Identification and Structural Elucidation of Acylsugars in Tomato Leaves Using Liquid Chromatography–Ion Mobility–Tandem Mass Spectrometry (LC-IM-MS/MS)

Abstract

Leaves of tomato plants contain various glandular trichomes that produce a wide range of metabolic products including acylsugars, which may serve as a defense mechanism against various insect pests. Acylsugars exhibit significant structural diversity, differing in their sugar cores, acylated positions, and type of acyl chains. This work demonstrated a comprehensive approach using multidimensional separation techniques, specifically liquid chromatography–ion mobility–tandem mass spectrometry (LC-IM-MS/MS), for structural characterization, and the discrimination of different tomato plants (one cultivar and five accessions) was demonstrated using tomato leaf extracts; six genotypes from five species of Solanum were represented. As a result, we identified 16 acylsugars through their molecular formulas and annotations using LC and MS analyses. The incorporation of ion mobility (IM) analysis revealed an additional 9 isomeric forms, resulting in a comprehensive total of 25 isomeric acylsugars identified. Furthermore, the experimental collision cross section (CCS_exp) values agreed reasonably well with the corresponding predicted values (CCS_pred), with an overall estimated error of less than 2%. These findings pave the way for research into how the different structural isomers of acylsugars might influence the self-defense mechanism in plants. Moreover, this work demonstrated that the investigated cultivar and accessions of tomatoes can be distinguished from each other based on their metabolite profile, e.g., acylsugars, with principal component analysis (PCA) and linear discriminant analysis (LDA) statistical models, yielding a prediction rate of 98.3%.

Graphical Abstract

INTRODUCTION

Plants in the genus Solanum, which include the commercial tomato (S. lycopersicum) and its wild relatives, potato (S. tuberosum), and eggplant (S. melongena), have different types of glandular trichomes that can produce a wide array of specialized metabolites, including acylsugars.¹ Acylsugars play significant biological roles, particularly in self-defense mechanisms against numerous insect and arachnid herbivores including whiteflies, aphids, spider mites, and leaf miners.² In the Solanum genus, acylsugars are typically characterized by glucose or sucrose cores esterified with C₂–C₁₂ aliphatic acyl chains.^1–3 These compounds exhibit chemical diversity due to variations in sugar core composition, acyl chain length, branching patterns, and the position and number of acyl groups. These variations pose significant challenges to structural elucidation.⁴ To date, approximately 400 acylsugars have been identified across different species within the Solanaceae family.^2,3,5–8 However, the potential permutations of acylsucrose structures exceed 46 billion.⁵ This suggests that even a single plant can produce a wide array of acylsugar structures. The biosynthesis of acylsugar involves a set of enzymes known as acylsugar acyltransferases (ASATs), which synthesize acylsugars from sugars and acyl-coenzyme A esters.^9,10 ASATs display diverse activities among tomato species due to ortholog divergence, gene duplication, and neofunctionalization, resulting in differences in the acceptor and donor substrate range of ASATs between tomato species.² The substrate specificity and promiscuity of these enzymes have been shown to influence both interspecific and intraspecific variations of acyl chain length and the acylation positions on the sucrose backbone.¹¹ Furthermore, the loss of an ASAT enzyme has been reported to lead to the diversification of acylsugar phenotypes within individuals of the same species.¹⁰

Several methods have been used for the identification and structural elucidation of acylsugars, including mass spectrumetry (MS) because of its high mass resolution and high mass accuracy, and the addition of tandem MS (MS/MS) provides fragmentation patterns, allowing the separation of isobaric acylsugars.⁴ However, the sole dependence on MS analysis often struggles to distinguish isomers that vary in the positions of acyl substitution or acyl chain branching.³

Gas chromatography (GC) coupled with MS is commonly used for the separation and detection of lower molecular weight and volatile molecules in complex plant extracts.^4,12,13 However, analyzing acylsugars using GC-MS often requires a derivatization step that converts the acyl groups into alkyl esters through transesterification, making the acylsugars suitable for GC-MS analysis is often required to enhance their volatility.^4,12 The disadvantage of this approach is that it does not reveal the number of acyl groups attached to each acylsugar or their specific positions of acylation.^4,12,13 In contrast, liquid chromatography (LC) is better suited for handling a broader range of compounds, including those that are less volatile or more thermally labile.¹⁴ The use of LC coupled with MS assists in separating the acylsugars from matrix components. This enhances sensitivity and precision while improving specificity by segregating interferences, including isobaric compounds, which cannot be distinguished by the MS alone.¹⁵

High-performance liquid chromatography with evaporative light scattering detection (HPLC-ELSD) has also been employed for acylsugar analysis, whereby complete separation and determination of monoester, diester, triester, and higher esters with different acyl chain lengths in each fraction was achieved in complementarity with ESI-MS by a single run.¹⁶ ELSD measures the intensity of light scattering by analyte droplets or particles formed after the rapid evaporation of the mobile phase.¹⁷ Although ELSD is a highly sensitive and universal detector, it does not provide structural information about acylsugars.⁴ To determine the structure of acylsugars, a time-consuming purification process is often required. This involves isolating milligram quantities of the analyte for analysis using nuclear magnetic resonance (NMR) spectroscopy.^3,18 NMR is a nondestructive method that offers detailed structural information, including the number of acyl chains and the specific positions of acylations. Even though NMR provides a comprehensive tool for elucidating the structure of acylsugars, it is not as sensitive as mass spectrometry.⁴

