Using Metabolomics to Identify the Exposure and Functional Biomarkers of Ginger

Abstract

Liquid chromatography–mass spectrometry (LC–MS)-based metabolomics has become an important tool to increase our understanding of how diet affects human health. However, public and commercial mass spectral libraries of dietary metabolites are limited, resulting in the greatest challenge in converting mass spectrometry data into biological insights. In this study, we constructed an LC–MS/MS ginger library as an example to demonstrate the importance of dietary libraries for discovering food biomarkers. The functional and exposure biomarkers of ginger were investigated using plasma samples from mice treated with control and ginger extract diets. Our results showed clear discrimination between the metabolome of mice on normal and ginger extract diets. Using the in-house ginger library, we identified 20 ginger metabolites that can be used as exposure biomarkers of ginger. However, without the LC–MS/MS ginger library, none of the ginger metabolites could be accurately identified based on online mass databases. In addition, ginger treatment significantly impacts the endogenous metabolome, especially the purine metabolism and phenylalanine, tyrosine, and tryptophan biosynthesis. Overall, we demonstrated that the construction of LC–MS/MS spectra dietary libraries would enhance the ability to identify potential dietary biomarkers and correlate potential health benefits associated with food consumption.

1. INTRODUCTION

Liquid chromatography–mass spectrometry (LC–MS)-based metabolomics is a powerful tool for profiling and measuring the levels of low-molecular-weight (<1500 Da) metabolites present in biological samples subjected to external stimuli.¹ Metabolomics analyses are classified into targeted and untargeted. Targeted metabolomics focuses on the quantification of defined annotated metabolites. In contrast, untargeted metabolomics focuses on analyzing all of the quantifiable metabolites in a sample, including unknown metabolites.² Because untargeted metabolomics is an exploratory technique that covers a wider range of metabolites, a large set of spectra is generated and annotated through mass spectral libraries. Although untargeted metabolomics offers the opportunity to annotate/discover metabolites using mass spectral libraries, the challenge lies in the remaining acquired unknown metabolites.³ Thus, metabolite annotation is the bottleneck in untargeted metabolomics analysis.⁴

A food biomarker is a metabolite detected in a physiological sample that is produced only after the ingestion of a specific food.⁵ Food biomarkers provide a more accurate assessment of nutritional intake than habitual self-reported methods, such as food frequency questionnaires and food diaries. The discovery of food biomarkers can also improve the ability to assess dietary intake. More specifically, food biomarkers can improve the assessment of the relationship between diet and health. Thus, the discovery of food biomarkers by metabolomics is of great interest in nutritional epidemiology, although the limitations of mass spectral libraries of dietary metabolites need to be addressed to reach its full potential as an emerging omics.

Even though the number of mass spectra libraries has been increasing and has attempted to provide a more complete, robust, and reliable collection of metabolites, the current public and commercial mass spectra libraries of dietary metabolites are limited. Therefore, there is a real need to construct mass spectral libraries of dietary metabolites to facilitate the annotation of food biomarkers in untargeted metabolomics related to food intake.⁶

Ginger, the rhizome of Zingiber officinale Roscoe, is a spice used in the food industry that possesses several pharmacological properties.^7,8 In traditional medicine, ginger has been used to treat common diseases and symptoms, such as nausea, emesis, and diarrhea.⁹ Likewise, studies describe that ginger exhibits antioxidant, anti-inflammatory, and anticarcinogenic activities.^10,11 These health benefits are mainly attributed to the phenolic compounds gingerols and shogaols.^12,13 However, the exposure and functional biomarkers of ginger are still largely unknown.

The goals of this study were to construct a mass spectral library of ginger metabolites as an example to demonstrate the importance of mass spectral dietary libraries in untargeted metabolomics for discovering food biomarkers and explore the metabolic changes associated with ginger intake regarding the potential health benefits. To the best of our knowledge, dietary metabolites of ginger intake have not yet been described in untargeted metabolomic studies. Thus, the construction of mass spectral dietary libraries and subsequent uploading of spectra into mass spectral repositories will improve metabolite annotation.

