Article pubs.acs.org/JAFC
High-Resolution Mass Spectrometry Associated with Data Mining Tools for the Detection of Pollutants and Chemical Characterization of Honey Samples Jérôme Cotton,†,§ Fanny Leroux,§ Simon Broudin,§ Mylène Marie,§ Bruno Corman,§ Jean-Claude Tabet,†,# Céline Ducruix,§ and Christophe Junot*,† †
CEA, iBiTec-S, Service de Pharmacologie et d’Immunoanalyse, Laboratoire d’Etude du Métabolisme des Médicaments, MetaboHUB Paris, 91191 Gif-sur-Yvette, France § Profilomic, 31 rue d’Aguesseau, 92100 Boulogne Billancourt, France # Laboratoire de Chimie Structurale Organique et Biologique, IPCM/CNRS UMR 7201, Université Pierre et Marie Curie, MetaboHUB Paris, 75252 Paris, France S Supporting Information *
ABSTRACT: Analytical methods for food control are mainly focused on restricted lists of well-known contaminants. This paper shows that liquid chromatography−high-resolution mass spectrometry (LC/ESI-HRMS) associated with the data mining tools developed for metabolomics can address this issue by enabling (i) targeted analyses of pollutants, (ii) detection of untargeted and unknown xenobiotics, and (iii) detection of metabolites useful for the characterization of food matrices. A proof-of-concept study was performed on 76 honey samples. Targeted analysis indicated that 35 of 83 targeted molecules were detected in the 76 honey samples at concentrations below regulatory limits. Furthermore, untargeted metabolomic-like analyses highlighted 12 chlorinated xenobiotics, 1 of which was detected in lavender honey samples and identified as 2,6-dichlorobenzamide, a metabolite of dichlobenil, a pesticide banned in France since 2010. Lastly, multivariate statistical analyses discriminated honey samples according to their floral origin, and six discriminating metabolites were characterized thanks to the MS/MS experiments. KEYWORDS: high-resolution mass spectrometry, honey, metabolomics, data mining, metabolite, xenobiotics, pollutants, liquid chromatography, multiresidue, food analysis, electrospray, veterinary drugs, pesticides, bees
■
INTRODUCTION The international globalization of the food market significantly increases the risk of contamination and fraud as illustrated by the several food scandals over the past 20 years, such as dioxin chicken in 1999 and milk adulterated with melamine in 2008. These crises have a huge impact on the economy and on the credibility of the agrifood business and may also have dramatic consequences for the health of consumers. Furthermore, pesticides and drugs are commonly used in agriculture and farming to increase production yields and treat infection or insect pests.1 Their extensive use has become a major environmental concern and a public health problem2 that has prompted health authorities to strengthen controls. Pesticides and drugs are traditionally detected using multiplexed methods involving gas or liquid chromatography coupled to triple-quadrupole (QqQ) mass spectrometers operated in the multiple reaction monitoring (MRM) mode.3−9 These methods are considered as a reference in terms of sensitivity and specificity.10,11 New simple and fast sample preparations (i.e., “QuEChERS” for quick, easy, cheap, effective, rugged, and safe) have extended the analytical possibilities to a wide variety of compounds with different physical−chemical properties.12−15 However, as more pesticides become available, the number of pollutants to be monitored increases significantly,1 and MRM-based methods, which focus on a limited number of targeted compounds, are © XXXX American Chemical Society
not adapted to the detection of untargeted and also unknown chemical contaminants.16 The implementation of high-resolution mass spectrometry (HRMS)-based methods, including time-of-flight (TOF), Fourier transform ion cyclotron resonance (FT-ICR), and Orbitrap-based instruments, which operate in the full scan mode and provide accurate mass measurements,1,17−21 have improved screening capabilities by enabling not only targeted but also retrospective analyses for the detection of nonpreselected molecules through their elemental composition. Thus, LC-HRMS approaches have been developed as an alternative to those relying on triple-quadrupole instruments for the screening and also quantification of xenobiotics in environment and food matrices and also mammal biofluids.1,18,22−25 Furthermore, the access to resolved isotopic patterns provides useful information to confirm metabolite identification,17,21,22,26 and HRMS instruments may be used as tandem to perform MS/MS experiments by collisional activation (CID) of selected precursor ions. This ion excitation proceeds either through “in axis” mode (CID in Qq/TOF and Qh/ICR and Received: September 15, 2014 Revised: October 25, 2014 Accepted: October 30, 2014
