Exact mass GC‐MS analysis: Protocol, database, advantages and application to plant metabolic profiling

Abstract

Plant metabolomics has been used widely in plant physiology, in particular to analyse metabolic responses to environmental parameters. Derivatization (via trimethylsilylation and methoximation) followed by GC‐MS metabolic profiling is a major technique to quantify low molecular weight, common metabolites of primary carbon, sulphur and nitrogen metabolism. There are now excellent opportunities for new generation analyses, using high resolution, exact mass GC‐MS spectrometers that are progressively becoming relatively cheap. However, exact mass GC‐MS analyses for routine metabolic profiling are not common, since there is no dedicated available database. Also, exact mass GC‐MS is usually dedicated to structural resolution of targeted secondary metabolites. Here, we present a curated database for exact mass metabolic profiling (made of 336 analytes, 1064 characteristic exact mass fragments) focused on molecules of primary metabolism. We show advantages of exact mass analyses, in particular to resolve isotopic patterns, localise S‐containing metabolites, and avoid identification errors when analytes have common nominal mass peaks in their spectrum. We provide a practical example using leaves of different Arabidopsis ecotypes and show how exact mass GC‐MS analysis can be applied to plant samples and identify metabolic profiles.

1. INTRODUCTION

Plant metabolome analysis is now of crucial importance to describe responses to environmental conditions and thus understand associated metabolic mechanisms. Standardised metabolomics protocols have been proposed to characterise crop metabolism (Zheng et al., 2021). The term ‘metabolomics’ is used to refer to techniques exploited to investigate ‘small’ biological molecules (metabolites), that is, extractible molecules with limited molecular weight, usually less than 1200 atomic mass units (a.m.u.) (Roessner & Bowne, 2009). Two mainstream mass spectrometry techniques can be used: gas chromatography coupled to mass spectrometry (GC‐MS) and liquid chromatography coupled to mass spectrometry (LC‐MS) (Allwood et al., 2011; Perez de Souza et al., 2019). The output of metabolomics is a data set of metabolic features (m/z metabolite peaks with retention time in LC‐MS; analytes resulting from metabolite derivatisation in GC‐MS) that can be used for statistics and detect metabolic changes between samples.

GC‐MS metabolomics (also sometimes referred to as metabolic profiling) has been used extensively in plants under many conditions, species, or genetic backgrounds: a simple search with a literature database with query keywords ‘plant’, ‘metabolomics’ and ‘gc‐ms’ returns 43,700 entries. When restricted to Arabidopsis, it returns 13,500 entries. It shows the massive utilisation of GC‐MS for plant physiology and molecular biology. In particular, this technique is very useful to have, in a single sample analysis, a relative quantitation of most important metabolites of plant primary metabolism, such as amino acids, small soluble sugars, polyamines or organic acids. It has thus been used to describe the response of C and N primary metabolism to major environmental cues, for example, herbivores (Jansen et al., 2009), CO₂ mole fraction (Högy et al., 2010; Misra & Chen, 2015), drought (Bowne et al., 2012; Sanchez et al., 2012), nutrient conditions (Cui, Abadie et al., 2019; Cui et al., 2021; Cui, Davanture, et al., 2019), or abiotic stress combinations (Ghatak et al., 2018; Nakabayashi & Saito, 2015; Shulaev et al., 2008).

To date, the vast majority of GC‐MS analyses for metabolic profiling utilise nominal mass acquisition (i.e. at a.m.u. resolution). Accordingly, databases associated with GC‐MS metabolomics such as the Golm Metabolomics Database provide spectral data at a.m.u. resolution, and in a recent review of metabolomics resources, only nominal mass databases are discussed for GC‐MS analyses (Vinaixa et al., 2016). In other words, to our knowledge, there is no directly accessible, high resolution (exact mass, i.e. at 0.0001 a.m.u. resolution or lower) and comprehensive GC‐MS resource for plant metabolomics. This lack of curated, accessible and available resource for GC‐MS analyses has three origins: (1) The availability of (affordable) exact mass GC‐MS instruments is relatively recent, since the implementation of the orbitrap technology took place in the 2000s (Makarov, 2000; Makarov et al., 2009; Peterson et al., 2010) and the description of standard practices for high resolution GC‐MS analyses has been proposed in 2021 only (Misra, 2021); (2) Many ordinary applications of GC‐MS metabolomics profiling do not require exact mass resolution since they are targeted on common, well‐known compounds; and (3) Whenever high resolution is required, LC‐MS can be used. The use of high resolution in LC‐MS may be important using full scan analyses, because there is a limited number of fragments (mainly parental ion and adducts) and therefore, identification essentially relies on both exact mass and isotopic pattern (De Vos et al., 2007; Kind & Fiehn, 2006). By contrast, in GC‐MS analyses, the fragmentation pattern along with the retention index are used to identify analytes, with generally good accuracy. Several tools have been recently proposed to automatically annotate ions or fragments in mass spectra, in particular from LC‐MS spectral data, for example in (Doerfler et al., 2014; Gaquerel et al., 2013; Matsuda et al., 2011; Qiu et al., 2016).

