrophobic) interactions between the H and L chains. Each peptide chain contains an N-terminal variable region (Fv), which is composed of three hypervariable or complementarity determining regions (CDRs) and four intervening framework regions (FR). The Fv regions form several antiparallel 6-sheets which are folded as a 6-sandwich. The antigen binding site is built from turns and surface loops which connect the antiparallel 6-sheets. These loop structures coincide with the hypervariable regions and are termed complementarity determining regions because of their contact with antigen epitopes. Comparative studies performed on empirically solved Fab structures have revealed that the conformations of the peptide backbone of the H and L chain variable regions tend to be highly conservative. There also appear to be a limited number of CDR main chain conformations found in the known structures. Chothia
’
Supported by grants from the Smokeless Tobacco Research Council (STRC-0136) and The American Heart Association (89-1066). Author to whom correspondence should be addressed at the Texas A and M University. 0952-3499/91/01oOo7-09 $05.OO 0 1991 by John Wiley & Sons, Ltd.
and Lesk (1987) and Chothia et al. (1989) identified particular amino acid residues that through their packing, hydrogen bonding or their ability to assume or x conformations are responsible for unusual rp, creating the backbone conformations of the CDRs. These CDR conformations were identified as being ‘canonical’and the use of canonical CDRs in computeraided modelling algorithms has been useful in predicting the CDR structures of several antibodies which had not been solved empirically. For example, using canonical CDR structures as a ‘parent’ scaffolding, Chothia et al. (1989) correctly predicted the conformations of five of the six CDR loops found in each of the following immunoglobulins: HyHEL-5, HyHEL-10, NC41 and NQ10. The root mean square (rms) difference in the atomic position of the main chain atoms, after optimal superposition, between predicted and observed CDRs ranged between 0.3 and 1.3 A. Another approach used to successfully model the antigen binding site uses a conformational search algorithm CONGEN developed by Bruccoleri et al. (1988). This algorithm samples the complete conformational space of a polypeptide chain segment and calculates low energy loop configurations. The lowest free-energy conformation should correspond to the naturally occurring one. This technique was able to correctly predict the conformations of the CDR loops of both McPC603 and HyHEL-5. The results for HyHEL-5 showed rms deviations of 0.5-2.1 A for backbone atoms and rms deviations of 1.7-4.1 A for all atoms. McPC603 had rms deviations of 0.7-2.6 A for backbone atoms and a deviation of 1.4-3.3 8, for all atoms. MAbs have proved useful as mimics of receptors in the study of the interactions that occur between a ligand and its natural biological receptor. Ronayne et al. (1990) used mAbs specific for the antidepressant drug imipramine, which binds to the muscarinic receptor, as a model of the receptor binding site. Glasel et al. (1983; Glasel, 1989) used mAb to morphine to study Received 22 January 1991 Accepted I9 May 1991
8
J. M. ANCHIN, S. SUBRAMANIAM AND D. S. LINTHICUM
residues. Examination of CDR L:l in mAb185 revealed that its size and key residues closely match that of human myeloma REI. Using this procedure we were able to choose an experimentally determined canonical structure for each loop (except H3). In the present study, we modelled the H and L chain CDRs in the order proposed by Bruccoleri ef al. (1988): L2, H1, L3, H2, H3, L1. Modelling the mAb in this sequence first modifies the loops that do not interact with others, thereby building the base scaffolding upon which the remaining loops are modelled. Random order modelling of the CDRs resulted in excessively high potential energies due to poor steric interactions between atoms in the CDR peptide backbones and side chains. Interactive graphics modelling was carried out on a Silicon Graphics IRIS 4D-220GTX workstation using QUANTA software (version 3.0) (Polygen Corp, Waltham, MA, USA); energy refinements were carried out using CHARMM developed by Brooks et al. (1983). Computer-aided stereo docking of the ligand was accomplished using continuous CHARMM energy functions. Two dimensional plots of the structures were made using the ARTIST program (Quanta version 3.0) for the IRIS workstation. The ‘parent’ structures (REI, Hy HEL-5, McPC603) have been previously solved by x-ray crystallography with a resolution below 3.0 A and the three-dimensional atomic coordinates for these structures were obtained from the Brookhaven Protein Data Bank (Bernstein et al., 1977). The kappa L chain from the human myeloma RE1 and the H chain from the mouse Ig HyHEL-5 were used as