A
dx.doi.org/10.1021/jf504400c | J. Agric. Food Chem. XXXX, XXX, XXX−XXX
Journal of Agricultural and Food Chemistry
Article
Table 1. Xenobiotics of the Targeted Analysis ID
name of molecule
formula
classification
ID
name of molecule
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42
5-hydroxythiabendazole acetamiprid albendazole aldicarb amitraz atrazine azoxystrobin boscalid carbaryl carbendazim carbofuran carfentrazone-ethyl chlorfenvinphos clothianidin coumaphos cyproconazol cyprodinil diazinon dichlorvos difenoconazol diflubenzuron dimethoate dimethomorph dimetridazole diphenylamine febantel fenbendazole fenoxycarb flubendazole flusilazole fluvalinate furazolidone hexaconazol imidacloprid indoxacarb iprodione malathion mebendazole metalaxyl methoxyfenozide metolachlor metronidazole
C10H7N3OS C10H11ClN4 C12H15N3O2S C7H14N2O2S C19H23N3 C8H14ClN5 C22H17N3O5 C18H12Cl2N2O C12H11NO2 C9H9N3O2 C12H15NO3 C15H14Cl2F3N3O3 C12H14Cl3O4P C6H8ClN5O2S C14H16ClO5PS C15H18ClN3O C14H15N3 C12H21N2O3PS C4H7Cl2O4P C19H17Cl2N3O3 C14H9ClF2N2O2 C5H12NO3PS2 C21H22ClNO4 C5H7N3O2 C12H11N C20H22N4O6S C15H13N3O2S C17H19NO4 C16H12FN3O3 C16H15F2N3Si C26H22ClF3N2O3 C8H7N3O5 C14H17Cl2N3O C9H10ClN5O2 C22H17ClF3N3O7 C13H13Cl2N3O3 C10H19O6PS2 C16H13N3O3 C15H21NO4 C22H28N2O3 C15H22ClNO2 C6H9N3O3
antibiotic pesticide antibiotic pesticide pesticide pesticide pesticide pesticide pesticide pesticide pesticide pesticide pesticide pesticide pesticide pesticide pesticide pesticide pesticide pesticide pesticide pesticide pesticide antibiotic pesticide antibiotic antibiotic pesticide antibiotic pesticide pesticide antibiotic pesticide pesticide pesticide pesticide pesticide antibiotic pesticide pesticide pesticide antibiotic
43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83
myclobutanil nifuroxazide oxfendazole oxibendazole penconazol pendimethalin piperonyl-butoxide pirimiphos-ethyl pirimiphos-methyl prochloraz propiconazole pyraclostrobin pyrimethanil pyriproxifen ronidazole simazine spinosad spirodiclofen sulfachloropyridazine sulfadiazine sulfadimethoxine sulfamerazine sulfamethazine sulfamethizole sulfamethoxazole sulfamethoxypyridazine sulfamonomethoxine sulfaphenazole sulfapyridine sulfaquinoxaline sulfathiazole sulfisoxazole tebuconazole tebufenozide tebuthiuron tetramethrin thiabendazole thiacloprid thiamethoxam trifloxystrobin trimethoprim
formula C15H17ClN4 C12H9N3O5 C15H13N3O3S C12H15N3O3 C13H15Cl2N3 C13H19N3O4 C19H30O5 C13H24N3O3PS C11H20N3O3PS C15H16Cl3N3O2 C15H17Cl2N3O2 C19H18ClN3O4 C12H13N3 C20H19NO3 C6H8N4O4 C7H12ClN5 C41H65NO10 C21H24Cl2O4 C10H9ClN4O2S C10H10N4O2S C12H14N4O4S C11H12N4O2S C12H14N4O2S C9H10N4O2S2 C10H11N3O3S C11H12N4O3S C11H12N4O3S C15H14N4O2S C11H11N3O2S C14H12N4O2S C9H9N3O2S2 C11H13N3O3S C16H22ClN3O C22H28N2O2 C9H16N4OS C19H25NO4 C10H7N3S C10H9ClN4S C8H10ClN5O3S C20H19F3N2O4 C14H18N4O3
classification pesticide antibiotic antibiotic antibiotic pesticide pesticide pesticide pesticide pesticide pesticide pesticide pesticide pesticide pesticide antibiotic pesticide pesticide pesticide antibiotic antibiotic antibiotic antibiotic antibiotic antibiotic antibiotic antibiotic antibiotic antibiotic antibiotic antibiotic antibiotic antibiotic pesticide pesticide pesticide pesticide pesticide pesticide pesticide pesticide antibiotic
