Untargeted Liquid Chromatography-Mass Spectrometry-Based Metabolomics Analysis of Wheat Grain

Summary

A method for the untargeted analysis of wheat grain metabolites and lipids is presented. The protocol includes an acetonitrile metabolite extraction method and reversed phase liquid chromatography-mass spectrometry methodology, with acquisition in positive and negative electrospray ionization modes.

Abstract

Understanding the interactions between genes, the environment and management in agricultural practice could allow more accurate prediction and management of product yield and quality. Metabolomics data provides a read-out of these interactions at a given moment in time and is informative of an organism's biochemical status. Further, individual metabolites or panels of metabolites can be used as precise biomarkers for yield and quality prediction and management. The plant metabolome is predicted to contain thousands of small molecules with varied physicochemical properties that provide an opportunity for a biochemical insight into physiological traits and biomarker discovery. To exploit this, a key aim for metabolomics researchers is to capture as much of the physicochemical diversity as possible within a single analysis. Here we present a liquid chromatography-mass spectrometry-based untargeted metabolomics method for the analysis of field-grown wheat grain. The method uses the liquid chromatograph quaternary solvent manager to introduce a third mobile phase and combines a traditional reversed-phase gradient with a lipid-amenable gradient. Grain preparation, metabolite extraction, instrumental analysis and data processing workflows are described in detail. Good mass accuracy and signal reproducibility were observed, and the method yielded approximately 500 biologically relevant features per ionization mode. Further, significantly different metabolite and lipid feature signals between wheat varieties were determined.

Introduction

Understanding the interactions between genes, environment and management practices in agriculture could allow more accurate prediction and management of product yield and quality. Plant metabolites are influenced by factors such as the genome, environment (climate, rainfall etc.), and in an agriculture setting, the way crops are managed (i.e., application of fertilizer, fungicide etc.). Unlike the genome, the metabolome is influenced by all of these factors and hence metabolomics data provides a biochemical fingerprint of these interactions at a particular time. There are usually one of two goals for a metabolomics-based study: firstly, to achieve a deeper understanding of the organism's biochemistry and help explain the mechanism of response to perturbation (abiotic or biotic stress) in relation to the physiology; and secondly, to associate biomarkers with the perturbation under study. In both cases, the outcome of having this knowledge is a more precise management strategy to achieve the goal of improved yield size and quality.

The plant metabolome is predicted to contain thousands¹ of small molecules with varied physicochemical properties. Currently, no metabolomics platforms (predominantly mass spectrometry and nuclear magnetic resonance spectroscopy) can capture the entire metabolome in a single analysis. Developing such techniques (sample preparation, metabolite extraction and analysis), which provide as great a coverage of the metabolome as possible within a single analytical run, is a key aim for metabolomics researchers. Previous untargeted metabolomics analyses of wheat grain have combined data from multiple chromatographic separations and acquisition polarities and/or instrumentation for greater metabolome coverage. However, this has required samples to be prepared and acquired separately for each modality. For example, Beleggia et al.² prepared a derivatized sample for the GC-MS analysis of polar analytes in addition to the GC-MS analysis of the nonpolar analytes. Das et al.³ used both GC- and LC-MS methods to improve coverage in their analyses; however, this approach would generally require separate sample preparations as described above as well as two independent analytical platforms. Previous analyses of wheat grain using GC-MS^2,3,4 and LC-MS^3,5 platforms have yielded 50 to 412 (55 identified) features for GC-MS, 409 for combined GC-MS and LC-MS and several thousand for an LC-MS lipidomics analysis⁵. By combining at least two modes into a single analysis, extended metabolome coverage can be maintained, increasing the richness of biological interpretation while also offering savings in both time and cost.