Ion mobility (IM) has been widely employed to distinguish isomers and conformers, as well as to elucidate their structures.¹⁹ IM separates ions based on their size, shape, and charge under the influence of an electric field and buffer gas.²⁰ In addition, a unique inherent physicochemical property of the molecules obtained from IM is the collision cross section (CCS), which is the rotationally averaged surface area of ions. Moreover, CCS is highly reproducible across instruments and laboratories and can be easily standardized, making it a reliable parameter to aid compound identification.^21–26 Furthermore, CCS can be obtained using experimental and theoretical approaches such as computational (IMoS, Sigma, etc.) and machine learning (AllCCS, LipidCCS, AlphaPeptDeep, etc.) approaches.²⁷ IM measurements can be performed using several different types of commercially available platforms, such as drift tube ion mobility spectrometry (DTIMS), traveling wave ion mobility spectrometry (TWIMS), trapped ion mobility spectrometry (TIMS), and field asymmetric ion mobility spectrometry (FAIMS).^27,28 Recently, we reported the successful separation of constitutional and geometrical isomers utilizing ambient ionization coupled with a commercially available DTIMS platform.²⁹ Moreover, the inclusion of IM into the typical MS-based workflows enabled the species and strain-level discrimination of closely related microorganisms in various matrices with ~100% prediction rates.^30–32 These studies among others show that multiple clinical and environmental applications are enabled by the coupling of multiple orthogonal separation methods (e.g., by using LC-IM-MS/MS workflows).^32–34 However, to the best of our knowledge, LC-IM-MS/MS has not been applied for analyzing acylsugars.

The LC-IM-MS/MS approach provides a comprehensive platform that integrates multiple dimensions of separation, significantly enhancing the resolution and identification of complex mixtures.^32–34 Herein, we systematically optimized and applied an LC-IM-MS/MS method to identify and characterize acylsugars from tomato leaf extracts from different species, such as Solanum lycopersicum, Solanum habrochaites, Solanum pennellii, Solanum pimpinellifolium, and Solanum galapagense. The identity of acylsugars was confirmed based on their retention time (RT), accurate mass, fragmentation patterns, and CCS values. For accurate identification, experimental CCS (CCS_exp) values were compared with theoretical or predicted CCS (CCS_pred) values using AllCCS.³⁵ In addition, multivariate statistical analysis was utilized to visualize the differences between various tomato leaf plants including one cultivar and five accessions from each other based on the data obtained from LC-IM-MS/MS analysis. The broader goal of this study is to establish a rapid, easy, and cost-effective LC-IM-IM/MS method that can be utilized efficiently to identify and characterize different structural isomers of plant metabolites. This advanced analytical technique not only addresses the limitations of traditional MS and GC-MS methods but also paves the way for exploring the structural complexity and biological functions of acylsugars in plant defense, ultimately contributing to advancements in agricultural and ecological research.

EXPERIMENTAL SECTION

Chemicals.

Ammonium formate (10 mmol/L) in water acidified with 0.05% formic acid and propyl-4-hydroxybenzoate (>99%) were purchased from Thermo Fisher Scientific (Carlsbad, CA). Methanol, formic acid, acetonitrile, and isopropanol (LC-MS grade) were purchased from VWR Chemicals (Radnor, PA). LC-MS grade acetonitrile (ACN) and low-concentration ESI tuning mix were purchased from Agilent Technology (Santa Clara, CA). All reagents and chemicals were used without additional purification.

Stock standard solution of propyl-4-hydroxybenzoate (1 mg mL⁻¹) was prepared in 100% methanol and stored at −18 °C. Working standard mixture solutions were prepared by appropriate dilution of the stock solution in 100% methanol.

Plant Materials.

Five wild tomato accessions were obtained for this study: Solanum habrochaites (PI209978 and PI199381) from the United States Department of Agriculture (USDA), Agricultural Research Service (ARS), Germplasm Resources Information Network (GRIN-Global); Solanum pennellii (LA0750), Solanum pimpinellifolium (LA1581), and Solanum galapagense (LA1137) from the University of California, Davis, Tomato Genetics Resource Center (TGCR). Additionally, a commercial tomato cultivar (COM), the beefsteak tomato Patsy (Bejo3353, Bejo Seeds, Oceano, CA, USA), from the Solanum lycopersicum species was acquired and used as a control.

All plants were grown in a greenhouse (Patterson Greenhouse Complex, Auburn University, Auburn, AL), under natural lighting conditions (approximately 12/12; light/dark) at a temperature of 30 °C (±5 °C) and relative humidity ranging from 50 to 80%. The seeds were planted into 200-cell trays containing soilless media. After germination, the seedlings were transferred into separate 3-L pots, in triplicate, each filled with soilless media. This resulted in three plants per sample, represented by _1, _2, and _3 in the sample name. For example, the S. habrochaites (PI209978) was named as PI209978_1, PI209978_2, and PI209978_3. The names of the tomato plants and their corresponding sample names are presented in Table S1. Throughout the growth period, fertilization and irrigation were provided as necessary. After 28 days of post-transplantation, four leaflets were collected from each plant. These leaflets were air-dried for 1 week before undergoing acylsugar extraction. The weight of the dried leaflets was then measured.