the ‘parent’ canonical structures for the modelling of mAb 185. In one instance the ‘parent’ model (HyHEL-5) did not possess the proper H3 region (in terms of size and sequence) and an alternative H3 ‘parent’ (McPC603) was chosen for the modelling. The amino acid homology between RE1 and mAb185 was 64% when all L chain residues were compared. The homology between HyHEL-5 and mAb185 H chain was 78%, excluding CDR H3. The protein database files containing the Fab structures were edited to EXPERIMENTAL remove the constant domains of the H and L chain; this file contains only the Fv structure (V, residues 1-107 and VH reidues 1-113). Polar hydrogens are not Monoclonal antibodies to the neuroleptic drug haloper(4‘-fluoro-4-(4-hydroxy-4-p-cholorophenyl- included in the PDB files and were added to the Fv idol structures using the CHARM-build hydrogen facility. piperidino)butyrophenone were developed in our Hydrogens are initially classified according to their laboratory using standard hybridoma cloning protocols respective environments and an optimum position is (Kussie et al., 1989). For the study presented herein, found taking into account the energy contributed by the we refined the computer-aided modelling of the van der Waals term, the electrostatic term, and the mAb185 (anti-haloperidol). This mAb had been predihedral term derived from the full energy expression viously studied by Sherman et al. (1986) using quantitative structure/activity analysis (QSAR), endogenous of CHARMM. The dihedral angle which yields the fluorescence quenching and computer-aided modelling lowest energy is chosen and the hydrogens are placed of the Fv regions. Sequence data for the heavy and light using this optimum dihedral angle. Harmonic conchain variable regions has been published (Sherman straints were applied to fix the entire protein backbone and Bolger, 1988). structure and the Fv fragment was then subjected to a The procedure we used for modelling the CDR loops CHARMM energy minimization procedure, which generally followed the strategies set forth by Chothia consists of 100 steps Steepest Descents, 100 steps and Lesk (1987) and Chothia et al. (1989). We exaConjugate Gradient and as many steps of mined the size of each CDR loop and determined Adopted-Basis Newton-Raphson as is needed to bring whether it was homologous to one of a known crystal the structure to convergence. The potential energy of structure. We then examined each CDR sequence to RE1 was reduced from an initial value of +974 to see whether it contained the set of residues responsible -3104 kcal/mol, and HyHEL-5 from +3289 to -1643 for giving the loop its particular conformation. For kcal/mol. The L chain alpha carbons of RE1 and HyHEL-5 example, Chothia and Lesk define four types of canoniwere overlapped using a least-square fitting technique cal structures for loop L:l based on size and key
the molecular features thought to be present in the opioid p recetpro. Studies carried out by Sherman ef al. (1986) Sherman and Bolger (1988) showed that predictions could be made about the pharmacophore of the neuroleptic drug haloperidol, which is an antagonist for the D-2 dopaminergic receptor, by examination of ligand/mAb interactions. Quantitative structure/ activity relationship (QSAR) studies indicated that mAb185 binds the folded axial conformer of haloperidol, and that the primary antigenic feature was the fluorophenyl ring of the drug. The antibody combining site was predicted to have a deep binding pocket for the ligand. Ligand induced quenching of internal tryptophan fluorescence was observed experimentally and it was predicted that if the fluorophenyl ring was proximal to a tryptophan residue this could account for the quenching. Sequencing of mRNA and computer-aided modelling of the mAb185 revealed a tryptophan residue in CDR H:2 that would allow haloperidol to be involved in a n-n interaction. Since mAb185 was first modelled (Sherman and Bolger, 1988), the number of Fab structures for which three-dimensional structures have been adequately resolved has greatly increased. There are now crystallographic coordinates for more than 10 Fab fragments and 3 light chain dimers, of which five have been cocrystallized with bound antigen (Amit et al., 1986; Colman et al., 1987; Padlan et al., 1989; Segal et al., 1974; Sheriff et al., 1987). Using the canonical structures represented in the expanded database of known structures, in addition to computational energy minimization procedures, we have refined the original model of mAb185. The refined model has allowed us to examine further the supramolecular interactions that are predicted to occur between the ligand and receptor site. We have now identified three key residues in the binding site which appear to account for the majority of the ligand binding energetics and specificity.
