Pure Shift Nuclear Magnetic Resonance: a New Tool for Plant Metabolomics

Summary

This paper presents the use of PSYCHE and SAPPHIRE-PSYCHE in the metabolic profiling of plants and includes detailed procedures for sample preparation and optimal Pure Shift NMR spectra recording. Examples through which the gain in resolution achieved by homonuclear decoupling allows a more comprehensive understanding of the system are discussed.

Abstract

Nuclear Magnetic Resonance (NMR) is one of the most powerful tools used in metabolomics. It stands as a highly accurate and reproducible method that not only provides quantitative data but also permits structural identification of the metabolites present in complex mixtures.

Metabolic profiling by ¹H NMR has proven useful in the study of various types of plant scenarios, which include the evaluation of crop conditions, harvest and post- harvest treatments, metabolic phenotyping, metabolic pathways, gene regulation, identification of biomarkers, chemotaxonomy, quality control, denomination of origin, among others. However, signal overlapping of the large number of resonances with expanded J-coupling multiplicities complicates the spectra analysis and its interpretation, and represents a limitation for classical ¹H NMR profiling.

In the last decade, novel NMR broadband homonuclear decoupling techniques through which multiplet signals collapse into single resonance lines - commonly called Pure Shift methods - have been developed to overcome the spectra resolution problem inherent to ¹H NMR classical spectra.

Here a step-by-step protocol of the plant extract preparation and the procedure to record optimal Pure Shift PSYCHE and SAPPHIRE-PSYCHE spectra in three different plant matrices - Vanilla plant leaves, potato tubers (S. tuberosum), and Cape gooseberries (P. peruviana) - is presented. The effect of the gain in resolution in metabolic identification, correlation analysis and multivariate analyses, as compared against classical spectra, is discussed.

Introduction

The complete set of metabolites that comprise an organism - substrates, intermediates, and end products of biological processes - was coined in 1998 with the term, metabolome. It is well known that the metabolome is closely related to the phenotype, and it is of particular interest in plants as it reflects the direct interaction between the genotype and the environment^1,2. Hence, the characterization of the metabolomic profile has become of paramount importance in plants. Through the identification and quantification of biomarkers (key metabolites) and metabolic patterns, the discrimination between species, cultivars, development stages, pathogenic diseases, or environmental conditions (daily and seasonal changes, soils, water stress, mechanical stress, harvest and post- harvest treatments), among others, has been possible^3,4,5.

Mass spectrometry (MS) and Nuclear Magnetic Resonance (NMR) spectroscopy are the most widely used analytical platforms for this purpose. Contrary to MS methodologies, NMR stands as a highly reproducible, non-biased, quantitative, accurate, and non-destructive technique that requires minimal sample preparation, making it suitable for metabolomics studies. However, when compared to MS methods, the inherent low sensitivity is a limitation. In recent years and through the use of high-field magnets, cryogenic probes, micro-coil devices, and Dynamic Nuclear Polarization (DNP) methods, the sensitivity of NMR has been greatly improved. In the case of the latter approach, for instance, the sensitivity gain was in the level of two to three orders of magnitude^6,7. To date, almost 20% of the published metabolomics studies are NMR-based and the number is rising⁷.

Even though Proton NMR is the most popular and sensitive experiment for NMR metabolomics fingerprinting, it has some drawbacks. First, all the ¹H NMR signals detected in the sample are distributed in a small window corresponding to the proton chemical shift window, which results in crowded spectra. Second, the homonuclear scalar coupling splits the signals into multiple components (signal multiplicity), spreading the proton signal over a wider frequency range, complicating furthermore the spectra reading by increasing crowding and signal overlapping. In addition, NMR metabolomics is employed in the analysis of mixtures usually containing 50 to 300 molecules at an NMR observable concentration, generating complex spectra comprised of 200 to 2000 peaks.

Homonuclear decoupling proton NMR, also known as Pure Shift, is a method that induces the collapse of a multiplet signal into a single peak. It stands as an excellent tool for increasing signal resolution in crowded spectra^8,9,10 and therefore represents a convenient tool for plants metabolomics¹¹.

In the last decade, new Pure Shift pulse sequences, increasing both sensitivity and decoupling performance, have emerged. Their range of applications have also expanded, from molecular structure elucidation^12,13, to fluxomics¹⁴, mixture assignment^15,16,17, translational diffusion measurements¹⁸, enantiomeric discrimination¹⁹, unit distribution in co-polymers²⁰, among others.

Historically, Broadband Pure Shift experiments suffer from low sensitivity and complicated processing methods, limiting their scope in the assessment of biological extracts⁸. In 2014, Foroozandeh et al. published a new Pure Shift experiment, PSYCHE (Pure Shift Yielded by Chirp Excitation), based on anti-z-COSY pulse sequence which yielded excellent homonuclear decoupling and improved sensitivity values²¹. However, as PSYCHE is a 2D interferogram experiment where chunks of time domain data are acquired, it suffers from periodic sideband artifacts that result from J-coupling modulation distortions at the edges of the chunk. In complex mixtures, these artifacts yield signals larger than those associated with metabolites present at very low concentrations, hindering the analysis¹¹. There are two methods to remove these artifacts - TSE-PHYCHE²² and a more recent modification of the PSYCHE experiment called SAPPHIRE-PSYCHE (Sideband Averaging by Periodic PHase Incrementation of Residual J Evolution)²³.

