May 5, 2016 - compared with an Average Danish Diet (ADD) in blood plasma and reveals ... healthy Nordic food cuisine of high organoleptic quality was.
Article pubs.acs.org/jpr
New Nordic Diet versus Average Danish Diet: A Randomized Controlled Trial Revealed Healthy Long-Term Effects of the New Nordic Diet by GC−MS Blood Plasma Metabolomics Bekzod Khakimov,*,† Sanne Kellebjerg Poulsen,‡ Francesco Savorani,† Evrim Acar,† Gözde Gürdeniz,‡ Thomas M. Larsen,‡ Arne Astrup,‡ Lars O. Dragsted,‡ and Søren Balling Engelsen*,† †
Department of Food Science, Faculty of Science, University of Copenhagen, Rolighedsvej 26, DK-1958 Frederiksberg C, Denmark Department of Nutrition Exercise and Sports, Faculty of Science, University of Copenhagen, Rolighedsvej 26, DK-1958 Frederiksberg C, Denmark
‡
S Supporting Information *
ABSTRACT: A previous study has shown effects of the New Nordic Diet (NND) to stimulate weight loss and lower systolic and diastolic blood pressure in obese Danish women and men in a randomized, controlled dietary intervention study. This work demonstrates long-term metabolic effects of the NND as compared with an Average Danish Diet (ADD) in blood plasma and reveals associations between metabolic changes and health beneficial effects of the NND including weight loss. A total of 145 individuals completed the intervention and blood samples were taken along with clinical examinations before the intervention started (week 0) and after 12 and 26 weeks. The plasma metabolome was measured using GC−MS, and the final metabolite table contained 144 variables. Significant and novel metabolic effects of the diet, resulting weight loss, gender, and intervention study season were revealed using PLS-DA and ASCA. Several metabolites reflecting specific differences in the diets, especially intake of plant foods and seafood, and in energy metabolism related to ketone bodies and gluconeogenesis formed the predominant metabolite pattern discriminating the intervention groups. Among NND subjects, higher levels of vaccenic acid and 3-hydroxybutanoic acid were related to a higher weight loss, while higher concentrations of salicylic, lactic, and N-aspartic acids and 1,5-anhydro-D-sorbitol were related to a lower weight loss. Specific gender and seasonal differences were also observed. The study strongly indicates that healthy diets high in fish, vegetables, fruit, and whole grain facilitated weight loss and improved insulin sensitivity by increasing ketosis and gluconeogenesis in the fasting state. KEYWORDS: weight loss, health benefit, biomarker, diet, metabolomics, PARAFAC2, ASCA, PLS-DA
■
INTRODUCTION
information about the effects of ingestion of multiple food components as well as signatures of whole diets.5 Diet is an important factor that influences human health and life expectancy and may significantly vary according to culture as well as socio-economic factors.5d,6 In 2003, a sustainable, healthy Nordic food cuisine of high organoleptic quality was formulated in a Manifest.7 A few years later a project was launched in Denmark called OPUS (defined as Optimal wellbeing, development and health for Danish children through a healthy New Nordic Diet) and focused on investigating possibilities to develop a healthy New Nordic Diet (NND) based on regional foods that is attractive to the public and environmentally friendly.8 This trial was registered at www. clinicaltrials.gov as NCT01195610 (https://clinicaltrials.gov/ ct2/show/NCT01195610). The NND was developed based on
Blood plasma constitutes up to 55% of the human total blood volume and holds blood cells and lipids in suspension, playing a key role transporting metabolites and excretory products in the body. Plasma mostly contains water (up to 95%), proteins, carbohydrates, hormones, clotting factors, and several hundreds of small metabolites that dynamically change during metabolism.1 The levels of several plasma components such as sugars and hormones reflect human health and are often used for diagnostic purposes in medicine. Likewise, the relative levels of small blood plasma metabolites represent a metabolic status that is largely determined by health, diet, and other life style and individual factors. Recent advances in “omics” technology have opened new horizons using blood plasma for understanding, for example, glucose homeostasis2 or metabolic perturbations during diabetes3 and even forecasting chronic disease.4 Moreover, the plasma metabolome also contains rich © 2016 American Chemical Society
Received: February 4, 2016 Published: May 5, 2016 1939
DOI: 10.1021/acs.jproteome.6b00109 J. Proteome Res. 2016, 15, 1939−1954
Article
Journal of Proteome Research
Figure 1. Overview of the intervention study design. A total of 145 volunteers (45 male and 100 female) followed either New Nordic Diet (NND) or Average Danish Diet (ADD) over 26 weeks between October 2010 and July 2011. One week run-in time served as a standardization period of the participants before starting the intervention in Week 0 (Time 0). Clinical examinations and blood sample collection took place at three time points, week 0 (Time 0), week 12 (Time 1), and week 26 (Time 2). Samples were collected in the morning at least after 8 h of fasting from last meal. Yellow, gray, green, and blue color codes correspond to the number of individuals, who have been examined in autumn, winter, spring, and summer seaons, respectively.