2. MATERIALS AND METHODS

2.1. Chemicals and Reagents.

LC–MS-grade water, methanol, acetonitrile, formic acid, ammonium formate, ammonium acetate, and ammonium hydroxide were obtained from Thermo Fisher Scientific (Waltham, MA). The mass spectrometry metabolite library of standards (MSMLS) kit was acquired from IROA Technologies (Boston, MA). Gingerol-enriched ginger (GEG) extract was provided as a gift by Sabinsa Corporation (Piscataway, NJ). Gingerols, shogaols, and their major metabolites were obtained from our previous studies.^14,15

2.2. Endogenous Library Construction.

The MSMLS kit, which contains 601 endogenous metabolites from biologically important chemical classes and pathways in seven 96-well plates, was used to build our in-house library of endogenous metabolites. The chemical classification of the MSMLS kit provides a broad chemical coverage that includes 7 superclasses, 23 classes, and 25 subclasses, of which the MSMLS kit mainly covers organic acids and derivatives; lipid and lipid-like molecules; organic oxygen compounds; nucleosides, nucleotides, and analogues; organoheterocyclic compounds; organic nitrogen compounds; and benzenoids. The metabolites (5 μg each) contained in the MSMLS kit were reconstituted as follows: metabolites in plates one to five, which are composed mainly of organic acids and derivatives; organic oxygen compounds; nucleosides, nucleotides, and analogues; and organoheterocyclic compounds, were dissolved in 100 μL of methanol (95%; v/v); metabolites in plate six, which are composed mainly of organic nitrogen compounds and benzenoids, were dissolved in 100 μL of methanol (40%; v/v), except for the water-soluble metabolites (sugars), which were dissolved in 100 μL of water; and metabolites in plate seven, which are composed mainly of lipid and lipid-like molecules, were dissolved in 100 μL of ethanol. Multiplexed standard mixtures were prepared from these plates by taking 10 microliters of each row and transferring to amber glass HPLC vials for injection. A volume of 3 μL of each multiplexed standard mixture was injected into all chromatographic columns described in Section 2.6. Metabolites that did not appear at the concentration of the multiplexed standard mixture (4.16 μg/mL) were injected directly at 50 μg/mL. Precursor masses (m/z), retention time (RT), and MS/MS spectra were collected to construct the mzRT (.txt) and MS/MS (.msp) libraries.

2.3. Ginger Exposome Library Construction.

Similar to the construction of the endogenous library, for the construction of the dietary library, the ginger metabolites were injected into a Q-Exactive Plus Orbitrap mass spectrometer, as described in Section 2.6. Ginger metabolites such as (5S)-[6]-gingerol (6G), (5S)-[8]-gingerol (8G), (5S)-[10]-gingerol (10G), [6]-shogaol (6S), [8]-shogaol (8S), [10]-shogaol (10S), (3R,5S)-[6]-gingerdiol, (3S,5S)-[6]-gingerdiol, and [6]-gingerdione were purified from the GEG extract.^14,15 On the other hand, 6S metabolites named 6S-M1, 6S-M2, 6S-M5, 6S-M6, 6S-M7, 6S-M9, 6S-M10, 6S-M11, 6S-M12, 6S-M20, and 6S-4′-glucuronide were synthesized from 6S.^16,17 Our previous studies have indicated that ginger glucuronate metabolites are mainly excreted in the urine, while in fecal samples in their free forms.¹⁸ Therefore, the spectra of the glucuronate ginger metabolites were obtained from a mouse urine sample and tentatively identified by analyzing their MS² and MS³ spectra.

2.4. Animal Treatments.

Five-week-old male CD-1 mice were obtained from the Charles River Laboratories (Morrisville, NC). Upon arrival, mice were kept under circadian rhythm with food and water ad libitum for acclimation and housing until 16 weeks old. Then, the mice were randomly divided into two groups and treated for 5 weeks. Mice in the control group (n = 5) were fed with the control diet (AIN-93G), whereas mice in the treatment group (n = 4) were fed with the AIN-93G diet containing 0.6% (w/w) GEG extract. The GEG extract contains 14.3% 6G, 2.2% 8G, 4.0% 10G, 2.2% 6S, 0.4% 8S, and 0.8% 10S. Body weight (BW) and food and fluid intake were recorded weekly. The average initial body weights of the control and GEG groups were 44.38 ± 1.32 and 43.85 ± 2.66 g, respectively. After 5 weeks of feeding, the average body weight of the GEG-treated mice was significantly lower than that of the control mice (43.55 ± 1.76 vs 45.98 ± 0.71 g; p < 0.01). The average daily food intake of GEG-treated mice was also lower than that of the control mice (3.82 vs 4.23 g/mouse/ day). However, the ginger-treated mice drank more water than the control mice (6.29 vs 4.10 mL/mouse/day). For the food and water intake, we were unable to analyze the statistical differences because only one cage per group was used.

Mice were euthanized after 5 weeks of treatment following the experimental protocol approved by the Institutional Animal Care and Use Committee of the North Carolina Research Campus (Protocol number 19-015). Blood was collected using the syringes moister with heparin. Plasma was obtained by centrifuging at 2000g for 20 min at 4 °C and then stored in a −80 °C freezer until further analysis.