~~
MOLECULAR MODELLING OF ANTI-HALOPERIDOL ANTIBODY
to optimize the overlap between the two structures. A fit of the C a atoms of residues 1through 105 gave a rms difference in the atomic positions of
Binding of the Neuroleptic Drug Haloperidol to a Monoclonal Antibody: Refinement of the Binding Site Molecular Model Using Canonical Structures? Jerry M. Anchin, Shankar Subramaniam and D. Scott Linthicum* Department of Veterinary Pathobiology, College of Veterinary Medicine, Texas A and M University, College Station,
TX 77843-4467, USA and Department of Chemistry, Princeton University, Princeton, NJ 08544, USA
The ligand binding site of a monoclonal antibody (185), which binds the neuroleptic drug haloperidol, has been modelled using canonical structures and energy minimization techniques. This refined modelling protocol has allowed us to predict the variable region loop conformations. Three key residues, H:50(W), H:lOOa(D)and L W Y )appear to create the basis of the electrostatic, n-n stacking interactionsand hydrogen bonding required for the high affinity binding site characteristics present in this antibody. The use of computer-aided graphics techniques and appropriate three-dimensional modeUing permits inspection of the predicted molecular recognition features of the ligand binding site.
~~
~
INTRODUCTION Much of our understanding of immunoglobulin (Ig) structure and function comes from numerous chemical, genetic and biophysical studies, but crystallization and x-ray diffraction analyses of Fab fragments derived from monoclonal antibodies (mAb) and myeloma proteins have provided the most detailed molecular picture of the antigen binding site architecture (Poljak et al., 1973; Poljak, 1975; Davies and Metzger, 1983; Padlan and Davies, 1975; Kabat et al., 1977; Alzari et al., 1988). Antibody molecules are heterodimeric proteins composed of two heavy (H) and two light (L) polypeptide chains. Assembly of the whole Ig molecule involves both covalent (disulfide) and non-covalent (hydrogen bonds and hydrophobic) interactions between the H and L chains. Each peptide chain contains an N-terminal variable region (Fv), which is composed of three hypervariable or complementarity determining regions (CDRs) and four intervening framework regions (FR). The Fv regions form several antiparallel 6-sheets which are folded as a 6-sandwich. The antigen binding site is built from turns and surface loops which connect the antiparallel 6-sheets. These loop structures coincide with the hypervariable regions and are termed complementarity determining regions because of their contact with antigen epitopes. Comparative studies performed on empirically solved Fab structures have revealed that the conformations of the peptide backbone of the H and L chain variable regions tend to be highly conservative. There also appear to be a limited number of CDR main chain conformations found in the known structures. Chothia
’
Supported by grants from the Smokeless Tobacco Research Council (STRC-0136) and The American Heart Association (89-1066). Author to whom correspondence should be addressed at the Texas A and M University. 0952-3499/91/01oOo7-09 $05.OO 0 1991 by John Wiley & Sons, Ltd.