■
existing scientific knowledge within health and nutrition9 and characterized by a high content of vegetables, fruits, whole grains, nuts, fish, and various seafood products.9 A randomized controlled dietary intervention study, SHOPUS (SHop in OPUS), was conducted within the OPUS project that included a comparison of the health effects of the NND and an Average Danish Diet (ADD) diet.10 The intervention study lasted for 6 months, where participants were given special foods for NND and ADD diets, free of charge, in a special shop organized by the University of Copenhagen. The study illustrated beneficial health effects of the NND in adults with increased waist circumference10 and a similar “Healthy Nordic Diet” decreased inflammatory gene expression in subcutaneous adipose tissue in individuals with features of the metabolic syndrome.11 An untargeted metabolomics study performed on urine samples from the SHOPUS study proved the presence of significant metabolic differences between NND and ADD diets and identified food related biomarkers that were used to estimate compliance to each dietary pattern.12 The present study investigates the use of blood plasma GC− MS metabolomics for discovering metabolic differences between individuals who followed NND and ADD diets during the 6 months of dietary intervention. In addition to diet, effects related to the season of the year when participants started and ended the intervention study (later called season), participants’ gender, and the weight loss within NND participants were also investigated. Blood plasma GC−MS metabolomics involved a newly developed methodology for derivatization of a broader spectrum of metabolites using trimethylsilyl cyanide (TMSCN)13 as well as an efficient multiway decomposition method, namely, PARAFAC2 for processing the GC−MS data.14 To the best of our knowledge, this is the first application of PARAFAC2 to process GC−MS metabolomics data from human blood plasma. Ultimately the extracted metabolite table was analyzed in context to the experimental design by employing ANOVA-Simultaneous Component Analysis (ASCA)15 and partial least squares discriminant analysis (PLS-DA).16
EXPERIMENTAL SECTION
Study Design
A 6-month nonblinded, parallel, randomized, controlled dietary intervention study was carried out to investigate the health effect of the NND compared with the ADD, as previously reported in detail.10 The intervention started with 1 week of run-in time when participants became familiar with the study shop, in which all foods were handed out free of charge throughout the intervention. During the last 3 days of the runin time all participants were provided with specific ADD foods in prespecific amounts. This ensured energy balance and served as a standardization period. The intervention study was carried out between October 2010 and July 2011, with 147 centrally obese Danish men and women. Clinical examinations and blood samples were collected at three time points, T0 (week 0, before starting intervention), T1 (after week 12), and T2 (after week 26) (Figure 1). Participants were randomly assigned to either the NND or the ADD diet in a 3:2 ratio, and 145 participants who completed the intervention (89 NND and 56 ADD) were included in this study. In terms of dietary intake the main difference between ADD and NND was macronutrient composition and intake of foods from 15 prespecified food groups identified as central for the NND.9 Detailed information on the percentage of consumers and average amount of consumption of the most explanatory foods for NND and ADD is published elsewhere.12 These prespecified food groups reflect a seasonal variation and, as a consequence, dietary intake within the NND differs during the four temperate seasons of the year. All participants in the current study started the intervention during the autumn or winter season; the majority of completed week 12 examinations (T1) during winter season and week 26 examinations (T2) were mostly conducted during spring, although some participants had their last examination in the winter or summer season. Hence, blood samples representing the same time point of intervention were collected during different seasons of the year and may have represented different foods as part of the NND. In contrast, the ADD diet was relatively insensitive to seasonal changes. Blood Plasma Metabolite Extraction
Ethylenediaminetetraacetic acid (EDTA) was used as an anticoagulant; blood plasma was separated from freshly collected samples and kept in −80 °C until analysis. Prior to 1940
DOI: 10.1021/acs.jproteome.6b00109 J. Proteome Res. 2016, 15, 1939−1954
Article
Journal of Proteome Research
manufacturer’s recommendation by using perfluorotributylamine (PFTBA). The MPS autosampler and GC−MS were controlled from ChemStation software (ver: E.02.02.1431, Agilent).
