Illustration of complex protein structure in a scientific setting with a beaker, molecular diagrams, and data graphs.

Membrane proteins comprise approximately 30% of the human proteome and are targets for over 60% of approved drugs[^1]. However, their inherent instability in aqueous environments has hindered detailed structural and functional analyses. Previous attempts to create soluble versions of membrane proteins have had limited success, often resulting in the loss of native structure or function. A groundbreaking study by researchers from the École Polytechnique Fédérale de Lausanne and the Swiss Institute of Bioinformatics, published in Nature[^2], demonstrates a novel computational approach to design soluble analogues of membrane proteins that retain their structural features and functionality. This study has opened up the potential for AI-driven protein engineering, new therapeutic approaches, and drug discovery.

Deep Learning-Based Membrane Protein Design

The researchers developed an innovative computational pipeline combining two powerful deep learning tools:

  1. AlphaFold2 (AF2): A highly accurate protein structure prediction algorithm.
  2. ProteinMPNN: A robust sequence design tool.

By inverting the AF2 network and coupling it with ProteinMPNN, the team created a flexible and generalizable method for designing protein sequences that adopt desired folds, including those of membrane proteins.

Designing Complex Membrane Protein Folds

The study successfully designed three challenging protein folds using AF2seq-MPNN:

  1. Ig-like Fold (IGF)
  2. β-Barrel Fold (BBF)
  3. TIM-Barrel Fold (TBF)

These folds are crucial for various biological functions and have complex structural topologies. The designed IGFs exhibited high thermostability and monodispersity in solution, while the BBFs and TBFs demonstrated precise hydrogen bonding patterns and stability.

Solubilizing AI-Designed Membrane Protein Folds

Membrane proteins exhibit unique topologies not found in soluble proteomes, raising the question of whether these folds possess intrinsic structural features precluding them from existing in soluble form. The study addressed this by designing soluble analogues of three membrane folds: claudin, rhomboid protease, and G-protein-coupled receptor (GPCR). The AF2seq-MPNNsol pipeline generated high-confidence designs with low surface hydrophobicity, resulting in soluble analogues that retained functional motifs. These designs exhibited high thermal stability and structural accuracy, demonstrating that membrane protein folds can be successfully adapted to soluble forms.

Structural Validation of AI-Designed Soluble Membrane Proteins

The researchers applied their method to create soluble versions of three distinct membrane protein folds:

  • Claudin Fold
  • Rhomboid Protease Fold
  • G-Protein-Coupled Receptor (GPCR) Fold

Remarkably, these designs maintained the overall topology of their membrane-bound counterparts while adopting hydrophilic surfaces suitable for aqueous environments.

High-resolution X-ray crystallography structures were obtained for several designs, demonstrating exceptional accuracy in both backbone and side-chain conformations compared to the computational models.

Functional Characterization of AI-Designed Soluble Membrane Protein Analogues

The researchers successfully designed and validated functional soluble analogues of membrane proteins, focusing on claudins and GPCRs.

Claudin Analogues

Soluble analogues of human claudin-1 and claudin-4 were designed to bind Clostridium perfringens enterotoxin (CpE), exhibiting binding kinetics and affinities comparable to their membrane-bound counterparts. The soluble claudin-4 formed oligomers similar to natural tight junctions, which could be disrupted by CpE. The cryo-EM structure of the soluble claudin-4 analogue complexed with CpE showed comparable topology and toxin binding mode to the natural complex.

GPCR Analogues

The researchers employed two distinct approaches to functionalize soluble GPCR analogues:

  1. Chimeric Proteins: Chimeric proteins with the ghrelin receptor were created by grafting intracellular loop 3 (ICL3) of the ghrelin receptor onto the GPCR-like fold (GLF) scaffold. Nine out of sixteen chimeric designs bound to an ICL3-targeting antibody.
  2. Conformation-Specific Designs: Adenosine A2A receptor analogues were designed for both active and inactive states. These analogues preserved the G-protein-binding site, including conserved sequences like the DRY motif. The active state designs (aGLFs) bound to the mini-Gs-414 protein, while the inactive state designs (iGLFs) showed no binding.

Implications of AI-Designed Soluble Membrane Proteins in Drug Discovery

The ability to design soluble analogues of membrane proteins opens new avenues for drug discovery and therapeutic development. Soluble analogues can facilitate the study of protein functions in biochemically accessible formats, accelerating the development of new drugs targeting membrane proteins. The precise conformational design of GPCR analogues, capable of binding or precluding G-protein interactions, underscores the potential for creating specific functional states for drug screening.

References

  1. Chemists develop a new drug discovery strategy for ‘undruggable’ drug targets | ScienceDaily.
  2. Goverde, C.A., Pacesa, M., Goldbach, N. et al. Computational design of soluble and functional membrane protein analogues. Nature 631, 449–458 (2024).

 

Drug Discovery & Development Assays Offered by Marin Biologic Laboratories