Accelerated Nuclear Magnetic Resonance Spectroscopy with Deep Learning

Nuclear magnetic resonance (NMR) spectroscopy serves as an indispensable tool in chemistry and biology but often suffers from long experimental times. We present a proof‐of‐concept of the application of deep learning and neural networks for high‐quality, reliable, and very fast NMR spectra reconstruction from limited experimental data. We show that the neural network training can be achieved using solely synthetic NMR signals, which lifts the prohibiting demand for a large volume of realistic training data usually required for a deep learning approach.     

NMR Characterization of Complex p‐Oligophenyl Scaffolds by Means of Aliasing Techniques to Obtain Resolution‐Enhanced Two‐Dimensional Spectra

The usefulness of computer‐assisted aliasing to secure maximal resolution of signal clusters in ¹H‐ and ¹³C‐NMR spectra (which is essential for structure determination by HMBC 2D NMR spectroscopy) in minimal acquisition time is exemplified by the complete characterization of the two complementary p‐octiphenyls 1 and 2 with complex substitution patterns. The need for digital resolution near 1 Hz/pt to dissect the extensive signal clusters in the NMR spectra of these refined oligomers excluded structure determination under routine conditions. High resolution was secured by exploiting the low signal density in the ¹³C dimension of HMBC spectra by using computer‐assisted aliasing to maximize signal density. Based on the observed shifts in DEPT and ¹H‐decoupled ¹³C‐NMR spectra of 1 and 2 , computer‐assisted aliasing allowed to reduce the number of required time increments by a factor of 20 to 30 compared to full‐width spectra with identical resolution. Without signal‐to‐noise constraints, this computer‐assisted aliasing reduced the acquisition time for high‐resolution NMR spectra needed for complete characterization of refined oligomers 1 and 2 by the same factor (e.g., from over a day to about an hour). With resolved signal clusters in fully aliased HSQC and HMBC spectra, unproblematic structure determination of 1 and 2 is demonstrated by unambiguous assignment of all C‐ and H‐atoms. These findings demonstrate that computer‐assisted aliasing of the underexploited ¹³C dimension makes extensive molecular complexity accessible by conventional multidimensional heteronuclear NMR experiments without extraordinary efforts.     

Study of near‐symmetric cyclodextrins by compressed sensing 2D NMR

The compressed sensing NMR (CS‐NMR) is an approach to processing of nonuniformly sampled NMR data. Its idea is to introduce minimal l_p‐norm (0 < p ≤ 1) constraint to a penalty function used in a reconstruction algorithm. Here, we demonstrate that 2D CS‐NMR spectra allow the full spectral assignment of near‐symmetric β‐cyclodextrin derivatives (mono‐modified at the C6 position). The application of CS‐NMR ensures experimental time saving and the resolution improvement, necessary because of very low chemical shift dispersion. In the overnight experimental time, the set of properly resolved 2D NMR spectra required for the unambiguous assignment of mono(6‐deoxy‐6‐(1‐1,2,3‐triazo‐4‐yl)‐1‐propane‐3‐O‐(phenyl)) β‐cyclodextrin was obtained. The highly resolved HSQC spectrum was reconstructed from 5.12% of the data. Moreover, reconstructed 2D HSQC–TOCSY spectrum yielded information about the correlations within one sugar unit, and 2D HSQC–NOESY technique allowed the sequential assignment of the glucosidic units. Copyright © 2013 John Wiley & Sons, Ltd.     

High‐field 95Mo and 183W static and MAS NMR study of polyoxometalates

The potential of high‐field NMR to measure solid‐state ⁹⁵Mo and ¹⁸³W NMR in polyoxometalates (POMs) is explored using some archetypical structures like Lindqvist, Keggin and Dawson as model compounds that are well characterized in solution. NMR spectra in static and under magic angle spinning (MAS) were obtained, and their analysis allowed extraction of the NMR parameters, including chemical shift anisotropy and quadrupolar coupling parameters. Despite the inherent difficulties of measurement in solid state of these low‐gamma NMR nuclei, due mainly to the low spectral resolution and poor signal‐to‐noise ratio, the observed global trends compare well with the solution‐state NMR data. This would open an avenue for application of solid‐state NMR to POMs, especially when liquid‐state NMR is not possible, e.g., for poorly soluble or unstable compounds in solution, and for giant molecules with slow tumbling motion. This is the case of Keplerate where we provide here the first NMR characterization of this class of POMs in the solid state. Copyright © 2017 John Wiley & Sons, Ltd.     

