CRAF与RAS结合的正构和别构决定因素

Intracellular protein-protein interactions are key drivers of oncogenic processes such as signal transduction and transcriptional rewiring. While protein-protein interfaces are traditionally hard to drug with small molecules, molecular glues are emerging as a promising modality to redirect the interactions of a target protein away from a pathogenic partner. However, rational discovery of molecular glues remains challenging due to a limited biophysical understanding of how proteins interact at the atomic level. To address this challenge, we developed a massively multiplexed approach to measure the impact of sequence variation on protein-protein interaction affinity inside of eukaryotic cells. Here, we applied this technology to map the binding energetics between CRAF and HRAS at single residue resolution through deep saturation mutagenesis of the CRAF Ras-binding domain (CRAF RBD ). In a single multiplexed assay, we measured binding affinity for 1,501 CRAF variants, including all possible single residue substitutions of the CRAF RBD . Free energy estimates were highly consistent with gold-standard biophysical measurements (Pearson R=0.93, P=6.8×10 -4 , N=8 mutations) and confirmed that our technology could sensitively detect high affinity (nanomolar K D ) and low affinity (micromolar K D ) binding events. Integrating the mutagenesis data with the crystal structure of the interface resulted in a 3D spatial map of binding energetics for the interaction. This map identified orthosteric interface and allosteric non-interface residues that destabilize the interaction. It also revealed a class of sequence variants that simultaneously decrease overall CRAF stability while increasing binding affinity for HRAS. Such mutations may be difficult to discover with traditional biophysical techniques due to the challenge of purifying unstable proteins. Moreover, they can directly inform rational molecular glue design by providing a roadmap for molecules that mimic these strengthening interactions. Finally, we demonstrated that rich mutational energetics data can improve accuracy of AI protein structure prediction models on protein complex prediction tasks. This advancement enables the possibility of generating 3D energetics maps for protein interfaces lacking a solved structure. Our results provide a map of orthosteric and allosteric determinants of CRAF RBD binding to HRAS. This map has direct application to the development of RAS and RAF modulators, a space of high relevance given the emerging landscape of RAS inhibitor resistance mechanisms. This work also demonstrates a highly scalable approach to deeply and rapidly characterize protein-protein interfaces. Future broad application of this technology will generate data atlases to directly inform rational molecular glue discovery.