2.5. Preparation of Metabolomic Samples.

One hundred microliters of plasma were vortex-mixed with 450 μL of ice-cold methanol containing cholesterol-d6, tridecanoic acid, dl-4-chlorophenylalanine, and dl-2-(4-chlorophenyl)glycine all at a concentration of 50 nM to assess recovery.¹⁹ After vortexing the mixture for a minute, the sample was centrifuged at 15,000g for 15 min at 4 °C. Supernatant was collected and dried using a SpeedVac. Finally, the precipitate was reconstituted in 100 μL of 50% methanol, containing 21 isotopically labeled internal standards, all at a concentration of 500 nM (Table S1). These isotopically labeled internal standards were used to assess instrument performance and execute retention time corrections for chromatographic alignment. A method blank was prepared using the same solvents, consumables, and standard operating procedures as the test samples, but without having the test sample. The quality control (QC) sample was prepared by taking 20 μL of each sample and pooling them. QC samples were analyzed every four samples to monitor drift, separate low- and high-quality data, and equilibrate the analytical platform.

2.6. Metabolomics Data Acquisition.

Analysis was carried out using a Vanquish UPLC system coupled to a Q-Exactive Plus Orbitrap mass spectrometer equipped with a HESI-II ion source (Thermo Fisher, Waltham, MA). To increase the number of features detected and prevent ion suppression, a C₁₈ column [Acquity UPLC Oligonucleotide BEH C₁₈ (100 mm × 2.1 mm; 1.7 μm)] in both positive and negative ionization modes and a HILIC column [Acquity UPLC BEH Amide (150 mm × 2.1 mm; 1.7 μm)] in negative ionization mode were used. A volume of 3 μL was injected into all chromatographic columns. The oven temperature was maintained at 40 °C, and the flow rate was set at 350 μL/min. The mobile phases for the C₁₈ column in positive ionization mode consisted of water with 0.1% v/v formic acid as solvent A and acetonitrile with 0.1% v/v formic acid as solvent B. The mobile phases for the C₁₈ column in negative ionization mode consisted of water with 10 mM ammonium bicarbonate as solvent A and methanol/water (95/5, v/v) with 10 mM ammonium bicarbonate as solvent B. The mobile phases for the HILIC column in negative ionization mode consisted of acetonitrile/water (50/50, v/v) with 10 mM ammonium formate as solvent A and acetonitrile/water/methanol (80/15/5, v/v/v) with 10 mM ammonium formate as solvent B. C₁₈ columns in positive and negative ionization modes share the same gradient elution program: 0 to 5.1 min, B increased linearly from 0.5 to 70%; 5.1 to 5.6 min, B increased linearly from 70 to 98%; 5.6 to 6.5 min, B remained isocratically at 98%; 6.5 to 6.7 min, B decreased linearly from 98 to 0.5%; and from 6.7 to 10 min, B remained isocratically at 0.5%. On the other hand, the gradient elution program for the HILIC columns was as follows: 0 to 5.6 min, B increased linearly from 2 to 50%; 5.6 to 7.6 min, B increased linearly from 50 to 95%; 7.6 to 8.6 min, B remained isocratically at 95%; 8.6 to 8.8 min, B decreased linearly from 95 to 2%, and from 8.8 to 10 min, B remained isocratically at 2%. Spectra were collected using the data-dependent mode (ddMS²). MS¹ acquisition parameters were as follows: resolution (70,000); AGC target (1e⁶); maximum IT (60 ms); scan range (70–910 m/z); and spectrum data type (centroid). MS² acquisition parameters were as follows: resolution (17,500); AGC target (1e⁵); maximum IT (50 ms); loop count (4); isolation window (1.5 m/z); (N)CE/stepped (10/30/55); spectrum data type (centroid); minimum AGC target (8.00e³); intensity threshold (1.6e⁵); exclude isotopes (on); and dynamic exclusion (3.0 s). For the positive ionization mode, the ion source conditions were set as follows: sheath gas flow rate (40 arbitrary units); aux gas flow rate (8 arbitrary units); sweep gas flow rate (0 arbitrary units); spray voltage (3.5 kV); capillary temp (300 °C); S-lens RF level (60); and aux gas heater temp (400 °C). Finally, the negative ionization mode ion source conditions were set as follows: sheath gas flow rate (40 arbitrary units); aux gas flow rate (10 arbitrary units); sweep gas flow rate (4 arbitrary units); spray voltage (2.5 kV); capillary temp (275 °C); S-lens RF level (60); and aux gas heater temp (400 °C).

