The metabolic analysis of psoriasis identifies the associated metabolites while providing computational models for the monitoring of the disease
The majority of studies on psoriasis have focused on explaining the genetic background and its associations with the immune system’s response. The aim of this study was to identify the low-molecular weight compounds contributing to the metabolomic profile of psoriasis and to provide computational models that help with the classification and monitoring of the severity of the disease. We compared the results from targeted and untargeted analyses of patients’ serums with plaque psoriasis to controls. The main differences were found in the concentrations of acylcarnitines, phosphatidylcholines, amino acids, urea, phytol, and 1,11-undecanedicarboxylic acid. The data from the targeted analysis were used to build classification models for psoriasis. The results from this study provide an overview of the metabolomic serum profile of psoriasis along with promising statistical models for the monitoring of the disease.
KeywordsMetabolomics Psoriasis Targeted analysis Untargeted analysis Computational model
Psoriasis is an immune-mediated skin disorder, where in addition to visible scaly inflamed plaques on the skin, the joints and nails might also be affected. The cause of psoriasis remains uncertain, but it is known that a genetic predisposition accompanied with environmental factors, such as stress, smoking, and alcohol abuse, can lead to the development of the disease. Psoriasis patients are also at a higher risk for metabolic syndrome, cardiovascular diseases, and overall morbidity . The past work on psoriasis has focused mainly on the genetical background and immunology of the disease with fewer papers published on metabolomics [6, 31]. Metabolomics is an emerging field in the omics family that concerns with the identification and quantitation of small molecules including amino acids, carbohydrates and their derivatives, biogenic amines, lipids, and more. Current work on the metabolomics of psoriasis is very limited with only a number of published papers. Kamleh and his colleagues have discovered changes in free-circulating amino acids, namely, arginine, proline, alanine, glutamate, aspartate, glycine, serine, and threonine, which levels are elevated in the plasma of patient psoriasis. The levels of amino acids revert to normal after the biological treatment with TNFα receptor blocker Etanercept . In another study by Armstrong et al., higher concentrations of alpha ketoglutaric acid and glucuronic acid and lower levels of asparagine and glutamine were determined in psoriasis patients’ serum. In addition, patients with both psoriasis and psoriatic arthritis showed higher levels of lignoceric acid and a lower level of alpha ketoglutaric acid in the serum when compared to patients with psoriasis alone .
Metabolomics employs many different methods to analyze samples which vary depending on the medium, collection methods and time [25, 60]. Therefore, using varied approaches to measurements can yield in a higher coverage of available metabolites. In this study, serum samples from psoriasis patients and controls were analyzed using a targeted approach that analyses the concentrations of known metabolites, e.g., amino acids, lipids, biogenic amines, etc. and an untargeted profiling that helps to discover metabolites not included in the targeted method.
Metabolomics produces a lot of data for every measured sample which makes data interpretation challenging. Many different mathematical methods have been used including partial least squares discriminant analysis or principal component analysis (PCA) . Using machine learning and algorithms for metabolomic profiling is a promising solution for data interpretation and automation [35, 41].
Materials and methods
Recruitment of volunteers
For untargeted analysis, a total of 40 volunteers were included in this study—20 diagnosed with plaque psoriasis and 20 age and sex-matched controls (13 men, 7 women, age range 20–75). For targeted analysis, the number of volunteers were expanded to 106—55 psoriasis patients (37 men, 18 women, age range 20–75) and 51 controls (15 women, 36 men, age range 23–75). The subjects with plaque psoriasis recruited for this study were patients in the Clinic of Dermatology at the University Hospital of Tartu. The participating subjects were diagnosed with plaque psoriasis by dermatologists. Psoriasis Area and Severity Index (PASI) scores were calculated ranging from 1 to 34 to allow the analysis of a wide disease span. Controls were recruited either from the same clinic or from the Clinic of Traumatology. Exclusion criteria for the patients and controls included any other skin diseases, diabetes, gout, hypertension, and the taking of prescription medication. Detailed questionnaires were filled out that covered age, sex, skin type, comorbidities, smoking status, and PASI scores.
Blood sample collection and storing
Fasting blood samples were collected before breakfast in the morning into 5 ml Vacutainer (REF 367614) tubes that have micronized silica particles for the acceleration of the clotting process. The collected blood was left to clot for 1 h at room temperature after which it was centrifuged at 1300×g for 20 min. The supernatant serum was aliquoted into 300 µl fractions and placed in the freezer at −80 °C.