Acylsugars’ Extraction.

Surface acylsugars were extracted using the leaf dip technique according to a previously developed method, with modifications.² Briefly, four dried leaflets per tomato plant were collected into a 15 mL polypropylene conical tube, 4 mL of extraction solvent (acetonitrile: isopropanol: water (3:3:2 v/v/v) containing 0.1% formic acid and 10 μM propyl 4-hydroxybenzoate as internal standard) was added, and the samples were vortexed for 2 min. After extraction, the leaflets were removed, and the extract was centrifuged at 16 000 g for 90 s. The resulting solution was filtered using a 0.45 μm nylon syringe filter and transferred into a glass autosampler vial (Figure 1).

Figure 1.

Schematic workflow for structural characterization of tomato leaves using LC-IM-MS/MS in the negative ion mode.

LC-IM-MS/MS Analysis.

Acylsugars were analyzed by LC-IM-MS/MS using an Agilent 1290 Infinity II LC System coupled with an Agilent 6560 IM-QTOF (Agilent Technologies, Santa Clara, CA). The acylsugars were separated using a reverse-phase column, Zorbax RRHD Extend-C18 (2.1 × 50 mm, 1.8 μm particle size; Agilent Technologies, Santa Clara, CA), coupled to a guard column UHPLC Zorbax SB-C18 (2.1 × 5 mm, 1.8 μm particle size; Agilent Technologies, Santa Clara, CA). The UHPLC parameters used were according to a previous method with some modifications.² Briefly, the column temperature was maintained at 40 °C with a maximum pressure of 1200 bar. Mobile phase A consisted of a mixture of water and 10 mM ammonium formate, acidified with formic acid until the pH reached 2.8 while mobile phase B consisted of pure acetonitrile. The flow rate was 0.3 mL/min with a starting condition of 95% solvent A and 5% solvent B. The gradient for elution was as follows: 5% B at 0 min; 60% B at 1 min; 100% B at 5 min, held at 100% B until 6 min; 5% B at 6.01 min and held at 5% B until 7 min; then a postrun of 2 min, making it a total run of 9 min per sample. The injection volume was 2 μL.

The effluent from the UHPLC was directed to the electrospray ionization source (ESI). All experiments were performed in the negative ion mode using data-independentacquisition (DIA) workflow in mass range of m/z 50–1000 under the following conditions: drying gas temperature and flow rate, 325 °C and 5 L/min, respectively; capillary voltage, 3500 V; nozzle voltage, 1000 V; fragmentor, 400 V; octupole RF, 750 V; nebulizer gas pressure, 30 psi; sheath gas temperature and flow rate, 275 °C, and 12 L/min, respectively. Collision energy was alternated between low (0 V) and high (20 V) fragmentation frames during MS/MS acquisition. IM spectra were acquired using a pseudorandom 5-bit sequence in multiplexing mode, with a trap fill time set at 1800 μs and a release time of 150 μs. MassHunter Acquisition Software B09.00 (Agilent Technologies, Santa Clara, CA) was used for the data acquisition. More details of the parameters used for the IM-MS/MS are provided in Table S2. Samples were injected into LC-IM-MS/MS in 7 replicates.

Data Processing.

PNNL PreProcessor 4.1, IM-MS Browser 10.0, and Agilent High-Resolution Demultiplexing 2.0 (HRdm 2.0) tools were used to demultiplex and increase the IM resolution of the raw data prior to data analysis.^31,36 Details on the parameters chosen for the conversion of the multiplexed data are given in Table S3. The MS/MS and IM data files were exported from MassHunter qualitative analysis and MassHunter IM-MS Browser software, respectively, and reconstructed using Origin 2018b.

Acylsugar Annotation.

Acylsugars were annotated based on RT, m/z of the precursor and fragment ions, isotope distributions, and CCS values. First, a transition list with 48 acylsugars containing their name, molecular formula, and fragment ions was created based on previous reports.^1–3 This document was imported into Skyline software³⁷ and was used as a transition list. Then, the converted data files from the seven replicates of LC-IM-MS/MS measurements of each sample were imported to Skyline for data analysis. The first 3 isotopic peaks of a precursor ion were selected for spectral matching with a mass tolerance of 0.05 m/z. The transition settings are presented in Table S4.

The peak areas were integrated automatically, and manual adjustments of peak area integration were required for a small number of acylsugars. Acylsugars with an Isotope Dot Product (idotp) less than 0.5 and with a precursor mass error higher than 10 ppm were excluded to obtain an accurate annotation. Acylsugars were annotated using the nomenclature proposed by Schilmiller et al.⁷ In this nomenclature system, the acylsugar name S3:22(5,4,5) indicates a sucrose backbone sugar (S), with 3 fatty acyl chains that contain a total of 22 carbons, and 5,4, and 5 are the length of the three individual acyl chains.

