Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2025 Jul 15.
Published in final edited form as: Food Chem. 2024 Feb 16;446:138744. doi: 10.1016/j.foodchem.2024.138744

Multidisciplinary approach combining food metabolomics and epidemiology identifies meglutol as an important bioactive metabolite in tempe, an Indonesian fermented food

Marvin N Iman 1,13, Danielle E Haslam 2,4,13, Liming Liang 5,6, Kai Guo 7, Kaumudi Joshipura 5,7, Cynthia M Pérez 8, Clary Clish 9, Katherine L Tucker 10, JoAnn E Manson 3, Shilpa N Bhupathiraju 2,4,14, Eiichiro Fukusaki 1,11,12,14, Jessica Lasky-Su 2,14, Sastia P Putri 1,12,14,*
PMCID: PMC11247955  NIHMSID: NIHMS2005009  PMID: 38432131

Abstract

This study introduces a multidisciplinary approach to investigate bioactive food metabolites often overlooked due to their low concentrations. We integrated an in-house food metabolite library (n=494), a human metabolite library (n=891) from epidemiological studies, and metabolite pharmacological databases to screen for food metabolites with potential bioactivity. We identified six potential metabolites, including meglutol (3-hydroxy-3-methylglutarate), an understudied low-density lipoprotein (LDL)-lowering compound. We further focused on meglutol as a case study to showcase the range of characterizations achievable with this approach. Green pea tempe was identified to contain the highest meglutol concentration (21.8±4.6 mg/100g). Furthermore, we identified a significant cross-sectional association between plasma meglutol and lower LDL cholesterol in two Hispanic adult cohorts (n=1,628) (β[SE]: −5.5(1.6) mg/dl, P=0.0005). These findings highlight how multidisciplinary metabolomics can serve as a systematic tool for discovering and enhancing bioactive metabolites in food, such as meglutol, with potential applications in personalized dietary approaches for disease prevention.

Keywords: Food metabolomics, molecular epidemiology, meglutol, functional food, bioactive compounds

1. Introduction

Functional foods, also known as nutraceuticals, are foods that provide health benefits beyond basic nutrition through the presence of physiologically active components. The utilization of functional foods as a means to manage disease risk has been increasingly popular in recent years (Iwatani & Yamamoto, 2019; Tadesse & Emire, 2020). Dietary changes, as opposed to drug intervention, may pose fewer adverse side-effects and are perceived to be a more “natural” approach (Megson, Whitfield, & Zabetakis, 2016). Accordingly, identifying and developing novel functional foods with specific bioactive properties enhance personalized dietary approaches (Birch & Bonwick, 2019).

Common approaches for investigating food bioactive properties often do not consider the vast array of identifiable metabolites in food (Barabási, Menichetti, & Loscalzo, 2020). Without this information of food’s metabolite constituents, less-common metabolites with bioactive properties present in low concentrations may be overlooked (Caesar, Kellogg, Kvalheim, & Cech, 2019). Even in cases where data capturing most of the food metabolome is available, assessing the bioactive properties of these metabolite constituents often poses a significant challenge (Rinschen, Ivanisevic, Giera, & Siuzdak, 2019). Applying conventional biochemical assays, which are typically confined to a restricted set of pre-selected targets and purified compounds, to assess the biological activities of the entire food metabolome would be impractical and costly (Faudone, Arifi, & Merk, 2021; Lelario et al., 2018; Sánchez-Ruiz & Colmenarejo, 2022).

Recently, cheminformatics approaches have been developed to predict the bioactivity of food metabolites, such as predictions based on metabolites’ chemical structures and interactions (Sánchez-Ruiz et al., 2022; Staszak, Staszak, Wieszczycka, Bajek, Roszkowski, & Tylkowski, 2022). While these approaches are more scalable and can be used to rapidly and systematically predict the bioactive properties of food metabolites, these in silico predictions and biochemical assays are not based on human health data, such as from epidemiological studies or clinical trials. The integration of human health data into biological activity prediction is crucial to provide physiological context and improve accuracy, as in silico and in vitro testing alone often demonstrate poor correlation with in vivo biological activities (Zhu, 2020).

To address these challenges, here, we introduce a multidisciplinary framework for investigating bioactive metabolites in foods. This framework integrates three key components: a food metabolome library, publicly available metabolite pharmacological databases, and a human metabolite library derived from epidemiological studies. By linking the data from these three sources, we enable rapid in silico identification of potential food bioactive metabolites that may otherwise be overlooked. Additionally, the multifaceted nature of the linked data allows for important characterizations of identified metabolites. For instance, the food metabolome library can be used to pinpoint food sources rich in the identified metabolite and investigate its modulability, while the epidemiological data can reveal potential associations with health outcomes.

2. Methods

2.1. Combining food metabolomics and molecular epidemiological data to screen for unique food-sourced bioactive compounds

2.1.1. Cross-referencing food metabolome and molecular epidemiological data

We constructed an internal food metabolite library containing 494 unique metabolites derived from previously analyzed food metabolome data within our group (Aditiawati et al., 2020; Amalia, Aditiawati, Yusianto, Putri, & Fukusaki, 2021; Dahlan, Nambu, Putri, & Fukusaki, 2022; Hanifah, Firmanto, Putri, & Fukusaki, 2022; M. Ikram, Ridwani, Putri, & Fukusaki, 2020; M. M. M. Ikram, Mizuno, Putri, & Fukusaki, 2021; Jumhawan, Putri, Bamba, & Fukusaki, 2016; Kadar, Aditiawati, Astawan, Putri, & Fukusaki, 2018; Kitani, Putri, & Fukusaki, 2022; Ohto, Putri, Suantika, & Fukusaki, 2021; A. A. Parijadi, Putri, Ridwani, Dwivany, & Fukusaki, 2018; A. A. R. Parijadi et al., 2022; S. L. E. Putri et al., 2021; S. L. E. Putri, Suantika, Situmorang, Putri, & Fukusaki, 2022; Sastia Prama Putri, Irifune, Yusianto, & Fukusaki, 2019; Sastia P Putri, Jumhawan, & Fukusaki, 2015; Rahmawati, Astawan, Putri, & Fukusaki, 2021; A. Sato, Astuti, Putri, & Fukusaki, 2020; A. Sato, Putri, Astuti, & Fukusaki, 2022; M. Sato, Ikram, Pranamuda, Agusta, Putri, & Fukusaki, 2021; Yamana et al., 2020). These data were obtained using gas chromatography-mass spectrometry (GC-MS) and annotated by comparing retention indices (RIs) and mass spectra with an in-house reference library (Riken, Kanagawa, Japan) (Tsugawa et al., 2015). The cohort study library comprises of 891 uniquely annotated metabolites from previous epidemiological studies (Andriankaja, Jiménez, Muñoz‐Torres, Pérez, Vergara, & Joshipura, 2015; Joshipura, Muñoz-Torres, Dye, Leroux, Ramírez-Vick, & Pérez, 2018; Litonjua et al., 2014; Pérez et al., 2017; Tucker et al., 2010). Metabolites from the food metabolite library and the cohort study library were cross-referenced to identify common metabolites found in both datasets. The cross-referencing was performed based on The Human Metabolome Database (HMDB) ID, the International Chemical Identifier (InChIKey), and common metabolite names.