2.7. Metabolomics Data Processing.

LC–MS data were processed using the open-source software MS-DIAL (version 4.70).²⁰ MS-DIAL was used for data collection, peak detection, compound identification, and alignment. Data collection and peak detection parameters were as follows: MS¹ tolerance (0.01 Da); MS² tolerance (0.025 Da); minimum peak height (50,000 amplitude); mass slice width (0.1 Da); smoothing method (linear weighted moving average); smoothing level (3 scans); and minimum peak width (5 scans). Identification parameters for the MS/MS file were as follows: accurate mass tolerance (MS¹): 0.005 Da; accurate mass tolerance (MS²): 0.01 Da; and identification score cutoff: 85%, whereas the identification parameters for the mzRT file were as follows: retention time tolerance (0.15 min); accurate mass tolerance (0.005 Da); and identification score cutoff (85%). For the positive ionization mode, the following adducts were included in the adduct ion setting: [M + H]⁺, [M + NH₄]⁺, [M + Na]⁺, [M + K]⁺, [M + H − H₂O]⁺, [M + H − 2H₂O]⁺, [2M + H]⁺, [2M + NH₄]⁺, [2M + Na]⁺, and [2M + K]⁺, whereas for the negative ionization mode, the following adducts were included in the adduct ion setting: [M − H]⁻, [M − H − H₂O]⁻, [M + Cl]⁻, [M − H + FA]⁻, [M − H + CH³COOH]⁻, and [2M − H]⁻. Finally, alignment parameters were as follows: retention time tolerance (0.1 min); MS1 tolerance (0.015 Da); remove features based on blank information; sample max/blank average (10-fold change); keep removable features and assign the tag; and gap filling by compulsion. Data from the three chromatographic conditions were processed separately and combined after data curation. Metabolites were annotated using the Metabolomics Standards Initiative (MSI) confidence levels by matching precursor masses, retention times, and MS/MS fragment data against the NIST20 mass spectral reference (National Institute of Standards and Technology, Gaithersburg, MD), MoNA (MassBank of North America, MassBank.us), and our internal library. Adducts, duplicate peaks, and isotopic features were flagged using the open-source software MS-FLO.²¹ For annotated compounds present in more than two chromatographic conditions, the criterion used to retain the compounds was the higher signal-to-noise ratio. Missing values were replaced by 1/5th of the minimum positive value of each variable.

2.8. Statistical Analysis.

Univariate analysis was carried out using GraphPad Prism (GraphPad software, version 9.3.1), while multivariate analysis was performed using the web-based metabolomics data-processing tool MetaboAnalyst 5.0.^22,23 Data were normalized and log₁₀-transformed. Multivariate exploratory analysis was performed using the unsupervised principal component analysis (PCA) and supervised orthogonal partial least squares-discriminant analysis (OPLS-DA). Model validation for the OPLS-DA was performed using a permutation test (n = 2000). Metabolites with a variable importance in projection (VIP) score higher than 1.5 and a p-value lower than 0.05 were considered biomarkers. Statistical significances were identified by * for p-value < 0.05, ** for p-value < 0.01, *** for p-value < 0.001, and **** for p-value < 0.0001.

3. RESULTS AND DISCUSSION

3.1. Metabolome Data Processing.

Plasma samples obtained from three liquid chromatographic conditions were analyzed via Orbitrap high-resolution mass spectrometer. Data obtained from the analyzed samples were processed using MS-DIAL.²⁰ MS/MS spectral information from the NIST20, MoNA, and our internal library was used to annotate 460 metabolites. To assess the precision of our untargeted metabolomics study, a quality control (QC) reference pool sample was made from all plasma extract to reflect an aggregated plasma metabolite composition. Data from this QC reference pool sample were utilized by SERFF software to normalize metabolite intensities and correct for potential drift or batch effect.²⁴ Our results indicated that the relative standard deviation (RSD) from the QC reference pool sample data was 4.40% (Figure S1), suggesting minimal residual deviation. Variation associated with sample processing and instrument performance for each sample, independent of the QC pool sample, was evaluated with the isotopically labeled internal standards. Our results indicate that these metabolites were detected at RSD <20% (data not shown), highlighting the quality of the data. Later, we found strong Spearman rank correlations within biological replicates of the same treatments (r_xy 0.80–0.88) (Table S2). Overall, our results showed excellent repeatability across experiments and were sufficient to ensure data quality.