HPLC-grade solvents [acetonitrile, water, and formic acid (FA)] were purchased from Sigma-Aldrich (Germany). For the targeted approach, an Agilent Zorbax Eclipse XDB C18, 3.0 × 100 mm, 3.5 µm with Pre-Column SecurityGuard, Phenomenex, C18, 4 × 3 mm was used with the AbsoluteIDQ p180 kit (Biocrates Life Sciences AG, Innsbruck, Austria). For chromatographic separation in the untargeted part, a SeQuant® ZIC®-pHILIC (5 µm polymer) PEEK 150 × 4.6 mm metal-free HPLC column and ZIC®-pHILIC Guard column PEEK 20 × 2.1 m were used.
For targeted analysis, the serum samples were thawed on ice and prepared according to the specifications detailed in the AbsoluteIDQ p180 kit’s user manual. Shortly, 10 µl of serum was pipetted onto 96-well plate filter inserts, internal standards added, and samples dried under nitrogen and derivatized using phenylisothiocyanate. The samples were measured using a combination of flow injection analysis and through a C18 column. The prepared kit plate was analyzed on a QTRAP 4500 (ABSciex, USA) mass spectrometer which was coupled to an Agilent 1260 series HPLC (USA). The results from the analysis were quantified concentrations of different acylcarnitines, amino acids, biogenic amines, hexose, glycerophospholipids, sphingolipids, and ratios of different metabolites.
Blood samples were left to thaw on ice. 100 µl of serum was pipetted into a new Eppendorf tube, where 400 µl of acetonitrile was added for protein precipitation. The tube was vortexed vigorously for 2 min and left for 15 min at room temperature. The mixture was centrifuged for 15 min at 15,800×g and 4 °C. The supernatant was transferred to a clean tube, the samples were randomized, and 10 µl of sample was used for analysis. A Shimadzu HPLC (Japan) was coupled to a 3200 QTRAP (ABSciex, USA) mass spectrometer, where the parameters were as follows: solvents used were acetonitrile + 0.1% FA, water + 0.1% FA, runtime 62 min, gradient flow rate 0.3 ml/min from 80% acetonitrile to 20% in 32 min, to 5% in 1 min, at 5% for 8 min, then to 100% in 5 min, at 100% for 8 min, then re-equilibration at 80% for 8 min. The turbo spray’s curtain gas was set to 10 au, collision gas to “High”, ionspray voltage to 4500 V, temperature to 300 °C, declustering potential to 20 V, entrance potential to 10 V, and collision energy to 10 V in measurements and 20, 30, or 40 V in fragmentation analysis. The samples were measured in both positive and negative mode from 50 to 1500 mass-to-charge ratios (m/z). Fragmentation analysis was performed using the same settings and column but in Enhanced Product Ion mode for the statistically significant masses.
Identification of metabolites
The spectra from the fragmentation analysis were compared to spectra from public databases METLIN , HMDB , MassBank , and LipidMaps . A compound was considered identified when the fragmentation spectra, its peaks, and relative heights of peaks of a certain m/z were identical to a spectrum from an online database.
For the untargeted analysis, the acquired .wiff files were converted to .mzXML using the MSConvert software . The data were analyzed in RStudio version 0.98.501 , where peaks were extracted using XCMS  and further processed using mzMatch.R  which included the combining of biological replicates, Reproducibility Standard Deviation filtering, retention time correction, blank filtering, gap filling, filtering on the number of detections (minimum of six), and matching of related peaks. After data processing, a Wilcoxon–Mann–Whitney test was performed to determine the peaks and their corresponding retention times that differ statistically differently between subjects with plaque psoriasis and controls. The differentiating m/z-s were selected for fragmentation analysis. For targeted analysis, a Wilcoxon–Mann–Whitney test was used to determine which metabolites differ statistically significantly between controls and subjects with plaque psoriasis.
Principal component analysis was applied for the visualization of general differences between the groups on data from both targeted and untargeted analysis measurements .