2.1.2. Metabolite bioactive properties prediction and filtering to target less-common metabolites

To predict bioactive properties of the annotated metabolites, we collected pharmacological data for each individual metabolite from PubChem (https://pubchem.ncbi.nlm.nih.gov/), DrugBank (https://go.drugbank.com/), and Chemical Entities of Biological Interest (ChEBI) (https://www.ebi.ac.uk/chebi/). To target less-commonly studied metabolites, we selected only metabolites with a total number of National Library of Medicine (NLM) Curated PubMed Citations of less than 1000. The NLM Curated PubMed Citations retrieves PubMed records (https://pubmed.ncbi.nlm.nih.gov/) tagged with MeSH terms associated with a compound. Metabolites identified in this literature search were considered for further characterization. We chose the most promising metabolite as a case study based on existing characterization in the literature amongst the identified metabolites.

2.2. Case Study: Evaluation and optimization of meglutol content in foods

2.2.1. Screening for foods containing meglutol

We used our in-house food metabolome database, which encompasses an extensive collection of metabolome profiles derived from various food samples to screen for foods containing meglutol. In addition, we incorporated foods that have been previously documented to contain meglutol in earlier studies into our sample set to enable a comparative analysis of their meglutol content (Farag, Khattab, Ehrlich, Kropf, Heiss, & Wessjohann, 2018; Klimek-Szczykutowicz, Szopa, & Ekiert, 2020; Thiombiano, Gontier, Molinie, Marcelo, Mesnard, & Dauwe, 2020; Xu, Chen, Lu, Zhao, Yang, & Liu, 2021).

2.2.2. Legume and tempe samples

In this study we prepared soybean tempe along with red kidney bean, green pea, white bean, and edamame tempes (n = 3) which have been reported to contain meglutol (Rahmawati et al., 2021). Sample details are presented in Table S1. All tempe samples were produced at the Laboratory of Bioresource Engineering, Osaka University, Japan in December 2021. Tempe samples were prepared according to the Rumah Tempe Indonesia (RTI) method using Raprima as the starter culture (Rahmawati et al., 2021). Briefly, 50 g of each legume was soaked, boiled, and dehulled before starter culture inoculation. Tempe samples were then packed and fermented in an incubator (EYELA, Tokyo, Japan) at 30°C for 48 hours.

2.2.3. Metabolite extraction and derivatization

All samples were lyophilized overnight and homogenized into fine powder using a multi-beads shocker homogenizer (Yasui Kikai, Osaka, Japan). Ten milligrams of lyophilized samples, taken in triplicates (n = 3), were weighed into a 2 ml tube. To each sample, 1 mL of methanol (Wako Pure Chemical Industries, Osaka, Japan), ultrapure water (Wako Pure Chemical Industries, Osaka, Japan), and chloroform (Kishida Chemical Co. Ltd., Osaka, Japan) in the ratio of 5:2:2 was added. An additional 100 μg/mL ribitol (Wako Pure Chemical Industries, Osaka, Japan) was also added. The mixture was incubated at 37°C for 30 minutes while being mixed at 2 × g agitation (Eppendorf Ltd., Hamburg, Germany). The mixture was then centrifuged at 4°C for 3 minutes at 14,000 × g. Four hundred microliters of the supernatant were then transferred to a new 1.5 mL tube. Into this, 300 μL ultrapure water was added, and the mixture was further centrifuged at 4°C for 3 minutes at 14,000 × g. Two hundred mL of the aqueous phase were transferred to a new 1.5 mL tube with a holed cap. Quality Control (QC) samples were prepared by pooling 200 μL of the previously obtained aqueous phase from all samples. All samples were centrifuged in vacuum condition using a Spin Dryer Standard (Taitec Co., Saitama, Japan) for 1 hour at room temperature and lyophilized overnight. One hundred microliters of 20 mg/mL methoxyamine hydrochloride (Sigma-Aldrich, Tokyo, Japan) were added to the lyophilized samples. The mixture was then incubated at 30°C for 90 minutes while being mixed with 2 × g agitation. Silylation was then performed by adding 50 μL of trifluoroacetamide (MSTFA) (GL Sciences, Tokyo, Japan) followed by incubation at 37°C for 30 minutes with 2 × g agitation. The mixture was then transferred to GC/MS vials for the subsequent analyses.

2.2.4. GC-MS profiling

GC-MS profiling was performed at the Laboratory of Bioresource Engineering, Osaka University, Japan, as previously reported by (Rahmawati et al., 2021), with minor adjustments. Briefly, GC/MS-QP2010 Ultra (Shimadzu, Kyoto, Japan) with an Inert-Cap MS/NS column (GL Sciences) was used in this analysis. AOC-20i/s autosampler (Shimadzu, Kyoto, Japan) was used to inject the samples in 25:1 (v/v) split mode. The injection temperature was measured at 230°C. Helium was used as the carrier gas with the flow rate measured at 1.12 mL/minute and the linear velocity measured at 39 cm/s. The column temperature was maintained at 80°C for 2 minutes, then increased to 330°C with an increment rate of 15°C/minute and maintained for 6 minutes. The temperatures of the ion source and transfer line used were 200°C and 250°C, respectively. Ions were generated using the electron ionization (EI) method. Mass spectra were recorded over the mass range of m/z 85–500 with a scan rate of 6.67 scans/s. Retention Indices (RI) were determined using a standard alkene mixture.

2.2.5. Data processing and statistical analysis of metabolite data

Baseline correction, denoising, peak detection, alignment, and annotation processes were performed using MS-DIAL 4.0 (RIKEN, Wako, Japan) (Lai et al., 2018). RI-based target compound identification and quantification was performed using MS-DIAL 4.0 based on the information from GC/MS-5MP spectral library (RIKEN, Wako, Japan) (Tsugawa, Tsujimoto, Arita, Bamba, & Fukusaki, 2011). Meglutol annotation was particularly confirmed and subsequently quantified by co-injection analysis using an authentic standard (Wako Pure Chemical Industries, Osaka, Japan).

The peak intensities of the annotated metabolites were normalized against the peak intensity of ribitol as the internal standard. Filtering was performed to only select metabolites with QC relative standard deviation values lower than 30%. Principal component analysis (PCA) was performed using SIMCA P+ ver. 13.0 (Sartorius, Göttingen, Germany). Partial least squares-discriminant analysis (PLS-DA) and heatmap analysis were performed using MetaboAnalyst 5.0 (Pang et al., 2021).

2.2.6. Meglutol quantification

Quantification of meglutol was performed using an external calibration curve constructed with standard solutions at calibration levels ranging from 0 to 20 μg/mL. Ribitol was also added as an internal standard to these solutions to correct for volume errors. The linearity of the constructed calibration curve was confirmed by calculating the relative standard deviation values of the response factors, which were found to be less than 15%. The R2 value for the constructed calibration curve was 0.9896 (Figure S1).