However, there are circumstances where high mass resolution may be desirable with GC‐MS, since (1) several compounds with the same retention time could generate fragments with the same nominal mass and (2) it could be useful to distinguish isotopic species (isotopologues) using their mass difference (e.g., there is a mass excess of +1.003355 Da with ¹³C while it is +1.006277 Da with ²H), and (3) one may desire to perform untargeted GC‐MS analyses with broad chemical coverage. Here, we describe an exact mass GC‐MS method for high resolution routine plant metabolic profiling and provide the associated curated database, checked with authentic standards. This allows us to address both (1) and (2). We also provide the list of current compounds having similar nominal‐mass fragments and similar retention time but can be distinguished easily with exact mass, avoiding quantification errors. We also take advantage of sulphur isotopes at natural abundance to allow the identification of S‐containing fragments in datasets. Finally, we used our protocol and the database using Arabidopsis leaves to show how it can be applied to real samples, allowing facile differentiation of genetic accessions.

2. MATERIALS AND METHODS

2.1. Chemicals

Chemicals were from Sigma Aldrich (Merck; with reference numbers between parentheses): Mass Spectrometry Metabolite Library (MSMLS, IROA Technologies), 2‐Oxoadipic acid (75447), 3‐ Iodo‐l‐Tyrosine (I8250), 5‐Aminolevulinic acid (08339), α‐Tocopherol (T3251), β‐Gentiobiose (G3000), Betain (W422312), Chlorogenic acid (C3878), Choline (C1879), Citraconic acid (C82604), D‐2‐Aminobutyric acid (116122), D‐3 phosphoglyceric acid (P8877), d‐Erythronic acid (75025), d‐Erythronic acid γ‐lactone (374385), d‐Galacturonic acid (48280), d‐Glucuronic acid (G5269), Diaminopimelate (33240), d‐Melibiose (92413), d‐Pinitol (441252), d‐Quinic acid (138622), d‐Threitol (377619), d‐Xylonic acid γ‐lactone (89339), Dulcitol (D0256), Galactinol (79544), Hydrocinnamic acid (135232), L‐α‐Glycerophosphocholine (G5291), Itaconic acid (I29204), l‐carnitine (C0283), l‐Cystathionine (C7505), Levoglucosan (06724), l‐Fucose (F2252), l‐Pyroglutamic acid (83160), Nicotinic acid (N4126), Norleucine (N6877), O‐Acetyl‐Serine (CDS020792), Phosphocholin (P0378), Stigmasterol (S2424), Sinapinic Acid (D7927), Myristic acid (M3128), Tryptamine (193747), Urea (U5378) and Xanthosine (CDS020790).

2.2. Plant material

The height Arabidopsis thaliana accessions were obtained from INRA Versailles Genomic Resource Centre VNAT collection: An‐1 (96AV), Bl‐1 (42AV), Col‐0 (186AV), Cvi‐0 (166AV), Ge‐0 (101AV), Mt‐0 (94AV), Oy‐0 (224AV) and Shahdara. Plants were grown in a growth chamber under short photoperiod (8 h light/16 h dark), PPFD = 150 µmol m⁻² s⁻¹, temperature 22°C/18°C (day/night) and relative humidity 60%. Plants were watered every 3 days and once a week with Plant‐Prod (15−10−30, N−P−K, Fertil). Fifty‐five days after sowing, two fully developed leaves were sampled on each rosette and instantly quenched in liquid nitrogen. Frozen samples were then frozen‐dried, dry material was grinded and 5 mg of dry powder used for metabolites extraction by adding 400 µl of methanol/water (90/10) with adonitol (55 µM) as an internal standard.

2.3. GC‐MS analyses

A detailed, step‐by‐step protocol is provided as a Supporting Information Material. This section summarises how exact mass GC‐MS analyses and data extraction with TraceFinder® were performed. GC‐MS analyses were carried out using a GC‐MS‐Orbitrap Q Exactive (Thermo Scientific). Fifty five µl of each standard solution (50 µM), or 4 µl from the plant leaf extract, were poured into a vial (with insert) and spin‐dried at 39°C. Samples were derivatized (automatically with a preparative robot) with 20 µl methoxyamine (20 mg ml⁻¹ in pyridine; 90 min at 37°C) and 30 µl N‐methyl‐N‐(trimethylsilyl)‐trifluoroacetamide (MSTFA) for 30 min at 37°C. Before injection, 5 µl of alkane mix (14 alkanes from C₉ to C₃₆, 3 µg µl⁻¹, Connecticut n‐Hydrocarbon Mix, Supelco) were added in each sample to compute the retention index. Analyses were performed by injecting 1 µl in splitless mode at 230°C (injector temperature) in a TG‐5 SILMS column (30 m × 0.25 mm × 0.25 µm; Thermo Scientific) set in a Trace 1300 Series GC (Thermo Scientific). Helium was used as gas carrier with a constant flow of 1 ml min⁻¹. After 1 min at initial GC oven temperature (70°C), temperature was raised to 325°C at 15°C min⁻¹ and finally kept at 325°C for 4 min. MS analyses were operated in positive polarity in full MS scan mode with the following source settings: mass scan range 50−750 m/z, resolution 60 000, AGC target 1E6, MS transfer line 300°C and filament delay 4.12 min. Ionisation by electron impact (70 eV) was performed at 250°C ion source temperature. Analytes were identified automatically using TraceFinder® (Thermo Scientific) using retention time, major characteristic fragment (m/z ion) and a confirmation fragment, with a maximum tolerance of 0.00007 Da via targeted screening.

