Creating a novel, selective binder to CTLA4 using AlphaFold2 and Rosetta
Background: CTLA4 is a single-chain protein that is involved in immune system checkpoint promotion and has become an important molecule for immunotherapies since CTLA4 inhibition can be a powerful tool to modulate immune responses. A search of the PDB showed that previous efforts had engineered an anti-CTLA4 nanobody KN044 (with PDB ID 6RQM). The goal of this project was to create a novel, selective binder by utilizing features present in the interaction surface between KN044 and CTLA4 using AlphaFold2, Rosetta Remodel, and Rosetta Scripts.
Methodology: The initial binder sequence was generated by AfDesign using hallucination and then fed to AlphaFold2 to produce an initial structure. This structure was then iteratively tuned using Rosetta Remodel and Rosetta Scripts for this specific docking task. More specifically, 4 structures were designed based on the constraints of this problem. The first was a length 30 naïve peptide that was allowed to dock anywhere on the CTLA4 structure, the second was a length 50 naïve peptide that was also allowed to freely dock, the third was a length 50 peptide that was designed to bind to hotspots on CTLA4 that were identified based on KN044’s interactions with CTLA4, and the fourth/last was a length 75 peptide that was also designed to bind to these hotspots.
After generating the initial structures for these designs from AlphaFold2, the designs were set up in Rosetta with constrained relaxation of the backbone and sidechains to see if the hallucinated proteins were reasonably stable (i.e., had a negative gibbs free energy). If the proteins had a positive energy, their structures were corrected by running unconstrained relaxation of the backbone and sidechains. In all cases where the constrained optimization led to a structure with positive energy, the unconstrained optimization approach produced a structure with negative energy.
These structures were then iteratively improved by cycles of Rosetta Remodel to re-design the interface followed by Rosetta Scripts to evaluate and modify the docked poses. Following this improvement, the final docked pose for each design was evaluated within Rosetta.
Metrics: The number of buried unsatisfied hydrogen bonds, the gibbs free energy at the interface, and the RMSD were calculated to assess the stability and the binding of the novel peptide to CTLA4.
Results: Design 4 had the least number of unsatisfied hydrogen bonds (12), the lowest gibbs free energy (-183.472), and the lowest RMS (6.811 A). Design 4 also showed tight packing and high residue density across the interface between hallucinated protein and CTLA4.
Conclusion: In this project, an initial modeling pipeline was developed to design and test hallucinated proteins to generate a binder to a target protein. With this pipeline, a relatively stable, selective binder was generated to a single-chain protein target.
Full project report (included figures and final pdbs) available upon request