Metadata and pre-processing
The collected data are suitable for the purposes of modeling, since the disease and control classes are of equal size. Although there are twice as many males, the distribution of ages and diagnosis is very similar between both sexes (Figure S1). Prior to analysis, metabolites, which showed zero or close to zero variance between samples, were removed. Scaling and centering was applied to all values in the data set.
For the accurate discrimination between psoriasis patients and controls, we used a popular machine learning and bioinformatics field’s method—random forest algorithm . It has been shown to work well in a wide variety of biological problems including the identifying of regulatory regions , classifying metabolomics data , and selecting highly reliable biomarkers for the diagnosis of Alzheimer’s disease . Random forest is an extension of another machine learning technique—decision tree  which builds a series of conditions (decisions) that best describe the underlying distribution of classes. Generated by the algorithm the series of decisions can be visualized as a tree-like structure. Random forest is an ensemble of decision trees built using random subsets of original training data.
Feature selection and cross validation
To select only relevant metabolite features for our classifications, we used three feature selection methods. Two were wrapper methods: a genetic algorithm  and a recursive feature elimination and one was a filter method . All three methods were used in parallel to cross-check the results. All of them used external cross validation  to avoid unwanted bias that could be introduced by aggressive feature selection procedures.
Due to the lack of training data, the use of fivefold cross-validation strategy was applied as suggested by Ambroise et al. . We only report average area under receiver operating characteristic (AUROC) values for our feature selection models. We used random forest with all three feature selection techniques. Modeling was implemented using R version 3.3.1 and package ‘caret’ 6.0.71 .
Statistically significantly different metabolites and their ratios from targeted analysis
Psoriasis mean μM ± SD
Control mean μM ± SD
0.88 ± 0.37
0.51 ± 0.27
Fraction of sulfoxidized methionine of the unmodified methionine pool
0.04 ± 0.02
0.02 ± 0.01
0.04 ± 0.01
0.05 ± 0.01
92.85 ± 66.43
49.06 ± 22.76
Ratio of citrulline to ornithine
0.37 ± 0.13
0.44 ± 0.12
ratio of acetylcarnitine to free carnitine
0.17 ± 0.08
0.22 ± 0.08
Ratio of short-chain acylcarnitines to free carnitine
0.18 ± 0.08
0.23 ± 0.08
Phosphatidylcholine diacyl C36:6
0.68 ± 0.27
0.89 ± 0.33
Ratio of esterified to free carnitine
0.25 ± 0.1
0.31 ± 0.1
Phosphatidylcholine acyl-alkyl C38:0
1.73 ± 0.5
2.17 ± 0.69
0.019 ± 0.006
0.024 ± 0.008
99.79 ± 29.44
82.28 ± 20.85
Phosphatidylcholine acyl-alkyl C40:6
3.39 ± 0.99
4.02 ± 1.01
Ratio of putrescine to ornithine
0.001 ± 0.001
0.002 ± 0.001
Phosphatidylcholine diacyl C36:5
24.78 ± 13.25
34.34 ± 19.98
82.91 ± 18.96
72.46 ± 13.51
[C16 + C18]/C0
Ratio of long-chain acylcarnitines to free carnitine
0.004 ± 0.001
0.005 ± 0.001
0.1 ± 0.036
0.124 ± 0.051
0.069 ± 0.021
0.076 ± 0.02
Phosphatidylcholine diacyls (PC aa) C36:5/C36:6 and phosphatidylcholine acyl-alkyls (PC ae) C38:0/C40.6 all showed higher levels in controls’ serum. Statistically significantly differing levels of amino acids were found to be for glutamate (Glu), ornithine (Orn), phenylalanine (Phe), and methioninesulfoxide (Met.SO). Amino acid concentrations in samples were statistically significantly higher in psoriasis subjects compared to controls. Ratios of acylcarnitine to free carnitine (C2…C0), short-chain acylcarnitines to free carnitine (X.C2.C3…C0), citrulline to ornithine (Cit…Orn), esterified to free carnitine (Total.AC…C0), putrescine to ornithine (Putrescine…Orn), and long-chain acylcarnitines to free carnitine ([C16 + C18]/C0) were all statistically significantly higher in controls, whereas the levels of the fraction of sulfoxidized methionine of the unmodified methionine pool (Met.SO…Met) were higher in psoriasis patients.
Statistically significantly different m/z-s from untargeted analysis
Intensity levels higher in psoriasis or controls?
