Exploring generality of experimental conformational changes with AlphaFold predictions

Abstract

Structural predictions have matched the accuracy of experimental structures in the case of close homologues, outperformed docking methods for multimeric complexes and helped sampling the conformational landscape of transporters and receptors. Such successes prompt the question whether predictions can be used to relate experimental structures in the context of available knowledge. LysR-type transcriptional regulators (LTTR) constitute the most common family of bacterial regulators. Intriguingly, their experimental structures are remarkably diverse. The active species, composed of flexible monomers dimerizing through their N- and C-terminal domains in a circular arrangement, differ across LTTR, due to intrinsic sequence differences or because crystals stabilize diverse snapshots of a common dynamic mechanism. We have used AlphaFold2 (AF) to interrogate the experimental AtzR structure in the context of predictions guided towards the different hetero-multimeric conformations known for other LTTR. Our approach drives AF prediction with the structure-based selection of the information input through sequence alignment and template conformation, linked to examination of the energy with PISA and interactions with ALEPH.

Verification: model-free phasing with enhanced predicted models in ARCIMBOLDO_SHREDDER

“…PDB entry 7z7j is AtzR, a LysR-type transcriptional regulator from Pseudomonas. The structure was determined with ARCIMBOLDO_SHREDDER using AlphaFold models (Castellvı ´et al, 2022).…”

“…This method is illustrated by the test case of the LysR-type transcriptional regulator AtzR (PDB entry 7z7j), which contains two copies of 300 residues in the asymmetric unit (Castellvı ´et al, 2022). The AlphaFold prediction yielded a model with an overall confidence of 92%.…”

Predicted models and CCP4

“…Crystal structures containing many copies in the asymmetric unit remain difficult to solve due to low signal to noise, even when using a highly accurate monomeric search model, although methods for predicting multimeric models are improving (Evans et al, 2022). There will also be crystal structures that may be difficult to solve with a predicted model due to the prediction favouring a particular conformation of a protein that can form different conformations (Chakravarty & Porter, 2022;Castellvı ´et al, 2022) or crystal structures containing complexes where one or more of the proteins is disordered unless in a complex (Dyson & Wright, 2005;Tunyasuvunakool et al, 2021). There remains a place for other methods that address the phase problem such as experimental phasing, but predicted models have made MR a reliable method in many cases that were previously the preserve of these other approaches (Terwilliger et al, 2023).…”