To permit the efficient separation of a wide range of lipid species by reversed-phase chromatography, modern lipidomics methodologies commonly use a high proportion of isopropanol in the elution solvent⁶, providing amenability to lipid classes that might otherwise be unresolved by the chromatography. For an efficient lipid separation, the starting mobile phase is also much higher in organic composition⁷ than the typical reversed phase chromatographic methods, which consider other classes of molecules. The high organic composition at the start of the gradient makes these methods less suitable to many other classes of molecules. Most notably, reversed phase liquid chromatography employs a binary solvent gradient, starting with a mostly aqueous composition and increasing in organic content as the elution strength of the chromatography is increased. To this end, we sought to combine the two approaches to achieve separation of both lipid and non-lipid classes of metabolites within a single analysis.

Here, we present a chromatographic method that uses a third mobile phase and enables a combined traditional reversed phase and lipidomics-appropriate chromatography method using a single sample preparation and one analytical column. We have adopted many of the quality control measures and data filtering steps that have previously been implemented in predominantly clinical metabolomics studies. These approaches are useful in determining robust features with high technical reproducibility and biological relevance and excludes those which do not meet these criteria. For example, we describe repeat analysis of the pooled QC sample⁸, QC correction⁹, data filtering^9,10 and imputation of missing features¹¹.

Protocol

This method is appropriate for 30 samples (approximately 150 seeds per sample). Three biological replicates of ten different field-grown wheat varieties were used here.

1. Preparation of grains

1. Retrieve samples (whole grains) from -80 °C storage.
   NOTE: Freeze-drying of seeds is recommended shortly after harvest if samples are being collected from multiple seasons. This minimizes any changes in metabolite concentration that may occur after varying periods of storage. To do this, transfer seeds to a 15 mL plastic centrifuge tube (approximately 300 seeds will fill the tube) and cover with aluminum foil. Pierce the foil two-three times using a pin and freeze dry the whole grains overnight (approximately 24 h). Samples can either be returned to the -80 °C freezer at this stage or the next step can be carried out.
2. Grind the seeds using a laboratory blender for two runs on high mode for 20 s.
   NOTE: The blender used for this protocol requires a minimum of approximately 150 seeds to fill the blender to blade height and give a relatively homogenously ground grain sample.
3. Remove the blender from the base and tap the side of the blender to bring any coarsely ground grain to the surface of the sample. Coarse material can be discarded or stored.
4. Transfer powder-like finely ground material from the blender to a 2 mL plastic microcentrifuge tube.
   NOTE: Wash the blender with deionized water and rinse with LC-MS-grade MeOH between samples. Ensure the blender is completely dry before proceeding to the next sample.
5. Return finely ground grain samples to the freezer or proceed to the next step (metabolite extraction).

2. Preparation of extraction solvent

NOTE: Prepare extraction solvent on the same day as performing the extractions.

1. Prepare at least 2 mL of 1 mg/mL of each standard. Use acetonitrile (ACN) to prepare 2-aminoanthracene, miconazole and d₆-transcinnamic acid. Use water to prepare ¹³C₆-sorbitol.
2. Take 2 mL of each 1 mg/mL standard and add to a 100 mL volumetric flask.
3. Fill the volumetric flask to the line with acetonitrile. Ensure that 100 mL of acetonitrile contains 20 µg/mL of each of the internal standards: 2-aminoanthracene, miconazole, ¹³C₆-sorbitol, d₆-transcinnamic acid.

3. Metabolite extraction

1. Weigh 200 mg of finely ground grain into a 2 mL microcentrifuge tube.
2. Add 500 µL of extraction solvent to 200 mg of finely ground grain sample.
3. Mix using a homogenizer for 2 runs of 20 s at 6,500 rpm.
4. Centrifuge at 4 °C for 5 min at 16,100 x g.
5. Transfer the supernatant to a 2 mL plastic tube.
6. Repeat this procedure from steps 3.1 to 3.5 twice more to give a total supernatant volume of approximately 1.5 mL.
7. Vortex to mix the supernatant.
8. Transfer an equal volume (55 µL) of each extract to a separate 2 mL tube to make a pooled grain extract sample.
9. Transfer a 50 µL aliquot of the extract to a glass vial.
   NOTE: Extracts can be frozen (-80 °C) at this point or proceed to the next step and follow through to the LC-MS analysis.

