Physics-based method to validate and repair flaws in protein structures Osvaldo A. Martina,b, Yelena A. Arnautovac, Alejandro A. Icazattia, Harold A. Scheragab,1, and Jorge A. Vilaa,b,1 aInstituto de Matemática Aplicada San Luis, Consejo Nacional de Investigaciones Científicas y Técnicas de Argentina, Departamento de Física, Universidad Nacional de San Luis, 5700 San Luis, Argentina; bDepartment of Chemistry and Chemical Biology, Cornell University, Ithaca, NY 14853; and cMolsoft LLC, San Diego, CA 92121 Contributed by Harold A. Scheraga, August 26, 2013 (sent for review April 4, 2013) A method that makes use of information provided by the provide complementary information regarding the quality of a α β combination of 13C and 13C chemical shifts, computed at the given structural model. β density functional level of theory, enables one to (i) validate, at It is worth noting that the observed 13C chemical shifts have the residue level, conformations of proteins and detect backbone so far been used predominantly to determine conformational or side-chain flaws by taking into account an ensemble average of preferences of the backbones of polypeptide chains (13), al- chemical shifts over all of the conformations used to represent though they also contain very valuable information about side- a protein, with a sensitivity of ∼90%; and (ii) provide a set of chain conformations, which would be a very important contri- (χ1/χ2) torsional angles that leads to optimal agreement between bution to accurate validation, determination, and refinement of α β the observed and computed 13C and 13C chemical shifts. The protein models. Despite this, the use of chemical shifts for de- method has been incorporated into the CheShift-2 protein valida- termination and refinement of protein structures (14–17) is not tion Web server. To test the reliability of the provided set of (χ1/ the aim of this work, although a simple refinement routine is χ2) torsional angles, the side chains of all reported conformations used here only as a tool to assess the reliability of a proposed of five NMR-determined protein models were refined by a simple methodology to repair flaws in proteins. α routine, without using NOE-based distance restraints. The refine- Evidence has been accumulated showing that the 13C chem- ment of each of these five proteins leads to optimal agreement ical shift is determined mainly by its amino acid residue without α β between the observed and computed 13C and 13C chemical shifts significant influence of the nearest-neighbor residues, except for for ∼94% of the flaws, on average, without introducing a signifi- residues preceding proline (18–21). There is also evidence that cantly large number of violations of the NOE-based distance not only the backbone but also the side-chain conformation α restraints for a distance range ≤ 0.5 Å, in which the largest number influences the 13C chemical shift to some extent (22, 23). β α of distance violations occurs. The results of this work suggest that However, the question whether 13C rather than 13C chemical use of the provided set of (χ1/χ2) torsional angles together with shifts are more sensitive to χ1/χ2 side-chain torsional angles re- α β other observables, such as NOEs, should lead to a fast and accurate mains to be investigated; that is, to what extent are the 13C and 13C refinement of the side-chain conformations of protein models. chemical shifts of an amino acid residue in a protein affected by its side-chain orientation? This query is relevant to the fact that ince the seminal observation by Kendrew (1) that “it is the the three torsion angles ϕ, ψ, and χ1 are not independent of α Sspatial relations between the side-chains which determine the each other because they involve a common N−C group (24, 25). α chemical behavior and biological specificity of the protein mol- To answer this important question, we have expanded our 13C - ecule as a whole and these relations cannot be determined, ex- based CheShift-2 Web server (12) to include the computation of 13 β 13 β cept in a fragmentary manner, by purely chemical techniques” C chemical shifts. This database of C chemical shifts con- ∼ interest has been focused on the development of accurate tains 600,000 conformations and can be used (together with 13 α methods to validate and determine side-chain conformations in the database for C ) for a detailed analysis of the combined 13 α 13 β χ proteins (ref. 2 and references therein). Side-chain chemical dependence of the C and C chemical shifts on the 1 tor- fi χ shifts have also been used for protein structure validation be- sional angle for a given xed 2 for all 20 naturally