No match in databases
PC(16:0/18:2) + FA
No match in databases
No match in databases
No match in databases
No match in databases
No match in databases
No match in databases
No match in databases
No match in databases
No match in databases
No match in databases
A random forest classifier with three different feature selection methods was trained. Each combination of random forest and feature selection method has been trained on the whole data set using fivefold cross validation to ensure fair model performance estimate.
The elevated levels of amino acids in the serum of psoriasis patients established in this present study can be linked to the higher demand for amino acids in the hyperproliferative epidermis, where de novo synthesis of proteins is upregulated and the rate of mitosis in basal keratinocytes is increased compared to non-lesional skin . In our study, all of the statistically significantly different levels of amino acids are from the non-essential amino acid pool with the exception of phenylalanine. Ornithine is part of the urea cycle and its levels in the serum were discovered to be over-expressed in psoriatic patients. Arginase I (EC 18.104.22.168) catalyzes arginine hydrolysis to urea and ornithine and has been demonstrated to be a limiting factor for cell proliferation . Higher activity of arginase has been demonstrated in psoriatic skin  which can be linked to the hyperproliferating keratinocytes and higher concentrations of ornithine. The end product of the ornithine cycle is urea that we discovered in the untargeted analysis to have higher levels in psoriatic patients’ serum. The ratio of the fraction of symmetrically dimethylated arginine of the unmodified arginine pool nor ADMA showed statistically significant differences in serums (p values 0.35 and 0.31, respectively), thus contradicting the recent results by Bilgic et al. . This illustrates the variability of psoriasis cohorts and the necessity for an even larger cohort studies to confirm the results of either study. Glutamate which level is significantly higher in psoriasis patients samples (p = 0.002) can be converted by glutamine synthetase (EC 22.214.171.124) to glutamine which is subsequently converted by carbamoyl-phosphate synthetase (EC 126.96.36.199) to carbamoyl-P which is then converted by ornithine carbamoyltransferase (EC 188.8.131.52) to citrulline that is an essential component of the urea cycle . It can be hypothesized that in addition to the overproduction of the urea cycle intermediates in psoriatic patients, the increase in concentrations of said metabolites come from changes in proteins that are deiminated (or citrullinated) in the skin mainly K1 , K10, and filaggrin proteins . The decreased deimination of psoriatic cornified cell layer proteins has been shown for K1 . Fewer cells that express K10 were detected in lesional skin . Filaggrin, filaggrin-2, and their mRNAs in psoriatic skin samples were found to be significantly reduced . Since less citrulline is produced in psoriatic skin, a cumulation of urea cycle intermediates can be explained. The ratio of citrulline to ornithine is lower in psoriasis patients, thus indicating the lower activity of ornithine carbamoyltransferase and the cumulation of ornithine. In addition, the ratio of putrescine to ornithine is lower in psoriasis patients showing a reduction in the activity of ornithine decarboxylase (EC 184.108.40.206) and to overall changes in the urea cycle. These findings correlate well with Kang et al.  who also noted the upregulation of metabolites in the urea cycle. Phenylalanine is a non-essential amino acid meaning that its uptake is dependent on diet. Phenylalanine is hydroxylated to tyrosine by phenylalanine hydroxylase (PAH, EC 220.127.116.11). PAH activity has been shown to be lower in psoriasis patients  and increases after UVB-light exposure , thus lowering the amount of Phe in the serum. Although the role of excess Phe in psoriasis is unclear, a connection with UV therapy and its effect can be noted.
Methioninesulfoxide (Met.SO) is the oxidized form of methionine that reacts with free radicals and goes through the oxidation process . The higher concentration of Met.SO and the ratio for the fraction of sulfoxidized methionine of the unmodified methionine pool are both indicative of oxidative stress in psoriasis patients’ serum.
