Article Cite This: J. Proteome Res. XXXX, XXX, XXX−XXX
pubs.acs.org/jpr
Vegetable Signatures Derived from Human Urinary Metabolomic Data in Controlled Feeding Studies Ke-Shiuan Lynn,† Mei-Ling Cheng,‡,§,⊥ Hsin-Chou Yang,∥ Yu-Jen Liang,∥ Mei-Jyh Kang,∇ Fong-Ling Chen,∇ Ming-Shi Shiao,‡ and Wen-Harn Pan*,∇,⊗ †
Department of Mathematics, Fu Jen Catholic University, New Taipei City 24205, Taiwan Department of Biomedical Sciences, College of Medicine, Chang Gung University, Taoyuan 33302, Taiwan § Metabolomics Core Laboratory, Healthy Aging Research Center, Chang Gung University, Taoyuan 33302, Taiwan ⊥ Clinical Metabolomics Core Laboratory, Chang Gung Memorial Hospital, Taoyuan 33305, Taiwan ∥ Institute of Statistical Science, Academia Sinica, Taipei 11529, Taiwan ∇ Institute of Biomedical Sciences, Academia Sinica, Taipei 11529, Taiwan ⊗ Institute of Population Health Sciences, National Health Research Institutes, Miaoli 35053, Taiwan
Downloaded via UNIV OF WINNIPEG on December 21, 2018 at 08:52:51 (UTC). See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.
‡
S Supporting Information *
ABSTRACT: Examination of changes in urinary metabolomic profiles after vegetable ingestion may lead to new methods of assessing plant food intake. To this regard, we developed a proofof-principle methodology to identify urinary metabolomic signatures for spinach, celery, and onion. Three feeding studies were conducted. In the first study, healthy individuals were fed with spinach, celery, onion, and no vegetables in four separate experiments with pooled urinary samples for metabolite discovery. The same protocol was used to validate the finding at the individual level in the second study and when feeding all three vegetables simultaneously in the third study. An LC−MSbased metabolomics approach was adopted to search for indicative metabolites from urine samples collected during multiple time periods before and after the meal. Consequently, a total of 1, 9, and 3 nonoverlapping urinary metabolites were associated with the intake of spinach, celery, and onion, respectively. The PCA signature of these metabolites followed a similar “time cycle” pattern, which maximized at approximately 2−4 h after intake. In addition, the metabolite profiles for the same vegetable were consistent across samples, regardless of whether it was consumed individually or in combination. The developed methodology along with the identified urinary metabolomic signatures were potential tools for assessing plant food intake. KEYWORDS: vegetable metabolomic signature, human urinary metabolite, liquid chromatography mass spectrometry
■
INTRODUCTION Plant-based foods are an indispensable part of healthy diets. Sufficient intake of these foods not only helps maintain normal bodily functions by providing essential nutrients, but also reduces the incidence of noncommunicable disease1−3 in part through the antioxidative and anti-inflammatory effects of a large amount of phytochemicals. Studies investigating the effect of phytochemicals have usually been limited to a single category of substances from a single plant food. Research has noted that many phytonutrients and dietary components likely act synergistically,1 and yet the health effects and mechanisms of phytonutrients in combination have not been carefully studied in populations. A bottleneck of such population-based epidemiological study is the inability of participants to recall a large variety of vegetables and fruits consumed. Therefore, we propose that exploring the change of metabolites in biospecimens after feeding may not only assist with understanding the © XXXX American Chemical Society
role of plant foods but also lead to advances in technology for assessing plant food intake. Current metabolomic technology enables the simultaneous measurement of thousands of small molecules in biological samples. Examining these metabolite profiles aids in the understanding of physiological states that arise under a variety of conditions or challenges. Each plant food potentially has its own unique metabolomic signature. These distinct patterns may be of use not only in the selection of target molecules with bioactivity but also as a means to assist in quantifying dietary intake.4−7 Controlled dietary interventions using high resolution spectroscopy quantitation (such as nuclear magnetic resonance or mass spectrometry (MS)) and multivariate statistical Received: June 19, 2018 Published: December 5, 2018 A
DOI: 10.1021/acs.jproteome.8b00470 J. Proteome Res. XXXX, XXX, XXX−XXX
Article
Journal of Proteome Research
Figure 1. Study design for the first exploratory study. In this study, a total of four experiments were conducted in which spinach, celery, onion, and no vegetables were used as the test vegetable, respectively. Each experiment consists of a pretest day and a test day followed by a 5-day washout period. The test vegetable was provided in a lunch, and five urine samples were collected during a certain time period before and after the meal on the test day.