4. Preparation of solutions for LC-MS analysis

CAUTION: For concentrated acid, always add acid to water/solvent.

1. Prepare 50 mL of 1 M ammonium formate stock solution. Weigh 3.153 g of ammonium formate and transfer to a 50 mL volumetric flask. Fill volumetric flask to the line with LC-MS grade H₂O.
2. Prepare 1 L of the mobile phase A consisting of 10 mM ammonium formate, 0.1% formic acid. To do so, add approximately 500 mL of LC-MS grade water to a 1 L volumetric flask. Add 10 mL of 1 M ammonium formate stock and 1 mL of formic acid. Fill volumetric flask to the line with LC-MS grade H₂O. Transfer to a 1 L bottle and sonicate for 15 min to degas.
3. Prepare 1 L of mobile phase B consisting of 10 mM ammonium formate in 79:20:1 acetonitrile:isopropyl alcohol:water, 0.1% formic acid ratio. Add 200 mL of isopropanol to a 1 L volumetric flask. Add 10 mL of 1 M ammonium formate stock and 1 mL of formic acid. Fill volumetric flask to the line with acetonitrile. Transfer to a 1 L bottle and sonicate for 15 min to degas.
   NOTE: Dilute the 1 M ammonium formate into the isopropyl alcohol (IPA) before adding the ACN. Ammonium formate is insoluble in CAN.
4. Prepare 1 L of mobile phase C consisting of 10 mM ammonium formate in the ratio of 89:10:1 isopropyl alcohol:acetonitrile:water. To do so, add approximately 500 mL of isopropanol, 10 mL of 1 M ammonium formate stock and 100 mL of acetonitrile to a 1 L volumetric flask. Fill to the line with isopropanol. Transfer to a 1 L bottle and sonicate for 15 min to degas.
5. Preparation of LC-MS system wash solvents
   1. Replace the pump-head wash solution with fresh solution. Use 50% methanol, 10% isopropyl alcohol or other as recommended by the manufacturer.
   2. Prepare strong and weak needle wash solutions for washing the injection fluidics prior to and following sample injection. For the strong wash, add equal volumes of ACN and IPA. For the weak wash, prepare a solution of 10% ACN (requiring approximately 500 mL and 1 L of each of the strong and weak washes respectively for this protocol and number of samples) in separate bottles.
   3. Set the needle wash volumes to 600 µL and 1800 µL for the strong and weak washes, respectively.

5. Preparation of samples for LC-MS analysis

1. As per the manufacturers standard operating procedure for the preparation of 400 ng/µL leucine enkephalin, pipette 7.5 mL of water into the 12 mL leucine-enkephalin vial containing 3 mg of leucine-enkephalin. Freeze at -80 °C in 50 µL aliquots.
2. Prepare 100 mL of 5% ACN containing 200 ng/mL leucine-enkephalin (50 µL of 400 ng/µL leucine enkephalin). Prepare on the same day as LC-MS analysis.
3. Add 950 µL of 5% acetonitrile containing the injection standard leucine-enkephalin to the 50 µL sample aliquot prepared from step 3.
4. Vortex to mix the prepared sample.

6. LC-MS setup

NOTE: A detailed description of instrument and acquisition method setup is described in the manufacturer's user guide. A general guide and the details specific to this protocol are outlined below. The following steps can be completed at any time prior to acquiring the data.