Semi‐preparative LC‐SPE‐cryoflow NMR for impurity identifications: use of mother liquor as a better source of impurities

Unambiguous structural elucidation of active pharmaceutical ingredients (API) impurities is a particularly challenging necessity of pharmaceutical development, particularly if the impurities are low level (0.1% level). In many cases, this requires acquiring high‐quality NMR data on a pure sample of each impurity. High‐quality, high signal‐to‐noise (S/N) one‐ and two‐dimensional NMR data can be obtained using liquid chromatography‐solid phase extraction‐cryoflow NMR (LC‐SPE‐cryoflow NMR) with a combination of semi‐preparative column for separation and mother liquor as a source of concentrated impurities. These NMR data, in conjunction with mass spectrometry data, allowed for quick and unambiguous structural elucidations of four impurities found at low level in the crystallized API but found at appreciable levels in the mother liquor that was used as the source for these impurities. These data show that semi‐preparative columns can be used at lower than ideal flow rates to facilitate trapping of HPLC components for LC‐SPE‐cryoflow NMR analysis without compromising chromatographic resolution. Also, despite the complex chromatography encountered with the use of mother liquor as a source of impurities, acceptably pure analytes were obtained for acquiring NMR data for unambiguous structure elucidations. Copyright © 2013 John Wiley & Sons, Ltd.     

Understanding J‐Modulation during Spatial Encoding for Sensitivity‐Optimized Ultrafast NMR Spectroscopy

Ultrafast (UF) NMR spectroscopy is an approach that yields 2D spectra in a single scan. This methodology has become a powerful analytical tool that is used in a large array of applications. However, UF NMR spectroscopy still suffers from an intrinsic low sensitivity, and from the need to compromise between sensitivity, spectral width, and resolution. In particular, the modulation of signal intensities by the spin–spin J‐coupling interaction (J‐modulation) impacts significantly on the intensities of the spectral peaks. This effect can lead to large sensitivity losses and even to missing spectral peaks, depending on the nature of the spin system. Herein, a general simulation package (Spinach) is used to describe J‐modulation effects in UF experiments. The results from simulations match with experimental data and the results of product operator calculations. Several methods are proposed to optimize the sensitivity in UF COSY spectra. The potential and drawbacks of the different strategies are also discussed. These approaches provide a way to adjust the sensitivity of UF experiments for a large range of applications.     

Ultrahigh-resolution (1)H-(13)C HSQC spectra of metabolite mixtures using nonlinear sampling and forward maximum entropy reconstruction

To obtain a comprehensive assessment of metabolite levels from extracts of leukocytes, we have recorded ultrahigh-resolution 1H-13C HSQC NMR spectra of cell extracts, which exhibit spectral signatures of numerous small molecules. However, conventional acquisition of such spectra is time-consuming and hampers measurements on multiple samples, which would be needed for statistical analysis of metabolite concentrations. Here we show that the measurement time can be dramatically reduced without loss of spectral quality when using nonlinear sampling (NLS) and a new high-fidelity forward maximum-entropy (FM) reconstruction algorithm. This FM reconstruction conserves all measured time-domain data points and guesses the missing data points by an iterative process. This consists of discrete Fourier transformation of the sparse time-domain data set, computation of the spectral entropy, determination of a multidimensional entropy gradient, and calculation of new values for the missing time-domain data points with a conjugate gradient approach. Since this procedure does not alter measured data points, it reproduces signal intensities with high fidelity and does not suffer from a dynamic range problem. As an example we measured a natural abundance 1H-13C HSQC spectrum of metabolites from granulocyte cell extracts. We show that a high-resolution 1H-13C HSQC spectrum with 4k complex increments recorded linearly within 3.7 days can be reconstructed from one-seventh of the increments with nearly identical spectral appearance, indistinguishable signal intensities, and comparable or even lower root-mean-square (rms) and peak noise patterns measured in signal-free areas. Thus, this approach allows recording of ultrahigh resolution 1H-13C HSQC spectra in a fraction of the time needed for recording linearly sampled spectra.     