of the present study is to demonstrate that LC-HRMS combined with a metabolomics data treatment workflow, which is mostly used for systems biology and biomarker discovery studies, is highly relevant to the field of food control by enabling the monitoring and identification of emerging pollutants and also by improving the chemical characterization of food matrices. We chose to focus on the analysis of honey as a proof-ofconcept study. Honey is a complex matrix obtained from bees during the harvest of nectar and pollen. Bees also play a major role in biodiversity as well as in the pollination31,32 of plants, which is very important in agriculture. However, they are exposed to pesticides sprayed on fields and fruit trees. The European Union has established maximum residue levels (MRLs) for some pesticides in honey, the lowest of them being 10 μg/kg.33 Three types of collateral damage may occur: (i) pollution of harvests and hence honey; (ii) poisoning of the
HCD in the LTQ/Orbitrap) or by resonance (radial excitation with LTQ and SORI-CID Qh/ICR). Although the recorded CID spectra are instrument-dependent, they are a very useful tool for confirming metabolite structure, especially when the product ions from CID are analyzed under high-resolution conditions yielding their respective elemental composition.27 Beside improvements brought in the fields of screening and multiplexed quantification of pollutants in various matrices, LCHRMS methods have also been coupled to untargeted data processing approaches28 or to effect directed analysis studies29 to highlight untargeted chemical contaminants in wastewater effluents and river sediments. Furthermore, HRMS is nowadays the tool most used for metabolomics, which deals with the large-scale detection and quantification of metabolites in biological media. Metabolomics is an interdisciplinary approach,30 which combines analytical chemistry, mathematics, statistics, and bioinformatics. In this context, the main objective B
dx.doi.org/10.1021/jf504400c | J. Agric. Food Chem. XXXX, XXX, XXX−XXX
Journal of Agricultural and Food Chemistry
Article
The 50 mL Falcon tubes were rotated end-over-end for 1 h. After centrifugation at 3500 rpm for 5 min, an aliquot of 4 mL of the upper layer extract (acetonitrile) was transferred into a 5 mL propylene tube. The eluents were evaporated to dryness under nitrogen. Samples were then reconstituted to 200 μL of H2O/ACN (95:5, v/v) before injection into the LC-HRMS system. Liquid Chromatography and Mass Spectrometry. Analyses were performed using a Nexera LC-30AD liquid chromatographic system (Shimadzu, Marne-la-Vallée, France) coupled to an Exactive mass spectrometer (Thermo Fisher Scientific, Les Ulis, France) fitted with an electrospray source operated in the positive ionization mode. The software interface was Xcalibur (version 2.1) (Thermo Fisher Scientific). To ensure the accuracy of the mass measurements, the instrument was calibrated every week in both polarities with a homemade modified Calmix solution (Thermo Fisher Scientific) with the addition of 100 μL of glycine solution (10 μg/mL) to 150 μL of Calmix. For singly charged ions, the mass resolution power of the analyzer was set to 25 000 (m/Δm, full width at half-maximum at m/z 200) and the mass accuracy over the range of m/z 70−1000 was below 5 ppm. The AGC target and the maximum injection time were set to 106 and 50 ms, respectively. The high-performance liquid chromatographic (HPLC) separation was performed on an Xterra C18 (5 μm, 150 mm × 2.1 mm) column (Waters, Guyancourt, France) with a Javelin column filter 0.5 μm (Thermo Fisher Scientific). The mobile phases were composed of water (A) and acetonitrile (B), both containing 0.1% formic acid. The elution consisted of an isocratic step of 2 min at 5% phase B, followed by a linear gradient from 5 to 100% of phase B from 2 to 22 min. These proportions were kept constant for 4 min before returning to 5% B from 26.1 to 30 min. The flow rate of the mobile phase was 0.3 mL/min, and the injection volume was 10 μL. The column and the automatic sample injector were kept at 30 and 4 °C, respectively. For the ESI source, the capillary voltage was set to 4800 V. For droplet evaporation, the sheath and auxiliary gas (N2) flows were fixed at 28 and 9 arbitrary units, respectively. The capillary temperature was set to 280 °C. For MS/MS experiments, a Q-Exactive (Thermo Fisher Scientific) was used with the same source parameters and chromatography conditions as previously described. Two scan events were used: (i) an MS scan with a mass resolution power, AGC target, and maximum injection time set to 70 000 (m/Δm, full width at half-maximum at 200 u), 106, and 50 ms, respectively; and (ii) an MS/ MS