metabolite extraction, the plasma samples were thawed at room temperature and vigorously vortexed for 20 s. To obtain clear plasma, we removed insoluble particles by centrifugation at 16 000 g at room temperature for 3 min. A pooled control sample was obtained by mixing 50 μL of plasma from each sample. Plasma (60 μL) was transferred into 0.6 mL Eppendorf tubes, followed by the addition of 180 μL of ice cooled acetonitrile, immediately vortexed for 10 s, and further mixed at the frequency 23 Hz for 3 min. In order to remove proteins, the plasma extracts were centrifuged at 20 000 g for 10 min at 4 °C and 50 μL of clear supernatant was completely dried in 200 μL glass inserts using SpeedVac vacuum centrifugation (Labogene, Lynge, Denmark) at 30 °C for 2 h at 1500 rpm. Metabolite extractions were performed in batches of 20 randomly selected samples at a time.
Data Preprocessing
To extract relative concentrations of peaks detected by GC− MS, we processed the raw data by PARAFAC2, as previously described in Khakimov et al.14a Prior to the PARAFAC2 modeling, the raw GC−MS data were arranged in a three-way array as, elution time points (4205) × mass spectra (450) × samples (435), and divided into 116 smaller data intervals in elution time dimension, and each interval was modeled individually. This ensured the necessary reduced complexity of the data to facilitate easy and faster PARAFAC2 model development and validation. For each data interval, PARAFAC2 models with one to ten components, depending on the complexity of the data, were developed. Then, all of these models were evaluated to find a single PARAFAC2 model per data interval with an optimal number of components. Several parameters of PARAFAC2 models were taken into account for selecting an optimal number of components, including the explained variance, core consistency, residuals, and comparison of elution time and mass spectral profiles resolved by PARAFAC2 against the raw data. PARAFAC2 concentration profiles, which represent relative concentrations of detected peaks, were extracted from validated models and used to construct a final metabolite table. To minimize nonsample related variations the final metabolite table was normalized in two steps: (1) normalization based on the area of the IS and (2) experimental variation derived from any GC−MS linear change between batches was removed by subtracting the mean of a variable within a batch from each of the corresponding variables.
GC−MS Analysis
After the addition of 2 ppm Internal Standard (IS), palmitic acid methyl ester, samples were derivatized in two steps: (1) addition of 10 μL of 20 mg mL−1 solution of methoxyamine hydrochloride in pyridine and agitation at 40 °C for 90 min at 750 rpm and (2) addition of 10 μL of trimethylsilyl cyanide (TMSCN)13 and agitation at 40 °C for 40 min at 750 rpm. Immediately after derivatization, 1 μL of sample was injected into a cooled injection system (CIS port) using splitless mode and the septum purge flow and purge flow to split vent (at 2.5 min after injection) were set to 25 and 15 mL min−1, respectively. Initial temperature of CIS was 40 °C, heated to 12 °C sec−1 after 30 s of equilibrium time, and kept for 5 min after reaching 320 °C. Then, the CIS port gradually cooled until 250 °C at 5 °C s−1, whereafter the temperature was kept constant during the run. The GC−MS consisted of an Agilent 7890A GC and an Agilent 5975C series MSD (Agilent Technologies, Glostrup, Denmark). GC separation was performed on a Phenomenex ZB 5MSi 5% Phe column (30 m × 250 μm × 0.25 μm) (Phenomenex ApS, Værløse, Denmark). A hydrogen generator (Precision Hydrogen Trace 500, Peak Scientific Instruments, U.K.) was used to supply a carrier gas, hydrogen, at the constant column flow rate of 1.2 mL min−1. The initial temperature of the GC oven was set to 60 °C and after 3 min of equilibrium time it was heated at the arte of 12 °C/min to reach 320 °C and hold for 8 min. The post run time at 60 °C was set to 5 min. All steps involving sample derivatization and injection were automated using a DualRait MultiPurpose Sampler (MPS) (Gerstel, Mülheim an der Ruhr, Germany). A deactivated glass wool packed liner of CIS was exchanged every 55 injections. Thus, the whole GC−MS data acquisition was performed within eight linear exchange batches, where all three samples (T0, T1, and T2) from the same individual were kept in the same batch. Samples within a single batch were randomized prior to derivatization and GC−MS analysis. One blank sample that contained only derivatization reagents and two control samples, a pooled sample and an alkane mixture sample (all even C10−C40 alkanes at 50 mg L−1 in hexane), was injected between each 5 or 10 samples. Blank samples were used for eliminating reagent-derived peaks, and pooled sample and alkane mixture samples assisted to monitor GC−MS stability over batches. Mass spectra were recorded in the range of 50−500 m/z with a scanning frequency of 3.2 scans s−1, and the MS detector and ion source were switched off during the first 7.6 min of solvent delay time. The transfer line, ion source, and quadrupole temperatures were set to 290, 230, and 150 °C, respectively. The mass spectrometer was tuned according to