2.3. Population-based examination of plasma meglutol and cardiometabolic risk

2.3.1. Study participants

We leveraged data from two well-characterized, population-based longitudinal cohort studies conducted among individuals of Puerto Rican descent to examine cross-sectional and prospective associations between plasma meglutol concentrations and cardiometabolic risk factors. The Boston Puerto Rican Health Study (BPRHS) includes 1,500 adults of Puerto Rican descent aged 45–75 y, living in or near Boston, MA and the San Juan Overweight Adult Longitudinal Study (SOALS) includes 1,300 overweight or obese [BMI ≥ 25 kg/m2] adults living in the San Juan, PR municipality and surrounding areas. We used data from BPRHS participants who completed a 2-year and 5-year follow-up visit and SOALS participants who completed a 3-year follow-up visit. Details about the design and data collection for these cohorts have been previously published (Andriankaja et al., 2015; Joshipura et al., 2018; Pérez et al., 2017; Tucker et al., 2010). Participants with available blood samples at baseline, plasma metabolomics profiling, and cardiometabolic risk factors were included in this cross-sectional analysis. Participants were excluded if they did not have the cardiometabolic risk factors available at follow-up or their standardized 2-year change in each outcome was greater than or less than 4 standard deviations (SD) from the mean.

Original study protocols and procedures were approved by The University of Puerto Rico Institutional Review Board (IRB) for SOALS and the IRB of Tufts Medical Center for BPRHS. All participants provided written informed consent for broad use of their data prior to participating in the study. This includes reuse of their data outside the original study. The IRB of the Harvard T.H. Chan School of Public Health approved the protocol for the current study.

2.3.2. Blood sample collection and plasma metabolomics profiling

Fasting blood collection was performed as part of the original BPRHS and SOALS protocols (Andriankaja et al., 2015; Joshipura et al., 2018; Pérez et al., 2017; Tucker et al., 2010). We leveraged meglutol concentrations in BPRHS and SOALS participants derived from previously conducted metabolomic profiling in plasma samples among 714 BPRHS (n=371 with T2D, n=339 free of T2D, and 4 missing T2D status) and 1,011 SOALS (n=77 with T2D, n=934 free of T2D) participants (Haslam et al., 2021). To account for skewed distribution of plasma meglutol, an inverse normal transformation was applied, and undetectable values were imputed at a value equal to half the minimum.

2.3.3. Outcome and covariate assessment

We leveraged data on cardiometabolic outcomes that were assessed at baseline, ~2–3-year, and 5–7-year visits among BPRHS participants and at baseline and 3-year visits among SOALS participants. Details about the methods to estimate waist circumference, height, weight, and plasma cholesterol, triglycerides (TG), HDL-C, LDL-C, plasma glucose, insulin, glycated hemoglobin A1c (HbA1c), and high-sensitivity C-reactive protein (CRP) concentrations have been previously reported (Andriankaja et al., 2015; Joshipura et al., 2018; Pérez et al., 2017; Tucker et al., 2010). The homeostatic model of insulin resistance (HOMA-IR) was calculated by multiplying fasting insulin (mU/L) by fasting glucose (mmol/L) divided by 22.5. Details about the collection of data on age, education, household income, medication use, family history, and health behaviors, including medication use and history of smoking, alcohol consumption, white blood cell counts, physical activity, acculturation (Marin & Gamba, 1996), perceived stress (Cohen, Kamarck, & Mermelstein, 1983), and diet quality (BPRHS only) (Tucker, Bianchi, Maras, & Bermudez, 1998) (Bhupathiraju, Lichtenstein, Dawson-Hughes, & Tucker, 2011) have also been previously reported (Andriankaja et al., 2015; Joshipura et al., 2018; Pérez et al., 2017; Tucker et al., 2010).

2.3.4. Statistical Analysis

Associations between meglutol and cardiometabolic risk factors were explored among 636 BPRHS and 1,011 SOALS participants with meglutol concentrations and cardiometabolic risk factors measured at baseline and follow-up exams (Figure S2). First, β coefficients and standard errors (SE) were extracted from multivariable linear regression models examining the cross-sectional associations between standardized plasma meglutol concentration (per 1-SD) and cardiometabolic risk factors (primary outcome: LDL-C; secondary outcomes: HDL-C, TG, WC, BMI, glucose, insulin, HbA1c, HOMA-IR, and CRP) at baseline. Models adjusted for the following covariates: Model (1) age and sex; (2) Model 1 plus smoking status (never, current, former), education (≤8th grade, 9th-12th grade or GED, college/some graduate school), physical activity (BPRHS: continuous score; SOALS: METs), alcohol intake (non-consumer, moderate consumer [females: 1 drink/day; males: 1–2 drinks/day], heavy consumer [females: >1 drink/day; males: >2 drinks/day]), T2D status, lipid-lowering medication status, hypertension medication status, income (<$20,000/year, ≥ $20,000/year), acculturation (BPRHS only: %), diet quality (BPRHS only: AHA-DS continuous score), perceived-stress score, family history of T2D (SOALS only), and white blood cell count (CRP outcome only: mm3); (3) Model 2 plus BMI (kg/m2) and WC (cm).

Second, we examined associations between baseline plasma meglutol concentration and longitudinal changes in cardiometabolic risk factors during the follow-up period. Standardized two-year changes in cardiometabolic risk factors were calculated by subtracting the baseline value from the prior exam, dividing by the follow-up time in years, and multiplying by two. β coefficients and standard errors (SE) were extracted from similar multivariable linear regression models that utilized the standardized two-year changes in cardiometabolic risk factors as the outcome. In longitudinal analyses, model 1 additionally adjusted for baseline values of each corresponding cardiometabolic risk factor and model 3 replaced adjustment for baseline waist circumference with standardized two-year changes in WC. In the BPRHS, linear mixed-effects regression models were utilized that accounted for multiple observations among individuals from baseline to 2–3-year and 2–3=5–7 -year visits.

Third, we examined effect modification by T2D status (BPRHS only due to the low number of participants with T2D in SOALS), lipid-lowering medication use, and BMI (obese [BMI ≥ 30 kg/m2] versus normal/overweight [BMI < 30 kg/m2] in both cross-sectional and longitudinal analyses of the primary outcome (LDL-C). In the fully adjusted models (Model 3), we extracted β coefficients and standard errors (SE) from stratified analyses and p-values for the multiplicative interaction terms. To combine cohort-specific estimates in ally analyses, we used inverse-variance weighted random effects meta-analyses. All statistical analyses were completed using R (version 3.6.0) statistical software (Team, 2013).

3. Results and discussion

3.1. Combining food metabolomics and nutritional epidemiological data to screen for unique food-sourced bioactive metabolites

Figure 1a shows the overview of the approach used to screen for unique food-sourced bioactive metabolites. We cross-referenced our internal food metabolite library (n=494)(Riken, Kanagawa, Japan) (Tsugawa et al., 2015), shown in Figure 1b, with the human metabolite library (n=891) (Andriankaja et al., 2015; Joshipura et al., 2018; Kelly et al., 2022; Litonjua et al., 2014; Pérez et al., 2017; Tucker et al., 2010). We identified 174 metabolites commonly shared between the two libraries (Figure 1d). To assess the bioactive properties of these 174 metabolites, we utilized the pharmacological information available in publicly accessible metabolite databases. Subsequently, we narrowed down our investigation to focus on less commonly reported metabolites, resulting in the identification of six unique metabolites with potential bioactive properties that may be important for human health and merit further study (Table S2).