provided with fried rice in a lunch. Urine samples were collected from the participants in the fasting state before breakfast, immediately before lunch, and 0−2, 2−4, and 4−7 h after lunch. The urine samples were pooled together at each time point. The experiment was carried out on Sunday, followed by 5 days of regular diet without the next experimental vegetable, and 1 day of a low-phytonutrient diet prior to the next experiment, as shown in Figure 1. The second confirmatory study tested whether the metabolomic signatures found in the first study’s pooled samples could be detected in the individual samples of the second study. The same protocol was followed; the same vegetables were given to three subjects, with a single vegetable provided each time. Urine samples were collected before lunch and 2−4 h after lunch, but were not pooled. For the third confirmatory study, the three vegetables (200 g in total) were prepared into a mixed dish and fed to three subjects in one experiment to determine whether all three vegetable metabolomic signatures could be detected. Individual urine samples were collected at time points identical to those of the second study.
analysis have successfully discovered biomarkers of the intake of certain foods, including cruciferous vegetable,8 citrus fruit,9,10 red meat,11,12 fish,13 coffee,14 and tea.15 However, without thorough validation, it is questionable whether the association between the discovered biomarkers and the food of interest is unique. Moreover, some biomarkers, although differentially measured in various dietary groups, may already exist in the human body; thus, their existence cannot directly reflect food intake.16,17 In the present study, we carried out three proof-of-principle short-term plant food feeding studies and applied a metabolomics methodology to identify the signatures of three vegetables for future consideration in appraising intake levels.
■
MATERIALS AND METHODS
Study Design
We designed three short-term feeding studies to identify the urinary metabolomic signatures of three vegetables: spinach, celery, and onion. The first was an exploratory study in which one vegetable at a time was fed to a group of 10 people and urine samples were collected. Metabolomic analyses were conducted on the pooled urinary samples. The second was a confirmatory study in which one vegetable at a time was fed to three individuals and individual urine samples were collected. Urinary metabolomic profiling was performed on each sample. The third was a further confirmatory study at an individual level in which three vegetables were fed together in a meal. Protocols, questionnaires, and informed consent forms were approved by the Institutional Review Board of the National Health Research Institutes in Miaoli County, Taiwan (EC0991003) and by that of the Tri-Service General Hospital in Taipei, Taiwan (099-05-211). All patients provided written, informed consent for their participation. Protocols were also registered at clinicaltrial.gov as NCT03483558. For the first exploratory study, participants were fed one of three selected vegetables for each experiment. A total of four experiments were conducted, including one control experiment without vegetables. On the day before the experiment, participants were asked to consume only the low-phytonutrient diet provided by the study, which comprised breakfast, lunch, and dinner. No vegetables or fruits were provided. Plain water was allowed but the consumption of other food or drink was not. On the day of the experiment, 200 g of a vegetable was
Participant Recruitment
We recruited 13 participants (10 for the first study and 3 for the second and third) using an online recruitment notice. According to MetSizeR,18 the sample size is adequate to control the false discovery rate (FDR) at 5% with 10% metabolites being significant. The experiments were conducted at the Tri-Service General Hospital. Inclusion criteria were as follows: (i) aged 18−60 years; (ii) body mass index (BMI) between 18.5 and 30 kg/m2; (iii) nonsmoker; and (iv) no alcohol abuse. Exclusion criteria were as follows: (i) those taking medication for hypertension or diabetes; (ii) patients with immune diseases (allergic or autoimmune), liver disease, metabolic disease (hyperthyroidism or hypothyroidism), and other fatal diseases such as cancer; (iii) those unwilling to cease taking dietary supplements; and (iv) patients with urinary tract infections or those who had taken antibiotics within 3 weeks of the study. The participant flowchart is provided in Text S1. When they reported to the hospital, the participants were asked whether any food, drink, or medication was consumed in the pretest day other than the provided standard diet and water. None have reported any. In addition, all participants B
DOI: 10.1021/acs.jproteome.8b00470 J. Proteome Res. XXXX, XXX, XXX−XXX
Article
Journal of Proteome Research Ultraperformance Liquid Chromatography−Mass Spectrometry
were required to start fasting from 10:00 PM of the pretest day and to report to the hospital’s clinical trial center and stay there during the test day, where only the designated meal and water were allowed. Therefore, the effects of alcohol and medication to this study, if any, should be minimal.
