Comparison of surface accessible residues in human and murine immunoglobulin Fv domains. Implication for humanization of murine antibodies.

Statistical analysis of a database of unique human and murine immunoglobulin heavy chain and light chain variable regions reveals that the precise patterns of exposed residues are different in human and murine antibodies, while most individual surface positions have strong preferences for a small number of residue types. Consideration of these surface patterns alone generates almost identical family groupings for light and heavy chain variable domain sequences to those produced by methods such as those of Kabat et al., where N-terminal framework sequences only are compared, or Tomlinson et al., in which entire variable region nucleotide sequences are used. This unexpected result suggests that the surfaces of V-regions are at least as well conserved as the core framework sequences. Furthermore, using these patterns of human and murine surface residues a novel method for the "humanization" of murine antibodies has been developed and tested.

Antibody design: beyond the natural limits.

Dissection of antibody-antigen interactions requires a knowledge of antibody structure, the ability to model accurately the conformation of antibody-combining sites, and an understanding of the energetic factors governing the interactions. When this understanding has reached the point where the molecular shape and chemical character of a combining site necessary to define a particular specificity and binding requirement can be designed, the antibody repertoire will have been extended 'beyond the natural limits'.

Humanization of murine monoclonal antibodies through variable domain resurfacing.

Two murine monoclonal antibodies, N901 (anti-CD56) and anti-B4 (anti-CD19), were humanized by a process we call "resurfacing." A systematic analysis of known antibody structures has been used to determine the relative solvent accessibility distributions of amino acid residues in murine and human antibody variable (Fv) regions and has shown that the sequence alignment positions of surface amino acids for human and murine variable region heavy (VH) and light (VL) chains are conserved with 98% fidelity across species. While the amino acid usage at these surface positions creates surface residue patterns that are conserved within species, there are no identical patterns across species. However, surprisingly few amino acid changes need to be made to convert a murine Fv surface pattern to that characteristic of a human surface. Resurfacing was used to change the patterns of surface accessible residues in the Fv regions of the N901 and anti-B4 antibodies to resemble those found on the Fv regions of human antibody sequences. Two different procedures for selecting a human sequence were compared. For anti-B4, a data base of clonally derived human VL-VH sequence pairs was used, while for N901, sequences for VL and VH were independently selected from the Kabat et al. data base [Kabat, E. A., Wu, T. T., Reid-Miller, M., Perry, H. M. & Gottesman, K. S. (1991) Sequences of Proteins of Immunological Interest (DHHS, Washington, DC), 5th Ed.]. Resurfaced N901 and anti-B4 antibodies had apparent affinities for their cell surface ligands that were identical to those of their respective parent murine antibodies. These data provide evidence that, despite the differences in the surfaces of mouse and human Fv regions, it is possible to substitute one for the other while retaining full antigen binding affinity.

A comparison of two murine monoclonal antibodies humanized by CDR-grafting and variable domain resurfacing.

The variable domain resurfacing and CDR-grafting approaches to antibody humanization were compared directly on the two murine monoclonal antibodies N901 (anti-CD56) and anti-B4 (anti-CD19). Resurfacing replaces the set of surface residues of a rodent variable region with a human set of surface residues. The method of CDR-grafting conceptually consists of transferring the CDRs from a rodent antibody onto the Fv framework of a human antibody. Computer-aided molecular modeling was used to design the initial CDR-grafted and resurfaced versions of these two antibodies. The initial versions of resurfaced N901 and resurfaced anti-B4 maintained the full binding affinity of the original murine parent antibodies and further refinements to these versions described herein generated five new resurfaced antibodies that contain fewer murine residues at surface positions, four of which also have the full parental binding affinity. A mutational study of three surface positions within 5 A of the CDRs of resurfaced anti-B4 revealed a remarkable ability of the resurfaced antibodies to maintain binding affinity despite dramatic changes of charges near their antigen recognition surfaces, suggesting that the resurfacing approach can be used with a high degree of confidence to design humanized antibodies that maintain the full parental binding affinity. By comparison CDR-grafted anti-B4 antibodies with parental affinity were produced only after seventeen versions were attempted using two different strategies for selecting the human acceptor frameworks. For both the CDR-grafted anti-B4 and N901 antibodies, full restoration of antigen binding affinity was achieved when the most identical human acceptor frameworks were selected. The CDR-grafted anti-B4 antibodies that maintained high affinity binding for CD19 had more murine residues at surface positions than any of the three versions of the resurfaced anti-B4 antibody. This observation suggests that the resurfacing approach can be used to produce humanized antibodies with reduced antigenic potential relative to their corresponding CDR-grafted versions.