Figure 1. A systematic approach to rapidly identify unique bioactive metabolites in foods.

Figure 1.

Open in a new tab

a) an overview of the approach used to identify and modulate unique metabolites with potential bioactive properties, including meglutol. The details of each stage are described in the Methods section. b) the metabolite composition in the utilized food metabolite library. c) the metabolite composition in the utilized human metabolite library from population studies. d) the number and composition of metabolites in each stage of filtering.

Given that these six identified metabolites emerged from the integration of food metabolome data and human metabolome data from epidemiological studies, it is possible for us to readily perform further characterizations on these metabolites in relation to food and potential health outcomes. This involves determining their concentrations in various foods using food metabolomics data and exploring their clinical potential by examining their associations with health outcomes through epidemiological data analysis. To demonstrate these characterizations, we select meglutol, one of the six identified metabolites, as a case study due to its comparatively limited existing characterization amongst the identified metabolites.

3.2. Screening for foods containing meglutol

We identified six food items containing meglutol in our internal database (Riken, Kanagawa, Japan), in addition to five other food items which have been reported to contain meglutol based on mass spectrometry (MS) or nuclear magnetic resonance (NMR) spectroscopy data (Table 1). Legumes made up a sizeable group of foods containing meglutol.

Table 1. Foods containing meglutol.

Legumes and legume-based products made up a significant part of this list. Tempe, an Indonesian traditional food made from fermented soybean, was found to contain meglutol.

Food product Detection method Reference
Orange GC-MS (Klimek-Szczykutowicz et al., 2020)
Flax seed LC-MS (Thiombiano et al., 2020)
Oat GC-MS (Xu et al., 2021)
Lentils GC-MS (Farag et al., 2018)
Lupinus seeds GC-MS (Farag et al., 2018)
Soybean GC-MS Internal database
Red kidney bean GC-MS Internal database
White bean GC-MS Internal database
Green pea GC-MS Internal database
Edamame GC-MS Internal database
Tempe GC-MS Internal database
Open in a new tab

We further measured the absolute concentrations of meglutol in these food samples (n=3) (Table 1). Quantification was performed using GC-MS with authentic meglutol standards. Results show that tempe contains significantly more meglutol than other foods (Figure 2a). On average, tempe had more than 3 times the concentration of meglutol than other foods previously reported to contain meglutol (n=3) (P = 6.763e-8). Tempe is an Indonesian traditional fermented food made from soybean (Figure 2b, 2c) and is a nutritious alternative food source rich in health-promoting bioactive compounds, such as isoflavones.

Figure 2. Concentrations of meglutol in previously reported foods.

Figure 2.

Open in a new tab

a) absolute quantification of meglutol concentration was performed using GC-MS with a meglutol standard compound (n=3). Groups labeled with different letters are significantly different using Tukey adjustment (P < 0.05). Error bars represent ±1SD from the mean. Tempe contained a significantly higher amount of meglutol relative to previously reported foods. Tempe is an Indonesian traditional food made from fermented soybean. b) soybean before tempe fermentation. c) soybean tempe.

Interestingly, comparing the meglutol concentrations in soybean before and after 48 hours of tempe fermentation showed significant modulation of meglutol concentrations by fungal fermentation. The meglutol content of tempe was significantly higher by over three-fold compared to its initial concentration in soybean before fermentation (P = 0.0009) (Table S3).

3.3. Tempe fermentation modulates meglutol concentrations in legumes

To further characterize the modulation of meglutol concentration by fermentation, we examined the effect of tempe fermentation in other legumes. We performed GC-MS-based metabolomic profiling on soybean, white bean, red kidney bean, green pea, and edamame before and after the tempe fermentation process (n=3) (Figure S3). We annotated 68 metabolites from these legumes and their resulting tempes by comparing the retention index (RI) values and mass spectra with our in-house library (Table S4). The detection and annotation of meglutol was specifically confirmed using a standard compound. The annotated metabolites were subjected to unsupervised principal components analysis (PCA) to observe the effect of tempe fermentation on the metabolome of these legumes (Figure 3a, 3b).

Figure 3. GC-MS metabolomic profiles of legumes before and after tempe fermentation.

Figure 3.

Figure 3.

Open in a new tab

a, Principal Component Analysis (PCA) score plot of legumes and tempes. Square points indicate legumes before fermentation. Triangle points indicate tempes (after fermentation). b, PCA loading plot of the metabolites obtained from GC-MS profiling of legumes and tempes. Light yellow point indicates meglutol. c, heatmap analysis of sugars and amino acid levels before and after tempe fermentation.

A distinct clustering pattern was observed along the PC1 axis of the score plot (Figure 3a), revealing two separate clusters. Prior to fermentation, all legumes were clustered together on the negative side of the PC1 axis, while the resulting tempe samples from the fermentation process were clustered together on the positive side of the PC1 axis, which suggests that tempe fermentation consistently modifies the metabolites of different legumes in a similar manner. The loading plot (Figure 3b) showed a clustering of sugars on the negative side of the PC1 axis, whereas amino acids were predominantly accumulated on the positive side of PC1. This indicates that fermentation effectively reduces sugar content while increasing amino acid content in legumes. Moreover, meglutol was found to be clustered on the extreme positive side of PC1 in the loading plot, providing further confirmation of our previous observation that tempe fermentation significantly enhances meglutol concentration. This modulation of meglutol might be attributed to the breakdown of amino acids, such as leucine, by fungal activity. Fungal metabolic models suggest the mitochondrial leucine degradation pathway may lead to meglutol (Rodríguez, Ruíz-Sala, Ugarte, & Peñalva, 2004). We measured the leucine content before and after fermentation in all five legumes. The results showed that the leucine content of tempes were significantly higher, at least five-fold, compared to their initial concentrations in legumes before fermentation (Figure S4). The increase of leucine as a meglutol precursor during fermentation may contribute to the elevated meglutol concentrations post-fermentation. Alternatively, Rhizopus oligosporus could also biosynthesize meglutol through its ergosterol pathway during tempe fermentation, thus increasing the meglutol content (Dupont, Lemetais, Ferreira, Cayot, Gervais, & Beney, 2012). The heatmap analysis further emphasizes the modulation of individual metabolites (Figure 3c).

Figure 4 shows the amount of meglutol in 100 grams of each tested legume and tempe. The meglutol content of each tested legume increased significantly after tempe fermentation (Figures 4a–d). All four tempes made from alternative legumes had higher meglutol content compared to traditional soybean tempe, regardless of their natural contents of meglutol before fermentation. Green pea contained the highest meglutol content among the tested legumes at 7.69 ± 0.63 mg meglutol/100 g bean (n=3), a number even higher than that of soybean tempe. Green pea tempe showed the highest meglutol content among the tested tempes at 21.8 ± 4.6 mg meglutol/100 g tempe (n=3) (Table S5). On the other hand, white bean showed the highest increase in meglutol concentration after tempe fermentation.

Figure 4. Tempe fermentation increased meglutol in all tested legumes.

Figure 4.