3. RESULTS

3.1. Overview of GC‐MS analysis

As an example, the chromatogram associated with an Arabidopsis leaf sample derivatised with methoxamine and N‐methyl‐N‐(trimethylsilyl)‐trifluoroacetamide (MSTFA) in pyridine is shown in Figure 1, where the signal shown in panel (a) is total ion current (TIC). Abundant compounds form peaks that can be easily seen on the chromatogram, including some amino acids (e.g., proline), organic acids (fumarate) and sugars (e.g., sucrose), as trimethylsilylated (–Si(CH₃)₃, abbreviated TMS) derivatives. The TIC signal does not directly reflect the quantity (in moles) in the sample since analytes do not have all the same response coefficient in the mass spectrometer. The internal standard used for semiquantitation (i.e., relative quantification) is adonitol (ribitol), the derivative of which (ribitol 5TMS) elutes at 10.68 min. As expected, there is some overlapping between analytes (i.e., several analytes have very close retention time and coelute). This is visible, for example, at 15.05 (panel (b) of Figure 1) where many m/z features related to sugars are visible together with 6‐phosphogluconate specific m/z features (e.g., at 387.14254 Da). Interestingly, a relatively close mass (387.32912 Da) can be found at a similar retention time (15.25 min) from arachidic acid 1TMS, but such a mass difference is in practice too large (nearly 0.18 Da) to allow confusion with exact mass analysis.

Figure 1.

Overview of a GC‐MS chromatogram from an Arabidopsis leaf extract (here, ecotype Cvi‐0). (a) total ion current chromatogram, showing annotation of typical peaks (in red). The observed mass spectrum at 15.05 min is shown in (b), with characteristic ions of 6‐phosphogluconate 3TMS annotated in red. Note the contribution of ion m/z = 387.14254 Da (fragment C₁₃H₃₂O₇PSi₂ ⁺). Note that a close, but easily distinguishable mass (387.32912 Da, fragment C₂₂H₄₇O₃Si⁺) can be observed at 15.25 min, from arachidic acid 1TMS; nominal mass analysis would have generated the same apparent mass at 387 (or 388) Da. [Color figure can be viewed at wileyonlinelibrary.com]

3.2. Mass resolution and isotopic pattern

Mass resolution across the m/z window (50−750 Da) was not found to be constant, varying from less than 0.0001 Da at low mass (<100 Da) to 0.0017 Da at higher mass (>350 Da), that is, within ≈1 and 5 ppm. In principle, this should allow to solve the natural isotopic pattern and distinguish ¹³C, ¹⁵N and ³³S isotopologues (m + 1) and ¹³C₂ and ³⁴S isotopologue (m + 2). However, this depends on whether analytes of interest are silylated or not, because Si has two isotopes (²⁹Si and ³⁰Si), with mass excess values (with respect to the monoisotopic species) very close to that of ³³S and ³⁴S (sulphur‐containing species are addressed further below). This is illustrated in Figure 2, using a non‐silylated compounds (naturally volatile and thus not requiring derivatisation), isopropyl‐4‐methylthiazole (IMT). In fact, using the example of the major fragment (molecule minus CH₃) (Figure 2a), ¹⁵N, ³³S, ²H and ¹³C isotopologues are resolved (Figure 2b,c). With methionine 2TMS (Figure 2d), the presence of Si does not allow full resolution of the isotopic pattern. This is shown using the m + 2 isotopologues (Figure 2e) where some isotopic species are found to be undistinguishable (³⁴S and ³⁰Si). Despite this limitation, the expected isotopologue abundance (predicted from isotope abundance and elemental composition) is very close to observed abundance (Figure 2f). A difference between observed and predicted abundance is visible for the deuterated isotopologue of IMT (red circle), likely due to the very small natural abundance of deuterium (0.015%) and thus the larger imprecision in quantitation. Also, it should be noted that ¹³C isotopologues are well‐separated (resolved) from other isotopologues, allowing facile monitoring of ¹³C molecules during plant labelling experiments, for example.

Figure 2.

Isotopic pattern in two N‐ and S‐containing molecules, isopropyl‐4‐methylthiazole (IMT) and methionine 2TMS. (a) overall mass spectrum of IMT, showing the prevalence of the demethylated fragment at 126.03770 Da. (b) magnification of the mass spectrum showing the isotopic pattern of the major fragment of IMT and (c) detailed view of m + 1 isotopologues showing the resolution of ¹⁵N, ³³S, ¹³C and ²H isotopologues. (d) overall spectrum of methionine 2TMS with two major fragments (128.08955 Da and 176.09290 Da) and (e) magnification showing m + 2 isotopologues. (f) relationship between expected and observed relative abundance of isotopologues in IMT using two fragments (blue and green) and methionine 2TMS (pink), plotted using a log scale. Note the underestimation of deuterium (²H) content (red circle). The black line stands for the 1:1 line. In (a) and (d), fragmentations are shown with chemical formulas to illustrate how the fragments of interest are formed. Note IMT does not require derivatisation while methionine is silylated (two TMS groups). In (e), note that there is a common peak for ³⁴S (+1.995796 Da) and ³⁰Si (+1.996844 Da) and also ³³S (+0.999387 Da) and ²⁹Si (+0.999568 Da). [Color figure can be viewed at wileyonlinelibrary.com]