Using the DT, CCS_exp was determined using the single-field (calibrant-dependent) method with a drift field of 18.5 V/cm, nitrogen (N₂) as the buffer gas, and the Agilent tunemix containing phosphazines.^27,38 Furthermore, AllCCS was utilized to obtain CCS_pred values, which were then compared with the experimental CCS values to identify the structure of the acylsugars. AllCCS is a machine learning open-source platform based on a support vector regression (SVR) algorithm, which is frequently used for metabolites, lipids, and small molecules.³⁵ For the CCS prediction of acylsugar, the Simplified Molecular-Input Line-Entry System (SMILES) was obtained from PubChem and inserted into AllCCS input, which generated CCS values for various adducts in the negative and positive ion modes.^27,35

Multivariate Statistical Analysis.

Acylsugars peak area obtained by Skyline was divided by the internal standard (IS) peak area, and the resulting values were normalized by the leaf dry weight. A heatmap was created in RStudio 4.2.2 to visually capture the inherent patterns within the data set.³⁹

The spectral data obtained from LC-IM-MS/MS were exported using IM Browser 10.0 and Mass Profiler 10.0.2.202, and then loaded into RStudio using an R script⁴⁰ to perform principal component analysis (PCA) and linear discriminant analysis (LDA) statistical analysis to visualize and quantify the differences among samples, enabling a clear understanding of the variations and relationships within data.^30,31 The code used for PCA and LDA was developed as part of our previous studies^30,31 which were performed using the packages “ggplot2”, “ggbiplot”, and “MASS”. First, the m/z and drift time values from 18 samples comprising seven replicates were selected from IM data. For each data set, the most abundant peaks (up to 100) were selected across all replicates, and the corresponding m/z and drift time values were recorded. Only those ions that were identified in at least four replicates were retained for further analysis. Next, the abundance filtering process was applied, where the drift times associated with selected m/z values and having abundances below 5% of the maximum abundance were excluded. This step ensured that each value in the reconstructed IM data matrix reflected the abundances of ions corresponding to specific drift times and m/z values. A similar approach was used for LC data, where peak identification and abundance-based filtering were performed. This resulted in a reconstructed LC data matrix capturing the abundances of ions corresponding to specific retention times and m/z values. Following the processing of both IM and LC data, the two matrices were combined into a single matrix. To ensure the comparability of features, min-max scaling was employed to normalize all the spectra prior to conducting PCA. The PCA was then performed on the resulting normalized matrix, and the eigenvectors derived from PCA were used for subsequent LDA. The first five principal components that explained the majority of variance in the data were selected. For LDA, a random test set index was generated, selecting one replicate from each of the 18 samples for testing, while the remaining samples formed the training set. Predictions were made on the testing set. The classification performance of LDA was evaluated using confusion matrices, which compared the predicted class labels (e.g., COM_1, LA0750_1, PI209978_1) with the actual class labels. Finally, the average classification accuracy and the confusion matrix across all 100 iterations were calculated to determine the model’s performance.

RESULTS AND DISCUSSION

Multidimensional Characterization of Acylsugars in Tomato Leaf Extracts.

In our previous studies, the volatile leaf extracts, such as terpenes, that also have a defense role in different tomato plants were studied using GC-MS.^41,42 In this study, LC-IM-MS/MS was utilized to further characterize and identify the nonvolatile leaf extracts, more specifically the acylsugars, from various tomato leaves, because of their self-defense role in plants and their wide structure diversity with differences in their sugar cores, acylation positions, and acyl chain type.^43,44 In addition, the acylsugar structural information could be used to differentiate various tomato leaves in the future and study how these variabilities can affect the self-defense mechanisms in these plants.

Using a 9-min LC-IM-MS/MS method, various tomato leaf plants including one cultivar and five accessions were distinguished from each other, revealing differences in their LC, MS, and IM profiles. The MS spectra presented in Figure 2 were obtained through the signal averaging of the entire ion chromatogram for each sample. These spectra reveal differences in the 400 to 1000 m/z range between tomato leaf extractions, which is typically the m/z range for acylsugars.

Figure 2.

MS spectra of tomato leaf plants (one cultivar and five accessions) obtained in the negative ion mode: commercial Solanum lycopersicum cultivar Patsy (COM), Solanum habrochaites (PI209978 and PI199381), Solanum pennellii (LA0750), Solanum pimpinellifolium (LA1581), and Solanum galapagense (LA1137).

In summary, 16 acylsugars presented in tomato leaf surface extracts were annotated (Figure 3), consisting of two acylglucoses and 14 acylsucroses, with PI209978 containing the greatest number of acylsugars. The MS and MS/MS analyses revealed significant diversity in both the number of acyl chains (2 to 5) and lengths of acyl chains (2 to 12 carbons). These results are similar to those previously reported for both commercial tomato cultivars and wild tomato accessions.^1,2,4,7 However, unlike previously published research, no acylsugar was identified in the S. pennellii (LA0750) and S. pimpinellifolium (LA1581) accessions.

Figure 3.