Open in a new tab

Absolute quantification of meglutol concentration was performed using GC-MS with a meglutol standard compound (n=3). Error bars represent ±1SD from the mean. Groups labeled with different letters are significantly different using Tukey adjustment (P < 0.05). a) comparison of meglutol content before and after fermentation in all tested legumes. b) comparison of meglutol naturally contained in all tested legumes (before fermentation). c) comparison of meglutol content of different tempes (after fermentation). d) the fold-change of meglutol content before and after fermentation in all tested legumes.

While all tempes made from alternative legumes showed better meglutol concentration than traditional soybean tempe, we found green pea tempe to contain the highest concentration of meglutol, suggesting that it can serve as an ideal source of meglutol in diet. Surprisingly, green pea tempe showed the lowest change of meglutol relative to its legume. In contrast, white bean tempe, which showed the lowest concentration of meglutol among the alternative tempes, had the highest change of meglutol relative to its legume. These findings suggest that green peas may possess limited bioavailability of meglutol precursors, potentially hindering the Rhizopus fungus’s access to sufficient precursors for producing high meglutol concentrations during the fermentation process. Conversely, white beans may contain abundant and readily bioavailable meglutol precursors, enabling the Rhizopus fungus to effectively generate a substantial amount of meglutol.

3.4. Population-based examination of plasma meglutol and cardiometabolic risk

In addition to its characterizations in foods, we further characterized meglutol’s clinical potential by leveraging the epidemiological data available to us. In the subsequent analyses, we investigated the associations between plasma meglutol concentration and cardiometabolic risk factors.

The characteristics of Boston Puerto Rican Health Study (BPRHS) and San Juan Overweight Adult Longitudinal Study (SOALS) participants are presented in Table S6. The majority were female (73%), and the mean age at baseline was 57.1 y in the BPRHS and 50.7 y in SOALS. In a meta-analysis of BPRHS and SOALS participants, baseline meglutol concentrations (per 1-SD) were significantly associated with lower LDL-C concentration (β [SE]: −5.5 [1.6] mg/dl, P = 0.0005) (Figure 5a and Table S7). These findings are novel and are in line with previous meglutol supplementation trials that have reported decreases in LDL and total cholesterol among hypercholesterolemia patients (Afifah, Nabilah, Supraba, Pratiwi, Nuryanto, & Sulchan, 2020; Lupien, Moorjani, Brun, & Bielmann, 1979).

Figure 5. Cross-sectional associations between meglutol and LDL-C concentration in US adults.

Figure 5.

Open in a new tab

a) in a random effects meta-analysis of the BPRHS and SOALS participants, baseline meglutol concentration (per 1-SD) was significantly associated with lower LDL-C concentration; b) the association between baseline meglutol and LDL-C concentration was stronger among individuals taking LDL-lowering medication (β [SE]: −7.6 (1.7) mg/dl, P < 0.0001) compared to those not taking LDL-C lowering medication (β [SE]: −3.7(1.7) mg/dl, P < 0.0001) (Pint = 0.002). c) among BPRHS participants only, baseline meglutol concentration was significantly associated with LDL-C among those with T2D (β [SE]: −10.4 (2.0) mg/dl, P < 0.0001), but not among those without T2D (β [SE]: −3.0 (2.3) mg/dl, P = 0.19) (Pint < 0.0001). d) an interaction by BMI approached significance (obese [BMI ≥ 30 kg/m2] vs. overweight/normal weight [BMI < 30 kg/m2]; Pint = 0.07). Abbreviations: BPRHS, Boston Puerto Rican Health Study; CI, confidence interval; LDL-C, low-density lipoprotein cholesterol; SOALS, San Juan Overweight Adult Longitudinal Study; SD, standard deviation.

However, no significant associations were observed between meglutol and longitudinal changes in LDL-C (Figure S5, Figure S6, and Table S8). This was likely due to large variability in 2-year changes in LDL-C, a lack of long-term follow-up, and the moderate sample sizes in the BPRHS and SOALS cohorts. In one previous study examining 1,300 metabolites in Qatari adults, the association between meglutol and LDL-C did not survive their correction for multiple testing, but was suggestive (β = −0.10; P = 0.01) (Yousri et al., 2022). In contrast, the same study highlighted a positive association between meglutol and prevalent T2D that was consistently observed in two independent population-based studies. In secondary analyses, we observed that baseline meglutol concentration (per 1-SD) was significantly associated with higher BMI (β [SE]: 0.3 [0.2] kg/m2, P = 0.04) in fully adjusted models, and no associations were observed between meglutol and other cardiometabolic risk factors. In previous studies among BPRHS and SOALS participants, we also observed positive associations between meglutol and prevalent T2D (Haslam et al., 2021), but not incident T2D (Rivas-Tumanyan et al., 2022). Another study observed a positive association between plasma meglutol and coronary artery calcium score among 236 European Americans (Chevli et al., 2021). These conflicting associations between meglutol and cardiometabolic risk factors suggest that further research is needed to understand the determinants of plasma meglutol and the role it may play in influencing cardiometabolic disease risk.

We observed statistically significant effect modification by both lipid-lowering medication use and T2D status for the cross-sectional association between meglutol and LDL-C concentration (Table S9). Among individuals taking LDL-lowering medication, the association between baseline meglutol and LDL-C was stronger (β [SE]: −7.6 [1.7] mg/dl, P < 0.0001) compared to those not taking LDL-C lowering medication (β [SE]: - 3.7 [1.7] mg/dl, P < 0.0001) (Pint = 0.002) (Figure 5b). In the BPRHS, baseline meglutol was significantly associated with LDL-C concentration among those with T2D (β [SE]: −10.4 [2.0] mg/dl, P < 0.0001), but not among those without T2D (β [SE]: −3.0 [2.3], P = 0.19) (Pint < 0.0001) (Figure 5c). The association between baseline meglutol and LDL-C was statistically significant only among individuals with obesity (BMI ≥ 30 kg/m2, β [SE]: −6.2 [1.1] mg/dl, P < 0.0001) and not among those without obesity (BMI< 30 kg/m2, β [SE]: −3.6 [2.5] mg/dl, P = 0.15), but the test for interaction was not statistically significant (Pint= 0.07) (Figure 5d). No statistically significant effect modificaiton by lipid-lowering medication use, T2D status, or BMI was observed in lognitudinal models (Figure S5 and Table S10).

As a whole, participants taking lipid-lowering medication or with T2D or obesity represent a less metabolically healthy group. Thus, among individuals who were less metabolically healthy, the difference in LDL-C by plasma meglutol concentration was greater than among those who were more metabolically healthy. In line with this finding, we also observed a modest positive association between meglutol and BMI in the BPRHS and SOALS cohorts. In this epidemiological approach, we cannot determine the cause of these associations, but we can hypothesize about what may be driving these observed associations. It is possible that upregulated inflammatory pathways may lead to endogenous upregulation of plasma meglutol production as a negative feedback loop to counteract the continual upregulation of cholesterol synthesis via inflammation (Lupien et al., 1979). Another potential explanation is that plasma meglutol concentration is upregulated among individuals taking LDL-C lowering medication, due to competitive binding of meglutol and drug substrates to enzyme active sites. Finally, it is possible that meglutol has additional unknown detrimental effects on metabolic health that may outweigh its potential LDL-C lowering effects. Only controlled metabolic experiments, including in vitro, animal, and human studies, can provide causal insights into the complete metabolism and regulation of plasma meglutol.