In 2019, we demonstrated for the first time¹¹ that the SAPPHIRE-PSYCHE Pure Shift method, which removes artifacts with almost no sensitivity penalty²³, could be employed for the analysis of complex biological mixtures, such as extracts of the fruits of Physalis peruviana, commonly known as Cape gooseberries¹¹. We showed that these methods increase the performance of metabolomics data analyses such as metabolic assignment, correlation analysis and multivariate coefficients analysis¹¹. Since then, several Pure Shift metabolomics studies on different biological matrices, such as soft corals²⁴, hypericum plants²⁵, honey^26,27, tea²⁷, peppermint oil²⁶, and walnuts²⁸ have been addressed, demonstrating its importance as a new tool for metabolomics analysis. Paradoxically, the vast majority of these studies employed the standard and easy to implement PSYCHE pulse sequence, available from any spectrometer library, instead of the SAPPHIRE-PSYCHE pulse sequence, which has been shown to perform better. However, it requires better understanding of the pulse sequence for proper setup.

This paper is intended to help new users to apply Pure Shift methods in the study of plants, in particular, leaves of Vanilla sp (V. planifolia and V. pompona)²⁹, potato tubers (S. tuberosum)³⁰, and Cape gooseberries (P. peruviana)³¹. Sample preparation, NMR experimental set up, data acquisition, and data analysis are described in detail. Moreover, the protocol includes key notes to help researchers, new to the field, to properly set up PSYCHE and SAPPHIRE-PSYCHE experiments in the metabolomic profiling of plants.

Protocol

1. Sample preparation

1. Cape gooseberries
   1. Place 100-200 g of fresh fruits in a blender vase. Keep at 4 °C for 30 min, and then homogenize in a laboratory blender.
   2. Immediately, transfer the juice to 50 mL plastic tubes, freeze them in liquid nitrogen and lyophilize to dryness for 4 to 5 days.
   3. Grind the lyophilized material to a fine powder using an electric grinder.
      NOTE: Handling of the dry material needs to be done quickly because the powder is highly hygroscopic.
   4. Weigh 1 g of the ground material and add 10 mL of ultrapure water. Vortex for 1 min.
   5. Sonicate for 20 min at 10 °C, and then centrifuge at 23,000 × g for 20 min at 10 °C.
   6. Recover the supernatant and filter it through a 13 mm polytetrafluoroethylene (PTFE) 0.45 µm syringe filter.
   7. Lyophilize 1 mL of the filtered extract to dryness and then re-suspend the obtained solid in 0.9 mL of 200 mM sodium oxalate buffer pH 4. Vortex.
   8. Lyophilize the resulting sample to dryness and dissolve it in 0.9 mL of deuterium oxide containing 5 mM of 3-(trimethylsilyl)propionic-2,2,3,3-d₄ acid sodium salt (TMSP-d₄).
   9. Fill the NMR tube with 0.6 mL of the sample using a micropipette.
2. Vanilla leaves
   1. Collect the leaves, clean them with damp tissue paper and freeze them whole in liquid nitrogen.
   2. Break the leaves into small pieces and lyophilize for 4 days until dryness.
   3. Grind the dry matter to a fine powder using an electric grinder.
   4. Weigh 50 mg of ground material and add 0.75 mL phosphate buffer pH 6.0 in deuterium oxide containing 0.1% of TMSP (w/w) and 0.75 mL methanol-d4. Vortex for 1 min.
   5. Sonicate for 20 min at 25 °C.
   6. Centrifuge at 13,000 × g for 10 min at 25 °C.
   7. Recover the supernatant (~1.3-1.4 mL) and filter it through a 13 mm PTFE 0.45 µm syringe filter.
   8. Fill the NMR tube with 0.6 mL of the filtered sample using a micropipette.
3. Potato tubers
   1. Peel and slice 4 to 8 tubers. Immediately, place approximately 125 g of material in stand-up bags and freeze them in liquid nitrogen.
      ​NOTE: To avoid oxidation during handling keep the potato damp.
   2. Lyophilize for 4 to 6 days until complete dryness.
   3. Grind the dry matter to a fine powder using an electric grinder.
   4. Weigh 160 mg of ground tuber and add 1.6 mL of deionized water. Vortex for 1 min.
   5. Sonicate for 45 min at 10 °C.
   6. Centrifuge at 23,000 x g for 20 min at 10 °C.
   7. Recover the supernatant (~1.5 -1.6 mL) and evaporate it to dryness in a refrigerated centrifugal vacuum concentrator for 16 h, at 10 °C.
   8. Add to the solid obtained (20-25 mg) 0.9 mL of 100 mM sodium oxalate buffer pH 4, vortex, and evaporate for 16 hours at 10 °C.
   9. Dissolve the solid obtained in 0.9 mL of deuterium oxide containing 3 mM of TMSP.
   10. Centrifuge at 23,000 x g for 5 min at 10 °C and filter the supernatant directly into the NMR tube through a 13 mm PTFE 0.45 µm syringe filter.
       ​NOTE: In this case, direct filtration into the NMR tube was performed to diminish steps in the preparation of more than 1000 samples.

