Engineering a protein-protein interface using a computationally designed library.

Computational algorithms for protein design can sample large regions of sequence space, but suffer from undersampling of conformational space and energy function inaccuracies. Experimental screening of combinatorial protein libraries avoids the need for accurate energy functions, but has limited access to vast amounts of sequence space. Here, we test if these two traditionally alternative, but potentially complementary approaches can be combined to design a variant of the ubiquitin-ligase E6AP that will bind to a nonnatural partner, the NEDD8-conjugating enzyme Ubc12. Three E6AP libraries were constructed: (i) a naive library in which all 20 amino acids were allowed at every position on the target-binding surface of E6AP (13 positions), (ii) a semidirected library that varied the same residue positions as in the naive library but disallowed mutations computationally predicted to destabilize E6AP, and (iii) a directed library that used docking and sequence optimization simulations to identify mutations predicted to be favorable for binding Ubc12. Both of the directed libraries showed > 30-fold enrichment over the naive library after the first round of screening with a split-dihydrofolate reductase complementation assay and produced multiple tight binders (K(d) < 100 nM) after four rounds of selection. Four rounds of selection with the naive library failed to produce any binders with K(d)'s lower than 50 μM. These results indicate that protein design simulations can be used to create directed libraries that are enriched in tight binders and that in some cases it is sufficient to computationally screen for well-folded sequences without explicit binding calculations.