TL;DR
Researchers have developed CompoMug, a computational tool that predicts point mutations to stabilize G protein-coupled receptors (GPCRs), a crucial class of drug targets. This approach significantly speeds up the process of engineering stable GPCRs, enabling their structural determination and paving the way for new drug discovery.
What Problem Does This Paper Address?
G protein-coupled receptors (GPCRs) are vital for cellular communication and are targets for nearly half of all therapeutic drugs. However, their inherent flexibility and low stability make them notoriously difficult to study structurally and biochemically. Traditional methods for stabilizing GPCRs involve laborious and costly trial-and-error mutations. This paper addresses the need for a more efficient, computational approach to predict stabilizing mutations, thereby accelerating the study of these important proteins and the development of new medicines.
Methods
The core of this work is the development of CompoMug, a computational tool that integrates four distinct modules to predict stabilizing point mutations in GPCRs:
- Knowledge-based module: Leverages a curated list of known, transferable stabilizing mutations that have previously aided GPCR crystallization.
- Sequence-based module: Analyzes multiple sequence alignments (MSAs) of homologous GPCR sequences to identify residues that deviate from conserved patterns, hypothesizing that restoring conserved amino acids in these positions can enhance stability.
- Structure-based module: Identifies potential stabilizing interactions, such as salt bridges ("ionic locks") and disulfide bonds, by analyzing 3D structural models of the receptor.
- Machine-learning-based module: Employs a support vector machine (SVM) trained on experimental mutagenesis data from other GPCRs to predict stabilizing mutations based on a comprehensive set of structural and energy-based features.
These modules work in concert, combining sequence, structural, and experimental data to generate a prioritized list of candidate mutations.
Datasets
- Training Benchmark: The machine learning module was trained on experimental alanine scanning mutagenesis data from three GPCR receptors: neurotensin receptor NTS1, A2A adenosine receptor, and β₁ adrenergic receptor ADRB1. This dataset comprised 79 stabilizing point mutations and 923 non-stabilizing point mutations.
- Application Target: The CompoMug tool was applied to the human serotonin 5-HT₂c receptor.
- Structural Data: Homology models of the 5-HT₂c receptor were generated using the structure of the 5-HT₂b receptor (PDB ID 4IB4) as a template.
Key Insights and Ideas
- Synergistic Approach: The CompoMug tool's strength lies in the integration of diverse predictive modules. The paper demonstrates that each module contributes uniquely, and their combined predictions are more effective than any single module alone.
- Predicting Specific Interactions: The structure-based module highlights the potential for designing specific stabilizing interactions like ionic locks and disulfide bonds, which can precisely constrain receptor conformations.
- Machine Learning for Unseen Mutations: The machine learning module is crucial for identifying stabilizing mutations that might be overlooked by sequence or knowledge-based approaches, especially for residues other than alanine.
- Conformational Selectivity: The study suggests that certain mutations can differentially stabilize specific receptor conformations (e.g., agonist-bound vs. antagonist-bound), which is critical for understanding receptor function and designing selective drugs.
Key Results
- High Success Rate: CompoMug successfully predicted 10 stabilizing mutations for the 5-HT₂c receptor, achieving a 25% hit rate for mutations that improved thermostability by at least 1.5°C. This is significantly higher than traditional experimental methods.
- Significant Thermostability Gains: The best single mutation (C₃₆₀⁷.⁴⁵ᴺ) increased the apparent melting temperature (Tm) by ~8.8°C. A triple mutant (C₃₆₀⁷.⁴⁵ᴺ, G₃₆₂⁷.⁴⁷ᴬ, A₁₇₁⁴.⁴²ᴸ) achieved a ~13°C increase in Tm.
- Ligand-Dependent Stabilization: For the triple mutant, binding of an antagonist (mesulergine) further increased thermostability by ~8°C, leading to a total gain of ~21°C compared to the wild-type apo receptor.
- Enabling Crystallization: The predicted stabilizing mutations were instrumental in obtaining high-resolution crystal structures of the 5-HT₂c receptor in both inactive and active-like conformations, including the first antagonist-bound structure of a serotonin receptor.
Limitations and Future Work
- Transferability: While some knowledge-based mutations are transferable, their effectiveness can vary across GPCRs. The paper notes that some previously known mutations did not yield significant improvements in the 5-HT₂c case.
- Machine Learning Training Data: The machine learning module's predictive power for residues other than alanine is limited by the current training set, which primarily consists of alanine scanning data.
- Structural Model Accuracy: The structure-based module's performance is highly dependent on the accuracy of the available 3D structural models.
- Future Work: The authors suggest that the CompoMug tool can be further improved by accumulating more experimental mutation data for GPCRs. They also propose that the structure-based and sequence-based modules could be adapted for other membrane protein families, and the machine learning feature set might have broader utility with retraining. The tool is already being applied to other GPCR targets.
Summary updated