The major strengths of the epidemiological design include the use of two complementary population-based epidemiological studies conducted in ethnic minority populations, the collection of detailed data on medical history, plasma cardiometabolic risk factors, and lifestyle factors to incorporate into multivariable regression models, and the availability of plasma meglutol concentration, a metabolite that has only recently been included in high-throughput plasma metabolomic profiling platforms. The possibility of residual confounding in these observational analyses is a limitation to our ability to infer causality. The modest sample size and follow-up in the longitudinal analyses limits our conclusions to focus primarily on the cross-sectional findings from the baseline exams, which exacerbates concern about potential residual confounding. However, the detailed data collection and adjustment for potential confounders minimizes the risk of residual confounding as much as possible. Larger studies examining the complex relationship between dietary intakes, plasma meglutol, and cardiovascular disease events during longitudinal follow-up will be necessary to further understand potential confounders in this relationship and the role of plasma meglutol concentrations on long-term health. The focus on individuals of Puerto Rican descent limits the generalizability of our findings, suggesting that future studies in other racial/ethnic groups are warranted to validate these findings.

4. Conclusion

Herein, we describe a multidisciplinary strategy to investigate the bioactive properties of foods by leveraging the integration of food metabolome data, metabolite pharmacological data, and human metabolome data from nutritional epidemiological studies. While extracting meaningful information from the vast amount of data generated through omics-based analyses often poses a significant challenge, the multidisciplinary strategy presented here enabled us to systematically survey hundreds of food metabolites to identify meglutol and five other bioactive metabolites for further study.

Using the food metabolome dataset and food engineering techniques, we were able to identify that tempe - green pea tempe in particular - was an excellent source of meglutol. Furthermore, we observed that plasma meglutol concentrations in adults were associated with lower LDL-C concentration among both BPRHS and SOALS participants at their baseline visit, particularly among those with T2D, with obesity, or taking LDL-lowering medication.

Through this study, we illustrate how adopting a multidisciplinary and complementary strategy involving food metabolomics and nutritional epidemiology brought to light meglutol as a potential target for further exploration of its impact on cardiometabolic health, along with tempe as a potential meglutol-rich functional food that could lead to food-based treatment of elevated LDL-C concentration. In the future, we can capitalize on the multifaceted strategy presented here to unravel hidden health benefits in various foods, especially unique and underexplored food commodities. The use of food metabolomics to exhaustively identify metabolite components of food commodities coupled with metabolite pharmacological data and nutritional epidemiology to confirm metabolites with significant health associations is a powerful tool for extracting information on health benefits in foods.

Supplementary Material

Supplement

Acknowledgement

The authors thank all BPRHS and SOALS participants and staff for their contribution to these studies. This study represents a portion of a dissertation submitted by M.N.I. to Osaka University in partial fulfillment of requirements for his PhD study.

Funding

This work is supported by The Japan International Cooperation Agency (JICA) (M.N.I), JSPS KAKENHI JPA20H025410 (S.P.P.), National Institutes of Health (NIH), 2T32CA009001 (D.E.H.), 1K01DK136968 (D.E.H.), 1K01DK107804-01A1 (S.N.B.), 1R01DK120560-01 (D.E.H., L.L., K.L.T., K.J., and S.N.B.), R01HL123915 (J.L-S), R01HL155742 (J.L-S), and R01DK125273 (J.L-S, S.N.B).

Guarantor Statement

M.N.I., D.E.H., S.S.P., S.N.B., and J.L-S. are guarantors of the work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and accuracy of the data analysis.

Data Availability Statement

Information on requesting data from the BPRHS (https://www.uml.edu/Research/UML-CPH/Research/bprhs/) and SOALS (http://soals.rcm.upr.edu/) can be found on their respective websites.