2. NMR Data Acquisition and Processing

1. NMR initial setup
   1. Transfer the samples to the NMR spectrometer.
   2. Tune and match the probehead.
   3. Lock and shim the sample.
   4. Calibrate the 90° hard pulse. Calibrate the 90° pulse using any of the standard procedures.
   5. Run a standard 1D proton NMR spectrum.
2. PSYCHE experiment
   1. Select the reset_psyche_1d pulse sequence from the Bruker Topspin library (Figure S1). Use the following standard parameters: 5 kHz spectral width (SW2), at least 1 or 2 seconds of relaxation recovery delay (D1), 16 dummy scans (DS), 64 or 128 complex data points per block (L31), and 64 or 128 scans (NS) (Figure S1).
      NOTE: L31 is the number of complex digital points acquired in each Pure Shift block, best to be set to a power of 2.²¹
   2. Set the desired CHIRP pulse flip angle excitation (CNST61) and 10 kHz for the CHIRP pulse bandwidth (CNST60) (Figure S2).
      NOTE: The PSYCHE experiment is based on an anti-z-COSY scheme; consequently the CHIRP pulse flip angle needs to be small to avoid recoupling artifacts (Figure 1). The absolute intensity increases with the excitation flip angle. The periodic artifacts are also enhanced, spreading into the spectrum and increasing the "noise" (Figure 1). The "noise" becomes a combination of standard noise and chuncking artifacts. A good compromise between sensitivity and low recoupling artifacts is to set CNST61 = 20°.^19,22
   3. Set the hard pulse length (P1) to the previously calibrated value and the PSYCHE shape pulse length to 30 ms (P49) (Figure S2).
      NOTE: It is very important to calibrate the hard pulse value as the shape pulse powers will be automatically calculated from this value.
   4. Choose the Crp_psyche.20 (SPNAM 37) shape pulse for the PSYCHE element (Figure S2).
   5. Set the strength of the pulse field gradient applied during the PSYCHE element (GPZ0). Choose RECT.1 for the gradient shape pulse (GPNAM 0) (Figure S2).
      NOTE: A weak magnetic field gradient is applied during the PSYCHE element, normally, between 1% to 4% of the maximum strength of the gradient, depending on the probe.
   6. Set the number of blocks to acquire in order to reconstruct the Pure Shift FID (TD1) (Figure S3).
      NOTE: PSYCHE is acquired as a pseudo-2D experiment where TD1 is the number of Pure Shift interferogram blocks. The spectrum resolution depends on the size of the spectral window (SW1) and the total number of acquired points, which is TD1*2*L31. Typically, 16 or 32 blocks with 64 or 128 complex points per block will provide enough digital resolution. As PSYCHE is recorded in an interferogram manner, a higher number of blocks increase the digital resolution, but, also the total acquisition time¹⁹. Homonuclear J-couplings evolve during each block resulting in an oscillating J modulation pattern^21,23. After Fourier transform, this generates periodic sideband artifacts that depend on the length of the block (Figure 2). To reduce artifacts, the duration of the block must be short, typically less than 16 ms (block duration = 2*in0: Figure S1). If the block duration is high, reduce L31.
   7. Process the data with Bruker's Proc_reset AU program and Fourier transform.
      NOTE: We recommend to transform the spectrum using zero filling and a sine bell apodization (Figure S4).
3. SAPPHIRE-PSYCHE experiment
   1. Select the SAPPHIRE-PSYCHE pulse sequence and set the pulse sequence parameters. Standard parameters would be the following: 5 kHz spectral width (SW3), at least 1 or 2 seconds of relaxation delay (D1), 16 dummy scans (DS), 8 or 16 scans per increment (NS) and D2 to 14 ms (Figure S5).
      NOTE: This sequence is not in Bruker's repertoire, however, the sequence and the processing programs may be obtained from the Manchester NMR Methodology Group website, (https://www.nmr.chemistry.manchester.ac.uk/?q=node/426)²³. The D2 delay ensures that T2 relaxation remains constant with each J modulation increment. D2 needs to be greater than 1/4*SW1+p16+2*d16.²³
   2. Set the desired CHIRP pulse flip angle excitation (CNST20) and 10 kHz for the CHIRP pulse bandwidth (CNST21) (Figure S6).
      NOTE: As in the regular PSYCHE experiment, the CHIRP pulse flip angle needs to be short to avoid recoupling artifacts. CNST20 = 20° is a good compromise between sensitivity and low recoupling artifacts^21,23,25.
   3. Set the hard pulse length (P1) to the previously calibrated value and the PSYCHE shape pulse length to 30 ms (P40) (Figure S6).
      NOTE: It is important to calibrate the hard pulse value as the shape pulse powers will be automatically calculated from it.
   4. Choose the PSYCHE_Saltire_10kHz_30m shape pulse for the PSYCHE element (Figure S6).
   5. Set the strength of the pulse field gradient applied during the PSYCHE element (GPZ10). Choose RECT.1 for the gradient shape pulse (GPNAM 10) (Figure S7).
      NOTE: A weak magnetic field gradient is applied during the PSYCHE element, normally, between 1% and 4% of the maximum strength of the gradient, value that depends on the probe.
   6. Set the number of SAPPHIRE J modulation increments in F2 (TD2) (Figure S7).
      NOTE: normally 8 increments ensure an excellent suppression of sideband artifacts (Figure 2 and 3). The total number of scans of the final Pure Shift FID is NS*TD2.²³
   7. Set the F1 and F2 spectral windows (SW1 and SW2) (Figure S5).
      NOTE: SW2=SW3/(2*TD2) and SW3/SW1 = TD2*N, were TD2 and N are even integers²³. The SAPPHIRE-PSYCHE experiment is acquired as a pseudo 3D where F2 encodes the J-coupling artifact phase modulation and F1 the Pure Shift interferogram acquisition²⁰. Since SAPPHIRE-PSYCHE removes J modulation sidebands, the interferogram Pure Shift block duration could be longer than regular PSYCHE (Pure Shift block duration = 1/SW1), typically between 20 to 40 ms (Figure 2). However, longer chunk data acquisition leads to higher J-coupling evolutions, which would require more J-coupling phase modulation increments to remove the stronger sidebands attained²³.
   8. Set the number of Pure Shift blocks (TD1) (Figure S7).
      NOTE: Since SAPPHIRE-PSYCHE needs to compensate the J-coupling phase modulation of the first block, an extra block needs to be acquired. Typically, 17 (16+1) or 33 (32+1) blocks give enough digital resolution²³.
   9. Process the data executing the pm_pshift and the pm_fidadd AU programs followed by Fourier transform²³.
      NOTE: We recommend to transform the spectrum using zero filling and a sine bell apodization (Figure S4).