3.2. Multivariate Statistical Analysis.

3.2.1. Principal Component Analysis.

PCA, the primary and most widely used unsupervised technique for reducing the dimensionality of high-dimensional data sets while preserving their original structure and relationships inherent in the data set, was used to reveal outliers and trends in metabolome data. The PCA score plot in Figure 1A showed that QC reference pool samples were aggregated into a tight cluster. Conversely, metabolic data of the treatment and control samples were scattered but within the colored ellipses representing the 95% confidence level, suggesting that there are no outliers and validating the robustness of our analytical platform. Considering the diets provided to the mice, the PCA score plot showed well-differentiated groups as a result of the metabolomic perturbations caused by the diets. Plasma from the control diet was located in the negative part of the PC1, while plasma from the treatment diet was located in the positive part of the PC1. Therefore, it is deductible that PC1 grouped samples based on diets with 38% of the total metabolic variance of the data set. The scree plot in Figure 1B showed that the first two principal components (PCs) explained 65.5% of the total metabolic variance of the data, and adding five PCs explained 86.8%.

Figure 1.

PCA score plot between control and treatment groups of plasma of healthy mice treated with a dose of 0.6% w/w GEG extract. Red dots stand for the control group, blue dots stand for group 0.6%, and green dots stand for the QC group. PC1 (x-axis) and PC2 (y-axis) accounted for 65.5% of the total variance (A). Scree plot showing the variance explained by the first five principal components. The green line on top shows the accumulated variance explained, whereas the blue line underneath shows the variance explained by each principal component (B).

3.2.2. Orthogenol Partial Least Square-Discriminant Analysis.

To further maximize the difference between control and treatment groups, an OPLS-DA a supervised technique, was employed. The OPLS-DA score plot in Figure 2A demonstrated that the metabolomic profiles of the control and treatment groups were entirely separated, suggesting that the diets induced significant biochemical changes. It is essential to mention that although variation within groups is appreciable in both clusters, samples were within the 95% confidence level. Results also showed that the two orthogonal T scores explained 50.4% of total variance. T score 1 (x-axis) explained 29.6% of the total variance, whereas T score 2 (y-axis) explained the additional 20.8% of the total variance. Overall, the OPLS-DA plot demonstrated good model adaptability. Subsequently, we used a permutation test (n = 2000) to assess the robustness of the OPLS-DA model (Figure 2B). High values of goodness-of-fit (R2Y; 0.991) and goodness-of-prediction (Q2; 0.84) indicate how good the separation between groups is. As shown in Figure 2B, the values of R2Y and Q2 are both above 0.84, which suggests excellent variation between groups. Moreover, R2Y and Q2 both had p-values lower than 0.05, suggesting that the model fits well and will reproduce mathematically the data.²⁵ The variable importance in projection (VIP) plot from the OPLS-DA model was employed to estimate the importance of each variable according to a VIP score (Figure 2C). Metabolites with VIP scores greater than 1.5 indicate that they could contribute significantly to the classification. In our study, 68 metabolites were selected as potential biomarkers based on a VIP greater than 1.5 and p-values less than 0.05 (Table 1).

Figure 2.

OPLS-DA score plot between control and treatment groups of plasma of healthy mice treated with a dose of 0.6% w/w GEG extract. Red dots stand for the control group, and green dots stand for group 0.6%. PC1 (x-axis) and PC2 (y-axis) accounted for 50.4% of the total variance (A). Permutation analysis with observed Q2 coefficients and cross-validated R2Y (B). Variable importance in projection (VIP) from the OPLS-DA model for the top 35 metabolites. Metabolites are ranked from most to least important. The colored boxes on the right indicate metabolite concentration in each group (C).

Table 1.

Potential Biomarkers from Mice Treated with a Control Diet and Diet with a Dose of 0.6% w/w of Gingerol-Enriched Ginger Extract Based on Criteria of the p-Value (<0.05) and VIP Score (>1.5)^a