References

  1. Aditiawati P, Astuti DI, Kriswantoro JA, Khanza SM, Kamarisima, Irifune T, … Putri SP (2020). GC/MS-based metabolic profiling for the evaluation of solid state fermentation to improve quality of Arabica coffee beans. Metabolomics, 16, 1–14. [DOI] [PubMed] [Google Scholar]
  2. Afifah DN, Nabilah N, Supraba GT, Pratiwi SN, Nuryanto N, & Sulchan M (2020). The effects of tempeh gembus, an Indonesian fermented food, on lipid profiles in women with hyperlipidemia.
  3. Amalia F, Aditiawati P, Yusianto, Putri SP, & Fukusaki E (2021). Gas chromatography/mass spectrometry-based metabolite profiling of coffee beans obtained from different altitudes and origins with various postharvest processing. Metabolomics, 17, 1–16. [DOI] [PubMed] [Google Scholar]
  4. Andriankaja OM, Jiménez JJ, Muñoz‐Torres FJ, Pérez CM, Vergara JL, & Joshipura KJ (2015). Lipid‐lowering agents use and systemic and oral inflammation in overweight or obese adult Puerto Ricans: the San Juan overweight adults longitudinal study (SOALS). Journal of clinical periodontology, 42(12), 1090–1096. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Barabási A-L, Menichetti G, & Loscalzo J (2020). The unmapped chemical complexity of our diet. Nature Food, 1(1), 33–37. [Google Scholar]
  6. Bhupathiraju SN, Lichtenstein AH, Dawson-Hughes B, & Tucker KL (2011). Adherence index based on the AHA 2006 diet and lifestyle recommendations is associated with select cardiovascular disease risk factors in older Puerto Ricans. The Journal of nutrition, 141(3), 460–469. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Birch CS, & Bonwick GA (2019). Ensuring the future of functional foods. International Journal of Food Science & Technology, 54(5), 1467–1485. [Google Scholar]
  8. Caesar LK, Kellogg JJ, Kvalheim OM, & Cech NB (2019). Opportunities and limitations for untargeted mass spectrometry metabolomics to identify biologically active constituents in complex natural product mixtures. Journal of Natural Products, 82(3), 469–484. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Chevli PA, Freedman BI, Hsu F-C, Xu J, Rudock ME, Ma L, … Shapiro MD (2021). Plasma metabolomic profiling in subclinical atherosclerosis: the Diabetes Heart Study. Cardiovascular diabetology, 20(1), 1–12. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Cohen S, Kamarck T, & Mermelstein R (1983). A global measure of perceived stress. Journal of health and social behavior, 385–396. [PubMed] [Google Scholar]
  11. Dahlan HA, Nambu Y, Putri SP, & Fukusaki E (2022). Effects of Soaking Tempe in Vinegar on Metabolome and Sensory Profiles. Metabolites, 12(1), 30. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Dupont S, Lemetais G, Ferreira T, Cayot P, Gervais P, & Beney L (2012). ERGOSTEROL BIOSYNTHESIS: A FUNGAL PATHWAY FOR LIFE ON LAND? Evolution, 66(9), 2961–2968. 10.1111/j.1558-5646.2012.01667.x. [DOI] [PubMed] [Google Scholar]
  13. Farag MA, Khattab AR, Ehrlich A, Kropf M, Heiss AG, & Wessjohann LA (2018). Gas chromatography/mass spectrometry-based metabolite profiling of nutrients and antinutrients in eight lens and lupinus seeds (Fabaceae). Journal of Agricultural and Food Chemistry, 66(16), 4267–4280. [DOI] [PubMed] [Google Scholar]
  14. Faudone G, Arifi S, & Merk D (2021). The medicinal chemistry of caffeine. Journal of Medicinal Chemistry, 64(11), 7156–7178. [DOI] [PubMed] [Google Scholar]
  15. Hanifah A, Firmanto H, Putri SP, & Fukusaki E (2022). Unique metabolite profiles of Indonesian cocoa beans from different origins and their correlation with temperature. Journal of Bioscience and Bioengineering, 134(2), 125–132. [DOI] [PubMed] [Google Scholar]
  16. Haslam DE, Liang L, Wang DD, Kelly RS, Wittenbecher C, Pérez CM, … Wong DT (2021). Associations of network-derived metabolite clusters with prevalent type 2 diabetes among adults of Puerto Rican descent. BMJ Open Diabetes Research and Care, 9(1), e002298. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Ikram M, Ridwani S, Putri S, & Fukusaki E (2020). GC-MS based metabolite profiling to monitor ripening-specific metabolites in pineapple (Ananas comosus). Metabolites, 10 (4), Article 134. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Ikram MMM, Mizuno R, Putri SP, & Fukusaki E (2021). Comparative metabolomics and sensory evaluation of pineapple (Ananas comosus) reveal the importance of ripening stage compared to cultivar. Journal of Bioscience and Bioengineering, 132(6), 592–598. [DOI] [PubMed] [Google Scholar]
  19. Iwatani S, & Yamamoto N (2019). Functional food products in Japan: A review. Food Science and Human Wellness, 8(2), 96–101. [Google Scholar]
  20. Joshipura KJ, Muñoz-Torres FJ, Dye BA, Leroux BG, Ramírez-Vick M, & Pérez CM (2018). Longitudinal association between periodontitis and development of diabetes. Diabetes research and clinical practice, 141, 284–293. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Jumhawan U, Putri SP, Bamba T, & Fukusaki E (2016). Quantification of coffee blends for authentication of Asian palm civet coffee (Kopi Luwak) via metabolomics: A proof of concept. Journal of Bioscience and Bioengineering, 122(1), 79–84. [DOI] [PubMed] [Google Scholar]
  22. Kadar AD, Aditiawati P, Astawan M, Putri SP, & Fukusaki E (2018). Gas chromatography coupled with mass spectrometry-based metabolomics for the classification of tempe from different regions and production processes in Indonesia. Journal of Bioscience and Bioengineering, 126(3), 411–416. [DOI] [PubMed] [Google Scholar]
  23. Kelly RS, Mendez KM, Huang M, Hobbs BD, Clish CB, Gerszten R, … Chu SH (2022). Metabo-Endotypes of Asthma Reveal Differences in Lung Function: Discovery and Validation in Two TOPMed Cohorts. American Journal of Respiratory and Critical Care Medicine, 205(3), 288–299. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Kitani Y, Putri SP, & Fukusaki E (2022). Investigation of the effect of processing on the component changes of single-origin chocolate during the bean-to-bar process. Journal of Bioscience and Bioengineering, 134(2), 138–143. [DOI] [PubMed] [Google Scholar]
  25. Klimek-Szczykutowicz M, Szopa A, & Ekiert H (2020). Citrus limon (Lemon) phenomenon—a review of the chemistry, pharmacological properties, applications in the modern pharmaceutical, food, and cosmetics industries, and biotechnological studies. Plants, 9(1), 119. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Lai Z, Tsugawa H, Wohlgemuth G, Mehta S, Mueller M, Zheng Y, … Fiehn O (2018). Identifying metabolites by integrating metabolome databases with mass spectrometry cheminformatics. Nature Methods, 15(1), 53–56. 10.1038/nmeth.4512. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Lelario F, Scrano L, De Franchi S, Bonomo M, Salzano G, Milan S, … Bufo S (2018). Identification and antimicrobial activity of most representative secondary metabolites from different plant species. Chemical and Biological Technologies in Agriculture, 5(1), 1–12. [Google Scholar]
  28. Litonjua AA, Lange NE, Carey VJ, Brown S, Laranjo N, Harshfield BJ, … Bacharier LB (2014). The Vitamin D Antenatal Asthma Reduction Trial (VDAART): rationale, design, and methods of a randomized, controlled trial of vitamin D supplementation in pregnancy for the primary prevention of asthma and allergies in children. Contemporary clinical trials, 38(1), 37–50. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Lupien P-J, Moorjani S, Brun D, & Bielmann P (1979). Effects of 3-Hydroxy-3-methylglutaric Acid on Plasma and Low-Density Lipoprotein Cholesterol Levels in Familial Hypercholesterolemia. The Journal of Clinical Pharmacology, 19(2–3), 120–126. 10.1002/j.1552-4604.1979.tb02469.x. [DOI] [PubMed] [Google Scholar]
  30. Marin G, & Gamba RJ (1996). A new measurement of acculturation for Hispanics: The Bidimensional Acculturation Scale for Hispanics (BAS). Hispanic Journal of Behavioral Sciences, 18(3), 297–316. [Google Scholar]
