Automatic assignment of methyl resonances using experimental NMR data and graph theory

• Main Author: Pritisanac, Iva
• Other Authors: Baldwin, Andrew ; Robinson, Carol
• Published: University of Oxford 2016
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.730004

Selective isotope labeling of methyl groups allows for atomic-resolution insight into the structures and dynamics of high molecular mass proteins by nuclear magnetic resonance (NMR) spectroscopy. Widespread application of the methodology has been limited due to the challenges and costs associated with assigning ¹H-¹³C resonances to particular amino acids within the protein. Here, I present a novel structure-based automatic assignment strategy, Methyl Assignment by Graph MAtching (MAGMA), which relates experimentally measured methyl-methyl connectivity (NOESY/DREAM) to inter-methyl distances extracted from a high-resolution protein structure. MAGMA features exact algorithms for graph-subgraph isomorphism and maximal common edge subgraph that can sample all theoretically possible methyl resonance assignments. MAGMA was applied to a benchmark of eight proteins for which high-resolution structures, NOESY or DREAM data, and methyl NMR assignments were available. On this benchmark, MAGMA provides a minimum of 31% and a maximum of 89% methyl residue assignments, with complete accuracy. On larger proteins in the benchmark MAGMA outperforms alternative automated methyl resonance assignment programs in both accuracy and coverage. I assessed the influence of different input structures on the accuracy of MAGMA assignments, and concluded that joining the assignments results from multiple structures guarantees accuracy in cases where the structural form in solution is unknown. Finally, I demonstrate the utility of MAGMA in two novel applications on molecular chaperones. In a ligand binding study involving the N-domain of human HSP90α, MAGMA confidently assigned 38% of the input methyl residues when the results were joined over two protein conformations. MAGMA's assignments were then combined with inter-molecular (ligand-methyl) NOEs and HADDOCK to generate models of the protein-ligand complex. The generated models, while featuring the correct ligand binding site, were of lower quality (0.5 - 2.5 Å higher ligand RMSD) when compared to models generated with previously published methyl assignments. Incorporating inter-molecular NOE restraints in the calculation enabled MAGMA to discriminate between certain ligand binding modes, and led to an increase in the number of confident assignments, demonstrating the potential for using such restraints in the future. Finally, a generic data collection protocol for MAGMA was established in a de novo assignment application on the 400 kDa archaeal small heat shock protein Hsp16.5. The protocol involves measurement of inter-methyl NOEs, determination of intra-residue NOEs between Leu-δ₁/δ₂ and Val-γ₁/γ₂ groups, and discrimination between Leu and Val resonances by preparation of an exclusively Val labelled protein sample. A high sparsity of inter-methyl NOE data obtained for this system led to predominantly ambiguous methyl resonance assignments, which can be used for guiding further experimental efforts.