Representative Results

NMR spectrum analysis
PSYCHE experiments increase spectra resolution by collapsing coupled resonances into singlets²¹, which in turn reduces overlap and facilitates assignment and data analysis. Pure Shift NMR can be applied to plant extracts. Here we demonstrate its use in three different matrices: vanilla leaves, potato tubers, and Physalis peruviana fruits. The resolution enhancement achieved in the spectra of these plant extracts is clear from Figures S8-S11.

In order to assess the effect of these pulse sequences in the resolution of the NMR signal, the width of the line in frequency units at 10% of the maximum height, W₁₀, were calculated for several types of resonances, all exemplified in spectra obtained with Vanilla sp extracts - Pure Shift against classical ¹H NMR - Figure 4. Overall, the expansion of coupled resonances reached W₁₀ values from 1 to 60 Hz, whereas singlet peaks varied between 1 to 10 Hz: the methyl group of acetic acid (1.97 ppm in Figure 4) attained a W₁₀ value of 2.0 Hz in classical ¹H-NMR and 2.1 Hz in the SAPPHIRE-PSYCHE (S) spectrum. The anomeric resonance of sucrose (at 5.40 ppm), a doublet with a J-coupling of 3.9 Hz, extends over a region equivalent to a W₁₀ of 6.5 Hz, with each individual peak accounting for 2.6 Hz of W₁₀ (Figure 4). This value was higher than that of the collapsed peak obtained in the SAPPHIRE-PSYCHE spectrum, W₁₀ = 1.9 Hz (Figure 4). In the case of the malic acid Hβ´ (at 2.55 ppm), a doublet of doublets with J =7.8 Hz, 15.6 Hz, extends over 29.4 Hz (Figure 4), with calculated W₁₀ values for each individual peak between 4.7 and 4.9 Hz. This multiplet signal collapsed by Pure Shift into a single line (at 2.55 ppm), expanding over a W₁₀ value of 4.7 Hz (Figure 4).

Signal multiplicity becomes more complex in highly coupled hydrogens where the constitutive peaks of the multiplet are no longer easily distinguishable, forming an almost continuous signal. This is the case for the homocitric acid hydrogens, Hγ (at 2.24 ppm), Hδ (at 2.02 ppm), and Hδ´ (at 1.91 ppm), Figure 4. The Hγ, Hδ, and Hδ´multiplets with W₁₀=38.3 Hz, W₁₀=41.5 Hz, and W₁₀=34.8 Hz, respectively, collapsed to singlets of W₁₀=7.8 Hz (Hγ, 2.24 ppm), W₁₀=8.0 Hz (Hδ, 2.02 ppm) and W₁₀=6.2 Hz (Hδ´, 1.91 ppm), Figure 4. This enhanced resolution allowed better discrimination of other overlapped signals corresponding to malic Acid (Hβ, 2.77 ppm Hβ´, 2.55 ppm), homocitric acid (Hα, 2.80 ppm and Hα´, 2.63 ppm), and homocitric acid lactone (Hα´, 2.84 ppm and Hδ, 2.65 ppm) (Figure 4).

The improvement in resolution in the SAPPHIRE-PSYCHE homonuclear decoupled spectra that permits better discrimination among signals is further demonstrated in two highly compromised regions, 4.85 - 5.08 ppm and 7.05 - 7.31 ppm, where important metabolites involved in vanilla fragrance metabolic pathway could be identified: glucoside A (see structure in Figure 5), CH-2´,6´ (at 7.28 ppm), CH-2",6" (at 7.18 ppm), CH-3´,5´ (at 7.09 ppm), CH-3",5" (at 7.07 ppm), CH₂-7´ (at 4.97 ppm and 5.05 ppm), CH₂-7'' (at 4.89 ppm), CH-Glc (at 4.96 ppm) and CH'-Glc (at 4.93 ppm), glucoside B (see structure in Figure 4), CH-2´,6´ (at 7.29 ppm), CH-2",6" (at 7.23 ppm), CH-3´,5´ (at 7.09 ppm), CH-3",5" (at 7.07 ppm), CH₂-7´ (at 4.99 ppm), CH₂-7'' (at 4.91 ppm), hydroxybenzyl alcohol (7.23 ppm) and hydroxybenzyl alcohol glucoside (7.10 and 4.51 ppm), Figure 4.

