The Metabolome Profiling and Pathway Analysis in Metabolic

ORIGINAL ARTICLE The metabolome proﬁling and pathway analysis in metabolic healthy and abnormal obesity

H-H Chen1, YJ Tseng2,3,4,5, S-Y Wang3,5, Y-S Tsai6, C-S Chang6,7, T-C Kuo4,5, W-J Yao8, C-C Shieh6, C-H Wu7,9 and P-H Kuo1,10

OBJECTIVES: Mechanisms of the development of abnormal metabolic phenotypes among obese population are not yet clear. In this study, we aimed to screen metabolomes of both healthy and subjects with abnormal obesity to identify potential metabolic pathways that may regulate the different metabolic characteristics of obesity. METHODS: We recruited subjects with body mass index (BMI) over 25 from the weight-loss clinic of a central hospital in Taiwan. Metabolic healthy obesity (MHO) is defined as without having any form of hyperglycemia, hypertension and dyslipidemia, while metabolic abnormal obesity (MAO) is defined as having one or more abnormal metabolic indexes. Serum-based metabolomic profiling using both liquid chromatography–mass spectrometry and gas chromatography–mass spectrometry of 34 MHO and MAO individuals with matching age, sex and BMI was performed. Conditional logistic regression and partial least squares discriminant analysis were applied to identify significant metabolites between the two groups. Pathway enrichment and topology analyses were conducted to evaluate the regulated pathways. RESULTS: A differential metabolite panel was identified to be significantly differed in MHO and MAO groups, including L-kynurenine, glycerophosphocholine (GPC), glycerol 1-phosphate, glycolic acid, tagatose, methyl palmitate and uric acid. Moreover, several metabolic pathways were relevant in distinguishing MHO from MAO groups, including fatty acid biosynthesis, phenylalanine metabolism, propanoate metabolism, and valine, leucine and isoleucine degradation. CONCLUSION: Different metabolomic profiles and metabolic pathways are important for distinguishing between MHO and MAO groups. We have identified and discussed the key metabolites and pathways that may prove important in the regulation of metabolic traits among the obese, which could provide useful clues to study the underlying mechanisms of the development of abnormal metabolic phenotypes. International Journal of Obesity (2015) 39, 1241–1248; doi:10.1038/ijo.2015.65

INTRODUCTION individuals share a similar total body fat percentage, but the MHO 6,7 Worldwide, obesity is associated with increased mortality and high group has lower visceral fat content than the MAO group. prevalence of metabolic-related diseases. In particular, there is an Moreover, the MHO group shows a significantly lower percentage 6,7 increased risk of insulin resistance, hypertension and dyslipidemia of ectopic fat, especially in the muscle and liver. The MHO group in the obese population.1 Nevertheless, about 10–30% of also exhibits a higher level of physical activity in comparison with obese individuals are reported to be insulin sensitive, having the MAO group.3 Our current knowledge about the causes and normal blood pressure and lipid profiles. In other words, a certain regulation pathways of different obesity-related metabolic profiles proportion of obese individuals possess a relatively healthy is still limited, despite the increasing awareness of disparate metabolic status.2,3 Previous studies have shown that the clinical outcomes of the two obese groups. Therefore, using a metabolically healthy obesity (MHO) group has a lower mortality more innovative and comprehensive screening tool is essential to and has a lower risk of developing metabolic diseases explore the differences between the MHO and MAO groups. (for example, diabetes and hypertension) compared with the Among numerous 'omics' technologies, metabolomics is often metabolic abnormal obesity (MAO)4,5 group. used to profile small endogenous molecules, or metabolites that Although the underlying mechanisms of the metabolic regula- are present in biological samples. Metabolites are the intermedi- tion are not yet clear, several attempts have been made to ates or products of different metabolic pathways. Therefore, their investigate relevant factors for distinguishing the MHO and MAO, concentrations could be influenced by innate genetic predisposi- such as adipose hormone, fat tissue distribution and life style. tion, environmental exposures or stimuli, as well as interactions For instance, higher level of adiponectin is associated with MHO between the two. Unlike genomics, the metabolome represents status, and this effect is independent of obesity severity.2 When the organisms’ conditions at any given time, and thus is able to considering body composition, metabolic healthy and abnormal capture the dynamic physiological condition corresponding to the