Carnitine is synthesized from lysine and methionine  and has a variety of functions that include the transport of different fatty acids to mitochondria for branched α-keto acid oxidation, mitochondrial fatty acid oxidation, and trapping of acyl-CoA metabolites that may impair gluconeogenesis, the citric cycle, and the urea cycle  among many other functions. In omnivores, up to 75% percent of carnitine comes from dietary sources or can be from endogenous origins in the case of strict vegetarians . In healthy individuals, up to 80% of carnitine from food is absorbed . The lower concentration values of various carnitines and acylcarnitines found in psoriasis patients can either be explained by the unlikely change in diet that is less abundant in dairy products, fish, meat, and poultry or the increase in fatty acid oxidation in lesional skin due to the increased energy consumption of rapidly proliferating cells. Caspary et al. have shown the latter , where the increased activity of carnitine palmitoyltransferase-1 (CPT-1) was demonstrated in lesional skin. CPT-1 is the enzyme responsible for the rate of transport of long-chain fatty acids into mitochondria. This correlates well with our results that show a significant decrease in the concentrations of C9, C12, C10.2, and C7.DC in psoriasis subjects compared to controls. On the opposite, CPT-1 and CPT-2 deficiency has been shown to increase the concentrations of C14.1, C12, C16, C18:1, and C0 . In addition to the changes in circulating carnitine levels, the ratios C2…C0, X.C2.C3…C0 and [C16 + C18]/C0 are all indicators of overall β-oxidation activity. Since the ratios are all higher in controls, it might be a sign of altered lipid use activity in the energy production in psoriasis.
The higher rate of proliferation for skin cells could also explain the lower concentrations of phosphatidylcholines in psoriasis patients’ blood serum that are essential parts of cell membranes. In targeted analysis, all of the phosphatidylcholines had lower concentrations in subjects with psoriasis. PC-s from untargeted analysis, however, had higher values in psoriasis patients’ serums which can be explained by the different methods and machinery used in the different approaches. The untargeted method strongly favors the solubility of lipophilic molecules, thus making some metabolites statistically different that were not in the targeted approach. LysoPC-s have been shown to induce the chemotaxis of T-lymphocytes , thus fueling the constant inflammation in the epidermis. A higher concentration of choline which is an essential precursor for phosphocholines has been noted in lesional psoriatic skin .
In targeted analysis, taurine had a higher concentration in psoriasis patients (122.6 µM) compared to controls (107.4 µM), but the change was not significant (p = 0.24). In untargeted analysis, taurine was also higher in psoriasis subjects and the change was considered statistically significant (p < 0.00159). Taurine is considered as a semi-essential amino acid that is not incorporated into proteins, but has different biological roles including the regulation of cell volume, anti-oxidative, anti-apoptotic, and anti-inflammatory effects . Higher concentrations of taurine were measured in psoriatic skin compared to uninvolved skin .
Phytol is the carbon side chain of chlorophylls that is released by bacteria present in the gut of ruminants’ intestines. From there, it becomes widely available in the adipose tissue and dairy products of the animal . This branched-fatty alcohol has been shown to activate peroxisome proliferator-activated receptor α in the liver followed by an effect on the control of lipid abnormalities in diabetes, obesity, and hyperlipidemia . In addition, phytol is converted into phytanic acid in the liver . Phytanic acid was demonstrated to increase oxidative stress and reduce the antioxidant potential in rats . Alpha-methylacyl-CoA racemase (AMACR) is responsible for the regulation and metabolism of dietary branched-chain lipids . In summary, it can be hypothesized that in psoriasis, the patients might have alpha and beta oxidation deficiencies which helps to explain the abnormal levels of acylcarnitines in the serum. The increase in the levels of phytol and phytanic acid contributes to oxidative stress and the continued inflammation of the skin in psoriatic patients. The single nucleotide polymorphism of AMACR in patients with psoriasis could also contribute to the increased level of phytol but that remains to be shown. Currently, a reduction of red meat and dairy products could potentially lower phytol levels in the serum. A meta-analysis has shown that dietary regulation and weight loss are important in achieving PASI 75 scores . A reduced intake of phytol from dairy products and red meat might certainly have an effect on that.
An excess of 1,11-undecanedicarboxylic acid has been found in patients with Zellweger syndrome and adrenoleukodystrophy . 1,11-undecanedicarboxylic acid together with phytanic acid could be indicative of peroxisome disorders, but the exact role remains unclear. Peroxisome proliferator-activated receptor-γ (PPARγ) has been demonstrated to be responsible for the regulation of lipid metabolism and glucose homeostasis along with cell differentiation and growth regulation . Since the activity of PPARγ is lower in psoriatic skin , then higher concentrations of phytol, 1,11-undecanedicarboxylic acid, and different phosphatidylcholines could be a direct result of that.