metabolite	class	sub-class	p-value	VIP	log2(FC)
guanosine	nucleotides	purine metabolism	3.68 × 10−3	1.52	1.68
2′-deoxyinosine			1.07 × 10−5	1.75	5.08
hypoxanthine			4.24 × 10−4	1.64	3.47
oxypurinol			2.03 × 10−5	1.74	4.01
inosine			6.03 × 10−5	1.71	4.24
xanthine			3.00 × 10−6	1.77	5.50
decenoic acid	lipids	fatty acids and conjugates	2.81 × 10−3	1.59	1.82
docosahexaenoic acid			3.14 × 10−3	1.60	−1.18
cholic acid		primary bile acid biosynthesis	2.40 × 10−3	1.61	1.12
cortisol		steroid hormone biosynthesis	1.00 × 10−5	1.77	3.04
glycerophosphate		glycerophospholipid metabolism	4.87 × 10−3	1.54	<1.00
tyrosine	amino acids	tyrosine metabolism	6.94 × 10−3	1.51	<1.00
acetyltyrosine			5.86 × 10−3	1.55	1.11
ophthalmatic acid		cysteine and methionine metabolism	5.84 × 10−4	1.71	1.22
acetylmethionine			9.10 × 10−3	1.52	<1.00
methylthioadenosine			1.00 × 10−3	1.70	<1.00
choline		glycine, serine, and threonine metabolism	6.09 × 10−3	1.53	<1.00
kynurenic acid		tryptophan metabolism	2.67 × 10−3	1.60	1.27
indole-3-acetic acid			1.28 × 10−3	1.70	<1.00
leucylaspartic acid		dipeptides	1.94 × 10−4	1.71	<1.00
glutamylvaline			9.44 × 10−4	1.65	<1.00
glutamyltyrosine			1.42 × 10−3	1.63	1.06
glutamylisoleucine			3.19 × 10−3	1.58	<1.00
rhamnose	carbohydrates	fructose and mannose metabolism	1.58 × 10−5	1.74	−1.67
fucose			4.58 × 10−5	1.72	−1.54
ribose 5-phosphate		pentose phosphate pathway	3.45 × 10−4	1.70	1.07
glucuronic acid		pentose and glucuronate interconversions	4.00 × 10−3	1.59	1.04
galactonic acid			4.00 × 10−3	1.54	1.11
sedoheptulose		carbohydrates	1.06 × 10−4	1.76	−1.27
galacturonic acid			1.36 × 10−5	1.79	1.52
2-keto-l-gulonic acid			4.00 × 10−3	1.59	1.04
3,5-di-tert-butyl-4-hydroxybenzoic acid	exogenous	food components	5.67 × 10−7	1.82	3.77
DHPPTA sulfate			2.88 × 10−7	1.80	2.69
vanillic acid 4-sulfate			8.94 × 10−7	1.78	3.03
4-methylcinnamic acid			9.23 × 10−7	1.81	4.26
2-methylcinnamic acid			9.44 × 10−7	1.81	6.73
dihydroferulic acid sulfate			1.86 × 10−6	1.80	2.88
3-hydroxy-4-methoxybenzenemethanol			2.24 × 10−6	1.80	3.68
3,5-DHPPA			5.91 × 10−3	1.52	1.00
4-hydroxybenzaldehyde			4.71 × 10−3	1.57	<1.00
1-(4-hydroxyphenyl)-2-phenyl-1-butanone			1.74 × 10−3	1.66	4.67
2-(6-hydroxyhexyl)-3-methylidenebutanedioic acid			5.55 × 10−3	1.54	1.04
N-(3-phenylpropyl)thiourea			2.49 × 10−6	1.80	4.42
N5-(1-iminoethyl)-l-ornithine			3.32 × 10−5	1.80	<1.00
3-(carboxymethyl)-1-β-d-glucopyranosyl-1H-indole			1.28 × 10−3	1.70	<1.00
propanedioic acid			3.30 × 10−3	1.59	3.95
indole-3-lactic acid			3.40 × 10−3	1.65	<1.00
3-(2-nitro-1-propenyl)indole			5.10 × 10−3	1.58	<1.00
(5S)-[6]-gingerol	ginger metabolites/exogenous	food components	2.95 × 10−7	1.82	2.68
[6]-isogingerol			2.42 × 10−5	1.77	3.10
(3R,5S)-[6]-gingerdiol			1.49 × 10−5	1.78	3.02
(3R,5S)-[6]-gingerdiol-4′-glucuronide			2.00 × 10−5	1.78	2.97
[6]-gingerdione			1.09 × 10−4	1.74	3.34
[6]-shogaol			2.89 × 10−5	1.76	2.89
[6]-shogaol-4′-glucuronide			7.57 × 10−5	1.74	3.13
(5S)-[8]-gingerol			6.46 × 10−5	1.74	3.02
(5S)-[8]-gingerdiol			2.97 × 10−6	1.79	2.83
(5S)-[10]-gingerol			3.09 × 10−5	1.77	3.22
[10]-shogaol			3.28 × 10−3	1.58	1.15
[10]-shogaol-4′-glucuronide			2.66 × 10−7	1.80	2.58
6S-M1-4′-glucuronide			7.23 × 10−4	1.66	3.83
6S-M5-4′-glucuronide			9.51 × 10−4	1.63	3.82
6S-M6			6.03 × 10−6	1.78	3.20
6S-M7			2.09 × 10−5	1.75	3.36
6S-M10–4′-glucuronide			1.31 × 10−7	1.80	2.76
6S-M12-4′-glucuronide			3.88 × 10−5	1.74	3.56
6S-M12-4′ glucuronide methyl ester			2.09 × 10−5	1.76	3.33
6S-M20-4′-glucuronide			1.50 × 10−4	1.69	3.60

^aDHPPTA sulfate, 5-(3,5-dihydroxyphenyl)pentanoic acid sulfate; 3,5-DHPPA, 3-(3,5-dihydroxyphenyl)-1-propanoic acid; 6S, [6]-shogaol; and M1, M5-7, M10, M12, and M20, metabolites of [6]-shogaol.