3.3. Differentiation of metabolites with exact mass

Exact mass allows one to distinguish m/z features with the same nominal mass but with very slight differences (well below 0.1 Da), reflecting differences in elemental composition. This is particularly useful when retention time of analytes at the origin of similar m/z features are very close. We took advantage of GC‐MS exact mass analyses of authentic standards used to construct the database (available as a Supporting Information Files), to tackle this problem. Amongst most common plant metabolites, many analytes were found to have both similar retention time and identical nominal mass m/z signals, and they are listed in Table 1. This problem can potentially lead to identification mistakes with nominal mass analyses especially when mass spectra are rather similar. Two examples are further illustrated in Figure 3. Ferulic acid 2TMS and tyramine 2TMS have the same retention time (12.09 min) and share the same nominal mass ion 338 a.m.u. However, there is a clear difference in exact mass (Figure 3a,b). Also, their spectra share little similarity (Figure 3a). Malic acid 3TMS and threitol 4TMS have a retention time difference of less than 4 s and share a feature at a nominal mass of 189. In addition, their mass spectra are partly similar, in particular for features <200 a.m.u (Figure 3c). However, the exact mass of features at 189 a.m.u. is readily distinguishable (189.11309 and 189.076741 Da) (Figure 3d).

Table 1.

Common analytes that share both similar nominal mass signals and retention time but can be distinguished using exact mass

RT window (min)	Nominal mass (a.m.u.)	Analyte	Fragment and exact mass (Da)	Analyte	Fragment and exact mass (Da)
4.40−4.42	160	Pentanoic acid 1TMS	13C‐C8H19OSi	Glyoxylic acid 1TMS	C5H10NO3Si
			160.12380		160.04299
8.00−8.32	174	O‐acetylserine 2TMS	C7H16NO2Si	β‐alanine 3TMS	C7H20NSi2
			174.09503		174.11343
8.80−8.86	189	Malic acid 3TMS	C8H21OSi2	Threitol 4TMS	C7H17O2Si2
			189.11309		189.07671
5.35−5.55	190	2‐hydroxybutyric acid 2TMS	C7H18O2Si2	Alanine 2TMS	C7H20NOSi2
			190.08453		190.10834
13.17−13.20	191	Caffeic acid 2TMS	C10H11O2Si	Myoinositol 6TMS	C7H19O2Si2
			191.05283		191.09236
9.85−10.0	192	Anthranilic acid 2TMS	C10H14NOSi	Phenylalanine 2TMS	C11H18Nsi
			192.08446		192.12085
12.21−12.24	203	Glucosamine 6TMS	C8H21NOSi2	Tryptophan 4TMS	13C‐C12H16NSi
			203.11617		203.10855
7.70−7.92	204	Serine 3TMS	C8H22NOSi2	N‐formylglycine 2TMS	C7H18NO2Si2
			204.12399		204.08760
9.4−9.64	205	Cinnamic acid 1TMS	C11H13O2Si	Glutathione 4TMS	C10H13N2OSi
			205.06848		205.07971
6.91–7.13	218	Hydroxypyruvic acid 2TMS	C9H22O2Si2	Isoleucine 2TMS	C8H20NO2Si2
			218.11583		218.10326
7.94−8.00	218	Threonine 3TMS	C9H24NOSi2	O‐acetylserine 2TMS	C8H20NO2Si2
			218.13964		218.10326
9.37−9.39	220	Threonic acid 4TMS	C8H20O3Si2	Cysteine 3TMS	C8H22NSSi2
			220.09510		220.10115
11.82−11.92	226	Methionine sulfoximine 2TMS 2MeOx*	C5H18N4O2SSi	Histamine 3TMS	C10H22N2Si2
			226.09197		226.13215
9.13−9.16	230	Pyroglutamic acid 2TMS	C9H20NO2Si2	Hydroxyproline 3TMS	C10H24NOSi2
			230.10326		230.13964
8.34−8.70	247	Erythronolactone 2TMS	C9H19O4Si2	Citramalic acid 3TMS	C10H23O3Si2
			247.08219		247.11857
10.31‐10.42	261	N‐acetylmethionine 2TMS	C10H23NO3Si2	N‐methylglutamic acid 3TMS	13C‐C11H26NO2Si2
			261.12165		261.15356
11.78−12.09	264	Adenine 2TMS	C10H18N5Si2	Tyramine 3TMS	C13H22NOSi2
			264.11007		264.12399
12.66−12.80	267	Urocanic acid 2TMS	C11H19N2O2Si2	Octopamine 4TMS	C13H23O2Si2
			267.09851		267.12366
11.93‐12.22	280	Pyridoxine 3TMS	C13H22NO2Si2	Tyrosine 3TMS	C14H26NOSi2
			280.11891		280.15529
12.09−12.26	308	Ferulic acid 2TMS	C14H20O4Si2	Coumaric acid 2TMS	C15H24O3Si2
			308.09001		308.12640
11.94−12.09	308	4‐hydroxyphenyl lactic acid 3TMS	C15H24O3Si2	Ferulic acid 2TMS	C14H20O4Si2
			308.12640		308.09001
12.13−12.38	319	Mannitol 6TMS	C13H31O3Si3	3‐indoleacetic acid 2TMS	C16H25NO2Si2
			319.15810		319.14238
12.09 (same RT)	338	Ferulic acid 2TMS	C16H26O4Si2	Tyramine 3TMS	C16H32NOSi3
			338.13696		338.17917