The same results were observed with Cape gooseberry extracts and in the potato tuber analysis. In these two cases, several crowded regions on regular ¹H-NMR, PSYCHE, and SAPPHIRE-PSYCHE have been expanded for comparisons (Figures 6 and 7).

It is clear that PSYCHE and SAPPHIRE-PSYCHE experiments clearly improved signal resolution of Cape gooseberry extracts for sucrose (4.04 ppm), β-fructose (4.02 ppm and 3.99 ppm), β-glucose (3.24 ppm), proline (2.34 ppm, 2.07 ppm, and 2.00 ppm), glutamic acid (2.16 ppm), and glutamine (2.13 ppm) (Figure 6), as we showed in previous work¹¹, as well as it did in the case of the S. tuberosum extracts: GABA (1.91 ppm, 2.33 ppm, and 3.02 ppm), pyroglutamic acid (2.04 ppm, 2.41 ppm, and 2.51 ppm), proline (2.00 ppm, 2.07 ppm, 3.33 ppm, and 3.42 ppm), glutamine (2.14 ppm and 2.46 ppm), valine (2.27 ppm), citric acid (2.64 ppm and 2.76 ppm), and β-glucose (3.25 ppm and 3.47 ppm), Figure 7.

The PSYCHE sequence is relatively easy to use and it has been successfully implemented in a wide range of applications⁹. To attain the decoupled spectrum, the pulse sequence acquires small chunks of FID refocussing the J-coupling at the middle of each block. However, a small J-coupling evolution occurs during each block and generates the periodic sidebands artifacts which normally represent less than 5 % of their parent peak²¹. The presence of these J modulation artifacts, inherent to PSYCHE experiments²¹, are evident in Figure 6 expanded zones. In the analysis of pure compounds these artifacts can be neglected. In biological samples, this may not be the case, because, as shown in Figure 6 for proline (4.13 ppm, 3.41 ppm, and 3.33 ppm), asparagine (3.97 ppm), myo-inositol (3.27 ppm), GABA (3.03 ppm), and malic acid (2.67 ppm), and in Figure 6 for pyroglutamic acid (2.04 ppm, 2.41 ppm, and 2.51 ppm) and valine (2.27 ppm), metabolites at higher concentrations generate artifacts as large as some of the signals belonging to compounds present at low concentrations compromising in this way the accuracy of the metabolic profiling.

The SAPPHIRE-PSYCHE experiment is a modification of the regular PSYCHE sequence in which these periodic artifacts are removed by systematic phase modulation, achieved by shifting the J refocusing point²³. Consequently, the SAPPHIRE-PSYCHE experiment allows to ensure a much cleaner Pure Shift spectrum, as shown in Figure 6 for proline (4.13 ppm, 3.41 ppm, and 3.33 ppm), asparagine (3.97 ppm), myo-inositol (3.27 ppm), GABA (3.03 ppm), and malic acid (2.67 ppm), in the case of Cape gooseberries¹¹, and in Figure 7 for pyroglutamic acid (2.04 ppm, 2.41 ppm, and 2.51 ppm) and valine (2.27 ppm), in the case of potatoes.

Another artifact from which all Pure Shift experiments suffer are those generated by the strong coupling effect. Some important primary metabolites such as citric acid (2.64 ppm and 2.76 ppm) and glutamine (2.14 ppm and 2.46 ppm), exhibit strong coupling artifacts, as shown in potato tuber extract (Figure 7). To date, there is no pulse sequence that can conveniently eliminate this problem; however, SAPPHIRE-PSYCHE performs better than regular PSYCHE^8,9,23.

Correlation matrix analysis
One of the main advantages of NMR spectroscopy is that the relative concentrations of metabolites in a mixture are directly proportional to the intensities of their signals. Then, a metabolite pairwise correlation matrix may be obtained directly from the spectra correlation matrix³².

The ¹H-NMR correlation matrix, commonly known as STOCSY³² (Statistical Total Correlation Spectroscopy), is usually represented as a pseudo-2D spectrum, where each cross peak is a correlation coefficient between two signals (Figure 8)¹¹. STOCSY displays high correlation between signals that belong to the same molecule, but also with signals from molecules that pertain to the same metabolic pathway^11,32. Hence, correlations patterns provide information about the physiological state of the system, and can therefore be employed as a physiological stage fingerprint³³.

The main limitation of STOCSY is signal overlapping, which decreases the pairwise correlation^11,32. Moreover, J-coupling multiplicity correlations, lead to highly complex patterns, further complicating the analysis^11,32. The use of SAPPHIRE-PSYCHE STOCSY, shown in Figure 8, enhances correlation values because it reduces J-coupling multiplicity correlations into a single peak, thereby reducing the overlapping¹¹.

