Reconstruction of ARNT PAS-B Unfolding Paths by Steered MD and Artificial Neural Network Reveals New Putative Binding Conformations

Abstract

Several proteins have been identified that can switch between different folds, expanding their functional role through previously unknown binding interactions. In this context, experimental evidence suggests that the PAS-B domain of the Aryl Hydrocarbon Receptor Nuclear Translocator protein (ARNT) has a fragile β-sheet structure that can undergo a +3 register shift upon a single mutation, changing its dimerization surface. Both unfolding and interconversion mechanisms require the same uncharacterized partially-unfolded state. Single-molecule Atomic Force Microscopy measurements (AFM) confirmed that the ARNT PAS-B domain is mechanically labile and revealed two distinct unfolding pathways, one following a single-stage mechanism, and the other showing an intermediate state. An atomistic characterisation of unfolding paths and relevant intermediate states has been carried out to identify new ARNT conformations possibly involved in partner recruitment. The unfolding pathways for ARNT PAS-B were simulated by steered Molecular Dynamics. Multiple replicas were recorded to explore different paths at low pulling speed. The overall model of unfolding was then reconstructed using a self-organizing map (SOM), a type of artificial neural network with an explicit visual representation. The map was trained on inter-residue distances of native contacts. The SOM model clearly depicts two alternative pathways in a human-understandable two-dimensional map. Retracing of single replicas on the map highlights a two-stage mechanism with initial unfolding of the C-terminal strand in 1/3 of the simulations, in agreement with AFM findings. Shortest path analysis on the map correctly detects replicas with minimal steering work. The partially-folded states extracted from the map could have a role in selectivity for partner recruitment during ARNT transcriptional activity.