Temperature as an Extra Dimension in Multidimensional Protein NMR Spectroscopy

NMR spectroscopy is a particularly informative method for studying protein structures and dynamics in solution; however, it is also one of the most time-consuming. Modern approaches to biomolecular NMR spectroscopy are based on lengthy multidimensional experiments, the duration of which grows exponentially with the number of dimensions. The experimental time may even be several days in the case of 3D and 4D spectra. Moreover, the experiment often has to be repeated under several different conditions, for example, to measure the temperature-dependent effects in a spectrum (temperature coefficients (TCs)). Herein, a new approach that involves joint sampling of indirect evolution times and temperature is proposed. This allows TCs to be measured through 3D spectra in even less time than that needed to acquire a single spectrum by using the conventional approach. Two signal processing methods that are complementary, in terms of sensitivity and resolution, 1) dividing data into overlapping subsets followed by compressed sensing reconstruction, and 2) treating the complete data set with a variant of the Radon transform, are proposed. The temperature-swept 3D HNCO spectra of two intrinsically disordered proteins, osteopontin and CD44 cytoplasmic tail, show that this new approach makes it possible to determine TCs and their non-linearities effectively. Non-linearities, which indicate the presence of a compact state, are particularly interesting. The complete package of data acquisition and processing software for this new approach are provided.     

Structural analyses of mannose pentasaccharide of high mannose type oligosaccharides by 1D and 2D NMR spectroscopy

NMR spectroscopy is a very important and useful method for the structural analysis of oligosaccharides, despite its low sensitivity. We first applied conventional measuring methods, 2D DQF COSY, ¹H–¹³C HSQC, and ¹H–¹³C HMBC, and also the Double Pulsed Field Gradient Spin Echo (DPFGSE)‐TOCSY and DPFGSE‐NOESY/ROESY techniques to analyze a branched mannose pentasaccharide as a model of high mannose type N‐glycans in natural abundance. The NMR spectra of the model compound are very complex and difficult to analyze owing to overlapping signals. The superior selective irradiation capability of the DPFGSE technique is useful for fine structural and conformational analyses of such complex oligosaccharides. We here introduce a novel technique called DPFGSE‐Double‐Selective Population Transfer (SPT)‐Difference and DPFGSE‐NOE/ROE‐SPT‐Difference spectroscopy. The DPFGSE‐Double‐SPT‐Difference method involves irradiation of two peaks from one proton and the subtraction of higher and lower peaks from each spectrum. The DPFGSE‐NOE/ROE‐SPT‐Difference method involves the transfer of the magnetization polarized by NOE/ROE from the nuclei to the spin‐coupled nuclei through scalar spin–spin interaction using the SPT method. Even if the signals in the NMR spectra overlap, each signal can be accurately assigned. In particular, DPFGSE‐NOE/ROE‐SPT‐Difference is very useful for identifying sugar connectivity. Copyright © 2012 John Wiley & Sons, Ltd.     

Complete 1H and 13C signal assignment of prenol‐10 with 3D NMR spectroscopy

The complete assignment of ¹H and ¹³C chemical shifts of natural abundance prenol‐10 is reported for the first time. It was achieved using 3D NMR experiments, which were based on random sampling of the evolution time space followed by multidimensional Fourier transform. This approach makes it possible to acquire 3D NMR spectra in a reasonable time and preserves high resolution in indirectly detected dimensions. It is shown that the interpretation of 3D COSY–HMBC and 3D TOCSY–HSQC spectra is crucial in the structural analysis of prenol‐10. Copyright © 2009 John Wiley & Sons, Ltd.     

Uniform Reduction of Scalar Coupling by Real‐Time Homonuclear J‐Downscaled NMR

Scalar coupling in proton NMR spectra provides important information for the structural analysis. However, the low resolution due to the resulting signal splitting, together with the rather narrow spectral range of hydrogen, often prevents the extraction of J‐coupling information. Here we present a method to achieve real‐time homonuclear J‐downscaling. Thereby, all J‐values are uniformly reduced by an arbitrary scaling factor. In the resulting one‐dimensional spectra, signal overlap is reduced, while scalar coupling information is still available.     

Accelerating Diffusion‐Ordered NMR Spectroscopy by Joint Sparse Sampling of Diffusion and Time Dimensions

Diffusion‐ordered multidimensional NMR spectroscopy is a valuable technique for the analysis of complex chemical mixtures. However, this method is very time‐consuming because of the costly sampling of a multidimensional signal. Various sparse sampling techniques have been proposed to accelerate such measurements, but they have always been limited to frequency dimensions of NMR spectra. It is now revealed how sparse sampling can be extended to diffusion dimensions.     

