Elimination of baseline artifacts in NMR spectra by oversampling

Abstract

Protein 3D structure determination by NMR spectroscopy in solution depends primarily on the collection of a large number of uniquely assigned intramolecular distance constraints ( I, 2). The ability to extract this information from NMR spectra, and in particular from mu ltidimensional spectra, may be lim ited by artifacts from instrumental instabilities (e.g., tl noise) and nonideal electronics (e.g., finite pulse length, filter characteristics). One problem that affects many NMR spectra in one way or another is baseline distortion. In order to evaluate such NMR spectra, automatic baseline correction routines are often applied. Such routines can lead to extensive calculations and they are not always successful in flattening the baseline. This Commun ication describes an alternative method which corrects the time-domain data to obtain a flat baseline in the resulting spectra. There are two ma jor sources leading to baseline distortions: phase correction of the resonance lines in the spectra and wrong intensities of the first few data points in the FID. Linear phase correction leads to baseline distortions (3) and should therefore be avoided. Depending on the experimental setup, even zeroth-order phase correction may distort the baseline if either single channel detection or quadrature channel detection with sequential sampling is used ( 4). The distorted data points at the beginning of a FID reflect the transient response of the spectrometer which is dominated by the characteristics of the audiofrequency filters used to prevent aliasing noise about the Nyquist frequency (5). Correcting the resulting errors by baseline correction in the frequency doma in has the disadvantage that the errors in only a few data points in the FID are spread over the whole spectrum of generally thousands of data points. Methods to correct the time-domain data seem therefore to be more suitable. The first few data points could be calculated by some extrapolation method from later, undistorted data points. Thus, for example, linear prediction methods were applied to reconstruct the beginning of the FIDs of 2D NMR measurements (6). Another method makes use of the filter characteristics to time properly the data acquisition (5 ) . It would also be possible to replace the narrowband analog audio filters by digital filtering, a solution which is not yet available on commercial spectrometers. Here an alternative method, based on oversampling ( 7)) is proposed. Oversampling allows one to open the audiofrequency filters and thereby reduce baseline distortions. Summing blockwise over n data points in the oversampled FID results in a “normal” size FID, where n stands for the ratio of the “normal” dwell time to the oversampling dwell time. If the oversampling ratio n is chosen large enough to guarantee a high-