1Department of Public Health and Institute of Epidemiology and Preventive Medicine, National Taiwan University, Taipei, Taiwan; 2School of Pharmacy, College of Medicine, National Taiwan University, Taipei, Taiwan; 3Department of Computer Science and Information Engineering, National Taiwan University, Taipei, Taiwan; 4Graduate Institute of Biomedical Electronic and Bioinformatics, National Taiwan University, Taipei, Taiwan; 5The Metabolomics Core Laboratory, Center of Genomic Medicine, National Taiwan University, Taipei, Taiwan; 6Institute of Clinical Medicine, National Cheng Kung University Medical College, Tainan, Taiwan; 7Department of Family Medicine, National Cheng Kung University Hospital, Tainan, Taiwan; 8Department of Nuclear Medicine, National Cheng Kung University Hospital, Tainan, Taiwan; 9Institute of Behavioral Medicine, National Cheng Kung University Medical College, Tainan, Taiwan and 10Research Center for Genes, Environment and Human Health, National Taiwan University, Taipei, Taiwan. Correspondence: Dr C-H Wu, Department of Family Medicine, National Cheng Kung University Hospital, 138 Sheng-Li Road, Tainan, Taiwan or Professor P-H Kuo, Department of Public Health, Institute of Epidemiology and Preventive Medicine, College of Public Health, National Taiwan University; Room 521, No. 17 Xuzhou Road, Taipei 100, Taiwan. E-mail: [email protected] or [email protected] Received 23 September 2014; revised 23 March 2015; accepted 12 April 2015; accepted article preview online 24 April 2015; advance online publication, 26 May 2015 Metabolome profiling for metabolic healthy obesity H-H Chen et al 1242 behavioral and clinical outcomes of interests. The development of except in the WC.16 On the contrary, individuals who had one or more obesity and its related metabolic phenotypes is a process involved abnormal metabolic indexes were placed in the MAO group. with both genetic and environmental (for example, diet and life style) factors. To investigate the complex molecular differences Metabolome profile between the MAO and MHO groups, conducting a global analysis The metabolomic profiling experiments were conducted by the Metabo- of metabolites provides an ideal way to uncover the underlying lomics Core Laboratory at Center of Genomic Medicine at National Taiwan mechanisms for the development of abnormal metabolic status University. Plasma samples were obtained after 8 h of fasting and stored at in obese individuals.8 Previously, metabolomics studies have been − 80 °C. LC-MS and GC-MS were used to perform the metabolomics profile applied to search for relevant metabolic pathways for obesity, analysis. The quality controls of samples were obtained by pooling aliquots diabetes and dyslipidemia.9–14 However, to the best of our from each plasma sample. All of the samples were extracted with 400 μlof knowledge, this approach has not yet been directly applied to methanol and analyzed using an Agilent 1290 UHPLC system coupled to a investigate the metabolomic profiles of the MAO and MHO 6540-QTOF (Agilent Technologies, Santa Clara, CA, USA). An Acquity HSS T3 μ groups. column (100 × 2.1 mm, 1.8 m; Waters, Milford, MA, USA) was used for the In the present study, we have aimed to identify important separation and the column was maintained at 40 °C. For sample ionization, a Jet Stream electrospray ionization source was used with a capillary metabolites to distinguish a normal metabolic state from an voltage of 4.0 kV in positive and negative mode. The MS parameters abnormal obesity metabolome. Both a targeted and untargeted were set as follows: gas temperature, 325 °C; gas flow, 5 l min−1; Nebulizer, analysis was employed in this study. The targeted analysis was 40 p.s.i.; sheath gas temperature, 325 °C; sheath gas flow, 10 l min−1; and compared against a library of 820 metabolites. We also mapped fragmentor, 120 V. A scan range of 50–1700 m/z was set. relevant metabolites to their corresponding metabolic pathways GC-MS was also applied. Plasma samples were extracted with 400 μl fi of methanol and derivatized using methoxyamine hydrochloride to pro le the underlying mechanisms of metabolic regulation −1 among obese individuals. To achieve our objective, we conducted (40 mg ml ) and derivatization agent (BSTFA+TMCS, 99:1). The instrument a metabolomics study in an age, sex and BMI paired sample of used an Agilent 7890 A gas chromatograph system (Agilent Technologies) coupled to an Agilent 5975 MSD mass spectrometer (Agilent Technolo- MAO and MHO individuals using both liquid chromatography/ gies). The column used for separation was a 10-m Duragard integrated fl time-of- ight mass spectrometry (LC-MS) and gas chromatogra- Agilent 122-5532G DB5-MS (30 m × 0.25 mm, 0.25 μm). The MS parameters phy/quadrupole mass spectrometry (GC-MS), for full assessment of were set as follows: transfer line temperature 280 °C; electron impact a wide range of metabolites. ionization at 70 eV; filament source temperature, 230 °C; quadrupole temperature of, 150 °C. A scan range of 50–550 μ at 2 spectra/s was set. SUBJECTS AND METHODS Statistical analysis Subjects There were 382 well-defined metabolites in the LC-MS library and 438 well- All participants were recruited from the Weight Management Outpatient defined metabolites in the GC-MS library17 in total. The metabolite intensities Clinic at National Cheng Kung University Hospital in southern Taiwan from were analyzed with a z-transformation. Because we used a matched study − 2009 to 2010. A body mass index (BMI) of over 25 kg m 2 was defined as design to select the groups of MHO and MAO individuals with similar sex, age being obese. To study the metabolomics profiles in the MAO and MHO and BMI distributions, we performed conditional logistic regression analysis to groups, we attempted to minimize potential confounding effects identify suggestive metabolites. Due to moderate sample size in the present by individually matching the two groups with similar age, sex and BMI study, we reported metabolites with P-value less than 0.10. The conditional in the current study. In addition, all of the participants were non-smokers, logistic regression estimates the effects of individual metabolites on health/ were not heavy-drinkers (defined as more than two drinks per day), were abnormal obesity status under our matched study design. Effect size was not diabetic, and had no other known inflammatory disease or cancer. estimated and presented as β coefficient to quantify the risk for metabolically Participants did not use hormone replacement therapy or any therapy that abnormal obesity. A positive β coefficient represents increased risk with might influence the metabolic results or the measurement of metabolites. higher metabolite levels; while a negative β coefficient represents decreased In total, 34 pairs of MAO and MHO subjects (N = 68) met the above criteria risk with higher metabolite levels. and were analyzed using both LC-MS and GC-MS. The study was approved The partial least square-discriminate analysis (PLS-DA) was applied to by the local institutional review board, and all participants gave written visualize how well the relevant metabolites could distinguish the MHO informed consent. from the MAO groups. PLS-DA is a supervised method to estimate the correlations between metabolites and outcome status by a covariance matrix, and is often applied in metabolomics studies for classification using Anthropometric and laboratory measurements identified metabolites to maximize the distance between defined groups We collected blood samples from each participant after 8 h of fasting. (that is, MHO and MAO in our study).18 In addition, we used the leave- Anthropometrical and biochemical indexes were assessed at baseline one-out validation with the maximum distance method to evaluate the examination. Anthropometrical (that is, body weight, height and waist accuracy to correctly identify individuals belonging to one of the MHO or circumference (WC)) measurements were done by trained staff. BMI was MAO group. The workflow of the current metabolomics study is shown in 2 calculated as body weight (kg)/(body height (m)) . The WCs were Figure 1. Additional metabolites, which were not included in the library, measured midway between the lateral lower rib margin and the superior were identified by their mass over charge value from the LC-MS analysis.19 anterior iliac crest with standard tape (Gulick II, Gays Mills, WI, USA, to the Analyses of condition logistic regression and PLS-DA were performed using nearest mm) at the end of a gentle expiration phase by the same trained R software (http://www.r-project.org/), with packages of 'survival' and staff. Systolic blood pressure (SBP) and diastolic blood pressure (DBP) were 'mixOmics', respectively. measured by a DINAMAP vital sign monitor (Model 1846SX, Critikon Inc., To examine the metabolomics profiles more comprehensively, all Tampa, FL, USA) as recommended by the American Heart Association. relevant detected metabolites (Po0.2) were mapped into metabolic Biochemical indexes included high-density lipoprotein cholesterol (HDLC), pathways using the Kyoto Encyclopedia of Genes and Genomes (KEGG) triglyceride (TG), and fasting blood glucose (BSAC) at fasting status, were database, and 80 human metabolism pathways were selected for further measured with an automated instrument biochemically (Roche Modulator analysis.20 We performed two different types of pathway analyses, DP, Roche, St Louis, MO, USA). including enrichment and topology analyses. The enrichment analysis The definition of abnormal metabolic was followed by the criteria of was conducted using a hypergeometric test, which compares the numbers metabolic syndrome according to the National Cholesterol Education of significant metabolites within a specific pathway with the expected Program Adult Treatment Panel III for Asian,15 including WC, SBP, DBP, value. In addition to considering the over-representation of significant HDLC, TG and BSAC. Abnormal metabolic indexes were defined as BSAC metabolites in examined pathways, the other important feature − − − over 100 mg dl 1, TG over 150 mg dl 1, HDLC lower than 40 mg dl 1 in of pathway analysis is the structure. The topology analysis considers the − males or 50 mg dl 1 in females, SBP over 130 mm Hg or DBP over position of significant metabolites in metabolic pathways that are 85 mm Hg. As nearly all the obese subjects had abnormal WC, the criterion annotated by KEGG. Two indexes were estimated in the topology analysis: of MHO was defined as having normal status in all of the above indexes, the centrality and pathway impact. The centrality measures the number of

