Metabolomics of Ramadan fasting: an opportunity for the controlled study of physiological responses to food intake

The Ramadan fasting study was conducted as a pilot study to the HuMet study [22]. The HuMet study protocol was approved by the ethical committee of the Technische Universität München (#2087/08). This study conformed with the Declaration of Helsinki. All study participants gave written informed consent. A total of eleven healthy male Muslim volunteers of varying BMI, age and nationality were enrolled at the Islamic Centre Munich in 2009. The study design is illustrated in Figure 1a. For the fast-breaking, study participants shared all the same meal in a joint cantina at the mosque. Trays with identical, pre-composed meals of substantial size were dispensed to all study participants. Some variation in food supply may have occurred, since for logistical reasons we could not verify that participants did not share some food with other members of the community who were not participating in the study. However, we estimate that this effect was minimal. Meal compositions and nutrient analysis are provided in Additional file 1: Table S1. Blood samples were collected before and after the fast-breaking at day 7 (time points 1 and 2) and day 26 (time points 3 and 4) from individuals who fasted from dawn to sunset during the month of Ramadan in 2009. A time interval of two hours between the pre- and postprandial sample collections (shortly before and after the meal) was applied. An additional blood sample was collected 4–8 weeks after Ramadan (time point 5) after an overnight fast. Venous blood samples were collected into 9 ml EDTA K₂-Gel tubes (Sarstedt, Nümbrecht, Germany), immediately centrifuged at 2,500 g for 10 min at 20°C. The obtained plasma was then aliquoted, frozen on dry ice and stored at −80°C until further analysis. Because this study was designed to be a proof-of-principle study, and also because we were unsure about how far we would interfere with the religious proceedings by doing so, no further data was collected from the study participants. However, we found that the community was very open to this kind of study, so that collecting additional phenotype data is not expected to pose any major obstacles in future studies.

The clinical biochemistry analysis was performed as previously described for the HuMet study [22]. Briefly, glucose and lactate concentrations were determined by using the Super GL easy ⁺ (Dr. Müller Geräte Bau, Freital, Germany) based on an enzymatic amperometric technique. Insulin level was measured by enzyme-linked immunosorbent assay (ELISA) (Dako #K6219, Glostrup, Denmark). The plasma concentration of triglycerides and non-esterified fatty acids (NEFAs) was measured using an enzymatic colorimetric method (Fluitest TG, Analyticon Biotechnologies AG, Lichtenfels, Germany) and (NEFA-HR, Wako Chemicals GmbH, Neuss, Germany), respectively. All of the above listed methods were used according to the manufacturer’s guidelines.