Applying predictive models is still very new, but could 1 day be used routinely in the clinic for a better monitoring of the treatment or to provide help with diagnosis. The models we reported could very well be used for both. Building classification models are simpler, since it is a binary problem, but for the PASI regression model, the problem of measuring PASI scores by different physicians arises. It has been reported that an intra-class correlation coefficient of 0.804 was achieved when different doctors measured PASI scores on the same patients , so a more precise method of measurement, e.g., on the molecular level, could provide better treatment monitoring options.
In conclusion, our analysis demonstrated that patients with psoriasis have an impaired amino acid and lipid metabolism and a disbalance in the components of the bilipid cellular membrane. These findings will help to understand the pathogenesis of the disease which could lead to the better treatment of patients in the clinic. Our reported models and the discovered metabolites could be useful in helping to diagnose plaque psoriasis or monitor treatment.
We would like to thank all of the participating patients and Jelena Lüter, Kadri Sikk, Ingrid Liimand, Katre Maasalu, and Aare Märtson for their help with the recruitment of patients and sample collection. This study was funded by the personal research Grant PUT1465, PUT177, Grant IUT20-42 from the Estonian Research Council and by Grant SP1GVARENG from the University of Tartu and by the European Union through the European Regional Development Fund.
Compliance with ethical standards
Conflict of interest
The authors state no conflict of interest.
This study was approved by the Research Ethics Committee of the University of Tartu. Permission number 245/M-18. The Declaration of Helsinki protocols were followed and patients gave their informed, written consent.
- 1.Abdullah MN, Yap BW, Zakaria Y, Abdul Majeed AB (2016) Metabolites selection and classification of metabolomics data on Alzheimer’s disease using random forest. In: Soft computing in data science SCDS 2016 communications in computer and information science, vol 652. Springer, SingaporeGoogle Scholar
- 7.Brot N, Weissbach H (2000) Peptide methionine sulfoxide reductase: biochemistry and physiological role. Biopolymers 55:288–296. doi:10.1002/1097-0282(2000)55:4<288:AID-BIP1002>3.0.CO;2-M CrossRefPubMedGoogle Scholar
- 10.Chambers MC, Maclean B, Burke R, Amodei D, Ruderman DL, Neumann S, Gatto L, Fischer B, Pratt B, Egertson J, Hoff K, Kessner D, Tasman N, Shulman N, Frewen B, Baker TA, Brusniak MY, Paulse C, Creasy D, Flashner L, Kani K, Moulding C, Seymour SL, Nuwaysir LM, Lefebvre B, Kuhlmann F, Roark J, Rainer P, Detlev S, Hemenway T, Huhmer A, Langridge J, Connolly B, Chadick T, Holly K, Eckels J, Deutsch EW, Moritz RL, Katz JE, Agus DB, MacCoss M, Tabb DL, Mallick P (2012) A cross-platform toolkit for mass spectrometry and proteomics. Nat Biotechnol 30:918–920. doi:10.1038/nbt.2377 CrossRefPubMedPubMedCentralGoogle Scholar
- 12.Gloerich J, van Vlies N, Jansen GA, Denis S, Ruiter JP, van Werkhoven MA, Duran M, Vaz FM, Wanders RJ, Ferdinandusse S (2005) A phytol-enriched diet induces changes in fatty acid metabolism in mice both via PPARalpha-dependent and -independent pathways. J Lipid Res 46:716–726. doi:10.1194/jlr.M400337-JLR200 CrossRefPubMedGoogle Scholar
- 13.Goto T, Takahashi N, Kato S, Egawa K, Ebisu S, Moriyama T, Fushiki T, Kawada T (2005) Phytol directly activates peroxisome proliferator-activated receptor alpha (PPARalpha) and regulates gene expression involved in lipid metabolism in PPARalpha-expressing HepG2 hepatocytes. Biochem Biophys Res Commun 337:440–445. doi:10.1016/j.bbrc.2005.09.077 CrossRefPubMedGoogle Scholar
