Rapid acquisition of high-resolution three and four-dimensional heteronuclear NMR experiments by nonlinear sampling

For the structure determination of small to medium size proteins by solution NMR, 3D NOESY experiments are usually analysed for the collection of NOE-derived distance restraints, even though assignment ambiguity due to degeneracy of ¹H resonances remains for one of the ¹H dimensions. Further separation of the ambiguous ¹H dimension with the chemical shifts of directly bound ¹³C or ¹⁵N nuclei in 4D spectra is a straightforward way to resolve the degeneracy. Despite this advantage in analysis, 4D NMR experiments are however infrequently used as a routine tool in protein NMR projects due to the long duration of the measurement and limited digital resolution. It would therefore be advantageous to be able to measure 4D NOESY spectra with good digital resolution in affordable measurement time. NMR spectroscopy is an inherently insensitive technique, thus new acquisition schemes for speeding up multidimensional NMR experiments [1] are demanded for dramatic improvements in both sensitivity and resolution. Recently, nonlinear sampling for indirectly acquired dimensions (also called non-uniform sampling or sparse sampling) [2-4] in combination with non-Fourier transform data processing methods have been proposed to be beneficial for 4D NMR experiments. Since discrete Fourier transform (FT) cannot be used for processing sparsely sampled data, maximum entropy (MaxEnt) methods [5,6] have been utilised for reconstructing nonlinearly sampled multi-dimensional NMR data. One of the major criticisms to non-FT methods is their questionable reliability in reproducing cross peaks with proper signal intensity, especially in the case of signals with a wide dynamic range as in NOESY-type experiments. However, the artefacts arising from MaxEnt processing in NOESY spectra have not yet been clearly assessed in comparison with other methods, such as quantitative maximum entropy [7], multidimensional decomposition [8], and compressed sensing [9,10]. I applied MaxEnt processing to 3D ¹⁵N-separated and ¹³C-separated NOESY of a small protein, the Thermus thermophilus HB8 TTHA1718 gene product, and compared its reliability in reproducing accurate signal intensity from nonlinearly sampled data with the alternative approaches. My results demonstrated that MaxEnt is robust, quick and competitive with other methods. Next, nonlinear sampling and 3D MaxEnt processing were applied to 4D ¹³C/¹⁵N-separated and ¹³C/¹³C-separated NOESY of TTHA1718. It is remarkable that 3D MaxEnt successfully reconstructed 4D NOESY, particularly the 4D ¹³C/¹³C-NOESY spectra. 4D ¹³C/¹³C-NOESY spectra have large diagonal (self-correlated) cross peaks, and their large dynamic range in peak intensity have been thought to be problematic for MaxEnt processing. The effect of the artefacts arising by employing nonlinear sampling and 3D MaxEnt processing to 4D NOESY spectra was evaluated by calculating structures based on the NOE-derived distance restraints obtained in the MaxEnt-processed 4D NOESY spectra with various reduced numbers of sampling points. My results indicated that sufficiently converged and accurate structures (RMSD of 0.91 Å to the mean and 1.36 Å to the reference structures) were obtained even with NOESY spectra reconstructed from 1.6% randomly selected sampling points for indirect dimensions. This suggests that 3D MaxEnt processing in combination with nonlinear sampling schedules is a useful and advantageous option for rapid acquisition of high-resolution 4D NOESY spectra of proteins. However, the mis-calibration of intensities started having a significant effect to decrease the numbers of picked NOE cross peaks in the case of much reduced data points, e.g. in the case of 1/128 random sampling points, thereby affecting the accuracy of the calculated structures. Nevertheless, my results clearly showed that 4D NOESY with nonlinear sampling is very advantageous to rapid determination of accurate global folds, particularly for cases suffering from short sample life times or low sensitivity.

収集根拠 : 博士論文（自動収集）
資料形態 : テキストデータ
コレクション : 国立国会図書館デジタルコレクション > デジタル化資料 > 博士論文
For the structure determination of small to medium size proteins by solution NMR, 3D NOESY experiments are usually analysed for the collection of NOE-derived distance restraints, even though assignment ambiguity due to degeneracy of ¹H resonances remains for one of the ¹H dimensions. Further separation of the ambiguous ¹H dimension with the chemical shifts of directly bound ¹³C or ¹⁵N nuclei in 4D spectra is a straightforward way to resolve the degeneracy. Despite this advantage in analysis, 4D NMR experiments are however infrequently used as a routine tool in protein NMR projects due to the long duration of the measurement and limited digital resolution. It would therefore be advantageous to be able to measure 4D NOESY spectra with good digital resolution in affordable measurement time. NMR spectroscopy is an inherently insensitive technique, thus new acquisition schemes for speeding up multidimensional NMR experiments [1] are demanded for dramatic improvements in both sensitivity and resolution. Recently, nonlinear sampling for indirectly acquired dimensions (also called non-uniform sampling or sparse sampling) [2-4] in combination with non-Fourier transform data processing methods have been proposed to be beneficial for 4D NMR experiments. Since discrete Fourier transform (FT) cannot be used for processing sparsely sampled data, maximum entropy (MaxEnt) methods [5,6] have been utilised for reconstructing nonlinearly sampled multi-dimensional NMR data. One of the major criticisms to non-FT methods is their questionable reliability in reproducing cross peaks with proper signal intensity, especially in the case of signals with a wide dynamic range as in NOESY-type experiments. However, the artefacts arising from MaxEnt processing in NOESY spectra have not yet been clearly assessed in comparison with other methods, such as quantitative maximum entropy [7], multidimensional decomposition [8], and compressed sensing [9,10]. I applied MaxEnt processing to 3D ¹⁵N-separated and ¹³C-separated NOESY of a small protein, the Thermus thermophilus HB8 TTHA1718 gene product, and compared its reliability in reproducing accurate signal intensity from nonlinearly sampled data with the alternative approaches. My results demonstrated that MaxEnt is robust, quick and competitive with other methods. Next, nonlinear sampling and 3D MaxEnt processing were applied to 4D ¹³C/¹⁵N-separated and ¹³C/¹³C-separated NOESY of TTHA1718. It is remarkable that 3D MaxEnt successfully reconstructed 4D NOESY, particularly the 4D ¹³C/¹³C-NOESY spectra. 4D ¹³C/¹³C-NOESY spectra have large diagonal (self-correlated) cross peaks, and their large dynamic range in peak intensity have been thought to be problematic for MaxEnt processing. The effect of the artefacts arising by employing nonlinear sampling and 3D MaxEnt processing to 4D NOESY spectra was evaluated by calculating structures based on the NOE-derived distance restraints obtained in the MaxEnt-processed 4D NOESY spectra with various reduced numbers of sampling points. My results indicated that sufficiently converged and accurate structures (RMSD of 0.91 Å to the mean and 1.36 Å to the reference structures) were obtained even with NOESY spectra reconstructed from 1.6% randomly selected sampling points for indirect dimensions. This suggests that 3D MaxEnt processing in combination with nonlinear sampling schedules is a useful and advantageous option for rapid acquisition of high-resolution 4D NOESY spectra of proteins. However, the mis-calibration of intensities started having a significant effect to decrease the numbers of picked NOE cross peaks in the case of much reduced data points, e.g. in the case of 1/128 random sampling points, thereby affecting the accuracy of the calculated structures. Nevertheless, my results clearly showed that 4D NOESY with nonlinear sampling is very advantageous to rapid determination of accurate global folds, particularly for cases suffering from short sample life times or low sensitivity.