Quantitative Protein Topography Measurements by High Resolution Hydroxyl Radical Protein Footprinting Enable Accurate Molecular Model Selection

Abstract

We report an integrated workflow that allows mass spectrometry-based high-resolution hydroxyl radical protein footprinting (HR-HRPF) measurements to accurately measure the absolute average solvent accessible surface area (<SASA>) of amino acid side chains. This approach is based on application of multi-point HR-HRPF, electron-transfer dissociation (ETD) tandem MS (MS/MS) acquisition, measurement of effective radical doses by radical dosimetry, and proper normalization of the inherent reactivity of the amino acids. The accuracy of the resulting <SASA> measurements was tested by using well-characterized protein models.Moreover, we demonstrated the ability to use <SASA> measurements from HR-HRPF to differentiate molecular models of high accuracy (< 3Å backbone RMSD) from models of lower accuracy (> 4Å backbone RMSD). The ability of <SASA> data from HR-HRPF to differentiate molecular model quality was found to be comparable to that of <SASA> data obtained from Xray crystal structures, indicating the accuracy and utility of HR-HRPF for evaluating the accuracy of computational models.

Accurate protein structure prediction with hydroxyl radical protein footprinting data

“…Notably, this agreement could be used to accurately distinguish models with backbone root-mean-square deviation (RMSD) greater than 4 Å from models with backbone RMSD less than 3 Å. For the first time, this validated the capability of using HRPF data to identify low RMSD computational models 19 . We have previously shown that the correlation between experimentally determined PFs and residue neighbor count can be exploited as a Rosetta scoring term to improve protein structure prediction 20 .…”

“…Our benchmark set consisted of four proteins (myoglobin, calmodulin, lysozyme, and low molecular weight protein tyrosine phosphatase (LMPTP)). These proteins had at least 15 labeled residues with residue-resolved PF data available 19,28,29 . To optimize the conical neighbor count calculation, we tested and identified an angle midpoint value of π/2 that balanced minimized NRMSE and larger R 2 values (Supplementary Table 1).…”

“…In this work, homology modeling was used to obtain models for lysozyme with RMSDs ranging from 1.2 to 4.6 Å 19 . In addition, it was possible to discriminate between models with backbone RMSD values greater than 4 Å and models with backbone RMSD values less than 3 Å for lysozyme 19 . As part of our ab initio modeling, we built models for lysozyme with RMSD values ranging from 4.24 to 28.14 Å.…”

“…We used similar templates for lysozyme as those used in ref. 19 . In addition to building homology models for lysozyme, we also built models for myoglobin, for which data was also presented in ref.…”

“…The marriage of HRPF data and computational techniques provides a unique opportunity for more accurate protein structure prediction. The recent innovative work of Xie and colleagues featured molecular dynamics (MD) simulations used cooperatively with HRPF protection factor (PF) data 19 . A strong correlation between experimentally determined PFs and solvent accessible surface area calculated from the MD simulations was observed.…”