and Lesk (1987) and Chothia et al. (1989) identified particular amino acid residues that through their packing, hydrogen bonding or their ability to assume or x conformations are responsible for unusual rp, creating the backbone conformations of the CDRs. These CDR conformations were identified as being ‘canonical’and the use of canonical CDRs in computeraided modelling algorithms has been useful in predicting the CDR structures of several antibodies which had not been solved empirically. For example, using canonical CDR structures as a ‘parent’ scaffolding, Chothia et al. (1989) correctly predicted the conformations of five of the six CDR loops found in each of the following immunoglobulins: HyHEL-5, HyHEL-10, NC41 and NQ10. The root mean square (rms) difference in the atomic position of the main chain atoms, after optimal superposition, between predicted and observed CDRs ranged between 0.3 and 1.3 A. Another approach used to successfully model the antigen binding site uses a conformational search algorithm CONGEN developed by Bruccoleri et al. (1988). This algorithm samples the complete conformational space of a polypeptide chain segment and calculates low energy loop configurations. The lowest free-energy conformation should correspond to the naturally occurring one. This technique was able to correctly predict the conformations of the CDR loops of both McPC603 and HyHEL-5. The results for HyHEL-5 showed rms deviations of 0.5-2.1 A for backbone atoms and rms deviations of 1.7-4.1 A for all atoms. McPC603 had rms deviations of 0.7-2.6 A for backbone atoms and a deviation of 1.4-3.3 8, for all atoms. MAbs have proved useful as mimics of receptors in the study of the interactions that occur between a ligand and its natural biological receptor. Ronayne et al. (1990) used mAbs specific for the antidepressant drug imipramine, which binds to the muscarinic receptor, as a model of the receptor binding site. Glasel et al. (1983; Glasel, 1989) used mAb to morphine to study Received 22 January 1991 Accepted I9 May 1991
8
J. M. ANCHIN, S. SUBRAMANIAM AND D. S. LINTHICUM
residues. Examination of CDR L:l in mAb185 revealed that its size and key residues closely match that of human myeloma REI. Using this procedure we were able to choose an experimentally determined canonical structure for each loop (except H3). In the present study, we modelled the H and L chain CDRs in the order proposed by Bruccoleri ef al. (1988): L2, H1, L3, H2, H3, L1. Modelling the mAb in this sequence first modifies the loops that do not interact with others, thereby building the base scaffolding upon which the remaining loops are modelled. Random order modelling of the CDRs resulted in excessively high potential energies due to poor steric interactions between atoms in the CDR peptide backbones and side chains. Interactive graphics modelling was carried out on a Silicon Graphics IRIS 4D-220GTX workstation using QUANTA software (version 3.0) (Polygen Corp, Waltham, MA, USA); energy refinements were carried out using CHARMM developed by Brooks et al. (1983). Computer-aided stereo docking of the ligand was accomplished using continuous CHARMM energy functions. Two dimensional plots of the structures were made using the ARTIST program (Quanta version 3.0) for the IRIS workstation. The ‘parent’ structures (REI, Hy HEL-5, McPC603) have been previously solved by x-ray crystallography with a resolution below 3.0 A and the three-dimensional atomic coordinates for these structures were obtained from the Brookhaven Protein Data Bank (Bernstein et al., 1977). The kappa L chain from the human myeloma RE1 and the H chain from the mouse Ig HyHEL-5 were used as the ‘parent’ canonical structures for the modelling of mAb 185. In one instance the ‘parent’ model (HyHEL-5) did not possess the proper H3 region (in terms of size and sequence) and an alternative H3 ‘parent’ (McPC603) was chosen for the modelling. The amino acid homology between RE1 and mAb185 was 64% when all L chain residues were compared. The homology between HyHEL-5 and mAb185 H chain was 78%, excluding CDR H3. The protein database files containing the Fab structures were edited to EXPERIMENTAL remove the constant domains of the H and L chain; this file contains only the Fv structure (V, residues 1-107 and VH reidues 1-113). Polar hydrogens are not Monoclonal antibodies to the