The targeted metabolomics assay AbsoluteIDQ™ p180 (BIOCRATES Life Sciences AG, Innsbruck, Austria) was applied for metabolite measurements in plasma samples. Sample preparation and metabolite detection was performed according to the AbsoluteIDQ™ assay kit instructions as previously described [25, 26]. Metabolite detection was performed by FIA – MS/MS (for carnitines and lipids) and LC – MS/MS (for amino acid and biogenic amines) on an API 4000 (AB Sciex) triple quadrupole mass spectrometer. Data evaluation was performed using the MetIDQ™ software package, which is part of the AbsoluteIDQ™ kit. The AbsoluteIDQ™ p180 assay enables the detection of 186 metabolites including: 40 acylcarnitines (free carnitine – C0 and acylcarnitines – Cx:y), 21 amino acids, 19 biogenic amines, 90 glycerophospholipids including lysophosphatidylcholines (LysoPC a Cx:y) and phosphatidylcholines with acyl (PC aa Cx:y) or ether (PC ae Cx:y) side chain, hexoses (H1) and 15 sphingolipids (SM Cx:y). As previously described [25, 26], the nomenclature used for lipid metabolites refers to the Lipid Maps comprehensive classification system [27]. The nomenclature for lipids is as follows: Cx:y, where “x” denotes the number of carbons (C) and “y” represents the number of double bonds. The protocol used for the analysis of bile-acids composition in the plasma samples was implemented as follows: In-house protocols were used for the analysis of bile-acids composition in the plasma samples. Plasma samples of 100 μl in Eppendorf tube were spiked with labeled internal standards d4-deoxycholic acid, d4-glycochenodeoxycholic acid, d4-glycocholic acid and d4-lithocholic acid (all internal standards were purchased from CDN Isotopes, Dr. Ehrendorfer) to yield concentrations of 50 ng internal standards/ ml plasma. After well mixing, the samples were transferred into a 96-well plate. In addition to samples from this study, 5 pooled human reference plasma samples and a series of 6 calibration samples with a concentration range of 5 to 150 ng bile acids/ ml plasma were also pipetted into the 96-well plate. The calibration samples were prepared by spiking certain concentrations of 12 bile acids standards (cholic acid, ursodeoxycholic acid, chenodeoxycholic acid, deoxycholic acid, lithocholic acid, glycocholic acid, glycochenodeoxycholic acid, glycodeoxycholic acid, taurocholic acid, taurochenodeoxycholic acid, taurodeoxycholic acid and taurolithocholic acid; all bile acids standards were purchased from Sigma-Aldrich) and labeled internal standards (50 ng/ml plasma) into human plasma-free endogenous bile acids. Bile acid standards and labeled internal standards used to spike the plasma samples were dissolved in 50% v/v methanol containing 0.012% formic acid and 5 mM ammonium acetate. These calibration samples were applied to develop a calibration curve for analyte quantifications. Human reference plasma samples, treated like the samples of this study, served as technical replicates throughout the data set to assess process variability. Besides the above mentioned samples, 100 μL of eluent buffer and 100 μl of plasma-free endogenous bile acids were also pipetted into the 96-well plate to serve as process blanks. Proteins were precipitated and bile acids were extracted from the samples with 1 ml of acetonitrile. After centrifugation, 800 μl of supernatant was transferred into another 96-well plate for evaporation in vacuum evaporator (Uniequip). Bile acids were analyzed with 4000 Q-Trap MS (AB Sciex) equipped with LC (Shimadzu) for chromatographic separation. Prior to analysis, the dried samples were reconstituted with 150 μl of 50% v/v methanol containing 0.012% formic acid and 5 mM ammonium acetate. The MS was operated in the ion spray ionization mode. Data were acquired and processed using Analyst 1.5.1 software (AB Sciex) and IntelliQuan MQll (AB Sciex) for the quantitative processing. The measurement was performed in negative ion mode with tandem MS via collision activated dissociation (CAD) with nitrogen as collision gas. Quantitative data were acquired in the multiple reaction monitoring (MRM) modes with specific precursor to product ion parameters for each analyte. Chromatographic separation was achieved using a reverse-phase C18 column, Onyx Monolithic, 130 Å pore size, 100 × 4.6 mm internal diameter (Phenomenex). After injection of the 40 μl sample extract, the column was developed with mobile phase of 10% methanol containing 0.012% formic acid and 5 mM ammonium acetate (eluent A) and 90% methanol with 0.012% formic acid and 5 mM ammonium acetate (eluent B) in a gradient of 30% eluent A to 95% eluent B in 10 min run time, and then at constant eluent of 95% B for 2 min before it gradually eluted to 70% eluent B. The total run time was 15 min at 1 ml/min flow rate. At analysis, the column temperature was maintained constant at 30°C and the auto sample temperature was at 4°C.