Several regions have been expanded in Figure 8. Malic acid shows complex J multiplet patterns associated with Hα (4.41 ppm), Hβ (2.82 ppm), and Hβ′ (2.67 ppm); in Pure Shift STOCSY those signals collapse into single correlation peaks (Figure 9A)¹¹. Same results are observed for β-glucose were intermolecular and intramolecular (α-glucose, α-fructose, and β-fructose) correlations are better depicted by SAPPHIRE-PSYCHE (Figure 9B)¹¹. Some amino acids also display strong intramolecular correlations in Cape gooseberry extracts. However, their identification in regular ¹H-NMR STOCSY is compromised by the overlap in crowded regions. With Pure Shift STOCSY, proline, alanine, glutamine, and glutamic acid intermolecular and intramolecular correlations are better depicted (Figures 9C and 9D)¹¹.

Multivariate analysis
Multivariate analysis is one of the main tools employed in addressing metabolomics data^34,35. While sample discrimination by PCA (Principal component analysis) or PLS-DA (Partial least squares-discriminant analysis) could easily be achieved through regular ¹H-NMR spectra, the interpretation of the loading is better addressed through pure-shift data^{11,24,25,27,28}.

In Figure 10, we show the PLS-DA score plot obtained using SAPPHIRE-PSYCHE (Figure 10A) and regular proton spectra (Figure 10B) in the discrimination of six ecotypes of P. peruviana extracts¹¹. Even though there are some studies which claim better discrimination performance when using Pure Shift data^24,25, our results show that the performance was barely affected by the homonuclear decoupling¹¹. In the case of the analysis of the loadings data, as shown in Figures 10C and 10D, the increased resolution attained with the SAPPHIRE-PSYCHE data simplified the analysis and allowed a better identification of the specific metabolites responsible of P. peruviana discrimination, namely, α-glucose, β-glucose, α-fructose, β-fructose, sucrose, citric acid, and alanine¹¹. This resolution gain was also critical when combining PLS coefficients analysis with STOCSY correlation, Figures 10C and 10D¹¹. The strong correlation between α- glucose (STOCSY vector at 5.23 ppm is color-coded on the first PLS component) and β- glucose, α-fructose and β- fructose - all metabolites which evolve from the same metabolic pathway - and its anti-correlation with respect to sucrose, are evident¹¹. With normal STOCSY analysis, the extensive overlap did not allow clear depiction among correlation coefficients and produce the loss of this particular metabolic pathway information (Figure 10)¹¹.

Figure 1. (A) PSYCHE spectra of Vanilla planifolia extracts using different CHIRP pulse flip angles for the PSYCHE element: left, flip angle values; right, intensity multiplication factor. (B) and (C) Graphs show the intensities and signal-to-noise of five decoupled peaks (black - 4.31 ppm, orange - 4.16 ppm, red - 2.65 ppm, yellow - 2.63 ppm, and blue 2.56 ppm) as a function of the flip angle of the PSYCHE CHIRP pulse, respectively. The signal-to-noise ratio (S/N) (more precisely, signal to noise + artifact ratio) was calculated using the signal of maximum intensity against that of the noise, value calculated over a 2 ppm range: from 7.75 ppm to 9.75 ppm (Figure S12). Please click here to view a larger version of this figure.

Figure 2. PSYCHE (A) and SAPPHIRE-PSYCHE (B) spectra of Vanilla planifolia extracts with different interferogram Pure Shift block duration. The experiments were acquired in order to maintain the same digital resolution and sensitivity. PSYCHE (A) parameter from bottom to top: 128 scans, 64 interferogram blocks, 6.4 ms block length, total acquisition time 6h07min; 128 scans, 32 interferogram blocks, 12.8 ms block length, total acquisition time 3h04min; 128 scans, 16 interferogram blocks, 25.6 ms block length, total acquisition time 1h32min; 128 scans, 8 interferogram blocks, 51.2 ms block length, total acquisition time 46min; 128 scans, 4 interferogram blocks, 102.4 ms block length, total acquisition time 24min. SAPPHIRE-PSYCHE (B) parameter from bottom to top: 16 scans, 8 J modulation increments, 65 interferogram blocks, 6.4 ms block length, total acquisition time 6h19min; 16 scans, 8 J modulation increments, 33 interferogram blocks, 12.8 ms block length, total acquisition time 3h13min; 16 scans, 8 J modulation increments, 17 interferogram blocks, 25.6 ms block length, total acquisition time 1h40min; 16 scans, 8 J modulation increments, 9 interferogram blocks, 51.2 ms block length, total acquisition time 53min; 16 scans, 8 J modulation increments, 5 interferogram blocks, 102.4 ms block length, total acquisition time 30min. Please click here to view a larger version of this figure.

Figure 3. (A) spectra of Vanilla planifolia extracts with 102.4 ms chunk length: top, PSYCHE 128 scans, 4 Pure Shift increments; middle top, 4 scans, 4 Pure Shift increments, 32 SAPPHIRE increments; middle bottom, 8 scans, 4 Pure Shift increments, 16 SAPPHIRE increments; bottom, 16 scans, 4 Pure Shift increments, 16 SAPPHIRE increments. (B) spectra of Vanilla planifolia extracts with 51.2 ms chunk length: top, PSYCHE 128 scans, 8 Pure Shift increments; middle top, 4 scans, 9 Pure Shift increments, 32 SAPPHIRE increments; middle bottom, 8 scans, 9 Pure Shift increments, 16 SAPPHIRE increments; bottom, 16 scans, 9 Pure Shift increments, 16 SAPPHIRE increments. Please click here to view a larger version of this figure.

