Affinity maturation of a computationally designed binding protein affords a functional but disordered polypeptide

Computational methods have been recently applied to the design of protein–protein interfaces. Using this approach, a 61 amino acid long protein called Spider Roll was engineered to recognize the kinase domain of the human p21-activated kinase 1 (PAK1) with good specificity but modest affinity (K_D = 100 μM). Here we show that this artificial protein can be optimized by yeast surface display and fluorescence-activated cell sorting. After three rounds of mutagenesis and screening, a diverse set of tighter binding variants was obtained. A representative binder, MSR7, has a >10²-fold higher affinity for PAK1 when displayed on yeast and a 6 to 11-fold advantage when produced free in solution. In contrast to the starting Spider Roll protein, however, MSR7 unexpectedly exhibits characteristics typical of partially disordered proteins, including lower α-helical content, non-cooperative thermal denaturation, and NMR data showing peak broadening and poor signal dispersion. Although conformational disorder is increasingly recognized as an important property of proteins involved in cellular signaling and regulation, it is poorly modeled by current computational methods. Explicit consideration of structural flexibility may improve future protein designs and provide deeper insight into molecular events at protein–protein interfaces.