Abstract
Antibody polyspecificity, i. e. the highly specific differential recognition of various epitopes, has been implicated in the onset of autoimmune diseases such as asthma and rheumatism. Little is established about the structural prerequisites that drive this biological phenomenon; only for the monoclonal antibody CB4-1 high resolution X-ray structures of complexes with non-related peptide epitopes are known. The aim of this work is to investigate the applicability of antibody-peptide complex structure prediction by means of molecular modeling for the characterization of polyspecific binding behaviour.
Three monoclonal antibodies were chosen as model systems for complex structure prediction by global stochastic optimization: the anti-p24 (HIV-1) antibody CB4-1 that binds to a number of sequentially related as well as unrelated peptides; the anti-TGFa antibody tAb2 which besides its linear wild type epitope also recognises similar cyclic peptides; and the anti-cholera toxin peptide 3 antibody TE33 whose polyspecificity was only recently demonstrated with ligands derived from a phage display library. The latter one binds not only to non-related linear but also to cyclic peptid epitopes.
In a first step, the conformational landscapes of the linear epitopes recognised by CB4-1 were analysed by modeling the unbound 11-meric peptides. Here, a correlation to the conformations of the antibody-bound species in known complex structures can be identified which leads to assumptions about its influence on the binding affinity. Extending these findings, antibody complexes with linear peptides were modelled and compared with published X-ray crystal structures. The results show that global conformational search simulations in the presence of the antibody yield solutions that only in exceptional cases superimpose tolerably with the target structure. Moreover, convergence of several independent simulations of the same peptide, each starting from a different random conformation, was generally very poor in structural terms. Only for one linear peptide, whose complex structure is yet unknown, repeated Monte Carlo calculations converged in a similar conformational cluster.
Simulation convergence is supposedly hampered by the vast number of degrees of freedom associated with linear peptides. The conformational space of peptides can be substantially reduced by means of cyclization. Therefore, can complex structure prediction of conformationally constrained peptides be achieved with greater verifiability?
The published three-dimensional structure of the antibody tAb2 in complex with a 7-meric lactam-circularized peptid was used to address this question. Again starting from randomly generated conformations, stochastic simulations of this peptide as well as homodetic and heterodetic cyclic homologues yielded structures that were in excellent agreement with the expected one while exhibiting very good convergence.
To validate this result, cyclic peptides with a larger number of residues within the ring structure had to be tested. Since no X-ray structures of antibody complexes with bigger peptide cycles containing all-natural residues are known, a number of cyclic analogues of an 11-meric linear peptide were constructed on-screen. The template peptide has been described in high resolution as exhibiting a quasi-cyclic conformation in the binding pocket of the antibody CB4-1. Complex structure predictions of the six disulfide and lactam-bridged analogues were able to identify the anticipated overall conformation with satisfying reproducibility. In addition, biochemical binding studies provided evidence that the constructed analoga bind to CB4-1 with comparable affinity while using identical functional groups.
The bound conformation of a cyclic peptide can be predicted with higher probability compared to linear entities even when modeling bigger ring sizes. This potential was used to suggest the complex structure of a cyclic 11-meric peptide binding to the antibody TE33 that was identified from a phage display library and shares no sequence similarity with the wild type epitope. The peptide complex structure found repeatedly in unbiased independent simulations displays an excellent consistency with the amino acid substitution pattern. Compared to the published complex structure of the linear wild type peptide, a surprisingly similar overall conformation of the cyclic binder is discernible in spite of very different individual contacts.
In conclusion it can be stated that a reliable structure prediction of antibody complexes by global stochastic optimization is feasible only for conformationally constrained peptides under the condition that no substantial antibody conformation change occurs upon binding. The data provided emphasize that polyspecific binding behaviour of antibodies is comprehensible only on the atomic level. The sequence of the binder merely provides a spatially and sterically restrictive framework of atom groups that can lead to a binding event if the dynamic behaviour is considered. Different peptide sequences can produce similar contact areas thus recognizing similar paratop regions; identical overall conformations of bound peptides do not indicate identical interactions even when the binders are closely related in sequence. |