1. Open an LC-MS hardware profile.
2. Set up the chromatographic method as outlined in Table 1. Ensure that the LC system is equipped with a quaternary solvent manager to set up this gradient.
   NOTE: IPA is a viscous solvent. It should be introduced at a low flow rate and a sufficient equilibration time should be used before increasing the composition to 98.0%. These steps will prevent the LC system from overpressuring and stopping.
3. Set up the mass spectrometer acquisition methods for each of the positive and negative ToF-MS modes over the m/z range 50-1,300.
   NOTE: The instrument used for the work presented here requires positive and negative methods to be calibrated and run individually (i.e., polarity switching within a method is not possible).
4. If the LC column is new, condition the column according to the manufacturer's recommendation.
   NOTE: The following steps should be completed directly before data acquisition.
5. In an 'MS only' hardware profile, calibrate the mass spectrometer according to the manufacturer's recommendations. Complete this step prior to each mode of acquisition, ensuring that the system has stabilized in each given modality before calibration.
6. Purge and flush the LC fluidics system using LC-MS grade solvents, including mobile phase and wash solvents.
7. Equilibrate the LC system using the LC method starting conditions, ensuring that column pressure has stabilized.
8. Inject sodium formate (0.5 mM in 90% IPA) at the beginning of the sample sequence (described below) to check the instrument calibration.
9. Set up the instrument sequence table so that solvent and preparative (extraction) blanks are analyzed first; followed by pooled QC samples (6-10) for system conditioning; then the randomized sample list with QC samples run at regular intervals (e.g., every fifth injection) as technical replicates. Run two QC samples at the end of the sequence.
   NOTE: It is helpful to include the date and injection/acquisition order in the sample filename as well as the sample ID. For example: YYYY MMDD_Injection number_Variety_Biological replicate. Before pressing start, ensure the LC column pressure is stable and that the LC is connected to the MS.

7. Data processing

NOTE: A general data processing workflow is presented in Figure 1.