Metabolite Identification
PARAFAC2-based deconvoluted mass spectra of each resolved peak were extracted and compared against NIST11 (version 2.0, NIST, USA). Retention indices (RIs) were calculated using the Van den Dool and Kratz equation17 and from retention times of C10−C40 all-even alkanes that were analyzed using the same GC−MS method. Metabolites were identified either at level 1 using authentic standards or at level 2 with an identification criterion of EI−MS match ≥80 (%) and RI match (±30). Furthermore, tentative identification (level 3) was performed using retention time and spectral similarity (EI−MS match ≥65 (%)) of observed peaks.18 The second identification criterion was that EI−MS matched with derivatized metabolites where labile protons being trimethylsilylated (TMS) or metabolites with aldehyde groups and labile protons being methoximated and trimethylsilylated (MEOX-TMS). Data Analysis
An effect of diet on the blood plasma metabolome was evaluated using two experimental data points per participant collected at week 12 (T1) and week 26 (T2). According to the experimental design (Figure 1), along with diet, effects derived from gender and season and their interactions may contribute to a significant variation. Subsequently, this variation may confound or interact with diet-related metabolomic changes. Therefore, we performed ANOVA-Simultaneous Component Analysis (ASCA) for decomposing the data matrix into several effect matrices (e.g., Xdiet, Xgender, Xseason) using the study design. ASCA can be regarded as an ANOVA that allows 1941
DOI: 10.1021/acs.jproteome.6b00109 J. Proteome Res. 2016, 15, 1939−1954
Article
Journal of Proteome Research
Figure 2. ASCA-based decomposition of three main effects: (1) diet, ADD vs NND, (2) season, Winter (Win) vs Spring (Spr), and (3) gender, Male (M) vs Female (F) and two factor interaction effects as well as the overall variance contribution of effects across 144 metabolites. The null hypothesis (H0) was checked for each main and interaction effects using sum of squares of effect matrices and p values were assessed by permutation test. The native unbalanced design (samples are not equally distributed between design blocks, levels of each factor) was balanced in a manner that ensured that each design block contained an equal number of samples by randomly removing samples from blocks containing more samples to investigate interaction effects. A total of 2000 fully balanced data sets were generated, and each of them was subjected to permutation test. Thus, the shown p values for main and two factor interaction effects (A) and (B) are the mean p values are calculated from the permutation tests of 2000 balanced data sets. To evaluate the main effects alone without balancing (no reduction in sample size), since in the case of nonbalanced design the variance distribution and p values remain valid for main effect only and not for interactions, the native unbalanced design was used directly (C).
ΔX = [ΔX1; ΔX2], meaning that the two delta matrices, having the same number of variables and a number of samples, were concatenated resulting a new ΔX matrix. The cutoff value of 6% for the high Weight Loss ↑(WL) subjects was determined based on the initial average body weight (89.7 ± 16.4 kg) of the NND subjects and the average change (−4.74 ± 0.48 kg) after the intervention.10 This showed that on average the NND subjects lost a significant body weight of ∼6%. The cutoff value for the low or no Weight Loss ↓(WL) subjects was determined as 2% of the initial body weight because this variation is close to the average daily body weight variation. The average Weight Loss among ADD subjects was −1.52 ± 0.45 kg, while the average initial body weight was 90.3 ± 18.2 kg, which corresponds to ∼1.7% of the initial body weight. Thus, the ADD subjects were not included in the Weight Loss study. A total of 80 NND samples (corresponding to 58 subjects), 40 high Weight Loss samples (corresponding to 31 subjects), and 40 low Weight Loss samples (corresponding to 27 subjects) were included in the PLS-DA modeling.