3.2.3. Hierarchical Clustering Heatmap.

To visualize the differences between the control and treatment groups, a hierarchical clustering heatmap was performed. As shown in Figure 3, two major clusters were formed. One cluster included the samples of the control group, and the other cluster included the samples of the treatment group. Figure 3 indicates that for the top 40 annotated metabolites, sedoheptulose, rhamnose, and fucose decreased in the treatment group, while all other metabolites increased in the treatment group.

Figure 3.

Hierarchical clustering heatmap between control and treatment groups of plasma of healthy mice treated with a dose of 0.6% w/w of GEG extract metabolites for the top 40 annotated metabolites using p-value criteria. Each row indicates a metabolite, and each column indicates a sample. The correlation degrees are shown using a color scale, the red color represents upregulation, while the blue color represents downregulation.

3.3. Identification of Ginger Exposure Biomarkers.

In LC–MS-based metabolomics, compound annotation from mass spectral data is the most challenging task. This challenge increases even more for dietary metabolites because standards are often not available and existing databases are limited.²⁶ For instance, most of the metabolites contained in the NIST20 mass spectral library are not dietary metabolites. In this study, our group constructed a mass spectral library containing both endogenous and dietary metabolites. Data processed and analyzed using these mass spectral libraries allowed the annotation of 460 metabolites. Among these 460 annotated metabolites, 68 were considered potential biomarkers based on the criteria of the p-value (<0.05) and VIP score (>1.5) (Table 1). Interestingly, 20 of these 68 metabolites were ginger dietary metabolites. Apart from this approach, data were also processed and analyzed using these libraries but without the dietary metabolites. Results from this approach indicated that only four ginger metabolites were annotated {(5S)-[6]-gingerol, (5S)-[8]-gingerol, (5S)-[10]-gingerol, and [6]-shogaol}, which demonstrates the importance of dietary libraries for discovering food biomarkers.

Although these four ginger metabolites were annotated using the NIST and MoNA libraries, none of them could be accurately identified. For example, when the data were processed and analyzed using the library containing the dietary metabolites, the dot- and reverse-dot product scores for (5S)-[6]-gingerol were excellent (>900). Furthermore, when comparing the experimental (red) and candidate (blue) MS/MS spectra, they were identical (Figure 4A). On the other hand, when the data were processed and analyzed using the online libraries, (5S)-[6]-gingerol did not match the top 10 choices of the NIST20 and MoNA libraries. As shown in Figure 4B, the first option for (5S)-[6]-gingerol using the online libraries was (5S)-[10]-gingerol, which obviously does not match the correct metabolite. This is also supported by the poor (<700) and fair (700–800) dot- and reverse-dot product scores, respectively. Furthermore, when comparing the experimental and candidate MS/MS spectra, the fragmentation pattern and peak intensities were not identical. Considering this example, it is recommended that when the NIST20 and MoNA libraries are used for the annotation of dietary metabolites, further validations using authentic standards must be conducted.

Figure 4.

Comparison of the dot- and reverse-dot product scores and experimental (red) and candidate (blue) MS/MS spectra using the in-house library containing the dietary metabolites (A) and the online database (B) for the annotation of (5S)-[6]-gingerol.

Table 1 lists the 20 potential ginger metabolites that can be used as exposure biomarkers of ginger. Among these 20 ginger metabolites, nine were identified as glucuronidated gingerols and shogaols. It has been described that glucuronidation is the most common phase II reaction occurring in mammals.²⁷ Thus, it can be deduced that ginger metabolites conjugated with glucuronic acid render them more hydrophilic and therefore excretable through urine or feces. In addition, we plotted the peak intensities for the 20 ginger metabolites, as shown in Figure 5. As expected, ginger metabolites were not detected in the control group, but they were significantly elevated in the treatment group. As shown in Figure 5, three ginger metabolites (6S-M1–4′-glucuronide, 6S-M5–4′-glucuronide, and 6S-M20-4′-glucuronide) had p-values lower than 0.001, while the remaining 17 ginger metabolites had p-values lower than 0.0001, suggesting that these dietary metabolites can be used as exposure biomarkers of ginger.