Statistical analysis was done using R version 3.0.2 (R Core Team (2013). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL http://​www.​R-project.​org/​). Linear mixed effects models using log-scaled data were fitted to the difference (MDIFF) between data of paired samples from same individuals, before and after fasting-break, and matched time-points of week 1 and 4 of Ramadan, respectively, using the lme4 package in R (Douglas Bates, Martin Maechler, Ben Bolker and Steven Walker (2014). lme4: Linear mixed-effects models using Eigen and S4. R package version 1.1-5. http://​CRAN.​R-project.​org/​package=​lme4). As fixed effect we entered the intercept and as random effect we used the participant identifier (PID) (model: MDIFF ~ 1 + (1|PID)). Significance was determined using the chi-square statistic from an ANOVA, using a null model with no intercept and the participant identifier as random effect (null model: MDIFF ~ 0 + (1|PID)). We hence test the hypothesis that the difference between the log-scaled metabolite concentrations is different from zero, which is equivalent to testing the hypothesis that the fold change of the metabolite concentrations is different from unity. Deviation from normality of the residuals was tested using the Shapiro-Wilk test. In cases where a deviation from normality was observed (p < 0.05) we conducted subsequently a paired Wilcoxon rank sum test (also known as ‘Mann-Whitney’ test). In Tables 1 and 2 we report all nominally significant associations (p < 0.05). In order to control the false discovery rate (FDR), we apply the Benjamini and Hochberg approach (as described in [28]) and focus the discussion on associations that have a FDRbelow 5%. The full association data is provided as Additional file 2: Table S2 and Additional file 3: Table S3.

Readouts from clinical biochemistry parameters and selected metabolites are presented in Figure 2. The corresponding p-values of associations are reported in Table 1. In accordance with the fasting state, glucose, lactate and insulin levels were low before fast break (time points 1 and 3) but increased and showed much larger variability after food intake (time points 2 and 4). In agreement with the expected more intense fasting during Ramadan, insulin and glucose levels at both Ramadan fasting time points were also lower than those observed after the overnight fast (time point 5). Levels of non-esterified fatty acids (NEFAs) at both Ramadan fasting time points were significantly higher compared to time points after food ingestion. Our results are in good accordance with the previously reported NEFA fluctuation after an oral glucose tolerance test and a meal tolerance test were applied [29]. In the fasting state, NEFAs are released from triacylglycerol stored in adipocytes (lipolysis), thereby increasing in plasma [30]. Consequently, increased postprandial insulin concentrations induce a decrease in NEFA levels by rapid suppression of lipolysis [31]. In the present study, we also observed this correlation between the elevated plasma insulin levels after ingestion of food and the decrease in concentrations of plasma NEFAs in the postprandial state. Plasma triglyceride levels increased after food intake, but only slightly on day 7 of Ramadan (time points 1 and 2) and moderately on day 26 (time points 3 and 4). Previous studies report triglyceride levels that doubled and reached its peak around three hours after ingestion [32]. In our study we did not observe such a strong effect, which may in part be due to differences in sampling time in our study and that of Coppack et al. [32], and also in part to difference in the meal composition. Overall, we find that the observed changes in levels of insulin, glucose, lactate, NEFAs and triglycerides are coherent with the pre-prandial and postprandial conditions of the eleven subjects, especially when considering the varying BMI, age and ethnicity of the participants. These observations thus demonstrate the reliability of our experimental setup.

To monitor the global metabolic events occurring in the body during Ramadan fast-breaking we applied targeted metabolomics approaches for the detection and quantification of a total of 202 metabolites. Individual data points for all measured variables (202 metabolites and 6 parameters determined using clinical chemistry methods) are provided in Additional file 4: Figure S1. At a nominal significance level of 0.05, the levels of 48 variables differed significantly between pre- and postprandial state (Table 1). These metabolites belong to different metabolic classes: amino acids, bile acids, acylcarnitines, and polyamines. To illustrate the direction of changes at different time points, the data were visualized in the form of box-plots (Figure 2) and line plots (Additional file 4: Figure S1). The levels of many amino acids increased after food consumption, including asparagine, arginine, alanine, glutamate, proline, and phenylalanine. Methionine sulfoxide (Met-SO), which is an oxidation product of methionine with reactive oxygen species (ROS) and has been suggested as a biomarker for oxidative stress [33], also showed a strong increase. Methionine oxidation is part of a highly regulated machinery, controlling the function of signaling factors under physiological challenge conditions and conditions of oxidative stress [34]. When the amino acids levels at time points 1 and 3 were compared with the overnight fasting situation at time point 5, asparagine, arginine and phenylalanine exhibited similar levels, in contrast to proline and alanine, which after overnight fasting were comparable to the post-prandial state in Ramadan. Our observations are in good agreement with a previous study that reported a decrease in amino acids levels in both plasma and urine samples of subjects after a prolonged fasting period of 36 h [35].