The samples from the exploratory study and two confirmatory studies were analyzed in the metabolomics core laboratory at Chang Gung University. The LC−MS analysis was performed using ultraperformance liquid chromatography (UPLC) (Waters ACQUITY UPLC System, Milford, MA, USA) with an electrospray ionization source coupled to a quadrupole time-of-flight mass spectrometer (Q-TOF MS) (Waters Synapt G1 HDMS System) in both positive and negative modes. The mobile phase was composed of the following two solutions: (A) 2 mM ammonium formate and 0.1% formic acid in water and (B) 0.1% formic acid in acetonitrile. Chromatographic separation was performed using a BEH C18 column (1.7 μm, 2.1 × 100 mm) at 45 °C and with gradient elution (1% B increasing from 1% to 48% over 2.5 min, followed by an increase from 48% to 98% B over 2.5−3.0 min and maintained at 98% over 3.0−4.2 min; the mobile phase was then returned to 1% B at 4.3 min for 1.7 min re-equilibration) occurring directly into the mass spectrometer at a flow rate of 500 μL/ min. We first performed LC−MS analysis on samples from the exploratory study. The samples from the two confirmatory studies were analyzed in another batch. During each batch, six replicates were generated for each sample to estimate the reliability of the measurements.
Vegetables and the Low-Phytochemical Diet
Studies have indicated that the human metabolism is influenced by numerous factors, including genes, age, sex, health status, diet, and physical activity.19−21 The provision of a standardized diet can normalize an individual’s metabolism and reduce interindividual variability.11,22,23 Accordingly, in our study design, we provided a low-phytochemical diet for 1 day prior to the experimental day to reduce the amount of interindividual variability at baseline. A low-phytochemical diet is characterized by a lack of vegetables, fruits, or their related products. On the basis of the weight, height, age, and sex of the participants at baseline, we calculated their daily caloric requirements and used the results to provide an appropriate amount of food to maintain the participants’ weight. The meals were categorized into four caloric levels: 1700, 1900, 2100, and 2400 kilocalories. For breakfast, we provided a steamed bun and milk; for lunch, we provided fried rice (containing rice, pork meatballs, fish cakes, and lean pork); and for dinner, we provided fried noodles (containing noodles, pork meatballs, fish cakes, and chicken breast). In terms of macronutrient proportions, the provided meals were 53% carbohydrate, 17% protein, and 30% fat. It has been recommended by the Taiwanese Food Guide that one should consume at least 3 servings of vegetables a day, and each serving is 100 g. Taiwanese people usually consume few vegetables in the morning, but one to two servings of vegetables at lunch and at dinner. We provided two servings of vegetables in our feeding study. This serving amount falls in the range of Taiwanese common practice.
Data Processing
We used Masslynx 4.1 (Waters) to generate peak quantitation results. Key parameters for the quantitation were as follows: mass tolerance = 0.03 Da; intensity threshold = 50 counts; mass window = 0.03 Da; RT window = 0.1 min; and noise elimination level = 6.0. An identical set of quantitation parameters was used for the two batches. We noted that a mass-to-charge ratio−retention time (m/z−RT) combination in a quantitation result may not always represent the precursor ion of a metabolite. In addition to precursor ions, derivative ions (including fragments, adducts, multiply charged ions, and multimers) are frequently seen in the quantitation results, mostly because of the ionization process.24−26 Consequently, we used the term “peak” to denote each m/z−RT combination in the quantitation results. The peaks in the two batches of quantitation results were matched using their m/z values (Δm/ z ≤ 0.03 Da), retention times (ΔRT ≤ 0.1 min), and mode (positive or negative). To ensure comparability among peaks in the statistical analyses, peak intensities were first converted to generalized log2, and then quantile-normalization was used to adjust the distribution and measurement scale. Subsequently, the six replicates for each peak were merged by taking their median value. To search for the vegetable signatures, we used the quantitation result of the exploratory study as the training data set. The results of the first confirmatory study (3 individuals × 4 vegetable groups × 2 collection time points) were used as the validation data set, and those of the second confirmatory study (3 individuals × 1 mixed vegetable group × 2 collection time points) was used as the test data set.