occurring cause these observables are highly sensitive to protein structural fi packing (3). The latter interest arises because it is largely ac- Signi cance cepted that a proper protein structure determination requires validation methods in which the observable values used to vali- Protocols for NMR determination of high-resolution protein date the structures are not used in their determination (4, 5), and structures in solution require, among other things, an accurate it will assure spectroscopists and other users that a given protein method with which to assess the quality of protein structures. It is important that such a validation method provide in- model is a good representation of the native structure in solu- formation as to where structural flaws are and how they can tion. However, the validation process involves two crucial steps: be repaired. So far, no generally accepted validation method (i) detecting flaws in the structure at the residue level and (ii) fl with these characteristics appears to exist for evaluation of providing details as to how these aws can be repaired. Existing protein structures in solution. As an approach to find a solution validation methods are mainly concerned with determining the to this long-standing problem, we developed a method to quality of the whole structure but only sometimes with high- detect flaws in protein structures at the residue level and also lighting where the flaws are located at the residue level (3, 6–12). to provide a way to repair these flaws. To the best of our knowledge, these validation methods do not provide detailed information as to how such flaws can be Author contributions: O.A.M., H.A.S., and J.A.V. designed research; O.A.M., Y.A.A., A.A.I., eliminated. Therefore, it is left to the spectroscopists to find and J.A.V. performed research; O.A.M., Y.A.A., A.A.I., H.A.S., and J.A.V. analyzed data; and O.A.M., Y.A.A., H.A.S., and J.A.V. wrote the paper. a reliable solution for the detected structural problems. Con- The authors declare no conflict of interest. sequently, to develop a validation method capable of detecting 1 fl To whom correspondence may be addressed. E-mail: [email protected] or jv84@cornell. and repairing structural aws at the residue level, we focused edu. 13 α 13 β our effort on a combined use of C and C chemical shifts, This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. 13 α 13 β rather than use of separate C or C chemical shifts, to 1073/pnas.1315525110/-/DCSupplemental. 16826–16831 | PNAS | October 15, 2013 | vol. 110 | no. 42 www.pnas.org/cgi/doi/10.1073/pnas.1315525110 Downloaded by guest on September 25, 2021 β amino acids, not including the 13C chemical shifts for Ala and Gly, which do not contain a side chain with χ1 and χ2, and Pro, for which χ1 and χ2 are fixed. Overall, use of the updated version of the CheShift-2 Web server α β containing both 13C and 13C chemical shifts will enable users to (i) assess the quality of protein structures and detect flaws at the residue level by making use of the ensemble average of the chemical shifts computed over all of the conformations used to represent a protein and not the chemical shift values for one conformation and (ii) obtain possible solutions as to how these flaws can be fixed by generating a list of side-chain χ1 and χ2 torsional angles that can be used to improve the agreement be- α β tween observed and predicted 13C and 13C chemical shifts. A series of tests of the methodology includes one to determine the sensitivity of the updated version of the CheShift-2 Web server to validate protein structures at the residue level. In addition, the reliability of the provided set of (χ1/χ2) torsional angles to repair existent flaws is assessed here by refinement of the NMR- determined structures of five proteins (26), namely, 1D3Z, 2KIF, 2LQ9, 2LU1, and 2M2J. Results and Discussion α β Comparison of the Dependence of Separate 13C and 13C Chemical Shifts on the Variation of Backbone and Side-Chain Conformations. To examine the relative dependence of these two chemical shifts for a given residue on its conformational changes, we selected a set of variable χ1 torsional angles in 30° intervals for each of the most frequently observed χ2 torsional angles. Then, the Fig. 1. Structures of ubiquitin in the left-hand column were determined by α β mean values of each of the 13C and 13C chemical shifts over all NMR spectroscopy; those in the right-hand column were determined by X- values of χ1 for fixed ϕ, ψ, and χ2 and the corresponding stan- ray crystallography. Structures in A and B pertain to validation of proteins α β 13 α dard deviations, σ and σ , were computed.