neuroleptic drug haloper(4‘-fluoro-4-(4-hydroxy-4-p-cholorophenyl- included in the PDB files and were added to the Fv idol structures using the CHARM-build hydrogen facility. piperidino)butyrophenone were developed in our Hydrogens are initially classified according to their laboratory using standard hybridoma cloning protocols respective environments and an optimum position is (Kussie et al., 1989). For the study presented herein, found taking into account the energy contributed by the we refined the computer-aided modelling of the van der Waals term, the electrostatic term, and the mAb185 (anti-haloperidol). This mAb had been predihedral term derived from the full energy expression viously studied by Sherman et al. (1986) using quantitative structure/activity analysis (QSAR), endogenous of CHARMM. The dihedral angle which yields the fluorescence quenching and computer-aided modelling lowest energy is chosen and the hydrogens are placed of the Fv regions. Sequence data for the heavy and light using this optimum dihedral angle. Harmonic conchain variable regions has been published (Sherman straints were applied to fix the entire protein backbone and Bolger, 1988). structure and the Fv fragment was then subjected to a The procedure we used for modelling the CDR loops CHARMM energy minimization procedure, which generally followed the strategies set forth by Chothia consists of 100 steps Steepest Descents, 100 steps and Lesk (1987) and Chothia et al. (1989). We exaConjugate Gradient and as many steps of mined the size of each CDR loop and determined Adopted-Basis Newton-Raphson as is needed to bring whether it was homologous to one of a known crystal the structure to convergence. The potential energy of structure. We then examined each CDR sequence to RE1 was reduced from an initial value of +974 to see whether it contained the set of residues responsible -3104 kcal/mol, and HyHEL-5 from +3289 to -1643 for giving the loop its particular conformation. For kcal/mol. The L chain alpha carbons of RE1 and HyHEL-5 example, Chothia and Lesk define four types of canoniwere overlapped using a least-square fitting technique cal structures for loop L:l based on size and key
the molecular features thought to be present in the opioid p recetpro. Studies carried out by Sherman ef al. (1986) Sherman and Bolger (1988) showed that predictions could be made about the pharmacophore of the neuroleptic drug haloperidol, which is an antagonist for the D-2 dopaminergic receptor, by examination of ligand/mAb interactions. Quantitative structure/ activity relationship (QSAR) studies indicated that mAb185 binds the folded axial conformer of haloperidol, and that the primary antigenic feature was the fluorophenyl ring of the drug. The antibody combining site was predicted to have a deep binding pocket for the ligand. Ligand induced quenching of internal tryptophan fluorescence was observed experimentally and it was predicted that if the fluorophenyl ring was proximal to a tryptophan residue this could account for the quenching. Sequencing of mRNA and computer-aided modelling of the mAb185 revealed a tryptophan residue in CDR H:2 that would allow haloperidol to be involved in a n-n interaction. Since mAb185 was first modelled (Sherman and Bolger, 1988), the number of Fab structures for which three-dimensional structures have been adequately resolved has greatly increased. There are now crystallographic coordinates for more than 10 Fab fragments and 3 light chain dimers, of which five have been cocrystallized with bound antigen (Amit et al., 1986; Colman et al., 1987; Padlan et al., 1989; Segal et al., 1974; Sheriff et al., 1987). Using the canonical structures represented in the expanded database of known structures, in addition to computational energy minimization procedures, we have refined the original model of mAb185. The refined model has allowed us to examine further the supramolecular interactions that are predicted to occur between the ligand and receptor site. We have now identified three key residues in the binding site which appear to account for the majority of the ligand binding energetics and specificity.
~~
MOLECULAR MODELLING OF ANTI-HALOPERIDOL ANTIBODY
to optimize the overlap between the two structures. A fit of the C a atoms of residues 1through 105 gave a rms difference in the atomic positions of
