Explaining Conformational Diversity in Protein Families through Molecular Motions

Valentin Lombard, Sergei Grudinin, Élodie Laine

TL;DR

This paper introduces DANCE, a computational pipeline that analyzes protein conformational diversity across entire families by clustering structures and identifying key "linear motions" that explain their variability. They demonstrate that most protein families exhibit low-dimensional conformational landscapes and that these insights can be used to generate plausible unseen conformations, paving the way for better protein function understanding and drug discovery.

What Problem Does This Paper Address?

Proteins are dynamic molecules, and their ability to adopt multiple shapes (conformations) is fundamental to their function. While experimental techniques and computational methods like AlphaFold2 have greatly expanded our knowledge of protein structures, understanding and systematically describing the full range of a protein's conformational variability, especially across entire families, remains a significant challenge. Existing methods often struggle with capturing subtle motions, predicting entirely new states, or require extensive manual curation. This gap hinders our ability to fully grasp protein function and design targeted therapeutics.

Methods

The core of this work is the DANCE (Dimensionality Analysis for protein Conformational Exploration) pipeline. DANCE operates in several key steps:

  1. Sequence Extraction and Clustering: It extracts amino acid sequences from input protein structures (experimental or predicted) and clusters them based on user-defined sequence similarity thresholds.
  2. Multiple Sequence Alignment (MSA): For each cluster, it generates an MSA to establish residue correspondence across different structures.
  3. Structure Extraction and Superimposition: It extracts backbone atom coordinates and then superimposes the structures within each cluster to minimize Root Mean Square Deviation (RMSD), creating a "conformational ensemble."
  4. Dimensionality Reduction (PCA): Principal Component Analysis (PCA) is applied to these ensembles to identify orthogonal directions of motion (principal components or "linear motions") that best explain the observed structural variability.
  5. Benchmarking: The authors also assess classical manifold learning techniques like kernel PCA (kPCA) and UMAP for their ability to reconstruct and generate unseen conformations.

A novel aspect is DANCE's ability to handle both experimental and predicted structures, its systematic approach to entire protein families, and its focus on characterizing variability through linear motions.

Datasets

  • Protein Data Bank (PDB): The primary dataset used is the PDB, containing over 748,000 protein chains with experimentally resolved 3D structures (as of June 2023). DANCE was applied to this entire dataset.
  • ABC Transporter Dataset: A curated set of 354 ABC transporter structures was used for a specific case study, leveraging structural similarity for grouping instead of sequence similarity.
  • Benchmark Set: A benchmark set of ten conformational ensembles with diverse motion complexities was created for evaluating manifold learning techniques. These ensembles represent proteins with significant conformational changes and are well-characterized in the literature.

Key Insights and Ideas

  • Proteins Live on Low-Dimensional Manifolds: A striking finding is that the vast majority of experimentally observed protein conformations for a given family can be explained by a surprisingly small number of linear motions (principal components). This suggests that protein conformational space is not chaotic but rather highly organized and constrained.
  • Sequence Similarity vs. Structural Similarity: When analyzing the ABC transporter superfamily, using structural similarity for clustering revealed a more continuous range of conformations and a richer description of functional motions compared to relying solely on sequence similarity, especially for highly divergent families.
  • Manifold Learning for Generating Plausible Conformations: Techniques like PCA and kPCA can not only describe observed motions but also generate novel, plausible conformations. This has significant implications for molecular dynamics simulations and drug design.
  • "Motion Simplification" with Expansion: Counterintuitively, as protein families are expanded to include more distant homologs (relaxing sequence similarity criteria), the conformational diversity can sometimes appear to simplify. This often occurs when a new, dominant motion emerges that explains a large portion of the variance, effectively overshadowing smaller, more localized motions.

Key Results

  • PDB-Wide Analysis: DANCE was successfully applied to the entire PDB, generating conformational collections for millions of protein chains.
  • Low Dimensionality: For most protein families, less than eight linear motions explain over 90% of the total positional variance. For example, bacterial adenylate kinase exhibits a one-dimensional motion between open and closed states.
  • ABC Transporter Insights: A structure-based analysis of ABC transporters revealed a highly collective opening-closing motion that explains a significant portion of the variance and is crucial for substrate binding pocket access.
  • Reconstruction Accuracy: PCA and kPCA demonstrated high accuracy in reconstructing conformations, with RMSD errors generally below 2 Å for most proteins. kPCA consistently outperformed linear PCA, increasing high-quality reconstructions by up to 82% for some proteins.
  • Interpolation Success: Linear interpolation between extreme states in PCA and kPCA representation spaces successfully generated known intermediate conformations for ATPase, suggesting these learned spaces capture biologically relevant pathways.

Limitations and Future Work

  • Superimposition Dependency: The current superimposition method relies on MSAs, which can lead to suboptimal coordinate matching and large deviations if sequence similarities are ambiguous or if fold changes occur. Future work aims to explore multi-reference or reference-free probabilistic frameworks.
  • Handling Complexes: DANCE primarily focuses on single polypeptide chains. Future improvements will include labeling complexes and accounting for them in clustering, as well as exploring more sophisticated ways to handle data uncertainty.
  • Non-linear Motions: While PCA captures dominant linear motions, some protein families exhibit complex, non-linear rearrangements that may require more advanced dimensionality reduction techniques for full characterization.
  • Drug Discovery Applications: The authors suggest using generated conformations as templates for drug design or molecular dynamics simulations, indicating a clear path for future application-oriented research.
  • Benchmarking Non-linear Methods: While kPCA and UMAP were tested, further exploration of other non-linear manifold learning techniques for conformation generation and analysis is warranted.

Summary updated

Abstract

Proteins play a central role in biological processes, and understanding their conformational variability is crucial for unraveling their functional mechanisms. Recent advancements in high-throughput technologies have enhanced our knowledge of protein structures, yet predicting their multiple conformational states and motions remains challenging. This study introduces Dimensionality Analysis for protein Conformational Exploration (DANCE) for a systematic and comprehensive description of protein families conformational variability. DANCE accommodates both experimental and predicted structures. It is suitable for analysing anything from single proteins to superfamilies. Employing it, we clustered all experimentally resolved protein structures available in the Protein Data Bank into conformational collections and characterized them as sets of linear motions. The resource facilitates access and exploitation of the multiple states adopted by a protein and its homologs. Beyond descriptive analysis, we assessed classical dimensionality reduction techniques for sampling unseen states on a representative benchmark. This work improves our understanding of how proteins deform to perform their functions and opens ways to a standardised evaluation of methods designed to sample and generate protein conformations.
BibTeX
@inproceedings{lombard2024explaining,
  author = {Valentin Lombard and Sergei Grudinin and Élodie Laine},
  title = {{Explaining Conformational Diversity in Protein Families through Molecular Motions}},
  booktitle = {Scientific},
  year = {2024},
  doi = {10.1038/s41597-024-03524-5},
}
Figure