Table 2. Results of signiﬁcant metabolites using conditional logistic regression analysis

Metabolite Conditional logistic regression

β s.e. P-value

LC-MS L-Kynurenine 1.386 0.630 0.028 Glycerophosphocholine − 1.500 0.714 0.036 N-Acetylserine 1.456 0.767 0.058 Decanoylcarnitine 0.537 0.285 0.060 L-Alloisoleucine/L-Norleucine/ 0.896 0.486 0.065 L-Isoleucine Figure 1. The workflow of the present metabolomics study. The gas Hexanoylcarnitine 0.464 0.252 0.066 chromatography–mass spectrometry was applied for target and 5,6-Dihydrothymine 0.607 0.368 0.099 non-target analysis and liquid chromatography–mass spectrometry was applied for target analysis. There were 23 metabolites GC-MS significantly identified in target analysis, and they were mapped Glycerol 1-phosphate − 1.124 0.376 0.003 into 9 metabolic pathways. Non-target analysis was based on these Glycolic acid − 0.970 0.329 0.003 9 candidate pathways in which 145 metabolites were detected, Tagatose − 0.897 0.314 0.004 and 49 metabolites were significant and used in further pathway Methyl palmitate 0.796 0.279 0.004 analysis. Uric acid 0.987 0.352 0.005 Dioctyl phthalate 1.519 0.596 0.011 Phosphoglycolic acid 0.570 0.256 0.026 Table 1. Demographic and clinical characteristics in MHO and MAO L-Threonine − 0.625 0.286 0.029 groups Palmitic acid 0.649 0.300 0.030 Stearic acid − 2.145 1.018 0.035 MHO (N = 34) MAO (N = 34) P-value L-Valine 0.628 0.316 0.047 Benzoic acid − 0.473 0.243 0.051 N-Methyl-DL-glutamic acid 2.707 1.416 0.056 Mean s.d. Mean s.d. L-(+) Lactic acid 0.538 0.306 0.078 L-Tyrosine 0.400 0.231 0.084 — Male, N (%) 17 (50%) 17 (50%) Isopropyl beta-D-1- 0.913 0.533 0.086 Age 32.12 7.92 34.68 8.73 0.210 thiogalactopyranoside Height (cm) 168.80 9.94 166.84 7.32 0.358 Weight (kg) 90.96 24.67 95.67 22.80 0.416 Abbreviations: GC-MS, gas chromatography/quadrupole mass spectro- − BMI (kg m 2) 31.53 6.03 34.16 6.92 0.099 metry; LC-MS, liquid chromatography/time-of-flight mass spectrometry. We WC (cm) 96.04 14.09 104.47 17.72 0.034 used conditional logistic regression models to quantify the effect of SBP (mmHg) 111.06 11.83 119.15 17.70 0.031 metabolites on metabolic abnormal obesity, effect size and the variation of DBP (mmHg) 69.38 8.52 75.24 12.16 0.025 parameter estimation were reported with β coefficient and standard error − BSAC (mg dl 1) 90.97 5.50 110.24 31.78 0.001 (s.e.), respectively. TG (mg dl−1) 82.29 26.17 172.38 137.95 o0.001 HDLC (mg dl−1) 53.53 10.80 42.18 8.68 o0.001 GOT (AST) U l−1 28.29 15.94 39.56 29.40 0.055 GPT (ALT) U l−1 40.21 40.51 54.82 47.49 0.177 The two groups were similar in sex, age, weight, height and BMI as expected by the matching design. However, the two groups were Abnormality, N (%) significantly different in all metabolic traits (Po0.05). The BP 0 11 (32.4%) proportion of abnormal metabolic status in the MAO group BSAC 0 18 (52.9%) ranged from one-third in blood pressure to two-thirds in HDLC. TG 0 13 (38.2%) The workflow of the current metabolomics study is shown in HDLC 0 24 (70.6%) Figure 1. Among the 382 metabolites in the library of LC-MS in the Abbreviations: BMI, body mass index; BSAC, fasting blood glucose; DBP, NTU core lab, 69 were detected in the studied subjects. Similarly, diastolic blood pressure; GOT, glutamate oxaloacetate transaminase; 69 metabolites out of 438 metabolites in the library of GC-MS were GPT, glutamate pyruvate transaminase; HDLC, high-density lipoprotein detected in the studied subjects. The intensity level of each cholestrol; MAO, metabolic abnormal obesity; MHO, metabolic healthy obesity; SBP, systolic blood pressure; TG, triglyceride; WC, waist circum- metabolite was analyzed using conditional logistic regression in stance. Abnormality: BP, SBP4130 mm Hg or DBP485 mm Hg; BSAC over 34 matched pairs of MHO and MAO subjects. In total, we included 110 mg dl−1; TG over 150 mg dl−1; HDL o40 for male and o50 for female. 138 metabolites that were identified from the LC-MS and GC-MS analysis. Results of 23 metabolites with a P-value of less than shortest paths going through the target metabolite, which is used to 0.10 are shown in Table 2. estimate the importance of each metabolite within a specific pathway. The Figure 2 shows a clear separation between MHO and MAO pathway impact is the cumulative value of the significant metabolites, and individuals using the 138 metabolites with three major compo- this value is normalized by summing the importance of each metabolite nents in PLS-DA. We found that most of the subjects were clearly in a particular pathway. These pathway analyses were conducted using separated into one of the two groups. In the PLS-DA analysis, 21 Metabolomics Pathway Analysis (MetPA) (http://metpa.metabolomics.ca/). most metabolites (91%) had loadings ranged from 0.2 to − 0.2, of which 11 metabolites from the top three PLS-DA components whose loadings are greater than 0.2 (or less than − 0.2) are listed in RESULTS Supplementary Table 1. In addition, results of leave-one-out The demographic characteristics and metabolic variables of validation demonstrated acceptable prediction error rates for 34 pairs of the MHO vs MAO subjects are presented in Table 1. the MHO and MAO groups. The error rate was 26.5% using the