Integrated and normalized acylsugar peak areas in different tomato leaves. The acylsugars peak area was divided by the internal standard’s peak area, and the resulting values were normalized by the leaf dry weight. The color scale ranges from no acylsugars (white) to the accession-specific maximum peak area (dark blue). Commercial Solanum lycopersicum cultivar Patsy (COM), Solanum habrochaites (PI209978 and PI199381), Solanum pennellii (LA0750), Solanum pimpinellifolium (LA1581), and Solanum galapagense (LA1137).

Acylsugars profiles in tomato leaves can vary even within a single plant and are affected by many factors such as the dynamic acylsugar phenotype, plant developmental stage, environmental conditions, presence of glandular trichomes, sample preparation, and the specific accessions and cultivar analyzed.⁴⁵ These findings highlight the complexity and variability of acylsugar profiles in tomato leaves, emphasizing the need for comprehensive characterization across different genotypes and conditions to better understand their biological roles and applications.

The identity of each acylsugar was determined by their RT, accurate mass, fragmentation patterns using collision-induced dissociation (CID) of 20 V, and isotope distributions for the precursor (Table S5). For example, at an RT of 4.08 min, the ion with m/z 737.39 was observed, which is assigned to the precursor ion [M+HCOO]⁻(Figure 4). From the MS/MS spectra, this acylsugar was further annotated as S3:22 (5,5,12), indicating that the acylsugar is from the acylsucrose class denoted as S and that it contains three acyl chains with a total length of 22 carbons (denoted as 5,5,12) at three different acylated positions (denoted as 3:22). The fragmentations observed in Figure 4A in the MS/MS spectra extracted from the LC and IM spectra supported these findings with fragment ions at m/z 101.06, 199.17, 341.11, 425.17, 509.22, 523.27, 607.33, and 691.39. These corresponded respectively to the carboxylate ions of C5:0 ([C₅−H]⁻), and C12:0 ([C₁₂−H]⁻), the loss of the carboxylate ions from the precursor ions such as [M−C₅−C₅−C₁₂−H]⁻, [M−C₅−C₁₂−H]⁻, [M−C₁₂−H]⁻, [M−C₅−C₅−H]⁻, [M−C₅−H]⁻, and the loss of CO₂ providing [M−H]⁻. These fragmentation patterns provide critical insights into the structural arrangement and acylation pattern of the acylsugar, supporting its precise identification and characterization.

Figure 4.

Acylsugar S3:22 (5,5,12) with m/z 737.39 in different tomato leaves in the negative ion mode at RT of 4.08 min. (A) MS and MS/MS spectra, (B) IM spectra, and (C) the structural conformers or isomers of S3:22 with an RSS score of around 0.76 and an error of around 1%.

To demonstrate the impact of including IM separations into the typical LC-MS/MS workflows, the CCS values were measured experimentally and compared to the predicted values using AllCCS. Table S6 shows the PubChem ID and the canonical SMILES of the acylsugars that were utilized to predict the corresponding CCS values using AllCCS. Table S7 shows a summary of the CCS values obtained for the acylsugars found. Careful examination of the IM spectra of S3:22 (5,5,12) with m/z 737.39 provided different drift times (DT) corresponding to different tomato leaf samples (Figure 4B). The IM peak with a DT of 32.6 ms (CCS_exp of 265.3 Ų) corresponded to the LA1137 accession, while a DT of 33.3 ms (CCS_exp of 270.1 Ų) corresponded to the PI209978 accession. However, both IM peaks were observed in the PI199381 accession, indicating that PI199381 contains two conformers of S3:22 (5,5,12). Noteworthily, these CCS values are in good agreement with the CCS_pred values obtained from AllCCS with a good representative structure similarity (RSS) score of around 0.8 and an error (ΔCCS) that varied from 0.1 to 1.4% respectively, when compared to the bottom and top molecular structure given in Figure 4C. This indicates that the top structure given in Figure 4C is associated with LA1137 and PI199381, and the bottom structure is associated with PI209978 and PI199381.

Furthermore, the CCS_exp values of other acylsugars found in different tomato leaf species were also in good agreement with the CCS_pred values, which are presented in Table S7. In addition to the RSS score, the molecular structure was generated with their corresponding CCS values and adducts in the positive and negative ion modes using AllCCS.³⁵ These adducts were [M+H]⁺, [M+Na]⁺, [M+NH₄]⁺, [M−H₂O+H]⁺, [M+Na−2H]⁺, [M−H]⁻, and [M+HCOO]⁻. However, since this study depends solely on the negative ion mode data and the [M+HCOO]⁻ precursor ions were dominant ions, we used the CCS_pred values of [M+HCOO]⁻ for comparison. Of note, it was reported previously that fragmentation patterns in the negative ion mode provide valuable information about the number and length of acyl chains present on each acylsugar, while fragmentation patterns obtained in the positive ion mode result in information about the glycosidic bond cleavage, which links the six-membered pyranose and five-membered furanose rings as well as the determination of the number and length of acyl chains present on each ring.⁴⁶ Additionally, since it was found that tomato leaves contain acylsugars with either a six-membered pyranose and/or a six-membered pyranose linked to a five-membered furanose as the sugar core,⁴³ performing additional fragmentation in the positive ion mode would not be needed, and it would add extra time for analysis. Moreover, since MS is unable to give other structural information about the acyl chains, such as position and branch, the addition of IM makes it possible to distinguish those isomers from each other. The integration of IM separations into the typical LC-MS/MS workflows significantly improves the characterization of acylsugars by providing an additional separation dimension that reveals subtle structural differences not detectable by LC-MS analysis alone. This approach allows for the differentiation of isomeric species and provides accurate CCS values, which are crucial for understanding the complex structural diversity of acylsugars in various tomato leaf accessions.