Note: For each example, the elemental composition of the fragment and its exact mass is shown. When the observed mass of interest belongs to the isotopic pattern, it is indicated with the isotope in front (¹³C). See also Figure 4 for a detailed analysis of two examples, with fragment chemical structure. This table shows couple of analytes with a difference in retention time of less than 0.4 min. *Note that m/z ions at ≈226.092 Da can also come from a fragment of allantoin 2TMS (C₄H₁₈N₄O₃Si₂, 226.09119 Da). All analytes and fragments in this table have been checked using authentic standards.

Abbreviation: a.m.u., atomic mass units.

Figure 3.

Illustration of some common nominal mass m/z ions between analytes with the same retention time: ferulic acid 2TMS and tyramine 2TMS, both at c. 12.09 min (a, b) and malic acid 3TMS and threitol 4TMS, both at (c) 8.85min (c, d). Comparisons of overall spectra (a, c) with nearly common masses distinguishable with exact mass (insets, in red), and origin of fragments (b, d). In (c), ions ‘other’ refer to C₇H₁₈O₂Si₂ ⁺ (190.083983 Da) and its protonated (+1.007825 Da) species (191.091808 Da). Note that tyramine yields a much higher signal (order of magnitude 10⁹) then ferulic acid (10⁷) despite the use of identical concentration in samples (500 ng ml⁻¹). m + 1 and m + 2 refer to +1 a.m.u. (mostly ¹³C, and ²⁹Si) isotopologues and +2 a.m.u. (mostly ¹³C₂, ³⁰Si and ¹³C‐²⁹Si) isotopologues, respectively. a.m.u., atomic mass units. [Color figure can be viewed at wileyonlinelibrary.com]

3.4. Arginine derivatives

In terms of biochemical analysis, one of the most problematic metabolites of plant primary N metabolism is arginine, because it can be converted to several by‐products upon derivatisation. It is particularly so during silylation, since the guanidium group of arginine may be cleaved to form ornithine, which in turn can lead to a cyclic form, ornithine lactam. In addition, one N atom of arginine can be lost and form citrulline. Taken as a whole, arginine derivatisation generates ornithine lactam 2TMS, ornithine 3TMS, arginine 3TMS and citrulline 3TMS (Figure 4a). Arginine 3TMS is the minor derivative and has a retention time extremely close to citrulline 3TMS, and therefore it has often been overlooked. Here, arginine 3TMS is visible on the TIC, can also be identified with certainty with exact mass and thus cannot be confused with citrulline 3TMS. Although small, features typical of arginine 3TMS and citrulline 3TMS can be found at 187.108676 Da (guanidium group 2TMS) and 171.08278 Da (decarboxylated citrulline), respectively (Figure 4b−e).

Figure 4.

Mass spectrum of arginine derivatives. (a) chromatogram showing the four products of arginine derivatisation: ornithine lactam 2TMS, ornithine 3TMS, arginine 3TMS and citrulline 3TMS. The mass spectrum of arginine 3TMS and citrulline 3TMS are shown in panels (b) and (c). Typical fragments are illustrated in (d) and (e), respectively. Note that arginine 3TMS and citrulline 3TMS have very close retention times (11.37 and 11.41 min) and thus might appear merged in the same peak under low resolutive conditions.

3.5. Identification of s‐containing analytes

The resolution of the isotopic pattern can be exploited to gain information on uncommon (or less common) elements contained by analytes. Nitrogen ¹⁵N is not sufficiently abundant (0.36% only) and can be overlapped by ²⁹Si (4.7%) and ³³S (0.76%) isotopes. Therefore, the use of the isotopic pattern to detect specific elements in analytes is only suitable for sulphur, provided (1) the signal coming from ³⁴S is high enough (³⁴S abundance is 4.2%), and (2) it can be somehow distinguished from ³⁰Si. The mass excess of the ³⁴S isotopologue is +1.995796 while that of ³⁰Si is +1.996844, meaning a difference of 0.001048 Da (i.e., 1.048‰). If mass resolution during analysis is sufficient, mass excess at nearly exactly +1.995796 can be found. A similar analysis with high resolution LC‐MS (without the issue of Si isotopes) has been undertaken previously (Nakabayashi & Saito, 2017). In Figure 5, automatic searching of mass excess of +1.995796 is shown, with the difference (in ‰) as a function of feature m/z. Using a mixture of authentic standards that includes methionine and cysteine (Figure 5a), only 9 (out of more than 2,000) features were found to fall in the mass excess +1.995796 ± 0.00005 (red shaded area). From these, manual checking allows confirmation of 3 of them, another one (labelled as 4 in Figure 5a) being ambiguous: 1 and 2 are S‐containing fragments of cysteine 2TMS (61.011195 and 131.035071 Da at 5.61 min), while 3 is a fragment of methionine 2TMS (176.092921 at 9.12 min). Four appears at the retention time of methionine 2TMS but in the spectrum, the peak is too wide to ascertain that it is not another fragment with ³⁰Si (i.e., other fragments without S with nearly the same exact mass). Using a plant sample (Arabidopsis leaf extract, Figure 5b), about 10 features fall in the ±0.05‰ window, of which two can be confirmed further: 1 is the same as in Figure 5a (S‐containing fragment of cysteine 2TMS) and 5 is C₃H₆S (74.018421 Da) and comes from methionine 2TMS, at 9.12 min. The appearance of this feature is further illustrated in Figure 5c, where it is found to be very close to the isotopic pattern of another methionine 2TMS fragment (C₂H₆OSi). Of course, looking at the mass excess (due to ³⁴S) with a high precision in Figures 5a,b is only possible if resolution is high enough to have a better probability to have an acquired m/z datapoint exactly at, or very close to +1.995796. This is illustrated in Figure 5d−f, where the observed signal (grey) is likely to show a mass signal coming from ³⁴S at medium and high resolution. This can optimally be achieved for relatively small features (m/z < 150 a.m.u.) for which resolution is larger.