Figure 2. The scatter plot of PLS-DA to distinguish MHO from MAO. The PLS-DA conducted based on 138 detectable metabolites. The results with three components can separate MHO (●) and MAO (▲) clearly.

138 metabolites, and 10.3% with the 23 relevant metabolites listed understanding in explaining the underlying metabolic regulation in Table 2. among obese individuals. Through mapping potential metabolites To perform further pathway analysis, we mapped 9 correlated to their respective metabolic pathways, several important path- metabolic pathways with the identified 23 metabolites listed in ways such as fatty acid biosynthesis, phenylalanine metabolism, Table 2, including fatty acid biosynthesis, pyrimidine metabolism, propanoate metabolism and valine, leucine and isoleucine galactose metabolism, valine, leucine and isoleucine degradation, degradation are identified. tryptophan metabolism, propanoate metabolism, phenylalanine Comparing the metabolome of MAO and MHO subjects, several metabolism, glyoxylate and dicarboxylate metabolism, citrate of the identified metabolites and related metabolism pathways cycle (Tricarboxylic acid cycle). Additional metabolites of these are related to energy balance and mitochondrial function. metabolic pathways, which were not originally included in the Metabolites such as acylcarnitines (including decanoylcarnitine library, were identified by their mass over charge value. More and hexanoylcarnitine) and branch chain amino acids (BCAA, detailed results of pathway analysis are listed in Table 3. Analysis including isoleucine and valine) demonstrate significantly differ- at the pathway level showed 407 metabolites involved in the 9 ent expression levels in the two groups (Table 2). Acylcarnitine is a pathways mentioned above. Among which, 145 were detected in mediator to transfer the catabolism products of fatty acid and the subjects (Figure 1). Conditional logistic regression was applied amino acid into mitochondria for β-oxidation,22 which is the key for the 145 metabolites, and the top 49 metabolites had a P-value transporter in the process of energy production. Excessive of less than 0.2 in the comparisons between MHO and MAO accumulation of acylcarnitine might indicate mitochondrial groups. Enrichment analysis was then conducted for the 49 dysfunction. We found higher levels of decanoylcarnitine and metabolites. Four pathways: fatty acid biosynthesis, phenylalanine, hexanoylcarnitine in MAO than in MHO individuals. Significantly propanoate, and valine, leucine and isoleucine degradation all had elevated levels of acylcarnitine, isoleucine and valine are also low P-values. In addition, two pathways, fatty acid biosynthesis reported in previous studies among obese and diabetes patients – – and phenylalanine, had high pathway impacts (over 0.2) obtained compared with healthy controls.12 14,23 27 Although diabetes from topology analysis (Figure 3). patients were excluded in this study, the MAO group exhibits a certain degree of dysglycemia and more than half of the MAO subjects have abnormal BSAC. Interestingly, the level of acylcarni- DISCUSSION tines is found to be correlated with BCAA supplement in rats,24 Among obese individuals, the fact that MAO has higher incidences and the intake of high protein in infants.28 Elevated BCAA levels of cardiovascular disease, diabetes and mortality compared with are also suggested to be a predictor of diabetes incidence.29 In that MHO,4,5 a lot of attention is directed to uncover the accordance with these findings, important metabolic pathways regulatory mechanisms that differentiate them. In this current (such as fatty acid biosynthesis, and valine, leucine and isoleucine study, we profiled the metabolomes of subjects with classified as degradation) are identified (Table 3). Moreover, the accumulation having MHO and MAO and identified several potential metabolic of the downstream of BCAA metabolism is reported to be a cause pathways that are involved in regulating metabolic traits among mitochondrial dysfunction,30 and decreased numbers of mito- obese. We reported a panel of metabolites that are important to chondria or abnormal mitochondria morphology are observed in distinguish MHO from MAO individuals, such as L-kynurenine, obese and diabetic patients.31 Results of these studies indicate the glycerol 1-phosphate, glycolic acid, tagatose, methyl palmitate, importance of mitochondrial function, the interrelationships uric acid, as well as several other metabolites. Using leave-one-out among our identified metabolites and their potential roles in cross validation, we also found that metabolome profiles could regulating abnormal metabolic status in obesity. successfully separate the two groups with a low error rate. In Propanoate metabolism, which was identified in our pathway addition to univariate analysis to estimate the effect of single analysis, also provides another evidence to link metabolically metabolites, pathway results may offer more biological abnormal obesity with mitochondrial dysfunction. It is important

Table 3. Results of pathway analysis

Pathway name Total number Detected metabolites Signiﬁcant metabolites Enrichment analysis Topology analysis P-value Impact factor

Fatty acid biosynthesis 49 10 7 0.022 0.29 Phenylalanine metabolism 45 23 10 0.242 0.26 Propanoate metabolism 35 18 8 0.259 0.14 Valine, leucine and isoleucine degradation 40 13 6 0.276 0.09 Pyrimidine metabolism 60 15 5 0.659 0.09 Citrate cycle (TCA cycle) 20 8 2 0.839 0.06 Galactose metabolism 41 20 5 0.900 0.17 Glyoxylate and dicarboxylate metabolism 50 14 2 0.984 0.03 Tryptophan metabolism 79 24 3 0.999 0.04