Figure 5.

Box plots of significant annotated ginger metabolites. Statistical significances were identified by (*) for significant (p-value < 0.05), (**) for significant (p-value < 0.01), (***) for significant (p-value < 0.001), and (****) for significant (p-value < 0.0001) differences.

3.4. Metabolic Pathway Analysis.

On the basis of annotated metabolites, a metabolic pathway analysis was performed by MetaboAnalyst’s tool to discover the metabolic pathways affected by the ginger diet (Figure 6). A total of 19 metabolic pathways showed diet-related changes. Metabolic pathways with high −log10(p) and pathway impact scores are significantly affected by the ginger diet. Thus, in terms of both significance and impact, the metabolic pathways impacted mainly were purine metabolism and phenylalanine, tyrosine, and tryptophan biosynthesis.

Figure 6.

Pathway analysis of significantly altered metabolites using MetaboAnalyst software. The color gradient and circle size indicate the significance of the pathway ranked by p-value (yellow: higher p-values and red: lower p-values) and pathway impact score (the larger the circle, the higher the impact score), respectively. Significantly affected pathways with high −log10(p) and high pathway impact scores are identified by name.

Because purine metabolism is the major metabolic pathway affected by the ginger diet, we selected and plotted their annotated metabolites. As shown in Figure 7, the six annotated metabolites (2′-deoxyinosine, xanthine, hypoxanthine, guanosine, oxypurinol, and inosine) in the treatment groups were significantly upregulated when compared to the control group. Previous studies have shown that purine metabolites may boost immune response, play an important role in energy metabolism, and serve as building blocks for DNA and RNA.^28,29 On the other hand, guanosine, oxypurinol, and inosine exert antioxidant properties.^30,31

Figure 7.

Box plots of significant annotated endogenous metabolite biomarkers associated with purine metabolism. Statistical significances were identified by (*) for significant (p-value < 0.05), (**) for significant (p-value < 0.01), (***) for significant (p-value < 0.001), and (****) for significant (p-value < 0.0001) differences.

Similar to the potential biomarkers of purine metabolism, metabolites from carbohydrate, amino acids, and lipid classes may also serve as potential functional biomarkers for ginger intake (Figure 8). For instance, in the class of amino acids, N-acetyl-l-tyrosine has shown suppressive activity on the growth of HCT116 human colorectal cancer cells. This finding was carried out by transplanting HCT116 cells into nude mice treated with and without N-acetyl-l-tyrosine and comparing the tumor volumes.³² N-Acetyl-l-tyrosine and l-tyrosine may be effective in treating mild-to-moderate depression because it is a precursor of dopamine and norepinephrine.³³ On the other hand, in the class of lipids, cortisol has anti-inflammatory effect and boosts the immune system.³⁴ Cholic acid assists in fat digestion by solubilizing fats for absorption.³⁵ This is supported by a previous study that demonstrated that a gingerol- and shogaol-enriched extract reduces total cholesterol in plasma by enhancing bile acid excretion.³⁶ Overall, our findings indicated that these biomarkers might ameliorate different pathogenesis disorders. However, further studies and evidence are needed to compare and validate these results in human samples to reveal their potential as biomarkers. For example, a recent study showed a data convergence of approximately 17% for metabolites differentiated from animal models to humans in metabolomics studies.³⁷ In addition, our study also has limitations. First, the number of mice was small, which could affect the detection of significant metabolites. Second, the percentage of GEG extract used in the treatment diet needs to be refined through efficacy and effectiveness trials.

Figure 8.

Box plots of significant endogenous annotated metabolite biomarkers associated with health benefits. Statistical significances were identified by (*) for significant (p-value < 0.05), (**) for significant (p-value < 0.01), (***) for significant (p-value < 0.001), and (****) for significant (p-value < 0.0001) differences.

In summary, we confirm the need to construct and continually expand mass spectral libraries of dietary metabolites for discovering food biomarkers. The use of our mass spectral library, containing both endogenous and dietary metabolites, combined with untargeted metabolomics allowed the annotation of 460 metabolites, of which 20 were identified as food biomarkers of ginger intake. Altogether, our results revealed that the plasma metabolome of mice on normal and ginger diets have different metabolic profiles, which are closely associated with purine metabolism and phenylalanine, tyrosine, and tryptophan biosynthesis. Further studies should evaluate the efficacy and effectiveness of trials. Overall, our findings suggest that more effort is needed by the metabolomic community to address the lack of dietary biomarkers, which in turn will allow researchers to correlate potential health benefits associated with food consumption.