Figure 5.

Identification of S‐containing fragments using exact mass data: appearance of m + 2 isotopologues at +1.995796 with a precision of 0.05‰ in a standard mixture containing methionine and cysteine (a) and an extract from Arabidopsis leaves (b). Ions that effectively contain S are circled in red. Other ions do not contain S despite their mass difference value (w.r.t. the isotopologue at +1.995796) close to zero (see main text for calculations). In (a) and (b), red frames illustrate the chemical structure of identified S‐containing fragments. (c) detailed observed isotopic pattern (m + 2 isotopologues) of the fragment ion with a monoisotopic mass of 74.018 Da, showing the ³⁴S isotopologue (observed at +1.99549) of structure 2 in (b), as well as ³⁰Si and ²⁹S + ¹³C isotopologues of C₂H₆OSi (74.018791 Da), the monoisotopic form of which also contributes to the observed signal at ≈74.018 Da. The signal at +2.00738 likely corresponds to another fragment rather than the ¹³C₂ isotopologue considering the relatively high signal. Other signals (na) are unrelated to the ion of interest. (d−f) simulation of the peak of m + 2 ³⁴S and ³⁰Si isotopologues using the arbitrary example of a monoisotopic mass at 156.04700 Da containing one Si and one S, and three levels of mass resolution. The observed signal (sum) is in grey, while the separate contribution of ³⁴S and ³⁰Si is in blue and orange, respectively. The sampled (i.e., measured by the mass spec) mass that is closest to ³⁴S is circled in red. It shows that there is no sampled mass at exactly +1.995796 (³⁴S mass excess w.r.t. monoisotopic) except at high resolution, and that ³⁴S can be easily confused with ³⁰Si due to their very small mass difference (0.001048 Da). [Color figure can be viewed at wileyonlinelibrary.com]

3.6. Application to plant samples

The compound database used here contains 1,064 target fragments that have all been confirmed with standards, representing 336 compounds with target identification exact mass features, and confirming exact mass features (see Supporting Information: File 1). Using Arabidopsis as a plant model, we found 234 compounds in leaves (266 analytes, some of them generating several derivatives). We assessed the usefulness and repeatability of exact mass analysis by conducting GC‐MS analyses on several Arabidopsis ecotypes (BL1, AN1, Col0, Cvi0, Oy0, Ge0, Mt0 and Sha). Results are shown in Figure 6 as a heatmap of analytes that are significantly different between ecotypes. Hierarchical clustering shows easy discrimination of samples (without any clustering error between ecotypes), suggesting good repeatability. Many metabolites were found to be different between ecotypes, and they are here subdivided into seven groups (Figure 6a). For example, Cvi0 was found to be enriched in various amino acid derivatives, while Shahdara was enriched in sugars. Repeatability (both technical and biological) was assessed by calculating standard errors (SE), expressed in % of mean value, for each genotype. For the vast majority of metabolites, there was a very good reproducibility, with a median value of SE of 9.7% only across all genotypes (Figure 6b). Some metabolites were associated with a high variability, larger than 40%. They were all metabolites present in very low amount, and/or having very poor ionisation, and/or encountering degradation during derivatisation, such that their relative signal was extremely small, 10⁻⁵ or less (Figure 6c).

Figure 6.

Overview of exact GC‐MS metabolic profiling of Arabidopsis genotypes. (a) heatmap with hierarchical clustering showing groups of metabolites that are significantly different (p < 0.0002, one‐way ANOVA) between genotypes. Numbers refers to groups named on top left. (b) Standard error between 6 biological replicates, in % of average value, in each genotype. The red dotted line stands for the median value(9.7%). The dotted frame comprises metabolites that are also framed and named in panel (c). Data are shown as whisker plots with median, 25% and 75% quartiles, and interquartile range. (c) Relationship between average standard error (in % of overall mean) and relative signal (relative to sucrose set at 1) obtained on the GC‐MS instrument (quantifying target mass). Signals were normalised to the internal standard (ribitol) and dry weight. Metabolites that appear highly variable are framed (and are the same as metabolites framed in panel [b]). Note that metabolite 14 is labelled ‘3‐iodotyrosine’ because it is identified with high certainty by Tracefinder® using the database although it is not expected in plants (see main text). ANOVA, analysis of variance. [Color figure can be viewed at wileyonlinelibrary.com]