Figure 3. The pathway impact in topology analysis and P-value in enrichment analysis conducted by Metabolomics Pathway Analysis (MetPA). Forty-nine top metabolites (P-valueo0.2) were involved in the pathway analysis. The size of pathway symbols represents significance level of enrichment analysis, and the color of pathway symbols represents the impact factor. in the catabolism of valine. One of the important intermediates, phosphate is the product of the catabolism of triglyceride and (S)-Methylmalonatesemialdehyde, from valine degradation to serves as a phosphate donor to regenerate NADH from NAD+ with propanoate metabolism is also expressed differently between glycerol 1-phosphate dehydrogenase (mGPDH).36 The activity of the MHO and MAO groups (Supplementary Table 3). Moreover, mGPDH shows a positive correlation with BMI in humans.37 Mice propionyl-CoA, an intermediate of propanoate metabolism, is with the deletion of mGPDH gene present higher glycerol and reported to have a toxic effect on mitochondrial function by triglyceride levels, and lower blood glucose than wild-type mice.38 suppressing pyruvate oxidation.32 Although we did not examine Moreover, mGDPH is thought to be a source of reactive oxygen propionyl-CoA in the current study, several upstream and down- species from mitochondria,39 and mice with high reactive oxygen stream metabolites of this pathway show different levels in the species expression usually present increased weight, reduced HDL two groups using our untargeted metabolome analysis (such as and impaired glucose tolerance. Thus, it is also suggested propanoate, propanoylphosphate and 2-oxobutanoate 2-methyl- that reactive oxygen species are involved in the development citrate in Supplementary Table 3). One disorder of deficient of obesity and metabolic syndrome.40 Glycolic acid (or called propanoate metabolism, propionic acidemia, has also been found glycolate) is a major precursor of oxalate.41 Oxalate has been to induce dysglycemia,33 further supporting the involvement of found to induce more reactive oxygen species from propanoate metabolism pathway in metabolic abnormality. mitochondria.42 The concentration of urinary oxalate is also Fatty acid biosynthesis has the highest impact score in our reported to be correlated with BMI.43 These aforementioned pathway analysis (Figure 3). Malonyl-acyl-carrierprotein, the study findings altogether support the relationship between immediate downstream metabolite of malonyl-CoA, exhibits mitochondrial function and abnormal metabolic status in our different levels between the MAO and MHO groups samples, and the alteration of identified metabolites or dysfunc- (Supplementary Table 2). Malonyl-CoA is an important regulator tion of related pathways may provide additional clues in under- for promoting fatty acid biosynthesis and to inhibit the transfer standing the regulatory mechanisms underlying metabolic of fatty acid into mitochondria for oxidization,34 which induces abnormal obesity. lipogenesis and stimulates glucose oxidation to response to Phenylalanine is identified as an important metabolite for acetyl-CoA carboxylase activated by insulin. Interestingly, higher distinguishing MAO from MHO (Supplementary Table 1), and the levels of malonyl-CoA and low rates of fatty acid oxidation have downstream product of phenylalanine catabolism, L-tyrosine, also been found in obese subjects.35 exhibits significant difference between the two groups (Table 2). Two other significant metabolites, glycerol 1-phosphate and High levels of phenylalanine are observed in obese subjects of glycolic acid, are also related to mitochondrial function. Glycerol 1- previous metabolomics studies.24,44 In addition, the phenylalanine