partitioning the sources of variance derived from main and interaction effects using all variables simultaneously; however, for interactions between effects, ASCA results are only valid if a study design is fully balanced, meaning that there is an equal number of samples in all levels among investigated factors (equal block size). In this study three main effects were evaluated, diet, gender and season, and also the most interesting two-factor interaction effects, diet × season and diet × gender (Figure 2). Prior to the investigation of the interactions between effects the basically unbalanced study design was balanced at the cost of sample size reduction. Some samples were left out in random order (to get equal number of samples in each block, for example, NND and ADD), and this procedure was repeated 2000 times generating 2000 balanced data sets. To evaluate the diet × season and diet × gender interactions these data sets were further subjected to an ASCA permutation test (2000 permutations)19 (Figure 2A,B). This procedure allowed evaluating the significance of interactions and their variance contributions. To evaluate the significance of the only three main effects, diet, gender, and season, without reduction in sample size, we performed ASCA permutation test by using a complete (unbalanced) data set because interpretation of main effects remains valid even though the data is not balanced (Figure 2C). In a similar manner, the PLS-DA based classification models were developed for diet, effects of gender (male vs female), and season (winter vs spring) using the metabolomics data corresponding to the week 12 (T1) and week 26 (T2) of the intervention period, and time before the intervention period, week 0 (T0), was excluded from the analysis. Discrimination of individuals who followed NND and lost ≥6% of their initial body weight (later referred as high Weight Loss ↑(WL)) from those who lost ≤2% or gained weight (later referred as low Weight Loss ↓(WL)) was performed using a delta metabolomics data set. The delta metabolomics data set, ΔX, was obtained as follows: ΔX1 = XT1 − XT0, ΔX2 = XT2 − XT0, and
PLS-DA Model Optimization and Validation
All PLS-DA models were optimized, and their performance was assessed in a similar fashion as the double cross-validation procedure described in Szymanska et al.20 The statistical significance of optimized final models was evaluated with permutation testing using two diagnostic statistics, the area under the receiver operating characteristics curve (AUC), and the misclassification rate.21 The final PLS-DA models were obtained in two steps, (1) model optimization, which involved assessment of the model complexity (number of latent variables) and variable selection, and (2) model assessment using the virgin test set samples. The initial data set was randomly divided into a virgin test set and a calibration set, in 1:3 or 1:4 ratio, keeping both time points, T1 and T2, corresponding to the same individual present either in the test set or in the calibration set. Further PLS-DA model 1942
DOI: 10.1021/acs.jproteome.6b00109 J. Proteome Res. 2016, 15, 1939−1954
Article
Journal of Proteome Research
Table 1. List of Discriminating Variables Selected in PLS-DA Variable Selection for Classifying Diet, NND vs ADD, and High Weight Loss (↑WL) (loss of ≥6% of pre-intervention time body weight) vs Low Weight Loss (↓WL) (≤2% of pre-intervention time body weight) among Subjects Who Followed NNDa
a
Metabolites selected as markers are highlighted with grey background. Relative differences of metabolites between NND and ADD individuals at the intervention study times T1 and T2 were calculated from the raw data as follows: T1DIET = (((mean(NNDT1) − mean(ADDT1))/ mean(ADDT1)) × 100) and T2DIET = (((mean(NNDT2) − mean(ADDT2))/mean(ADDT2)) × 100). The percentage differences of metabolite levels between two diets are represented with up (↑) and down (↓) arrows. Up (↑) arrow indicates positive change, meaning that relative concentration of a metabolite in NND individuals on average was higher than in ADD individuals. Down (↓) arrow indicates exactly the opposite meaning that relative amount of a metabolite was greater in ADD than in NND. Relative differences of metabolites between high weight loss (↑WL) and low weight loss (↓WL) NND individuals were calculated as follows: T1↑WL = (((mean(↑WLT1) − mean(↑WLT0))/mean(↑WLT0)) × 100) and T1↓WL = (((mean(↓WLT2) − mean(↓WLT0))/mean(↓WLT0)) × 100), followed by T1WL = T1↑WL − T1↓WL. T2WL has been calculated in the same way as T1WL. The percentage differences of metabolite levels between high weight loss (↑WL) and low weight loss (↓WL) are represented with up (↑) and down (↓) arrows. Up (↑) arrow indicates positive change, meaning the relative concentration of a metabolite in high weight loss (↑WL) individuals on average was higher than that in the low weight loss (↓WL) individuals. Down (↓) arrows indicate the opposite meaning that a metabolite level was greater in ↓WL than in ↑WL. Bigger bold arrows illustrate metabolites, where relative differences were ≥20% and smaller arrows were