1. Check the data quality (internal standard mass accuracy (calculation below) and signal reproducibility) while the sequence is running. To check signal reproducibility, visual inspection of overlaid spectra should suffice.
   NOTE: Mass error (ppm) = ((Theoretical mass – measured mass) / theoretical mass) x 10⁶
2. Generate an aligned peak intensity matrix containing samples x internal standards (aligned by retention time and m/z values).
   1. Open the data processing software (see Table of Materials). Under Home > Open, click Data. Navigate to the appropriate file location and open all data files.
   2. Under Home > Sequence > Processing Type, select Quantitation from the drop-down menu.
   3. Under Home > Method > Quan, click Calibration Components. Fill in each field using the details provided in Table 2. Click OK.
   4. On the left panel in the columns next to the data files, fill in the sample type by right clicking on the cell and selecting Unknown. Fill in the level as n/a.
   5. Under Home > Processing, select Sequence. Choose a location to save the sequence and then click Process.
   6. Under Home > Results, select Quan. From the Quan viewer, select Export runs to matrix analyzer.
   7. From matrix analyzer results viewer, click Export to csv. Save the file as a spreadsheet file.
   8. In spreadsheet software, calculate the average, standard deviation and relative standard deviation of the intensity (peak area) of each internal standard.
3. Generate an aligned peak intensity matrix containing samples x untargeted features (aligned by retention time and m/z values).
   1. Open the small molecule discovery analysis software (see Table of Materials) and select File > New to create a new experiment. Name the experiment and choose the location to save and store experiment files. Click Next.
   2. Select the type of instrument used (high resolution mass spectrometer), data format (profile), and the polarity (positive or negative). Click Next. Select all adducts available in the library and edit adduct library as required. Click Create Experiment. A new page will load where the rest of the data processing will continue/occur.
   3. Import data files. Select the file format and then select import. Browse to data location and select the data files to be imported. The progress will be shown for each file on the left panel of the page.
   4. Once data files are imported, select Start Automatic Processing. Choose a method for selecting an alignment reference. Either let the software assess all runs for suitability, give a list of suitable samples (i.e., QC samples) or choose the reference most suitable i.e. a mid-sequence QC sample. Select Yes, automatically align my runs > Next.
   5. Select Next on the experiment design page (this can be set up later).
   6. Select Perform Peak Picking and then Set Parameters. Under the Peak Picking Limits tab, select Absolute Ion Intensity and enter 100. Select apply a minimum peak width and enter 0.01 min. Select OK > Finish. When processing is complete, select Close.
      NOTE: Peak picking limits can be optimized for other data file types as necessary.
   7. On the bottom right of the screen, select Section Complete. Review aligned runs and make sure each sample is aligned to the reference. The alignment scores were >90% for the data presented here. Select Section Complete.
   8. On the next page, select between subject design. Name the design. Select Group the runs manually and Create design. Add condition, click on Condition 1 and name the group appropriately. Click Section Complete.
      NOTE: Continue to add conditions as appropriate to use statistics within the software. Since we only used the software to generate an untargeted matrix, we used a single condition labelled 'all'.
   9. On the next page, select Section Complete. Do not re-do peak picking.
   10. Review the deconvolution and then click Section Complete.
   11. Go to File > Export Compound Measurements. Deselect any properties not wanted in the output. Click OK. Choose a location to save the .csv file. Click Save > Open File > Open Folder or Close.
4. Filter the data using the extraction blanks to remove artefacts (spreadsheet software).
   1. For each RT x m/z feature, in a new column, calculate the average response in extraction blanks.
   2. For each RT x m/z feature, calculate the average response in all other samples (including QC samples).
   3. Calculate the % peak intensity of blanks in samples (average response in blank/average response in samples x 100).
   4. Sort the percent contribution column from lowest to highest values.
   5. Remove features which have >5% intensity contribution from blanks.
5. Filter missing values and correct the feature signals to signals in pooled QC samples.
   1. Open the data processing software (Table of Materials). Click the View MatrixAnalyzer button. Click the Open Data File button.
   2. Navigate to the .csv file location containing the peak intensity of RT x m/z features for each sample (untargeted matrix). Select Open.
   3. Under the QC samples tab, fill out each parameter as required. For the data set described here, use QC category=QC, ignore categories=empty, impute type=none, coverage threshold=80, scale using=no scaling, uncheck log transform, correct using=smoothing spline, smoothing=0.25.
   4. On the top right of the matrix panel, click the play button QC correction.
   5. After the correction has been performed on the top right of the matrix panel, click Save Results. Navigate to an appropriate location and save the .csv results file.
6. Filter the data to remove features which have >20% RSD in QC samples.
   1. Arrange the data so that samples are in rows and features are in columns.
   2. In a new column, calculate the average peak intensity for QC samples.
   3. In a new column, calculate the standard deviation of the peak intensity for QC samples.
   4. Calculate the relative standard deviation of the peak intensity for QC samples: (QC standard deviation/QC average) x 100.
   5. Sort the features from lowest to highest %RSD and remove features which have a QC RSD >20%.
7. Filter the data to remove features with low RSD_sample/RSD_QC ratios (e.g., <1).
   1. In a new column, calculate the average peak intensity for samples.
   2. In a new column, calculate the standard deviation of the peak intensity for samples.
   3. Calculate the relative standard deviation of the peak intensity of the samples: (sample standard deviation/sample average) x 100.
   4. In a new column, divide the samples RSD by the QC RSD.
   5. Sort the values (RSD_sample/RSD_QC ratios) from highest to lowest and remove features with a ratio <1.
8. Impute missing values (several methods available online).
   1. Format the spreadsheet so that samples are in rows and features are in columns. The first column should be the samples filename. Create an additional column next to the filenames and enter the sample groupings. In this case, the samples were grouped by variety.
   2. Save the spreadsheet as a .csv file.
   3. Go to the homepage for the web-based analytical pipeline for high-throughput metabolomics (see Table of Materials) and click Click Here to Start.
   4. Click Statistical Analysis. Under Upload Your Data, select Data Type: peak intensity table, Format: Samples in rows (unpaired) and then Choose File.
   5. Navigate to .csv file, select Open and then select Submit.
   6. On the next page, select Missing Value Estimation. Uncheck step 1 (this was performed in AnalyzerPro XD). In step 2, choose a method to estimate missing values.
      NOTE: For the data presented here - 'estimate missing values' with 'KNN' was selected.
   7. At this stage, the data matrix can be downloaded (select Download from the left panel of the web page) or proceed to perform further statistical analyses.