© 2015 Macmillan Publishers Limited International Journal of Obesity (2015) 1241 – 1248 Metabolome profiling for metabolic healthy obesity H-H Chen et al 1246 metabolism pathway has a high score in topology analysis (0.26), metabolites we reported are consistent with findings from indicating that significant metabolites in this pathway have a high previous metabolomic studies of patients with diabetes or obesity. degree of interaction with other molecules in the same pathway. Third, even with a library containing more than 800 metabolites The liver is an important organ for phenylalanine metabolism, and coupled with non-targeting platforms, it is possible that more the rate of phenylalanine metabolism is often used to test liver metabolites can be identified with better instruments and function.45 Our finding in phenylalanine metabolism might analytical conditions. Fourth, the pathway analysis is based on indicate the involvement of liver dysfunction in abnormal the KEGG data set. Although KEGG data set maximizes the metabolic status among obese individuals. Alanine aminotransfer- collection of metabolites and reactions, some of them might not ase (ALT) is an important enzyme of the liver and a biomarker for be identified in humans and require manually checking. In liver function. The concentration of ALT is strongly correlated with addition, in untargeted analysis, one peak may correspond to the incidence of cardiovascular disease, metabolic syndrome and several metabolites with a similar mass over charge value, which every component in metabolic syndrome in a longitudinal study.46 may result in less inaccurate pathway analysis results. Finally, the Another study reported a strong association between the definition of MAO may result in heterogeneous metabolic concentration of blood phenylalanine and ALT.44 In addition, phenotypes in our samples. The different abnormal metabolic phenylalanine is higher in obese subjects than in normal traits may have different underlying causes. Although we aimed to controls.24,44 Both phenylalanine and ALT levels decrease after identify important metabolites and common causes for abnormal performing bariatric surgery.44 Therefore, liver function is impli- metabolic status in obese subjects, it is likely that some important cated in metabolic regulation in obesity and may involve the metabolites/pathways that are responsible are unique for the pathway of phenylalanine metabolism. In our study samples, we individual metabolic trait and cannot be directly examined in our also observed slightly higher ALT and aspartate aminotransferase study design. levels in the MAO compared with the MHO groups (Table 1). In conclusion, using a matched study design, we have There are several other metabolites we reported in Table 2 successfully identified a handful of relevant metabolites that worth further discussion for their associations with metabolic showed differences between MAO and MHO groups, including status. We reported that L-kynurenine and methyl palmitate are L-kynurenine, GPC, glycerol 1-phosphate, glycolic acid, tagatose, increased in the MAO group, while benzoic acid shows decreased methyl palmitate and uric acid. These metabolites robustly levels. Kynurenine is derived from tryptophan, and methyl discriminate MAO from MHO with a low prediction error rate, palmitate is a perivascular adipose tissue-derived relaxing and are involved in several important metabolic pathways, such as factor.47 Both metabolites exert their functions in vascular tone fatty acid biosynthesis, phenylalanine metabolism, propanoate to reduce blood pressure in rats.47 Elevated kynurenic acid, the metabolism, and valine, leucine and isoleucine degradation. Our downstream metabolite of kynurenine, is also found to increase results are consistent with findings of previous metabolomic the risk of preeclampsia, which is characterized by high blood studies related insulin resistance, diabetes and obese subjects. 48 pressure. On the other hand, we found that glycerophosphocho- So far, the mechanisms for the changes of metabolites from line (GPC) is significantly decreased in MAO subjects. GPC is normal to abnormal metabolic phenotypes are not yet clear. We 49 enriched in pancreatic islets cells in both human and rats, and is suggest that mitochondrial and liver functions may be involved in 50 reported to be involved in β-cell signaling. The concentration of the metabolic regulation in obese subjects via metabolomics GPC is correlated with the development of insulin resistance in profiles identified in this current study. More replication and 51 rats. Additionally, we found significant differences in tagatose validation studies are needed to better understand the underlying and uric acid between the MAO and MHO groups. Uric acid is a mechanisms for the dysregulation of metabolic traits present in 52 well-known predictor of cardiovascular diseases, and is related obesity. to obesity, metabolic syndrome and hypertension.53,54 Tagatose is an epimer of fructose, which exists in dairy products.55 A significant weight loss and HDL improvement are observed in CONFLICT OF INTEREST 56 diabetes patients diet with supplement of tagatose. Tagatose is The authors declare no conflict of interest. a drug currently under development for the treatment of diabetes and to control obesity.57 Finally, benzoic acid presents low level in the MAO group. Benzoic acid is a gut microbial co-metabolite of ACKNOWLEDGEMENTS hippurate, and the lower level of hippurate in urine has been This study was supported by grants NCKUH-9801002, NSC 99-2314-B-002-140-MY3 reported in obese subjects and diabetes patients.58,59 Because the and 102-2314-B-002-117-MY3. We thank Ms. Yu-Chen Shih, Sheng-Chi Lee and Shih- intestinal microbiota holds the key to affect nutrition extraction, Han Hsu at National Cheng Kung University for administrative assistance, data different compositions of microbiota have been found in lean collection and sample preparation and experiments. We also thank Prof. Ching-hua versus obese individuals.60 The diets of vegetables are reported to Kuo, Mr. Wen-Hsin Huang, Han-Chun Kuo, Jiawei Liu and Cheng-En Tan at the be correlated with elevated extraction of hippurate.61 Thus, our Metabolomics Core Laboratory at the Center of Genomic Medicine of National Taiwan University for performing metabolomics experiments and analysis. finding of different levels of benzoic acid may reflect the differences in dietary patterns and microbiota between the MHO and MAO groups, though we did not collect these data in the REFERENCES current study. 1 Mokdad AH, Serdula MK, Dietz WH, Bowman BA, Marks JS, Koplan JP. The spread Despite several interesting findings that explain the differences of the obesity epidemic in the United States, 1991-1998. JAMA 1999; 282: between the MAO and MHO groups, there are several limitations 1519–1522. in the current study. First, we had a moderate sample size for 2 Aguilar-Salinas CA, Garcia EG, Robles L, Riano D, Ruiz-Gomez DG, performing the metabolomics experiments. This might reduce our Garcia-Ulloa AC et al. High adiponectin concentrations are associated with the power to detect metabolites with minor effects. To reach a power metabolically healthy obese phenotype. J Clin Endocrinol Metab 2008; 93: of 0.8, the corresponding effect size (β) is 0.68 with our sample 4075–4079. size. Second, subjects were recruited from the weight manage- 3 Wildman RP, Muntner P, Reynolds K, McGinn AP, Rajpathak S, Wylie-Rosett J et al. ment clinic, and our results may not be able to directly generalize The obese without cardiometabolic risk factor clustering and the normal weight with cardiometabolic risk factor clustering: prevalence and correlates of 2 phe- to obesity individuals in the general population. Although we had notypes among the US population (NHANES 1999-2004). Arch Intern Med 2008; matched several potential confounding factors, such as sex, age 168: 1617–1624. and BMI in the current study, it is not guaranteed that our findings 4 Hamer M, Stamatakis E. Metabolically healthy obesity and risk of all-cause and are free of potential bias. Nevertheless, several significant cardiovascular disease mortality. J Clin Endocrinol Metab 2012; 97: 2482–2488.