  31. Megson IL, Whitfield PD, & Zabetakis I (2016). Lipids and cardiovascular disease: where does dietary intervention sit alongside statin therapy? Food & Function, 7(6), 2603–2614. 10.1039/c6fo00024j. [DOI] [PubMed] [Google Scholar]
  32. Ohto Y, Putri SP, Suantika G, & Fukusaki E (2021). Investigation of the characteristics of different shrimps by species and habitat using gas chromatography/mass spectrometry based metabolomics. Journal of Bioscience and Bioengineering, 132(3), 258–264. [DOI] [PubMed] [Google Scholar]
  33. Pang Z, Chong J, Zhou G, David A, Chang L, Barrette M, … Xia J (2021). MetaboAnalyst 5.0: narrowing the gap between raw spectra and functional insights. Nucleic Acids Research, 49(W1), W388–W396. 10.1093/nar/gkab382. [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Parijadi AA, Putri SP, Ridwani S, Dwivany FM, & Fukusaki E (2018). Metabolic profiling of Garcinia mangostana (mangosteen) based on ripening stages. Journal of Bioscience and Bioengineering, 125(2), 238–244. [DOI] [PubMed] [Google Scholar]
  35. Parijadi AAR, Yamamoto K, Ikram MMM, Dwivany FM, Wikantika K, Putri SP, & Fukusaki E (2022). Metabolome analysis of banana (Musa acuminata) treated with chitosan coating and low temperature reveals different mechanisms modulating delayed ripening. Frontiers in Sustainable Food Systems, 6, 835978. [Google Scholar]
  36. Pérez CM, Muñoz F, Andriankaja OM, Ritchie CS, Martínez S, Vergara J, … Joshipura KJ (2017). Cross‐sectional associations of impaired glucose metabolism measures with bleeding on probing and periodontitis. Journal of clinical periodontology, 44(2), 142–149. [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Putri SLE, Suantika G, Situmorang ML, Christina J, Nikijuluw C, Putri SP, & Fukusaki E (2021). Shrimp count size: GC/MS-based metabolomics approach and quantitative descriptive analysis (QDA) reveal the importance of size in white leg shrimp (Litopenaeus vannamei). Metabolomics, 17, 1–12. [DOI] [PubMed] [Google Scholar]
  38. Putri SLE, Suantika G, Situmorang ML, Putri SP, & Fukusaki E (2022). Metabolomics approach to elucidate the importance of count size in commercial penaeid shrimps: white leg shrimp (Litopenaeus vannamei) and black tiger shrimp (Penaeus monodon). Journal of Bioscience and Bioengineering, 133(5), 459–466. [DOI] [PubMed] [Google Scholar]
  39. Putri SP, Irifune T, Yusianto, & Fukusaki E (2019). GC/MS based metabolite profiling of Indonesian specialty coffee from different species and geographical origin. Metabolomics, 15, 1–11. [DOI] [PubMed] [Google Scholar]
  40. Putri SP, Jumhawan U, & Fukusaki E (2015). Application of GC/MS and GC/FID-based metabolomics for authentication of Asian palm civet coffee (Kopi Luwak). Journal of Bioscience and Bioengineering, 120(5), 33–41. [DOI] [PubMed] [Google Scholar]
  41. Rahmawati D, Astawan M, Putri SP, & Fukusaki E (2021). Gas chromatography-mass spectrometry-based metabolite profiling and sensory profile of Indonesian fermented food (tempe) from various legumes. Journal of Bioscience and Bioengineering, 132(5), 487–495. [DOI] [PubMed] [Google Scholar]
  42. Rinschen MM, Ivanisevic J, Giera M, & Siuzdak G (2019). Identification of bioactive metabolites using activity metabolomics. Nature reviews Molecular cell biology, 20(6), 353–367. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Rivas-Tumanyan S, Pacheco LS, Haslam DE, Liang L, Tucker KL, Joshipura KJ, & Bhupathiraju SN (2022). Novel Plasma Metabolomic Markers Associated with Diabetes Progression in Older Puerto Ricans. Metabolites, 12(6), 513. [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. Rodríguez JM, Ruíz-Sala P, Ugarte M, & Peñalva MÁ (2004). Fungal Metabolic Model for 3-Methylcrotonyl-CoA Carboxylase Deficiency. Journal of Biological Chemistry, 279(6), 4578–4587. 10.1074/jbc.m310055200. [DOI] [PubMed] [Google Scholar]
  45. Sánchez-Ruiz A, & Colmenarejo G (2022). Systematic Analysis and Prediction of the Target Space of Bioactive Food Compounds: Filling the Chemobiological Gaps. Journal of Chemical Information and Modeling, 62(16), 3734–3751. [DOI] [PubMed] [Google Scholar]
  46. Sato A, Astuti DI, Putri SP, & Fukusaki E (2020). Quality improvement of semi-wet terasi by optimizing the starter culture ratio of controlled fermentation. HAYATI Journal of Biosciences, 27(4), 320–320. [Google Scholar]
  47. Sato A, Putri SP, Astuti DI, & Fukusaki E (2022). Metabolome analysis to investigate the effect of controlled fermentation on taste-related metabolites in terasi. Metabolomics, 18(7), 44. [DOI] [PubMed] [Google Scholar]
  48. Sato M, Ikram MMM, Pranamuda H, Agusta W, Putri SP, & Fukusaki E (2021). Characterization of five Indonesian mangoes using gas chromatography–mass spectrometry-based metabolic profiling and sensory evaluation. Journal of Bioscience and Bioengineering, 132(6), 613–620. [DOI] [PubMed] [Google Scholar]
  49. Staszak M, Staszak K, Wieszczycka K, Bajek A, Roszkowski K, & Tylkowski B (2022). Machine learning in drug design: Use of artificial intelligence to explore the chemical structure–biological activity relationship. Wiley Interdisciplinary Reviews: Computational Molecular Science, 12(2), e1568. [Google Scholar]
  50. Tadesse SA, & Emire SA (2020). Production and processing of antioxidant bioactive peptides: A driving force for the functional food market. Heliyon, 6(8), e04765. [DOI] [PMC free article] [PubMed] [Google Scholar]
  51. Team, R. C.(2013). R: A language and environment for statistical computing.
  52. Thiombiano B, Gontier E, Molinie R, Marcelo P, Mesnard F, & Dauwe R (2020). An untargeted liquid chromatography–mass spectrometry‐based workflow for the structural characterization of plant polyesters. The Plant Journal, 102(6), 1323–1339. [DOI] [PubMed] [Google Scholar]
  53. Tsugawa H, Cajka T, Kind T, Ma Y, Higgins B, Ikeda K, … Arita M (2015). MS-DIAL: data-independent MS/MS deconvolution for comprehensive metabolome analysis. Nature Methods, 12(6), 523–526. [DOI] [PMC free article] [PubMed] [Google Scholar]
  54. Tsugawa H, Tsujimoto Y, Arita M, Bamba T, & Fukusaki E (2011). GC/MS based metabolomics: development of a data mining system for metabolite identification by using soft independent modeling of class analogy (SIMCA). BMC bioinformatics, 12(1), 1–13. [DOI] [PMC free article] [PubMed] [Google Scholar]
  55. Tucker KL, Bianchi LA, Maras J, & Bermudez OI (1998). Adaptation of a food frequency questionnaire to assess diets of Puerto Rican and non-Hispanic adults. American journal of epidemiology, 148(5), 507–518. [DOI] [PubMed] [Google Scholar]
  56. Tucker KL, Mattei J, Noel SE, Collado BM, Mendez J, Nelson J, … Falcon LM (2010). The Boston Puerto Rican Health Study, a longitudinal cohort study on health disparities in Puerto Rican adults: challenges and opportunities. BMC Public health, 10(1), 1–12. [DOI] [PMC free article] [PubMed] [Google Scholar]
  57. Xu Z, Chen X, Lu X, Zhao B, Yang Y, & Liu J (2021). Integrative analysis of transcriptome and metabolome reveal mechanism of tolerance to salt stress in oat (Avena sativa L.). Plant Physiology and Biochemistry, 160, 315–328. [DOI] [PubMed] [Google Scholar]
  58. Yamana T, Taniguchi M, Nakahara T, Ito Y, Okochi N, Putri SP, & Fukusaki E (2020). Component profiling of soy-sauce-like seasoning produced from different raw materials. Metabolites, 10(4), 137. [DOI] [PMC free article] [PubMed] [Google Scholar]
  59. Yousri NA, Suhre K, Yassin E, Al-Shakaki A, Robay A, Elshafei M, … Fakhro KA (2022). Metabolic and Metabo-Clinical Signatures of Type 2 Diabetes, Obesity, Retinopathy, and Dyslipidemia. Diabetes, 71(2), 184–205. [DOI] [PMC free article] [PubMed] [Google Scholar]
  60. Zhu H (2020). Big data and artificial intelligence modeling for drug discovery. Annual review of pharmacology and toxicology, 60, 573–589. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplement
NIHMS2005009-supplement-Supplement.docx (808.7KB, docx)

Data Availability Statement

Information on requesting data from the BPRHS (https://www.uml.edu/Research/UML-CPH/Research/bprhs/) and SOALS (http://soals.rcm.upr.edu/) can be found on their respective websites.

RESOURCES