Figure 4. Selected expansion regions of ¹H NMR (1H) and SAPPHIRE-PSYCHE (S) spectra V. planifolia (1.85 - 2.9 ppm) and V. pompona (4.85 - 7.31 ppm), showing signal assignments. Please click here to view a larger version of this figure.

Figure 5. Structures of vanilla fragrance precursors: 4-hydroxybenzyl alcohol (4-HBA), 4-HBA glucoside, glucoside A, and glucoside B. Please click here to view a larger version of this figure.

Figure 6. Selected expansion regions of ¹H NMR (1H), PSYCHE (P), and SAPPHIRE (S) spectra of an aqueous extract of Cape gooseberry (Bambamarca I Peruvian Andean region²⁸) showing signal assignments (Reprinted with permission from Lopez et al.¹¹ ). Please click here to view a larger version of this figure.

Figure 7. Selected expansion regions of ¹H NMR (1H), PSYCHE (P), and SAPPHIRE (S) spectra of an aqueous extract of potato showing signal assignment. Please click here to view a larger version of this figure.

Figure 8. Selected expanded regions (3.20 ppm-4.30 ppm) of two-dimensional STOCSY NMR spectra obtained with data from six Cape gooseberry extracts showing correlation values (r²) above 0.85: (A) regular ¹H NMR STOCSY and (B) SAPPHIRE-PSYCHE STOCSY. Reprinted with permission from Lopez et al.¹¹ Please click here to view a larger version of this figure.

Figure 9. STOCSY representations of NMR spectra of six different Cape gooseberry extracts showing correlations (r²), in the left without homodecoupling and in the right with homodecoupling for regions: (A) 4.38-4.42 ppm and 2.40-4.42 ppm with r² above 0.80 for malic acid (MA) signal (Hα-MA); (B) 3.21-3.27 ppm and 3.21-4.67 ppm with r² above 0.85 for β-glucose signal (H2-β-Gluc); (C) 2.30-2.38 ppm and 1.25-4.36 with r² above 0.93 for proline (Pro) signal (Hβ′-Pro); (D) 2.15-2.17 ppm and 1.25-4.5 ppm with r² above 0.90 for glutamic acid (Glu) signal (Hβ-Glu); α-glucose, α-fructose, β-fructose are symbolized as α-Gluc, α-Fruc, and β-Fruc, respectively. Reprinted with permission from Lopez et al.¹¹ Please click here to view a larger version of this figure.

Figure 10. PLS scores plot of Cape gooseberries extracts grown in six different Andean regions^11,28 (San Marcos: red circles, Celendin III: brown triangles, Bambamarca I: blue stars, Celendin I: yellow triangles, Bambamarca II: green squares, Celendin II: magenta diamonds) based on (A) classical ¹H NMR and (B) SAPPHIRE-PSYCHE experiments. Hotelling's T² ellipses were set to 95% confidence level. Combination of PLS1 loadings and 1D STOCSY for α-glucose correlation using the STOCSY signal at 5.23 ppm as driver peak. The coefficient of determinations (r²) have been color coded and projected on the coefficients of the first PLS component: (A) 1D STOCSY obtained with SAPPHIRE-PSYCHE data (top) and its expansion 3.15-4.17 ppm (bottom); (B) 1D STOCSY obtained with ¹H NMR data (top) and its expansion 3.15-4.17 ppm (bottom); α-glucose, β-glucose, α-fructose, β-fructose, and sucrose are symbolized as α-G, β-G, α-F, β-F, and S, respectively. PLS-DA and STOCSY analysis was perform using MATLAB Version R2018a. (Reprinted with permission from Lopez et al.¹¹) Please click here to view a larger version of this figure.

Discussion

Metabolite structural identification and quantitation are key issues in the characterization of the metabolome, data that when subjected to multivariable analyses permits to better understand the biological system under study. Sample preparation and data acquisition are critical aspects that need optimization in order to provide reliable results.

In this article, we describe and illustrate the sample preparation for NMR analysis of three different plant matrices. As with any extraction procedure, the amount of solvent per gram of material and the physical properties of the selected solvent will determine the chemical composition of the final extract and the concentration of the extracted metabolites. In the case of NMR metabolomic profiling, pH, reproducibility between independent sample extractions, and final amount of extract in the NMR tube are also aspects that need optimization. The importance of reproducibility in metabolomics is to avoid the introduction of uncorrelated variance, which could lead to unreliable results. In our experience, optimal extraction conditions were attained with the dry and grinded plant material. In the case of Cape gooseberries, the dry product was very difficult to handle (highly hygroscopic) so the fresh berries were homogenized first, prior to lyophillization.

In the case of spectra acquisition, Pure Shift experimental setup is of particular importance, as wrong parameters can lead to chunking and recoupling artifacts. The theory behind the principles of Pure Shift experiments, extensively reviewed elsewhere^8,9,10, is important to understand how to correctly configure the pulse sequence and implement it as a routine experiment.

In brief, most Pure Shift experiments are based on refocusing the J-coupling evolution during chemical shift recording. This is typically accomplished by a J-coupling refocusing element that selectively inverts "passive" spins, while the "active" spins remain unaffected. PSYCHE and SAPPHIRE-PSYCHE are based on an anti-z-COSY experiment where "passive" spins are statistically inversed.

