Automated NMR structure refinement via NOE peak volumes. Application to a dodecamer DNA duplex

Abstract

In NMR structure determination, the initial estimation of distances from NOE cross peaks usually involves the use of a two-spin approximation and the resulting distance geometry initial structures satisfy a set of initial distances that contain some spin-diffusion errors. In order to account for spin diffusion in a multiple-spin system, NOE back-calculation methods are frequently employed, using either numerical integration ( I, 2) or relaxation-matrix methods (3). Iterative back-calculation structure refinements have been carried out by visual comparison of simulated cross-peak intensities (4-6) and by full relaxation-matrix methods ( 7-10) to produce refined structures whose simulated spectra better match the experimental spectra. For visual comparison against experimental data, simulated spectra are generated with the appropriate chemical shifts, linewidths, and lineshapes. NOESY cross-peak volumes depend not only on peak height but also on linewidths and lineshapes (multiplet splitting) that, if incorrect, can lead to volume errors in matching peaks of similar intensity. The relaxation-matrix approaches IRMA ( 7, 8) and MORASS ( 10) use a “hybrid matrix,” that is, a combination of experimental volumes and simulated volumes generated from the coordinates of the current model structure. In the hybrid-matrix approach, distance constraints are not determined by the actual intensity differences between experimental and calculated cross peaks, but by the distance matrix derived from the hybrid volume matrix; because of this, the hybrid-matrix method does not minimize the R factor (the average NOE intensity difference between experimental and calculated cross-peak volumes) very rapidly or efficiently. For example, Koning et al. (8) reported that the average R factor for a simulation of an octamer DNA duplex was about 0.5 after the first cycle of refinement and improved relatively little during further iterations. Furthermore, the relaxation-matrix structural-refinement procedure is confined to fully resolved NOE cross peaks (unless independent spectral deconvolution methods are applied to partially overlapped peaks), while visual inspection can interpolate partially overlapped multiple peaks reasonably accurately. We have developed an automated back-calculation refinement method that operates on all NOE cross peaks, including overlapped cross peaks, by using experimental