We have developed a generally applicable experimental procedure to fin
d functional proteins that are many mutational steps from wild type. O
ptimization algorithms, which are typically used to search for solutio
ns to certain combinatorial problems, have been adapted to the problem
of searching the 'sequence space' of proteins. Many of the steps norm
ally performed by a digital computer are embodied in this new molecula
r genetics technique, termed recursive ensemble mutagenesis (REM). REM
uses information gained from previous iterations of combinatorial cas
sette mutagenesis (CCM) to search sequence space more efficiently. We
have used REM to simultaneously mutate six amino acid residues in a mo
del protein. As compared to conventional CCM, one iteration of REM yie
lded a 30-fold increase in the frequency of 'positive' mutants. Since
a multiplicative factor of similar magnitude is expected for the mutag
enesis of additional sets of six residues, performing REM on 18 sites
is expected to yield an exponential (30 000-fold) increase in the thro
ughput of positive mutants as compared to random [NN(G,C)]18 mutagenes
is.