International Journal of Obesity (2015) 1241 – 1248 © 2015 Macmillan Publishers Limited Metabolome profiling for metabolic healthy obesity H-H Chen et al 1247 5 Appleton SL, Seaborn CJ, Visvanathan R, Hill CL, Gill TK, Taylor AW et al. Diabetes 29 Wang TJ, Larson MG, Vasan RS, Cheng S, Rhee EP, McCabe E et al. and cardiovascular disease outcomes in the metabolically healthy obese Metabolite profiles and the risk of developing diabetes. Nat Med 2011; 17: phenotype: a cohort study. Diabetes Care 2013; 36: 2388–2394. 448–453. 6 Karelis AD, Faraj M, Bastard JP, St-Pierre DH, Brochu M, Prud'homme D et al. 30 Lynch CJ, Adams SH. Branched-chain amino acids in metabolic signalling and The metabolically healthy but obese individual presents a favorable inflammation insulin resistance. Nat Rev Endocrinol 2014; 10: 723–736. profile. J Clin Endocrinol Metab 2005; 90: 4145–4150. 31 Ritov VB, Menshikova EV, He J, Ferrell RE, Goodpaster BH, Kelley DE. Deficiency of 7 Brochu M, Tchernof A, Dionne IJ, Sites CK, Eltabbakh GH, Sims EA et al. What are subsarcolemmal mitochondria in obesity and type 2 diabetes. Diabetes 2005; 54: the physical characteristics associated with a normal metabolic profile despite a 8–14. high level of obesity in postmenopausal women? J Clin Endocrinol Metab 2001; 86: 32 Schwab MA, Sauer SW, Okun JG, Nijtmans LGJ, Rodenburg RJT, van den Heuvel LP 1020–1025. et al. Secondary mitochondrial dysfunction in propionic aciduria: a pathogenic 8 Preet A, Karve TM, Rizk N, Cheema AK. Metabolomics: approaches and applica- role for endogenous mitochondrial toxins. Biochem J 2006; 398: 107–112. tions to diabetes research. J Diabetes Metab 2012; S6: 001. 33 Dweikat IM, Naser EN, Abu Libdeh AI, Naser OJ, Abu Gharbieh NN, Maraqa NF 9 Connor SC, Hansen MK, Corner A, Smith RF, Ryan TE. Integration of metabolomics et al. Propionic acidemia mimicking diabetic ketoacidosis. Brain Dev 2011; 33: and transcriptomics data to aid biomarker discovery in type 2 diabetes. Mol 428–431. BioSyst 2010; 6: 909. 34 Foster DW. Malonyl-CoA: the regulator of fatty acid synthesis and oxidation. J Clin 10 Gu S, A J, Wang G, Zha W, Yan B, Zhang Y et al. Metabonomic profiling Invest 2012; 122: 1958–1959. of liver metabolites by gas chromatography-mass spectrometry and 35 Bandyopadhyay GK, Yu JG, Ofrecio J, Olefsky JM. Increased malonyl-CoA levels in its application to characterizing hyperlipidemia. Biomed Chromatogr 2010; 24: muscle from obese and type 2 diabetic subjects lead to decreased fatty acid 242–252. oxidation and increased lipogenesis; thiazolidinedione treatment reverses these 11 Kim JY, Park JY, Kim OY, Ham BM, Kim H-J, Kwon DY et al. Metabolic profiling of defects. Diabetes 2006; 55:2277–2285. plasma in overweight/obese and lean men using ultra performance liquid 36 Mracek T, Drahota Z, Houstek J. The function and the role of the mitochondrial chromatography and Q-TOF mass spectrometry (UPLC-Q-TOF MS). J Proteome Res glycerol-3-phosphate dehydrogenase in mammalian tissues. Biochim Biophys Acta 2010; 9: 4368–4375. 2013; 1827:401–410. 12 Mihalik SJ, Goodpaster BH, Kelley DE, Chace DH, Vockley J, Toledo FGS et al. 37 Swierczynski J, Zabrocka L, Goyke E, Raczynska S, Adamonis W, Sledzinski Z. Increased levels of plasma acylcarnitines in obesity and type 2 diabetes Enhanced glycerol 3-phosphate dehydrogenase activity in adipose tissue of and identification of a marker of glucolipotoxicity. Obesity (Silver Spring) 2009; 18: obese humans. Mol Cell Biochem 2003; 254:55–59. 1695–1700. 38 Alfadda A, DosSantos RA, Stepanyan Z, Marrif H, Silva JE. Mice with deletion of the 13 Mihalik SJ, Michaliszyn SF, de Las Heras J, Bacha F, Lee S, Chace DH et al. Meta- mitochondrial glycerol-3-phosphate dehydrogenase gene exhibit a thrifty phe- bolomic profiling of fatty acid and amino acid metabolism in youth with obesity notype: effect of gender. Am J Physiol Regul Integr Comp Physiol 2004; 287: and type 2 diabetes: evidence for enhanced mitochondrial oxidation. Diabetes R147–R156. Care 2012; 35: 605–611. 39 Mracek T, Pecinova A, Vrbacky M, Drahota Z, Houstek J. High efficiency of ROS 14 Wang C, Feng R, Sun D, Li Y, Bi X, Sun C. Metabolic profiling of urine in young production by glycerophosphate dehydrogenase in mammalian mitochondria. obese men using ultra performance liquid chromatography and Q-TOF mass Arch Biochem Biophys 2009; 481:30–36. spectrometry (UPLC/Q-TOF MS). J Chromatogr B Analyt Technol Biomed Life Sci 40 Youn JY, Siu KL, Lob HE, Itani H, Harrison DG, Cai H. Role of vascular oxidative 2011; 879: 2871–2876. stress in obesity and metabolic syndrome. Diabetes 2014; 63: 2344–2355. 15 Tan C-E, Ma S, Wai D, Chew S-K, Tai E-S. Can we apply the National Cholesterol 41 Poore RE, Hurst CH, Assimos DG, Holmes RP. Pathways of hepatic oxalate Education Program Adult Treatment Panel definition of the metabolic syndrome synthesis and their regulation. Am J Physiol 1997; 272: C289–C294. to Asians? Diabetes Care 2004; 27: 1182–1186. 42 Cao L-C, Honeyman TW, Cooney R, Kennington L, Scheid CR, Jonassen JA. Mito- 16 Kuk JL, Ardern CI. Are metabolically normal but obese individuals at lower risk for chondrial dysfunction is a primary event in renal cell oxalate toxicity. Kidney Int all-cause mortality? Diabetes Care 2009; 32: 2297–2299. 2004; 66:1890–1900. 17 Ho TJ, Kuo CH, Wang SY, Chen GY, Tseng YJ. True ion pick (TIPick): a denoising 43 Taylor EN, Curhan GC. Body size and 24-hour urine composition. Am J Kidney Dis and peak picking algorithm to extract ion signals from liquid chromatography/ 2006; 48:905–915. mass spectrometry data. J Mass Spectrom 2013; 48:234–242. 44 Swierczynski J, Sledzinski T, Slominska E, Smolenski R, Sledzinski Z. Serum phe- 18 Bartel J, Krumsiek J, Theis FJ. Statistical methods for the analysis nylalanine concentration as a marker of liver function in obese patients before of high-throughput metabolomics data. Comput Struct Biotechnol J2013; 4: and after bariatric surgery. Obes Surg 2008; 19:883–889. e201301009. 45 Lara Baruque S, Razquin M, Jimenez I, Vazquez A, Gisbert JP, Pajares JM. 13C- 19 Dunn WB, Erban A, Weber RJM, Creek DJ, Brown M, Breitling R et al. Mass appeal: phenylalanine and 13C-methacetin breath test to evaluate functional capacity of metabolite identification in mass spectrometry-focused untargeted metabo- hepatocyte in chronic liver disease. Dig Liver Dis 2000; 32:226–232. lomics. Metabolomics 2012; 9:44–66. 46 Goessling W, Massaro JM, Vasan RS, D'Agostino RB Sr., Ellison RC, Fox CS. 20 Kanehisa M, Goto S, Sato Y, Furumichi M, Tanabe M. KEGG for integration and Aminotransferase levels and 20-year risk of metabolic syndrome, diabetes, and interpretation of large-scale molecular data sets. Nucleic Acids Res 2011; 40: cardiovascular disease. Gastroenterology 2008; 135: 1935–1944 1944 e1. D109–D114. 47 Lee YC, Chang HH, Chiang CL, Liu CH, Yeh JI, Chen MF et al. Role of Perivascular 21 Xia J, Wishart DS. MetPA: a web-based metabolomics tool for pathway analysis Adipose Tissue-Derived Methyl Palmitate in Vascular Tone Regulation and and visualization. Bioinformatics 2010; 26: 2342–2344. Pathogenesis of Hypertension. Circulation 2011; 124:1160–1171. 22 Eaton S, Bartlett K, Pourfarzam M. Mammalian mitochondrial beta-oxidation. 48 Nilsen RM, Bjorke-Monsen AL, Midttun O, Nygard O, Pedersen ER, Ulvik A et al. Biochem J 1996; 320:345–357. Maternal tryptophan and kynurenine pathway metabolites and risk of 23 Gall WE, Beebe K, Lawton KA, Adam K-P, Mitchell MW, Nakhle PJ et al. α-Hydro- preeclampsia. Obstet Gynecol 2012; 119: 1243–1250. xybutyrate is an early biomarker of insulin resistance and glucose intolerance in a 49 Ramanadham S, Hsu FF, Bohrer A, Nowatzke W, Ma Z, Turk J. Electrospray ioni- nondiabetic population. PLoS ONE 2010; 5: e10883. zation mass spectrometric analyses of phospholipids from rat and human 24 Newgard CB, An J, Bain JR, Muehlbauer MJ, Stevens RD, Lien LF et al. A branched- pancreatic islets and subcellular membranes: comparison to other tissues and chain amino acid-related metabolic signature that differentiates obese implications for membrane fusion in insulin exocytosis. Biochemistry 1998; 37: and lean humans and contributes to insulin resistance. Cell Metab 2009; 9: 4553–4567. 311–326. 50 Konrad RJ, Major CD, Wolf BA. Diacylglycerol hydrolysis to arachidonic acid is 25 Wahl S, Yu Z, Kleber M, Singmann P, Holzapfel C, He Y et al. Childhood obesity is necessary for insulin secretion from isolated pancreatic islets: sequential associated with changes in the serum metabolite profile. Obes Facts 2012; 5: actions of diacylglycerol and monoacylglycerol lipases. Biochemistry 1994; 33: 660–670. 13284–13294. 26 Tai ES, Tan ML, Stevens RD, Low YL, Muehlbauer MJ, Goh DL et al. Insulin resis- 51 Hsu FF, Bohrer A, Wohltmann M, Ramanadham S, Ma Z, Yarasheski K et al. Elec- tance is associated with a metabolic profile of altered protein metabolism in trospray ionization mass spectrometric analyses of changes in tissue phospholipid Chinese and Asian-Indian men. Diabetologia 2010; 53:757–767. molecular species during the evolution of hyperlipidemia and hyperglycemia in 27 Shaham O, Wei R, Wang TJ, Ricciardi C, Lewis GD, Vasan RS et al. Metabolic Zucker diabetic fatty rats. Lipids 2000; 35:839–854. profiling of the human response to a glucose challenge reveals distinct 52 Gagliardi ACM, Miname MH, Santos RD. Uric acid: a marker of increased axes of insulin sensitivity. Mol Syst Biol 2008; 4: 214. cardiovascular risk. Atherosclerosis 2009; 202:11–17. 28 Kirchberg FF, Harder U, Weber M, Grote V, Demmelmair H, Peissner W et al. 53 Choi HK, Atkinson K, Karlson EW, Curhan G. Obesity, weight change, hypertension, Dietary protein intake affects amino Acid and acylcarnitine metabolism in infants diuretic use, and risk of gout in men: the health professionals follow-up study. aged 6 months. J Clin Endocrinol Metab 2015; 100: 149–158. Arch Intern Med 2005; 165:742–748.

