(169b) Antibody Affinity Maturation Using Computational Protein Design

In the development of a therapeutic protein, it is often necessary to engineer a high-affinity interaction. Computational protein design and directed evolution provide complementary and potentially synergistic approaches to protein engineering. Computations can in principle search a vastly larger sequence space, albeit with an approximate energy description serving as a surrogate for biochemical function. Directed evolution is limited by experimental library size, with random mutagenesis generally covering most single mutations and only sampling larger combinations of mutations.

We present the affinity maturation of the anti-lysozyme antibody D44.1 using computational protein design. Single mutations predicted to improve binding affinity were rationally designed and then validated experimentally. The design process began with a 2.5 Å crystal structure of the wild-type complex. For each antibody complementarity determining region (CDR) position, mutation to the other 19 amino acids was independently evaluated. The dead-end elimination and A* algorithms were used to search a fixed-backbone, discrete-rotamer conformational space, using an approximate, pairwise-additive energy function. The lowest-energy structures were re-evaluated using more accurate models and energy functions, including Poisson-Boltzmann continuum electrostatics.

Of 9 single mutants tested spanning 7 amino-acid positions, 6 mutants bound tighter than the wild type, with an 8-fold improvement for the best single mutant. Combinations of single mutants could produce larger affinity enhancements, transitioning from the nanomolar (nM) to picomolar (pM) affinity range. Affinity gains predicted through improved electrostatic and solvation effects are emphasized when selecting mutants to test experimentally. Designed interactions include removal of unsatisfied hydrogen-bonding groups and addition of charged residues. The overall procedure is fast and for this case yields a high hit-rate of improvements.

This work was supported by an NSF fellowship to SML and a grant from the NIH (CA096504).