4. DISCUSSION

4.1. Exact mass GC‐MS database for routine analysis

The present curated database contains 336 compounds, 234 of them being identified and quantified in Arabidopsis leaves. This might appear relatively small compared with the total estimated number of small metabolites (several thousands) in plants. However, this compares well with most targeted routine GC‐MS analyses for metabolic profiling, which yield a list of about 80−100 metabolites in the vast majority of cases (e.g., there are 162 metabolites in (Cui, Davanture, et al., 2019), and 178 in (Cui et al., 2021) found in leaves). Of course, the ability of instruments and softwares to extract a proper data set from raw data using the database depends on the quality of analyses. In effect, despite the considerable dynamic range of modern instruments (here, six orders of magnitude in peak height), precise quantification can only be achieved when analytes are not too concentrated (inadequate peak shape does not allow peak extraction by softwares like Tracefinder®) (Kaufmann & Walker, 2017). This can be challenging when some metabolites are present in high amounts (e.g., sucrose or proline) while others are present in trace amounts or generate a weak signal (e.g., salicylamide) (Figure 6). It should be noted that data extraction from raw data can also be processed via untargeted peak searching, providing a much more powerful way to appreciate the diversity of molecules present in extracts (Perez de Souza et al., 2019). However, this has two drawbacks: (1) processing time is very long, at least 20 times slower with Tracefinder®. Note however that other publicly available softwares such as MS‐DIAL (Tsugawa et al., 2015) or MZmine (Pluskal et al., 2010) can be used for high‐resolution data. (2) many peaks would appear as unidentified, with only the m/z value and retention time (and thus post‐hoc identification is required using exact mass and potentially, co‐occurring fragments). Therefore, for routine analyses, it is probably more convenient to rely on targeted analyses with the database we propose here.

4.2. Advantages of exact mass analysis

Nominal mass GC‐MS analyses rely on two main criteria: retention index and mass spectrum (Garcia & Barbas, 2011). A well‐recognised difficulty with mass spectrum analysis and comparisons in biological samples is coelution of analytes, leading to mixed mass spectra. To address this problem, deconvolution algorithms can be used (e.g., in ChromaTOF® or the deconvolution plugin in TraceFinder®) (Du & Zeisel, 2013; H. Lu et al., 2008) or alternatively, data extraction can focus on individual target m/z, for example, in MetabolomeExpress (Carroll et al., 2010). However, there remains a risk that coelution leads to erroneous identification and quantitation when common m/z nominal mass features are present. Here, we have covered this aspect by searching systematically coeluting nominal masses m/z (Table 1, Figure 3). Rather common plant metabolites are concerned by this problem (e.g., malate, serine, etc.). Exact mass offers an efficient way to avoid it since exact mass values of targeted m/z features are clearly different. In addition, the isotopic pattern reflects elemental composition (Figure 2) and can be used as a post‐hoc verification. Erroneous identification is therefore much less probable. Here, using Arabidopsis samples, there was only one case of misidentification: iodotyrosine 3TMS has been identified by TraceFinder® at 16.12 min (Figure 6c). This molecule is very unlikely since enzymes of iodotyrosine metabolism (thyroperoxidase and iodotyrosine deiodinase) are absent in plants (Phatarphekar et al., 2014; Taurog, 1999). Our Arabidopsis analyte list available in Supporting Information Material has been curated accordingly. Coincidentally, several unusual sugar‐containing molecules with similar retention time can generate fragments (m/z features) with the same exact mass. For example, rutinose 6TMS 1Meox elutes just before sucrose 8TMS (16.17 min here) and can form a target fragment at 218.1033 like iodotyrosine (Supporting Information: Figure S1). The observed signal is very low (10⁻⁶ that of sucrose 8TMS), suggesting it is a minor compound. Further work is needed (e.g., using MS² on a purified fraction where this compound is more concentrated and thus generates well‐visible MS² fragments) to determine what compound generates a signal similar to iodotyrosine in Arabidopsis samples.

Another advantage of exact mass analyses is the resolution of isotopologues. This is of particular interest to identify ¹³C‐molecules upon double‐labelling (typically ¹³C‐¹⁵N or ¹³C‐³³S) since the signal associated with ¹³C mass excess (+1.003355 Da) is readily visible. This allows direct quantitation of ¹³C isotopologues unlike in nominal mass analyses where all m + 1 isotopologues are under the same peak. In a recent study, the systematic, nontargeted analysis of ¹³C isotopologues (+1.003355 Da, +2.006710 Da, +3.010065 Da, etc.) has been used with high resolution LC‐MS to look at variations in leaf metabolites with CO₂ and O₂ gaseous conditions (Abadie et al., 2021). For ³⁴S‐isotopologues, there is an interference with ³⁰Si (silicium being carried by trimethylsilyl groups). Despite this limitation, it is worth noting that whenever mass resolution is sufficiently high (in particular for low m/z values), it is possible to identify S‐containing fragments (Figure 5). Resolution is less of an issue upon labelling since ³⁴S would prevail over ³⁰Si and thus the peak apex would be closer to the mass excess of ³⁴S (+1.995796 Da).