Similar observations as Figure 4 were found for S4:24 at an RT of 4.55 min and a precursor ion of m/z 779.41 [M+HCOO]⁻ (Figure S1A), providing the annotation to be S4:24 (2,5,5,12) due to the fragment ions shown in Figure S1B: [C₅−H]⁻ (m/z 101.06), [C₁₂−H]⁻ (m/z 199.17), [M−C₂−C₅−C₁₂−H]⁻ (m/z 425.17), [M−C₅−C₁₂−H]⁻ (m/z 467.17), [M−C₂−C₁₂−H]⁻ (m/z 509.22), [M−C₂−C₅−H]⁻ (m/z 607.33), [M−C₅−H]⁻ (m/z 649.34), [M−C₂−H]⁻ (m/z 691.39), and [M−H]⁻ (m/z 733.40). Furthermore, S4:24 (2,5,5,12) was only found in LA1137, PI209978_2, and PI209978_3. IM was able to discriminate S4:24 (2,5,5,12) in PI209978_2 and PI209978_3 accessions from LA1137, indicating the presence of multiple conformers for PI209978 (CCS_exp of 270.9 Ų and 275.9 Ų) and one conformer for LA1137 (CCS_exp of 270.9 Ų) as shown in Figure S1A. The possible molecular structures found for S4:24 (2,5,5,12) in PI209978_2, PI209978_3, and LA1137 are shown in Figure S1C with an error of ~0.3–1.5% compared to CCS_pred (Table S7). These results reveal acylsugar diversity across the tomato leaf cultivar and accessions, which can result from differences in the concentration of acyl CoA or divergent substrate specificities of individual ASATs.⁴

In contrast, S4:20 at RT of 2.88 min with a precursor ion of m/z 723.34 [M+HCOO]⁻ (Figure S2A) provided the fragment ions m/z 101.06, 425.17, 509.22, 593.28, and 677.34 which correspond to acyl chain of C5:0, the loss of acyl chains C5:0 as [M−C₅−C₅−C₅−H]⁻, [M−C₅−C₅−H]⁻, [M−C₅−H]⁻, and the loss of CO₂ ([M-H]⁻), respectively (Figure S2B). This provided the annotation of the acylsucrose as S4:20 (5,5,5,5). Further examination of the IM spectra provided a DT around 32.4 ms with a CCS_exp of 263.0 Ų (Figure S2A) that corresponds to the structure of PubChem ID 91754201 with an error ~0.6% (Figure S2C and Table S7), indicating the presence of the structure shown in Figure S2C in PI209978 and PI199381_1. Of note, LC-IM-MS/MS could not distinguish between these two tomato leaf accessions as they had the same result. However, the structure of S4:20 (5,5,5,5) could still be identified.

Discrimination of Tomato Leaf Extracts by LC-IM-MS/MS.

The previous results showed that structural characterization of acylsugars is possible using our optimized workflow (LC-IM-MS/MS). IM was able to discriminate the various tomato leaf accessions (LA1137, PI209978, and PI199381) from each other, while the LC and MS/MS had similar results for each sample. However, in some cases, LC and MS/MS were able to provide additional information about the discrimination between various tomato leaf cultivars and accessions such as PI209978 and PI199381 accessions, and COM_3 cultivar. In this case, S3:21 was used to show the high selectivity of LC-IM-MS/MS for the discrimination of tomato leaves by investigating their metabolites, such as acylsugars. S3:21 was identified by its precursor ion of m/z 723.38 [M+HCOO]⁻ (Figure 5). In Figure 5A, it is shown that S3:21 eluted at an RT of 3.59 min for PI199381 and COM_3, while it eluted at RTs of 3.59 and 3.75 min for PI209978, showing that tomato leaves can be distinguished in the LC dimension. Furthermore, in the IM dimension clear separation of these samples were observed, with PI209978_2 and PI209978_3 showing one conformer at a DT around 33.1 ms (CCS_exp = 268.9 Ų), and PI199381, COM_3, and PI209978_1 showing two conformers at a DT around 32.6 ms (CCS_exp = 264.5 Ų) and 33.1 ms (CCS_exp = 268.9 Ų).

Figure 5.

Acylsucrose S3:21 with m/z 723.38 in different tomato leaves in the negative ion mode. (A) LC-IM spectra and (B) MS/MS spectra extracted from IM.

