Binding Paths: Describing Small Molecule Interactions with Disordered Proteins via a Markov State Model

Adelie Auriane Berenice Louet, Gerhard Hummer, Michele VendruscolobioRxiv 2025 · DOI

TL;DR

This paper introduces a novel "binding path" framework to understand how small molecules interact with intrinsically disordered proteins, which lack traditional binding pockets. By using Markov State Models (MSMs) and knowledge graphs, they visualize these dynamic pathways, offering a new paradigm for designing drugs against these challenging targets.

What Problem Does This Paper Address?

Intrinsically disordered proteins (IDPs) are crucial in many biological processes but are notoriously difficult to target with drugs. Unlike well-structured proteins with defined binding pockets, IDPs are highly flexible and lack stable, predictable interaction sites. This makes traditional drug discovery methods, which rely on identifying and exploiting these pockets, largely ineffective. The challenge is to understand and leverage the dynamic, transient interactions that occur when small molecules bind to these flexible proteins.

Methods

The researchers developed a framework that combines several advanced computational techniques:

  • Molecular Dynamics (MD) Simulations: These simulations provide detailed, time-resolved trajectories of how the small molecule (10074-G5) interacts with the disordered protein (Aβ42).
  • VAMPNet Clustering: A deep learning approach used to identify potential "binding pockets" by clustering configurations from MD simulations. However, they found it struggled to distinguish distinct states in this highly dynamic system.
  • Handoff Matrix: A novel method to track the stochastic diffusion of the ligand across the protein surface. It quantifies how the ligand transitions its contacts from one residue to another over time, revealing "binding hotspots."
  • Markov State Models (MSMs): Built upon the handoff matrix analysis, MSMs represent the binding process as a network of discrete "states" (defined by groups of residues in contact with the ligand) and the probabilities of transitioning between these states. This allows for a kinetic and thermodynamic description of binding.
  • Master Equation Analysis: Used to quantitatively describe the binding dynamics captured by the MSM, allowing for the calculation of equilibrium populations and dissociation constants.
  • Knowledge Graphs: Visualizations of the MSM, where nodes represent binding states and edges represent transitions. This provides a high-level map of the binding landscape, revealing connectivity and affinity.

Datasets

The primary dataset used was generated from molecular dynamics simulations of the small molecule 10074-G5 interacting with the Aβ42 peptide. This dataset was generated in silico for the purpose of this study.

Key Insights and Ideas

  • Shifting from "Pockets" to "Paths": The core idea is to move beyond the concept of static binding pockets and instead describe binding as a dynamic "path" or trajectory across the protein surface.
  • Dynamic Binding Hotspots: Instead of fixed pockets, the ligand interacts with transiently formed clusters of residues, which the authors term "binding hotspots." These are regions of elevated interaction probability.
  • Gliding vs. Hopping: The analysis revealed that ligand movement on the disordered protein surface is primarily a "gliding" motion between similar residue clusters, with rare "hops" to distant regions.
  • Affinity Correlates with Connectivity: In the knowledge graph visualization, states that are more centrally connected (higher degree centrality) tend to exhibit stronger ligand binding affinity.
  • Programmable Surface: The dynamic binding paths can be viewed as a "programmable map" on the protein surface, which can be exploited for highly selective drug design.

Key Results

  • The developed MSM framework successfully characterized the binding mechanism of 10074-G5 to Aβ42 as a series of transient contacts and diffusion along the protein surface.
  • The analysis quantified binding affinities for different states, yielding an overall binding affinity of 47 µM, consistent with experimental measurements.
  • Knowledge graphs provided an intuitive visualization of the binding landscape, showing how different binding states are interconnected and how affinity relates to network centrality.
  • The approach was validated on other systems, including α-synuclein with three different small-molecule binders.

Limitations and Future Work

  • VAMPNet Limitations: The initial attempt to use VAMPNet for clustering binding pockets revealed its limitations in distinguishing subtle variations in contact patterns within highly dynamic systems.
  • Computational Cost: Generating and analyzing the extensive MD simulations and building MSMs are computationally intensive.
  • Future Work: The authors suggest that this "binding path" concept can be generalized to design small molecules that specifically follow these paths, leading to improved selectivity and reduced off-target interactions. This framework opens new avenues for targeting previously "undruggable" disordered proteins.

Summary updated

Abstract

Abstract Disordered proteins are challenging targets for drug discovery because they lack well-defined binding pockets. Although small molecules can form relatively stable complexes with disordered proteins, the highly dynamic nature of these proteins complicates the understanding of their binding mechanism. To address this problem, we analyze the binding of the small molecule 10074-G5 to Aβ42, which results in the formation of a disordered complex. We describe the binding mechanism in terms of binding paths, which are stochastic trajectories along which small molecule diffuse across disordered protein surfaces, forming transient contacts with overlapping groups of residues. To identify these binding paths, we define them as realizations of a stochastic process defined by a Markov State Model (MSM). The MSM is built from distinct states of the disordered complex and their corresponding transition probabilities, enabling both the dynamic mapping of binding hotspots and targetable regions where static pockets cannot be defined-extending the notion of binding pockets to disordered systems - and the quantitative calculation of binding affinities. The visualization of the MSM via knowledge graphs provides an intuitive representation of the binding paths. We further validated this approach across four additional systems, comprising C-terminal α-synuclein with three distinct small-molecule binders, and the full-length peptide (140 residues) with binder fasudil. By generalizing the concept of static binding pockets to dynamic binding paths, our approach rationalizes small-molecule recognition by disordered proteins and establishes a framework for identifying druggable regions on systems that were previously considered intractable, providing insights for future drug design programs.
BibTeX
@misc{louet2025binding,
  author = {Adelie Auriane Berenice Louet and Gerhard Hummer and Michele Vendruscolo},
  title = {{Binding Paths: Describing Small Molecule Interactions with Disordered Proteins via a Markov State Model}},
  year = {2025},
  doi = {10.1101/2025.11.19.688850},
  howpublished = {bioRxiv},
}
Figure