The computational algorithm SCHEMA was developed to estimate the disruption caused when amino acid residues that interact in the three-dimensional structure of a protein are inherited from different parents upon recombination. To evaluate how well SCHEMA predicts disruption, we have shuffled the distantly-related -lactamases PSE-4 and TEM-1 at 13 sites to create a library of 2 14 (16,384) chimeras and examined which ones retain lactamase function. Sequencing the genes from ampicillin-selected clones revealed that the percentage of functional clones decreased exponentially with increasing calculated disruption (E ס the number of residue-residue contacts that are broken upon recombination). We also found that chimeras with low E have a higher probability of maintaining lactamase function than chimeras with the same effective level of mutation but chosen at random from the library. Thus, the simple distance metric used by SCHEMA to identify interactions and compute E allows one to predict which chimera sequences are most likely to retain their function. This approach can be used to evaluate crossover sites for recombination and to create highly mosaic, folded chimeras.Keywords: Chimera; lactamase; PSE-4; recombination; schema; TEM-1; directed evolution Numerous protein traits can be improved or altered when recombination is coupled with a screening or selection strategy (for review, see Minshull and Stemmer 1999; Arnold 2001a,b). During laboratory evolution, as in nature, recombination promotes the rapid accumulation of beneficial mutations from multiple parents onto a single offspring (Stemmer 1994b;Moore et al. 1997;Crameri et al. 1998). It also explores a part of sequence space that is particularly rich in folded and functional proteins. Recombination plays a key role in natural evolution of proteins through the swapping of well-defined structural domains . Where a domain structure is not obvious, however, how recombination contributes to the evolution and diversification of protein sequence and function is less well understood.Recently we developed an algorithm, called SCHEMA, for predicting which fragments of homologous proteins can be recombined without disturbing the integrity of the structure (Voigt et al. 2002). Based on the 3D structures of the parent proteins, the algorithm identifies pairs of amino acids that are interacting, defined as those residues within a cutoff distance of 4.5 Å, and determines the net number of interactions broken when a chimeric protein inherits portions of its sequence from different parents (defined as E). Because calculating E for all possible crossover combinations is computationally intractable, it is difficult to identify which crossover locations are optimal with respect to their ability to yield folded chimeras. One version of SCHEMA circumvents this computational difficulty by finding compact, con-