Under the postprandial conditions, the glycine/taurine conjugated bile acids showed increased levels, whereas in the pre-prandial state at time points 1 and 3 their concentrations were similar to those measured after on overnight fast. Bile acids are produced from cholesterol in the liver and stored in the gall bladder. The gall bladder contracts in the postprandial state and releases bile acids into the intestine, which are reabsorbed through the enterohepatic circulation [36, 37]. Thus, the postprandial plasma bile acid levels are determined by the balance between intestinal input and hepatic clearance. Bile acid levels were shown to increase rapidly within 30 minutes after meal intake in the HuMet study [22] and are in agreement with our expectations.

Not all metabolites were up-regulated after food ingestion. A significant decrease in the concentration of long-chain acyl-carnitines was visible in the postprandial stage. The decrease in long-chain acyl-carnitine levels was expected since carnitine-bound fatty acids are imported into the mitochondria for beta-oxidation and represent an energy-source under fasting conditions. Consequently, decreases in long-chain acyl-carnitine levels reflect a switch of energy metabolism from fatty acid to glucose oxidation in agreement with the altered substrate availability in the postprandial state. We also observed a decrease in the levels of the polyamines spermidine and putrescine after food ingestion. Dietary polyamines are absorbed shortly after a meal by the small intestine and degraded in the gut before reaching systemic circulation [38]. However, recently an association between fasting plasma glucose and total polyamine concentration was reported [39], which suggests a fluctuation in polyamine concentrations as a response to the nutrition stimuli. Taken together, the observed changes in the different metabolic classes reflect the expected physiological response of the human body to the break of the fasting state.

We also studied the influence of long-term fasting on the human metabolic profile by comparing samples taken before fast breaking on week 1 to those on week 4, and those taken after fast breaking on week 1 to those of week 4, respectively. Overall, these changes displayed a lower level of significance than the changes observed with respect to fast-breaking. Twenty metabolites were significantly changed between week 1 and 4 of Ramadan, many of which were phosphatidylcholines (PCs) (Table 2). Many of these PCs are composed of long-chain fatty acids (C12-C18) and polyunsaturated omega-3 and omega-6 fatty acids (C18:2 and above). PC levels in the fourth week of Ramadan were higher than PC levels at the start of Ramadan, while PC levels at the start of Ramadan were very similar to those observed after an overnight fast. A recent study in a cohort of 173 families in Saudi Arabia showed that approximately two thirds of the respondents (59.5%) reported weight gain after Ramadan [40]. Hence, some of the changes observed in these metabolites may be attributed to a change in nutrition regimen during Ramadan. Our results clearly suggest that differences between the effects of short- and long- term fasting, changes in nutrition pattern, and adaptation of human metabolism to prolonged fasting conditions can be studied in a sufficiently powered metabolomics study during Ramadan.