Additionally, MS/MS spectra also provided separation between the samples. For example, in Figure 5B, the MS/MS spectrum extracted from the first LC peak (RT of 3.59 min) in the samples PI199381, PI209978, and COM_3 provided the annotation of the acylsucrose to be S3:21 (5,5,11). This isomer was confirmed by the fragment ions [C₅−H]⁻ (m/z 101.06), [C₁₁−H]⁻ (m/z 185.15), [M−C₅−C₅−C₁₁−H]⁻ (m/z 341.11), [M−C₅−C₁₁−H]⁻ (m/z 425.17), [M−C₁₁−H]⁻ (m/z 509.23), [M−C₅−H]⁻ (m/z 593.31), and [M−H]⁻ (m/z 677.38). This acylsucrose, S3:21 (5,5,11), was also found in both IM peaks (DTs of 32.6 and 33.1 ms). By combining these results, we identified three potential structure isomers of S3:21 (5,5,11) presented in Figure 6A–C. Furthermore, another isomer was identified, which is annotated as S3:21 (4,5,12) and supported by the fragment ions [C₄−H]⁻ (m/z 87.05), [M−C₅−C₁₂−H]⁻ (m/z 411.14), [M−C₁₂−H]⁻ (m/z 495.21), and [M−C₅−H]⁻ (m/z 593.31). This acylsucrose, S3:21 (4,5,12), was only found in PI209978 accessions at RT of 3.75 min (second LC peak) and DT 32.6 and 33.1 ms. This corresponds to the potential molecular structures shown in Figure 6D and Figure 6E, respectively. As seen in Figures 4 and 6, the molecular structures with a higher CCS value (longer DT) are more elongated than the molecular structures with lower CCS values (shorter DT). This agrees with the theory behind DTIMS that structures that are more compact tend to drift through the IM cell faster.^27,28,47 Unfortunately, the currently available database containing a variety of acylsugars is very limited. Therefore, the structures obtained in Figures 6B–E were generated in AllCCS using the SMILES calculated in ChemDoodle via scribbling (Chem-Doodle Web Components | Demos > SMILES). Then the calculated SMILES were inserted as input in AllCCS. These possible structures, illustrated in Figure 6B–E, show more accurate structure identification of S3:21 (4,5,12) and S3:21 (5,5,11) with lower ΔCCS of less than 0.3%. Furthermore, since S3:21 (4,5,12) was not found in the PubChem database, drawing these possible structures in ChemDoodle provided possible structure identification for S3:21 (4,5,12).This indicates the need for theoretical and computational studies on the theoretical CCS values of all possible structures of different acylsugars, such as using molecular dynamics (MD) to create a database as a future reference to compare CCS_exp values of acylsugars. To demonstrate that the structure shown in Figure 6E is a viable orientation for S3:21 (4,5,12), we calculated the CCS predictions for an alternative structural representation of S3:21 (4,5,12) by varying the positions of the acyl groups. The SMILES representations of these potential structures were generated using ChemDoodle and inputted into the AllCCS machine learning tool for analysis. As illustrated in Figure S3, altering the positions of the acyl groups resulted in varied CCS values, which exhibited little to no correlation with the experimental CCS values obtained. Additionally, more experimental studies on the CCS of acylsugars need to be conducted to create an extensive database with CCS_exp, which can be used to train machine learning (ML) CCS prediction models for more accurate reference in the future. Moreover, the database with CCS_exp would provide more confidence in the identification of acylsugars.

Figure 6.

Structure isomers of S3:21 with m/z 723.38 in different tomato leaves. Potential structure of S3:21 (5,5,11) at (A) CCS_exp of 264.5 Ų and (B, C) CCS_exp of 269.3 Ų. Potential structure of S3:21 (4,5,12) at (D) CCS_exp of 264.5 Ų and (E) CCS_exp of 269.3 Ų.

Multivariate Analysis.

As our results show, integrating various orthogonal separation techniques (e.g., LC-IM-MS/MS) enables multidimensional complementary separation. Since the amount of extracted acylsugars in tomato leaves is known to be low, using LC as a preionization technique before MS analysis can improve ion suppression.⁴⁶ Therefore, it may help to analyze the acylsugars with low intensity in tomato plants while using IM further separates the isomers and conformers that have similar RTs. Therefore, the isomers of acylsugars in tomato leaves were separated in at least one separation dimension.

