| dc.description.abstract | The traditional view in Structural Biology centered on the “one sequence–one structure–one function” paradigm, which emphasized a unique, folded state of protein as the basis for its biological activity. However, this perspective evolved with the discovery of intrinsically disordered proteins (IDPs) which are biologically active proteins lacking stable tertiary structure. IDPs exist as dynamic ensembles governed by shallow, frustrated energy landscapes. Their sequence features, such as low hydropathy, high charge, and enrichment in polar or structure-breaking residues, lead to conformational heterogeneity and responsiveness to cellular cues like pH, binding partners, and post-translational modifications. Functionally, IDPs contribute to diverse biological roles including signaling, molecular recognition, scaffolding, and phase separation. These capabilities arise from their ability to modulate structure upon binding or remain dynamic even in complex forms. To understand IDPs at atomic resolution, we employed advanced molecular dynamics simulations, combining enhanced sampling with rigorous structural analysis to capture and validate their
heterogeneous ensembles. In our first work, we conducted simulations of the low-complexity domain (LCD) of hnRNPA1 using an in-house enhanced sampling method called Replica Exchange with Hybrid Tempering (REHT). Starting from predicted models, we generated microsecond-scale trajectories that revealed transient local structures and modularity within the LCD. Structural validation was performed using chemical shifts, SAXS profiles, and inter-residue distances. We observed persistent β-structures and identified a modular dumbbell-like architecture composed of the RGG-box and PrLD
regions. Clustering using dimensionality reduction showed distinct conformational subpopulations, supporting the modular organization. Independent simulations of the RGG and PrLD domains recapitulated their respective structural features, confirming their domain-like behavior. This study offers a comprehensive conformational ensemble of hnRNPA1-LCD and insights into its structural features. In second work, we developed a classification framework to distinguish kinked from extended β-strands using bend and rotation angles as key structural features. This method, based on a support vector machine classifier, allows unambiguous identification of kinked motifs that are critical for the reversibility of amyloid-like fibrils. Application of this classifier to simulation data revealed that kinked motifs are present across reversible fibrils formed by IDPs and are structurally distinct from those in irreversible aggregates. We constructed a motif library from simula tion trajectories and incorporated sequence-based filters to identify low complexity aromatic-rich kinked segments or LARKs. Further, we used metadynamics simulations to show how disease-linked mutations shift the energetic landscape to favor irreversible aggregation. This approach provides a robust tool for investigating the molecular determinants of fibril reversibility in condensate-forming proteins. In third work, we studied the RGG-box domain of hnRNPA1 to uncover its role in remodeling telomeric G-quadruplexes (GQs). Enhanced sampling simulations captured the structural ensemble of the RGG domain, which was used for docking with a diverse set of DNA and RNA GQs. Molecular dynamics simulations revealed few distinct binding modes: end-stacking and groove- and loop- binding, both driven by π-stacking and electrostatics. In specific cases, RGG-box alone could destabilize GQs through amino acid sidechain insertions or K+ ion ejection from GQ core. We also examined the role of a short boundary region (BR) at the interface of ordered UP1 domain and disordered domain of hnRNPA1. Simulations of multiple UP1 constructs with and without BR revealed its importance in nucleic acid binding affinity and binding mode modulation. Interdomain electrostatics and structural rearrangements further highlighted how BR contributes to telomere recognition and
remodeling. In fourth work (ongoing work), we discuss about how capturing the full ensemble of long IDPs remains a computational challenge due to the interplay of entropic complexity, force field biases, and sampling inefficiencies. We attempted to simulate the long LCD of hnRNPA1 using enhanced sampling but found that certain global features, such as radius of gyration, were not recovered. Mutational scans and force field optimization partially improved local accuracy but could not resolve long-range conformational discrepancies. As a solution, we propose combining back-mapped coarse-grained and all-atom simulations through Bayesian maximum entropy reweighting, using experimental data as a guide. This strategy allows generation of refined ensembles that balance physical realism and computational feasibility, and can serve as a training set for machine learning-based force field development. Taken together, this thesis presents a multi-scale computational exploration of intrinsically disordered proteins, focusing on the hnRNPA1 low-complexity domain and its subregions. The goal is to advance our understanding of IDP structure–function
relationships and establish methodological frameworks that are broadly applicable to the growing class of disordered proteins. | en_US |