Our study was conducted in a small and heterogeneous group of eleven healthy male individuals with varying BMI, age and ethnicity and, therefore has considerable limitations. Nevertheless, the experimental design (Figure 1a) we applied in this proof-of-principle study already revealed sound metabolic signatures of the pre- and post-prandial stages as well as the effect of longer-term fasting (several weeks). Implementation of a full-fledged research study during the month of Ramadan and collection of additional phenotype information of the participants may reveal valuable information on body adaptation and the plasticity of metabolic control regarding anabolic/catabolic conditions. Countries with a large proportion of Muslims in their population provide an excellent opportunity for such studies. An example of a sample collection schedule is illustrated in Figure 1b. Time series may be collected along the month of Ramadan and also during a single day, e.g. respecting prayer times in order to reliably sample blood from study participants at different time points. Different depths of sampling density may be considered, including use of blood spots rather than venipuncture, and enrolling participants from the medical community at large clinical centers for more dense sampling protocols. These individuals may be more easily available for frequent blood sampling. Local communities often organize joint Iftar festivities in order to break the fast collectively. Such a set-up can be used for future research studies since the food consumption can then be monitored and drawing of samples from volunteers be made at dedicated mobile sampling and storage units. Furthermore, sample collection procedures can be simplified and conducted by the individual at home or work environment via blood spotting on filter paper after finger prick. Metabolic measurements in dried blood spots (DBS), covering a panel of 42 metabolites including acyl-carnitines and amino acids, were already applied in the clinic to screen infants for metabolic diseases [41]. The panel of metabolites detected from DBS can be extended to 188 small molecules as previously described [42]. Such a simplified sampling procedure could also be used in order to increase the number of samples taken during the day or to allow volunteers taking samples by themselves during the day without the requirement of venipuncture.

One aspect that may need specific consideration is circadian rhythm. During the Ramadan time the sleep patterns of some individuals may interchange day and night activities. Therefore, it is extremely important to conduct the study in a well controlled group of participants, where such changes are documented and ideally measured. Moreover, it may be important to document changes in overall activity pattern, since many Muslims use the Ramadan period to take vacation and visit family. Notwithstanding, this pilot study represents a starting basis for the design of a large-scale Ramadan metabolomics study to address the important issue of how cultural and religious fasting may affect metabolism and disease risk in a large group of individuals with variable preconditions.

Recently, obesity defined by the World Health Organization (WHO) as abnormal or excessive fat accumulation that represents a risk of health [43], reached an epidemic level by affecting nearly half a billion of the world’s population [44]. Obesity is recognized as a major risk factor of chronic diseases, especially cardiovascular disease, hyperglycemia [45], and type 2 diabetes, [46]. In Arab countries, the prevalence of overweight, obesity and type 2 diabetes increased drastically over the last three decades [47]. Therefore, implementation of a metabolomics study during the month of Ramadan in countries of dominantly Muslim culture, such as the Gulf countries, offers the possibility to monitor large groups of patients with morbid obesity and diabetes under strictly controlled conditions. A large-scale population study on the metabolic regulation during the month of Ramadan opens several possibilities to explore not only the metabolic plasticity in healthy individuals, but also to monitor subjects diagnosed with complex diseases. The experimental set-up can reveal novel biomarkers for these diseases, which might only be unmasked under physiological challenge. Furthermore, this study can also be used for classification of nutrients required for energy production and for investigating the proper functioning of the human body under prolonged fasting in both healthy and diabetic individuals. The importance of dietary guidelines to control and reduce the global disease burden including diet-related complex chronic diseases has already been emphasized [48].

Last but not least, we like to draw attention to the advantages of combining metabolomics and genome-wide association studies (GWAS) in the Arab population during the month of Ramadan. The capability of GWAS in the identification of genes associated with type 2 diabetes, cardiovascular disease, as well as with clinically relevant intermediate traits, such as cholesterol and triglyceride levels has already been reported [49‐51]. The first large-scale metabolomics project including genome wide association study data (mGWAS) [52] revealed relations between loci such as FADS1, ELOVL2 or SLC16A9 and changes in specific lipid species that are products and substrates of these enzymes and transporters. The advantages of applying metabolomics together with GWAS to overcome the limitations of GWAS-alone studies was recently discussed [53, 54]. The physiological and metabolic consequences, with respect to genetic variation analysis, triggered by such an extreme change in dietary habits (Ramadan) have not been studied before. Subsequently, by controlling for dietary patterns during the fasting month, we shall be able to investigate the effect of genetic variations in combination with various defined nutritional intakes (gene-environment interaction) on the metabolic profile of the subjects.