Chemical‐Shift‐Resolved 19F NMR Spectroscopy between 13.5 and 135 MHz: Overhauser–DNP‐Enhanced Diagonal Suppressed Correlation Spectroscopy

Overhauser–DNP‐enhanced homonuclear 2D ¹⁹F correlation spectroscopy with diagonal suppression is presented for small molecules in the solution state at moderate fields. Multi‐frequency, multi‐radical studies demonstrate that these relatively low‐field experiments may be operated with sensitivity rivalling that of standard 200–1000 MHz NMR spectroscopy. Structural information is accessible without a sensitivity penalty, and diagonal suppressed 2D NMR correlations emerge despite the general lack of multiplet resolution in the 1D ODNP spectra. This powerful general approach avoids the rather stiff excitation, detection, and other special requirements of high‐field ¹⁹F NMR spectroscopy.     

High magnetic field solid‐state NMR analyses by combining MAS,MQ‐MAS,homo‐nuclear and hetero‐nuclear correlation experiments

A general strategy of structural analysis of alumina silicate by combining various solid‐state NMR measurements such as single pulse, multi‐quantum magic angle spinning, double‐quantum homo‐nuclear correlation under magic angle spinning (DQ‐MAS), and cross‐polarization hetero‐nuclear correlation (CP‐HETCOR) was evaluated with the aid of high magnetic field NMR (800 MHz for ¹H Larmor frequency) by using anorthite as a model material. The high magnetic field greatly enhanced resolution of ²⁷Al in single pulse, DQ‐MAS, and even in triple‐quantum magic angle spinning NMR spectra. The spatial proximities through dipolar couplings were probed by the DQ‐MAS methods for homo‐nuclear correlations between both ²⁷Al–²⁷Al and ²⁹Si–²⁹Si and by CP‐HETCOR for hetero‐nuclear correlations between ²⁷Al–²⁹Si in the anorthite framework. By combining various NMR methodologies, we elucidated detailed spatial correlations among various aluminum and silicon species in anorthite that was hard to be determined using conventional analytical methods at low magnetic field. Moreover, the presented approach is applicable to analyze other alumina‐silicate minerals. Copyright © 2012 John Wiley & Sons, Ltd.     

Long‐Lived Spin States for Low‐Field Hyperpolarized Gas MRI

Parahydrogen induced polarization was employed to prepare a relatively long‐lived correlated nuclear spin state between methylene and methyl protons in propane gas. Conventionally, such states are converted into a strong NMR signal enhancement by transferring the reaction product to a high magnetic field in an adiabatic longitudinal transport after dissociation engenders net alignment (ALTADENA) experiment. However, the relaxation time T₁ of ～0.6 s of the resulting hyperpolarized propane is too short for potential biomedical applications. The presented alternative approach employs low‐field MRI to preserve the initial correlated state with a much longer decay time T_LLSS=(4.7±0.5) s. While the direct detection at low‐magnetic fields (e.g. 0.0475 T) is challenging, we demonstrate here that spin‐lock induced crossing (SLIC) at this low magnetic field transforms the long‐lived correlated state into an observable nuclear magnetization suitable for MRI with sub‐millimeter and sub‐second spatial and temporal resolution, respectively. Propane is a non‐toxic gas, and therefore, these results potentially enable low‐cost high‐resolution high‐speed MRI of gases for functional imaging of lungs and other applications.     

Kinetics from indirectly detected hyperpolarized NMR spectroscopy by using spatially selective coherence transfers