© 2015 Macmillan Publishers Limited International Journal of Obesity (2015) 1241 – 1248 Metabolome profiling for metabolic healthy obesity H-H Chen et al 1248 54 Coutinho Tde A, Turner ST, Peyser PA, Bielak LF, Sheedy PF 2nd, 58 Calvani R, Miccheli A, Capuani G, Tomassini Miccheli A, Puccetti C, Delfini M et al. Kullo IJ. Associations of serum uric acid with markers of inflammation, metabolic Gut microbiome-derived metabolites characterize a peculiar obese urinary syndrome, and subclinical coronary atherosclerosis. Am J Hypertens 2007; 20: metabotype. Int J Obes (Lond) 2010; 34: 1095–1098. 83–89. 59 Salek RM, Maguire ML, Bentley E, Rubtsov DV, Hough T, Cheeseman M et al. A 55 Levin GV. Tagatose, the new GRAS sweetener and health product. J Med Food metabolomic comparison of urinary changes in type 2 diabetes in mouse, rat, 2002; 5:23–36. and human. Physiol Genomics 2007; 29:99–108. 56 Donner TW, Magder LS, Zarbalian K. Dietary supplementation with d-tagatose in 60 Turnbaugh PJ, Hamady M, Yatsunenko T, Cantarel BL, Duncan A, Ley RE et al. A subjects with type 2 diabetes leads to weight loss and raises high-density core gut microbiome in obese and lean twins. Nature 2009; 457: 480–484. lipoprotein cholesterol. Nutr Res 2010; 30: 801–806. 61 Walsh MC, Brennan L, Pujos-Guillot E, Sebedio JL, Scalbert A, Fagan A et al. 57 Lu Y, Levin GV, Tagatose Donner TW. a new antidiabetic and obesity control drug. Influence of acute phytochemical intake on human urinary metabolomic profiles. Diabetes Obes Metab 2008; 10:109–134. Am J Clin Nutr 2007; 86:1687–1693.

Supplementary Information accompanies this paper on International Journal of Obesity website (http://www.nature.com/ijo)