Sample Collection
Participants reported to the clinical research center at 7:30 am and were asked to hand in their first morning urine sample. They were subsequently provided with breakfast, which they consumed by 9:00 am. Between 11:00 am and 11:30 am, prelunch urine samples were collected from the participants, after which they could begin lunch. They were expected to complete lunch within 30 min. Postlunch urine samples were collected at 0−2, 2−4, and 4−7 h. All urine samples were kept at −80 °C. A few participants in the exploratory study were incapable of attending all four experiments, and thus, the actual numbers of urine samples collected for the control, spinach, celery, and onion groups were 10, 9, 7, and 6, respectively. In both confirmatory studies, the three participants attended all of the experiments; thus, three urine samples were collected at each time point in these studies. Sample Preparation
The stored urine samples were thawed at 4 °C for 30 min and centrifuged at 12 000 rpm. We used a creatinine assay kit (Cayman Chemical, Ann Arbor, MI, USA) to measure the creatinine concentration in each sample. After converting the creatinine concentration to 100 μg/mL, we pooled the samples based on vegetable group and collection time. We then filtered the samples (pooled or individual) by using a syringe filter (polyvinylidene fluoride, 0.22 μm, Millpore-SLGV013NL, Billerica, MA, USA) to obtain the clear liquid upon which the analysis was carried out.
Significant MS Peaks for Each Vegetable Group
We expected to discover significant peaks for each vegetable group in the training data set. Because each peak in a vegetable group consisted of five measurements at five collection time points, to quantify the peak intensity change over time, we estimated the total change in peak intensities 0−7 h after lunch by calculating the net incremental area under the curve (net C
DOI: 10.1021/acs.jproteome.8b00470 J. Proteome Res. XXXX, XXX, XXX−XXX
Article
Journal of Proteome Research
Figure 2. Principal component analysis score plot of all urine metabolomes in the exploratory study (pooled samples). Data point color indicates vegetable group: orange for spinach, blue for celery, purple for onion, and green for control. Data point size, from small to large, indicates the time of sample collection: fasting, before lunch, 0−2 h postlunch, 2−4 h postlunch, and 4−7 h postlunch.
the vegetable, the minimum intensity of the peak must be at least five times greater than the maximum intensity in the second criterion. The feature peaks found in the training data set were further tested and filtered using the validation data set with the same selection criteria to improve their generalizability.
AUC) formed by the intensity time series. With the peak intensity from the prelunch urine specimens as the baseline of metabolic variation, the net AUC was computed by subtracting the area below the baseline from the area above the baseline. To search for significant peaks between a vegetable intake group and the control group, we first carried out two-way ANOVA to test the mean net AUC for each peak in the 6 sample replicates, and subsequently used the false discovery rate method to correct for the frequency of type I errors in multiple testing. Scheffe’s method was further applied for the post hoc comparisons between the vegetable and control experiments. In the search for significant peaks between groups, the significance was defined as p < 0.05; furthermore, the peaks with negative net AUC were not considered because they were counter to the purpose of this study, corresponding to metabolites that decrease in intensity during the dietary process.
Peak Group Refinement
As previously mentioned, many peaks in a quantitation result are derivative ions of the underlying metabolites. We grouped the peaks in the quantitation results on the basis of their retention times and intensities across samples for subsequent metabolite identification.27 Ideally, if a metabolite is related to the intake of a vegetable, all peaks in the associated peak group would be selected as feature peaks. However, in reality, some small peaks in the group may not be selected due to noise, coelution, or batch effects (peaks present in one experiment but not in the other). Such peaks were added to the feature peaks for metabolite identification and total intensity calculation. Conversely, in some peak groups, only a few small peaks were selected as feature peaks. Such peaks were also likely to be contaminated and passed the selection criteria by chance. Therefore, they were removed from the feature peaks.
Peaks Whose Presence Reflects a Vegetable Intake in Both Training and Validation Data
Peaks with significant differences between a vegetable group and the control group may not necessarily be associated with the metabolism of the vegetable. Instead, the peak may be related to multiple vegetables, other nutrition factors, or noise (e.g., contaminants) in the experiment. Furthermore, some peaks may correspond to metabolites that are already present in human urine and merely change intensity because of vegetable intake. Thus, the presence of such metabolites in urine may not reflect vegetable intake. We used the term “feature peaks” to denote peaks in a quantitation result that appear only after the consumption of the designated vegetables and searched for these peaks according to the following three criteria: (i) The net AUC of the peak must exhibit significant differences (adjusted p-value