4.3. Application to plant samples

Here, we took advantage of Arabidopsis samples to show how exact mass GC‐MS analysis can be performed routinely on plant samples, with a rather good outcome, that is, identification and quantification of more than 200 analytes. Another attempt to carry out high‐resolution GC‐MS for primary metabolites and pesticides analysis in Arabidopsis can be found in, for example, (Peterson et al., 2010). In terms of methodology and instrumentation, high resolution exact mass GC‐MS is similar to nominal mass GC‐MS analysis in that it requires standardised derivatisation (optimally with a robotic facility) ensuring that all samples are treated evenly (including the same time lag between the end of derivatisation and injection), and quality controls to assess reproducibility and quantitativity (for a specific discussion about automation of derivatisation for GC‐MS, see (Zarate et al., 2016)). Of course, a specific feature of exact mass analysis is that it requires checking mass accuracy (W. Lu et al., 2017), and here this was performed on a day‐to‐day basis with calibration gas FC43 containing perfluorotributylamine PFTBA (from ThermoFisher Scientific). Also, it is extremely useful to have specific samples to appreciate mass accuracy, for example with a compound that does not require derivatisation like IMT (Figure 2a−b). This allows one to check not only mass accuracy but also precision of mass excess values associated with all relevant isotopes (C, N, S isotopes; Figure 2c).

It should be noted that exact mass GC‐MS, however, does not solve the issue of multiple derivatives for some specific metabolites. In particular, amino acids can form by‐products upon silylation. So is the case of glutamic acid and glutamine, that can both yield pyroglutamate (for a recent discussion on this issue, see (Miyagawa & Bamba, 2019)). It is also the case of arginine, which is known to yield several products (Molnár‐Perl & Katona, 2000), and here it formed four products (Figure 4). High resolution analysis nevertheless provides a method to have a better picture of derivatives and here, we identified four main products of arginine derivatisation: ornithine lactam 2TMS, ornithine 3TMS, arginine 3TMS and citrulline 3TMS. Since these different compounds do not have the same response coefficient in the mass spectrometer (i.e., distinct observed signal response curve to concentration), arginine cannot be quantified very precisely by this method.

Basically, GC‐MS analysis provides semiquantitative information on plant samples. A good method to have information on effective quantity (i.e., absolute quantity) of analytes is GC‐C‐IRMS (GC coupled to combustion and isotope ratio mass spectrometry). In effect, the same sample (same derivatisation) can be injected and the GC‐C‐IRMS converts quantitatively each peak to CO₂ and N₂ and thus can provide direct information on the amount of carbon (via mass 44 monitoring) or nitrogen (via mass 28 monitoring) in each peak of interest. This method has been used recently (using alfalfa seed protein extracts) and shown to provide satisfactory estimates of absolute amino acid contents (Domergue et al., 2022). More classical method can also be used, such as external calibration curves, or deuterated standards (internal references). The use of deuterated standards has three drawbacks: Firstly, it complicates mass spectra, with some probability to coincide with m/z features of interest. An example with plant samples is cysteine 3TMS where the target peak (C₈H₂₂NSSi₂) at 220.1009 Da might coincide with a deuterated fragment ([²H]‐C₇H₂₁NOSi₃ and 220.0995 Da). Secondly, having all deuterated standards is tedious and expensive. Thirdly, deuteration can cause an isotope effect in either chromatography or ionisation efficiency so that the GC‐MS signal differs between deuterated and protiated forms (Alzweiri et al., 2015; Caban & Stepnowski, 2020; Matucha et al., 1991; Ripszam et al., 2013).

4.4. Perspectives

Our results show that high resolution exact mass GC‐MS can be used in plants, as an efficient way to carry out metabolic profiling. This technique shows several advantages over classical, nominal mass GC‐MS, such as identification less prone to errors thanks to exact mass m/z features. The database we propose here to perform targeted analyses covers more than 200 metabolites of different chemical families. This database is evolutive, in that it will be enriched over the coming years and updated versions kept accessible via the journal website. We recognise that plants contain a lot of sugars, with many isomers. Some of them have identical retention times, and thus cannot be resolved with exact mass since m/z features have the same elemental composition. To solve this problem, specific devices like ion mobility may be required (Morrison & Clowers, 2018; Mu et al., 2018; Przybylski & Bonnet, 2021). In terms of isotopic analysis via exact mass GC‐MS, Si isotopes represent a limitation to distinguish isotopologues. We showed that high mass precision allows reasonable access to ³⁴S isotopologues, but less so for ¹⁵N or ³³S that are much less abundant. When isotopic enrichments are modest, GC‐C‐IRMS analysis is the best alternative, but this technology is presently not able to perform ³⁴S analysis. Therefore, exact mass LC‐MS seems to be better suited to isotopic analysis of sulphur‐containing compounds.

CONFLICT OF INTEREST

The authors declare no conflict of interest.

Supporting information

Supplementary information.

Supplementary information.

Supplementary information.

Abadie, C. , Lalande, J. & Tcherkez, G. (2022) Exact mass GC‐MS analysis: protocol, database, advantages and application to plant metabolic profiling. Plant, Cell & Environment, 45, 3171–3183. 10.1111/pce.14407