Results

The plant metabolome is influenced by a combination of its genome and environment, and additionally in an agricultural setting, the crop management regime. We demonstrate that genetic differences between wheat varieties can be observed at the metabolite level, here, with over 500 measured compounds showing significantly different concentrations between varieties in the grain alone. Good mass accuracy (<10 ppm error) and signal reproducibility (<20% RSD) of internal standards (

Discussion

Here, we present an LC-MS-based untargeted metabolomics method for the analysis of wheat grain. The method combines four acquisition modes (reversed phase and lipid-amenable reversed phase with positive and negative ionization) into two modes by introducing a third mobile phase into the reversed phase gradient. The combined approach yielded approximately 500 biologically relevant features per ion polarity with roughly half of these significantly different in intensity between wheat varieties. Significant changes in metab...

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
13C6-sorbitol	Merck Sigma-Aldrich	605514
2-aminoanthracene	Merck Sigma-Aldrich	A38800-1 g
Acetonitrile	ThermoFisher Scientific	FSBA955-4	Optima LC-MS grade
Ammonium formate	Merck Sigma-Aldrich	516961-100 mL	>99.995%
Analyst TF	Sciex		Version 1.7
AnalyzerPro software	SpectralWorks Ltd.		Data processing software used for step 7.2. Version 5.7
AnalyzerPro XD sortware	SpectralWorks Ltd.		Data processing software used for step 7.5. Version 1.4
Balance	Sartorius. Precision Balances Pty. Ltd.
d6-transcinnamic acid	Isotec	513962-250 mg
Formic acid	Ajax Finechem Pty. Ltd.	A2471-500 mL	99%
Freeze dryer (Freezone 2.5 Plus)	Labconco	7670031
Glass Schott bottles (100 mL, 500 mL, 1 L)
Glass vials (2 mL) and screw cap lids (pre-slit)	Velocity Scientific Solutions	VSS-913 (vials), VSS-SC91191 (lids)
Installation kit for Sciex TripleToF	Sciex	p/n 4456736
Isopropanol	ThermoFisher Scientific	FSBA464-4	Optima LC-MS grade
Laboratory blender	Waring commercial		Model HGBTWTS3
Leucine-enkephalin	Waters	p/n 700008842	Tuning solution
Metaboanalyst	https://www.metaboanalyst.ca/MetaboAnalyst/faces/home.xhtml		Web-based analytical pipeline for high-throughput metabolomics. Free, web-based tool. Version 4.0.
Methanol	ThermoFisher Scientific	FSBA456-4	Optima LC-MS grade
Miconazole	Merck Sigma-Aldrich	M3512-1 g
Microcentrifuge (Eppendorf 5415R)	Eppendorf (Distributed by Crown Scientific Pty. Ltd.)		5426 No. 0021716
Microcentrifuge tubes (2 mL)	SSIbio	1310-S0
Microsoft Office Excel	Microsoft
Peak View software	Sciex		Version 1.2 (64-bit)
Pipette tips (200 uL, 100 uL)	ThermoFisher Scientific	MBP2069-05-HR (200 uL), MBP2179-05-HR (1000 uL)
Pipettes (200 uL, 1000 uL)	ThermoFisher Scientific
Plastic centrifuge tubes (15 mL)	ThermoFisher Scientific	NUN339650
Progenesis QI	Nonlinear Dynamics		Samll molecule discovery analysis software. Version 2.3 (64-bit)
Sciex 5600 triple ToF mass spectrometer	Sciex
Screw-cap lysis tubes (2 mL) with ceramic beads	Bertin Technologies
Sodium formate	Merck Sigma-Aldrich	456020-25 g
Tissue lyser/homogeniser	Bertin Technologies		Serial 0001620
Volumetric flasks (10 mL, 50 mL, 100 mL, 200 mL, 1 L)
Vortex mixer	IKA Works Inc. (Distributed by Crown Scientific Pty. Ltd.)		001722
Water	ThermoFisher Scientific	FSBW6-4	Optima LC-MS grade
Water's Acquity LC system equipped with quaternary pumps	Waters
Water's Aquity UPLC 100mm HSST3 C18 column	Waters	p/n 186005614