The PSYCHE element, which consists of two swept-frequency low flip angle pulses in the presence of a weak pulsed field gradient, induces frequency spatiotemporal averaging, selecting the anti-diagonal COSY terms while suppressing zero quantum and cross correlation terms. Accordingly, to avoid the inherent recoupling artifacts, the CHIRP pulse flip angle needs to be short (Figure 1). Typically, a 20° flip angle is a good compromise between sensitivity and decoupling performance (Figure 1). Therefore, pulse calibration is critical for the quality and sensitivity of the spectrum.

Metabolomic studies are usually associated with spectra recording on a large number of samples, which implies that pulse calibration must be fast or automatic. In our experience, if the samples are prepared in exactly the same way, the hard pulse length variability among samples is less than ± 0.2 µs. We normally calibrate between 6 and 12 samples and then use the average value attained to set-up the whole group of samples. In the case that the pulse length variability from sample to sample were higher, automatic calibration of each sample should be performed using Topspin pulsecal automation program on Bruker spectrometer.

The second important parameter to consider is the length of the recorded blocks during the stepwise interferogram acquisition^21,23. Interferogram acquisition consists in recording the FID by small chunks, with the refocusing point of the J-coupling evolution always coinciding with the center of the acquired chunk. The decoupled FID is constructed by concatenating each successive chunk^8,9,10. To ensure that the block acquisition does not truncate chemical shift evolution, the start of each data recording must exactly match the end of the previous chunk.

Although this procedure allows us to obtain a homodecoupled spectrum, the small J-coupling evolution during each block generates periodic sidebands artifacts directly dependent on the chunk length. On the other hand, the spectral digital resolution depends on the spectral window and on the total duration of the decoupled FID, which, in turn, depends on the block length and number of recorded blocks. Therefore, to reduce periodic artifacts without sacrificing resolution, the block duration should be short, and the total number of recorded blocks should be high. These conditions, however, will highly increase the total experimental acquisition time without increasing sensitivity (Figure 2). Typically, a PSYCHE experiment acquired with 16 to 32 blocks of 10 to 16 ms duration gives enough digital resolution in a reasonable experimental time (30 min to 5 h) (Figure 2).

In the case of SAPPHIRE-PSYCHE, an experiment that is acquired as a pseudo 3D, one of the indirect dimensions encodes for the Pure Shift interferogram acquisition and the other for the phase modulation of the periodic artifacts through the systematic shift of the J refocusing point in each chunk.

As periodic artifacts are strongly suppressed by SAPPHIRE, the block lengths could be longer; however, very long chunks strongly affect the intensity of the signals (Figures 2 and 3). In SAPPHIRE, J modulation increments contribute to spectrum sensitivity, therefore, the resulting decoupled FID total number of scans is equal to TD2 * NS (Figures 2 and 3)²³. Generally speaking, eight J modulation increments ensure an excellent periodic artifact suppression and more than eight increments have very little effect on the quality of the spectrum, even if long chunk lengths are used (Figure 3)²³. Pure Shift increments of 33 or 17 with durations between 20 to 40 ms ensure a good spectral digital resolution.

One limitation of both of these Pure Shift pulse sequences, PSYCHE and SAPPHIRE-PSYCHE, is quantification by absolute metabolite-internal standard integration. In regular ¹H-NMR, the integrated intensity is directly proportional to the concentration of each metabolite. In PSYCHE, this is no longer the case, because a number of phenomena distort the signals and affect integration. For instance, the total integral value diminishes due to T2 relaxation during the pulse sequence spin selection. Also, the truncated J coupling evolution during the chunk acquisition which generates sidebands artifacts, disrupts the Lorentzian shape of the signal. Hence, the integral is now comprised by areas under the main peak and under all the sidebands, complicating signal integration^8,9,21,23. The frequency and magnitude of the sidebands are directly related to the chunk length but also to intrinsically molecular properties such as relaxation and the J coupling magnitude and multiplicity: higher J coupling magnitudes and higher multiplicities, lead to more distorted signals. In the case of SAPPHIRE, even though this NMR experiment efficiently removes the sideband artifacts, the signal intensities are compromised by the truncated J coupling evolution. The sum of each J modulated increment, generates an averaged decoupled FID where signal diminishment is directly related to the chunk length and the J coupling magnitude and multiplicity²³. Moreover, CHIRP pulse flip angles generate recoupling artifact which also affects each signal differently, further complicating the quantification²¹. The magnitude of the effect of these pulse sequences in quantitative analysis was assessed in our earlier Cape gooseberry study yielding errors of around 10% to 30%¹¹.

Finally, we can conclude that Pure Shift is an excellent new tool for plant metabolomics, since it drastically increases the spectrum resolution, allowing a finer correlation matrix analysis and a better interpretation of multivariate analyses^{11,24,25,27,28}.

Materials

Name	Company	Catalog Number	Comments
77500 Series Freezone 4.5 Liter benchtop	Labconco	77500
Bruker Avance III 500 MHz equiped with a 5 mm TCI Z-gradient cryogenic probe	Bruker Corporation
Centrivap Refrigerated Centrifugal Concentrators Labconco 7310000 Series	Labconco	7310000
Deuterium oxide	Sigma-Aldrich	151882
Grinder machine MKM6003	Bosch	MKM6003
Licuadora Blender 8011S model Hgb2wts3	Waring	Hgb2wts3
Methanol-d4	Sigma-Aldrich	151947