Multivariate statistical analysis was performed; it utilizes the spectral information from LC, IM, and MS/MS to discriminate tomato samples and reveal structural differences of acylsugars in these plants. The PCA score plot of the six tomato plants using LC-IM-MS/MS analysis is presented in Figure 7. As shown in Figure 7, the clusters of PI209978_1 and PI209978_2 were completely separated from each other and other tomato plants. Moreover, the clusters of COM and LA1581_1 are close to each other, indicating some similarity between these tomato species and that they share several acylsugar metabolites. Similarly, the clusters of LA1137 and LA1581 (_2 and _3) are close to each other. Of note, a strong overlap was observed between the clusters of PI199381 and PI209978_3 accessions which reflects the similarity of their composition. From the PCA plot, it was noticeable that the samples associated with PI209987_3 were positioned apart from the other samples of the same accession. This can be attributed to the inconsistent distribution of acylsugars in each plant. Therefore, by collecting leaves from three different plants of the same species, with each plant providing an individual extract there is a possibility of a dissimilar distribution of acylsugars in each sample. The PCA score plot, which represents the variance captured by two principal components, indicates that the overall variance explained by the PC1 and PC3 was 18.1%. This relatively low percentage of variance explained by PCA reflects a moderate ability to separate the tomato samples in the reduced-dimensional space. To further explore the discriminative power of the data, LDA was performed based on the first five principal components obtained from PCA. Furthermore, examining the PCA results reveals that some accessions overlap in their respective regions. For instance, the accession LA1137 overlaps with LA1581, and a plant from PI209978 (PI209978_3) overlaps with PI199381. This clustering explains why LDA did not achieve 100% prediction accuracy, yet it still achieved a discrimination accuracy of 98.3% as indicated by the cross-validation (CV) confusion matrices shown in Table S8. However, when we analyze the PCA from another perspective, it becomes evident that the replicates of each individual plant—regardless of their accession—are closely clustered together, distinctly separated from other plants. The confusion matrices reveal that two samples, PI209978_3 and PI199381_3, presented significant misclassification rates. Specifically, PI199381_3 had 19% of instances incorrectly predicted as PI199381_2, while PI209978_3 had 11% incorrectly predicted as PI199381_1. Despite this misclassification, the high overall prediction accuracy underscores the effectiveness of using LDA in conjunction with PCA to classify tomato plants, even with the relatively low percentage of variance explained by PCA alone.

Figure 7.

PCA plots of the six-tomato leaf plants (one cultivar and five accessions) in the negative ion mode of the LC-IM-MS/MS method.

The success of LDA in achieving high classification accuracy suggests that while PCA alone might not capture all the discriminative features, the combination of PCA with LDA can be utilized effectively to distinguish between the tomato plants. This highlights the power of multidimensional separation techniques combined with advanced statistical methods to effectively distinguish even closely related tomato plants.

CONCLUSION AND FUTURE OUTLOOKS

This work demonstrates the use of a highly selective method, LC-IM-MS/MS, for the structural characterization of acylsugars in tomato leaves and the differentiation of tomato leaf plants including one cultivar and five accessions. Multidimensional LC-IM-MS/MS was used to detect isomers and conformers of acylsugars in tomato leaf extracts. A total of 16 acylsugars were found through their molecular formulas and annotations using LC and MS analyses, and 9 isomeric forms were revealed by incorporating IM into the workflow, resulting in a comprehensive total of 25 isomeric acylsugars identification. In addition to performing multivariate statistics, a prediction rate of 98.3% was achieved with LC-IM-MS/MS, indicating good accuracy and selectivity of this method. This indicates that this method can be used as a proof-of-concept to characterize structural isomers and even help to discriminate acylsugars in different plants. Moreover, the accuracy of statistical analysis is expected to increase upon including more plants than the three individual plants used for each accession in the current study as this would allow for a more thorough investigation into the possibility that during sowing, one of the plants from the same accession may not have thrived or produced sufficient acyl sugars as the other plants from the same accession. In some cases, the combination of MS/MS and LC with IM proved to be more effective in separating tomato leaf samples in addition to structural characterization, while in other cases, it could only help with structural characterization. Interestingly, our CCS_exp results are in good agreement with the CCS_pred values obtained with AllCCS. This allows us to characterize the structures of the acylsugars with more confidence.

Because our results show the presence of acylsugar isomers and conformers in tomato leaf plants, future studies should focus more on investigating the differences in acylsugar structures, and how these differences affect the self-defense mechanism of the plant. In addition, the method can be improved by using a longer LC gradient,⁴⁸ trying different acylsugar extraction methods,⁴ and using higher resolution IM such as atmospheric pressure drift tube ion mobility spectrometry (AP-DTIMS, Rp ~ 250),⁴⁹ trapped ion mobility spectrometry (TIMS, Rp ~ 400),⁴⁹ cyclic traveling wave ion mobility spectrometry (cTWIMS, Rp ~ 750),⁵⁰ and structures for lossless ion manipulations (SLIM, Rp ~ 1860).^51,52 In this way, the sensitivity, accuracy, and selectivity of the method could be improved. Furthermore, additional research is needed to develop a comprehensive database for acylsugars, as there is currently limited information available regarding these compounds.

In this article, AllCCS, a machine learning tool, was used to calculate CCS values theoretically using SMILES. However, one compound can have many structures if they are oriented differently, providing hundreds of plausible SMILES. Additionally, because the database of acylsugar is very limited, these SMILES structures are not easily available. Therefore, MD would be mandatory to acquire all these possible structures. Then, the computational approach should be taken into consideration for the determination of CCS values to characterize their structures, such as IMoS, Sigma, MOBCAL, etc. Nonetheless, MD simulations and computational methods can be demanding, often requiring extensive time and resources to generate accurate CCS values.²⁷

Supplementary Material

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/jasms.4c00376.

Additional annotation of acylsugars, PubChem ID of acylsugars, LC, IM, MS, and MS/MS spectra, confusion matrices, measured and predicted CCS values, and experimental settings (PDF)

Tables S5–S6–S7 consolidated (XLSX)