Transformer-based framework uncovers state-dependent modular organization and conformational landscapes of the β-arrestin 1 C-terminal tail

Abstract

{beta}-arrestins ({beta}arr) regulate signaling and trafficking of G protein-coupled receptors (GPCRs) across diverse physiological and pathological processes. However, mechanistic understanding of how ligand-activated GPCRs engage and activate {beta}arr remains limited, with the conformation of the entire {beta}arr tail in the active state still unknown. Here, by comparatively analyzing temperature replica-exchange molecular dynamics simulation data of {beta}arr1 in basal and active states, we investigated the conformational landscape of the {beta}arr1 tail to elucidate its role in activation. To overcome limitation of conventional analyses in characterizing the vast conformational space sampled by the 62-residue tail, we developed a transformer-based autoencoder (TAE) framework that integrates attention-derived residue relationships and latent-space clustering to identify the tails modular organization and conformational substates, providing an interpretable description of how local residue interactions couple to large-scale conformational rearrangements. Using the conformationally constrained basal state as a control, we validated the framework by showing that it recovers interpretable conformational features. In the active state, the framework revealed a reorganized modular architecture, recovered key interactions identified through manual analysis, and uncovered segment-specific dynamics inaccessible to conventional approaches. Our findings show that, in the active state, the {beta}arr1 tail preferentially engages the back side of the main body and forms substates in which the middle segment occupies the central crest crevice, suggesting that the released tail can self-engage functionally critical surfaces and influence the balance between tail-only and core-engaged receptor complexes. Together, this work establishes a TAE framework for analyzing large-scale conformational ensembles and advances our understanding of {beta}arr1 activation.