- 17.Ho TK (2013) Random decision forests. In: 12th international conference on document analysis and recognition, vol 01, p 278. doi:10.1109/icdar.1995.598994
- 18.Horai H, Arita M, Kanaya S, Nihei Y, Ikeda T, Suwa K, Ojima Y, Tanaka K, Tanaka S, Aoshima K, Oda Y, Kakazu Y, Kusano M, Tohge T, Matsuda F, Sawada Y, Hirai MY, Nakanishi H, Ikeda K, Akimoto N, Maoka T, Takahashi H, Ara T, Sakurai N, Suzuki H, Shibata D, Neumann S, Iida T, Tanaka K, Funatsu K, Matsuura F, Soga T, Taguchi R, Saito K, Nishioka T (2010) MassBank: a public repository for sharing mass spectral data for life sciences. J Mass Spectrom JMS 45:703–714. doi:10.1002/jms.1777 CrossRefPubMedGoogle Scholar
- 20.Jolliffe I (2014) Principal component analysis. In: Wiley StatsRef: Statistics Reference Online. Wiley. doi:10.1002/9781118445112.stat06472
- 21.Kamleh MA, Snowden SG, Grapov D, Blackburn GJ, Watson DG, Xu N, Stahle M, Wheelock CE (2015) LC-MS metabolomics of psoriasis patients reveals disease severity-dependent increases in circulating amino acids that are ameliorated by anti-TNFalpha treatment. J Proteome Res 14:557–566. doi:10.1021/pr500782g CrossRefPubMedGoogle Scholar
- 25.Kim K, Mall C, Taylor SL, Hitchcock S, Zhang C, Wettersten HI, Jones AD, Chapman A, Weiss RH (2014) Mealtime, temporal, and daily variability of the human urinary and plasma metabolomes in a tightly controlled environment. PLoS ONE 9:e86223. doi:10.1371/journal.pone.0086223 CrossRefPubMedPubMedCentralGoogle Scholar
- 29.Leipnitz G, Amaral AU, Zanatta A, Seminotti B, Fernandes CG, Knebel LA, Vargas CR, Wajner M (2010) Neurochemical evidence that phytanic acid induces oxidative damage and reduces the antioxidant defenses in cerebellum and cerebral cortex of rats. Life Sci 87:275–280. doi:10.1016/j.lfs.2010.06.015 CrossRefPubMedGoogle Scholar
- 33.Mitchell M (1996) An introduction to genetic algorithms. MIT press, Cambridge, MAGoogle Scholar
- 41.Ravanbakhsh S, Liu P, Bjorndahl TC, Mandal R, Grant JR, Wilson M, Eisner R, Sinelnikov I, Hu X, Luchinat C, Greiner R, Wishart DS (2015) Accurate, fully-automated NMR spectral profiling for metabolomics. PLoS ONE 10:e0124219. doi:10.1371/journal.pone.0124219 CrossRefPubMedPubMedCentralGoogle Scholar
- 50.Shigematsu Y, Hirano S, Hata I, Tanaka Y, Sudo M, Tajima T, Sakura N, Yamaguchi S, Takayanagi M (2003) Selective screening for fatty acid oxidation disorders by tandem mass spectrometry: difficulties in practical discrimination. J Chromatogr B Anal Technol Biomed Life Sci 792:63–72CrossRefGoogle Scholar
- 54.Tanaka T, Masuzaki H, Hosoda K, Nakao K (2010) Critical roles of PPARgamma in every aspect of the metabolic syndrome. Nihon Rinsho Jpn J Clin Med 68:203–209Google Scholar
- 55.Team RC (2014) R: a language and environment for statistical computing. R Foundation for Statistical Computing, ViennaGoogle Scholar
- 59.Wishart DS, Tzur D, Knox C, Eisner R, Guo AC, Young N, Cheng D, Jewell K, Arndt D, Sawhney S, Fung C, Nikolai L, Lewis M, Coutouly MA, Forsythe I, Tang P, Shrivastava S, Jeroncic K, Stothard P, Amegbey G, Block D, Hau DD, Wagner J, Miniaci J, Clements M, Gebremedhin M, Guo N, Zhang Y, Duggan GE, Macinnis GD, Weljie AM, Dowlatabadi R, Bamforth F, Clive D, Greiner R, Li L, Marrie T, Sykes BD, Vogel HJ, Querengesser L (2007) HMDB: the human metabolome database. Nucleic Acids Res 35:D521–D526. doi:10.1093/nar/gkl923 CrossRefPubMedPubMedCentralGoogle Scholar
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.