An important recent development in NMR spectroscopy is the advent of ex situ dynamic nuclear polarization (DNP) approaches, which are capable of yielding liquid‐state sensitivities that exceed considerably those afforded by the highest‐field spectrometers. This increase in sensitivity has triggered new research avenues, particularly concerning the in vivo monitoring of metabolism and disease by NMR spectroscopy. So far such gains have mainly materialized for experiments that focus on nonprotonated, low‐γ nuclei; targets favored by relatively long relaxation times T₁, which enable them to withstand the transfer from the cryogenic hyperpolarizer to the reacting centers of interest. Recent studies have also shown that transferring this hyperpolarization to protons by indirectly detected methods could successfully give rise to ¹H NMR spectra of hyperpolarized compounds with a high sensitivity. The present study demonstrates that, when merged with spatially encoded methods, indirectly detected ¹H NMR spectroscopy can also be exploited as time‐resolved hyperpolarized spectroscopy. A methodology is thus introduced that can successfully deliver a series of hyperpolarized ¹H NMR spectra over a minutes‐long timescale. The principles and opportunities presented by this approach are exemplified by following the in vitro phosphorylation of choline by choline kinase, a potential metabolic marker of cancer; and by tracking acetylcholine’s hydrolysis by acetylcholine esterase, an important enzyme partaking in synaptic transmission and neuronal degradation.     

Ultrafast-based projection-reconstruction three-dimensional nuclear magnetic resonance spectroscopy

Recent years have witnessed increased efforts toward the accelerated acquisition of multidimensional nuclear magnetic resonance (nD NMR) spectra. Among the methods proposed to speed up these NMR experiments is "projection reconstruction," a scheme based on the acquisition of a reduced number of two-dimensional (2D) NMR data sets constituting cross sections of the nD time domain being sought. Another proposition involves "ultrafast" spectroscopy, capable of completing nD NMR acquisitions within a single scan. Potential limitations of these approaches include the need for a relatively slow 2D-type serial data collection procedure in the former case, and a need for at least n high-performance, linearly independent gradients and a sufficiently high sensitivity in the latter. The present study introduces a new scheme that comes to address these limitations, by combining the basic features of the projection reconstruction and the ultrafast approaches into a single, unified nD NMR experiment. In the resulting method each member within the series of 2D cross sections required by projection reconstruction to deliver the nD NMR spectrum being sought, is acquired within a single scan with the aid of the 2D ultrafast protocol. Full nD NMR spectra can thus become available by backprojecting a small number of 2D sets, collected using a minimum number of scans. Principles, opportunities, and limitations of the resulting approach, together with demonstrations of its practical advantages, are here discussed and illustrated with a series of three-dimensional homo- and heteronuclear NMR correlation experiments.     

Development of variance‐filtered instrumental transfer methods for high‐resolution NMR spectroscopy

The introduction of a variance‐filter to both direct standardization (DS) and piece‐wise direct standardization (PDS) instrumental transfer methods for the analysis of NMR spectral data is described. The variance‐filter modification allows for the identification of regions in the NMR spectra that are not adequately represented by the limited number of transfer calibration samples used during the calculation of the instrument‐to‐instrument transfer matrix. For these spectral frequencies, the corresponding portion of the transfer matrix is replaced by identity (or scaled identity) prior to the secondary instrumental data sets being transferred to the target instrument response. The spectral matching performance of the variance‐filtered instrumental transfer method as applied to high‐resolution ¹H NMR spectra is presented along with possible uses and limitations. Copyright © 2010 John Wiley & Sons, Ltd.     

On‐Line Monitoring of Chemical Reactions by using Bench‐Top Nuclear Magnetic Resonance Spectroscopy

Real‐time nuclear magnetic resonance (NMR) spectroscopy measurements carried out with a bench‐top system installed next to the reactor inside the fume hood of the chemistry laboratory are presented. To test the system for on‐line monitoring, a transfer hydrogenation reaction was studied by continuously pumping the reaction mixture from the reactor to the magnet and back in a closed loop. In addition to improving the time resolution provided by standard sampling methods, the use of such a flow setup eliminates the need for sample preparation. Owing to the progress in terms of field homogeneity and sensitivity now available with compact NMR spectrometers, small molecules dissolved at concentrations on the order of 1 mmol L^?1 can be characterized in single‐scan measurements with 1 Hz resolution. Owing to the reduced field strength of compact low‐field systems compared to that of conventional high‐field magnets, the overlap in the spectrum of different NMR signals is a typical situation. The data processing required to obtain concentrations in the presence of signal overlap are discussed in detail, methods such as plain integration and line‐fitting approaches are compared, and the accuracy of each method is determined. The kinetic rates measured for different catalytic concentrations show good agreement with those obtained with gas chromatography as a reference analytical method. Finally, as the measurements are performed under continuous flow conditions, the experimental setup and the flow parameters are optimized to maximize time resolution and signal‐to‐noise ratio.