PSI Structural Biology Knowledgebase

PSI | Structural Biology Knowledgebase
Header Icons

Related Theme

Removing the NMR bottleneck

PSI-SGKB [doi:10.1038/th_psisgkb.2010.13]
Technical Highlight - April 2010
Short description: Protein backbone information in combination with the structure-prediction method Rosetta are sufficient for accurate structures up to 25 kDa.

Ensemble of lowest energy Rosetta structures for ALG13

Protein structure determination using nuclear magnetic resonance (NMR) usually relies on establishing the distances between protons in side chains. It can be very hard to pinpoint these proton resonances in spectra, and it is possible to make mistakes, which makes assigning side-chain resonance a slow process. A paper in Science shows that combining the structure-prediction method Rosetta with relatively simple-to-obtain NMR data produces accurate structures for proteins of up to 200 residues.

David Baker, from the University of Washington, Seattle, and his team developed Rosetta. This structure-prediction method relies on the native structure almost always having the lowest energy. The difficulty is in finding that structure, because the computer algorithm can be tricked into following the second or third lowest energy structure and never finding the lowest one of all.

This problem can be solved by including NMR chemical-shift information, particularly long-range data from residual dipolar couplings (RDC). The Rosetta method uses a Monte Carlo approach — in which random or almost random changes are made to the virtual protein structure over and over again to find a solution. If RDCs are added in, each Monte Carlo change can be compared with the NMR information and accepted if it fits the data better and rejected if not.

Using RDCs in this way, proteins of up to 120 residues can generate accurate models. But for more than 120 residues, RDC information is not sufficient, and so backbone NOE (nuclear Overhauser effect) interactions are added in. An iterative procedure using the lowest energy conformations from one round as a starting point for subsequent rounds is used.

When this new method was blind-tested on five data sets from PSI NESG before the NMR structures were determined, the team found that for four of these, the RDC-dependent protocol worked well and produced structures very similar to those later solved by NMR. For the fifth, a combination of RDC and backbone NOEs was needed, but it too resulted in a structure close to the experimentally solved one. For all these structures, side-chain positions were accurately modeled, without inputting any data that could define their positions.

This combination of NMR constraints and the Rosetta method is set to speed up NMR protein solution and open up the possibilities of routinely solving much larger structures than we are currently able to do.

Related articles

NMR has its wiki way

How does Dali work?

The future of NMR

GPCR modeling: any good?

Maria Hodges


  1. S Raman et al. NMR structure determination for larger proteins using backbone-only data.
    Science 327, 1014-1018 (2009). doi:10.1126/science.1183649

Structural Biology Knowledgebase ISSN: 1758-1338
Funded by a grant from the National Institute of General Medical Sciences of the National Institutes of Health