Sequence-specific targeting of intrinsically disordered protein regions
Kejia Wu,Hanlun Jiang,Derrick R. Hicks,Caixuan Liu,Edin Muratspahic,Theresa A. Ramelot,Yuexuan Liu,Kerrie McNally,Amit Gaur,Brian Coventry,Wei Chen,Asim K. Bera,Alex Kang,Stacey Gerben,Mila Ya-Lan Lamb,Analisa Murray,Xinting Li,Madison A. Kennedy,Wei Yang,Gudrun Schober,Stuart M. Brierley,Michael H. Gelb,Gaetano T. Montelione,Emmanuel Derivery,David Baker
DOI: https://doi.org/10.1101/2024.07.15.603480
2024-09-02
Abstract:A general approach to design proteins that bind tightly and specifically to intrinsically disordered regions (IDRs) of proteins and flexible peptides would have wide application in biological research, therapeutics, and diagnosis. However, the lack of defined structures and the high variability in sequence and conformational preferences has complicated such efforts. We sought to develop a method combining biophysical principles with deep learning to readily generate binders for any disordered sequence. Instead of assuming a fixed regular structure for the target, general recognition is achieved by threading the query sequence through diverse extended binding modes in hundreds of templates with varying pocket depths and spacings, followed by RFdiffusion refinement to optimize the binder-target fit. We tested the method by designing binders to 39 highly diverse unstructured targets. Experimental testing of ~36 designs per target yielded binders with affinities better than 100 nM in 34 cases, and in the pM range in four cases. The co-crystal structure of a designed binder in complex with dynorphin A is closely consistent with the design model. All by all binding experiments for 20 designs binding diverse targets show they are highly specific for the intended targets, with no crosstalk even for the closely related dynorphin A and dynorphin B. Our approach thus could provide a general solution to the intrinsically disordered protein and peptide recognition problem.
Biochemistry