scan (in HCD mode without LMCO limit) at normalized collision energies of 20 and 40%, with a mass resolution power, AGC target, maximum injection time, isolation width set to 17500 (m/Δm, full width at half-maximum at 200 u), 106, 250 ms, and 0.4 m/z, respectively. Experimental Design. Samples were randomized and analyzed in three batches of 152 samples (76 spiked and nonspiked samples). Within each batch, nonspiked samples were injected first. A blank and a pool of the 83 compounds of interest (in buffer at 100 ng/mL) were injected every 5 and 10 biological samples, respectively. Data Processing. Targeted Data Processing. All raw data were processed with two application managers of Xcalibur 2.1 software: Qualbrowser (for peak detection) and Quanbrowser (for peak integration). The identification of the xenobiotics of interest from raw data was based on four criteria: (1) retention time of the expected molecule compared with the retention time of standard compound in a time window of ±0.15 min; (2) accurate mass measurement of the analyte (mass tolerance ±5 ppm); (3) reliability of the isotopic pattern compared with that of the reference compound; and (4) increase of the chromatographic peaks area in honey spiked with xenobiotics. Data Processing for the Nontargeted Approach. Peak detection, alignment, and integration were performed with the XCMS software package. Raw files were first converted to mzXML format with MSconvert (ProteoWizard35). Data were processed using XCMS version 1.30.3 running under R version 3.0.0. The R software was installed on a Lenovo ThinkStation C20X 24 core Intel Xeon E5645 2.4 GHz with 24 Go RAM running Linux (Centos release 6.4 Final x86_64). The CentWave algorithm was used.36 The list of the parameters used is available as Supporting Information (Table 3s). The resulting peak table was then processed as described below. First,
bees and decimation of their colonies; and (iii) chronic intoxication of consumers. The decline in the number of bee colonies in Europe in general and France in particular has considerably reduced honey production, whereas consumption has remained stable. As a consequence, most of the honey consumed in France is imported from around the world, and advanced methods are required to improve the chemical characterization of honey to highlight adulteration or contamination. We report here the development, validation, and application of an LC-HRMS approach enabling the detection of 83 pollutants known to occur in honey samples and indicating whether or not their concentrations are above or below regulatory limits. We show that the same raw data can be processed using metabolomics like workflow, including automatic peak detection and alignment software, data mining, and multivariate statistical analysis tools to highlight untargeted and unknown chlorinated chemical contaminants and also to discriminate honey samples according to their floral origin.
■
MATERIALS AND METHODS
Chemicals and Reagents. All standards (see Table 1, 28 antibiotics and 55 pesticides, >95% purity), magnesium sulfate salt (MgSO4, >99%), sodium acetate (CH3COONa, >97%), and glycine (>99% purity) were from Sigma-Aldrich (Saint-Quentin Fallavier, France). HPLC grade acetonitrile (ACN), formic acid (HCOOH, content >99%), dichloromethane (CH2Cl2), methyl tert-butyl ether (MTBE), sterile Falcon tubes (50 mL), and propylene tubes (5 mL) were from VWR International (Fontenay-sous-Bois, France). HPLC grade methanol (MeOH) was from Carlo Erba reagents (Val de Reuil, France). HPLC grade water was obtained by purifying demineralized water in a Milli-Q system (Millipore, Molsheim, France). The standard mixtures used for the external calibration of the MS instrument (Calmix-positive, for the positive ion mode, consisting of caffeine, Lmethionylarginyl-phenylalanyl-alanine acetate, and Ultramark 1621; and Calmix-negative, for the negative ion mode, consisting of the same mixture plus sodium dodecyl sulfate and sodium taurocholate) were from Thermo Fisher Scientific (Les Ulis, France). Preparation of Stock Solutions. Stock solution (1 mg/mL when possible) of each standard was prepared by dissolving 1 mg of each substance in 1 mL of an appropriate solvent consisting of H2O, ACN, MeOH, H2O/ACN, or H2O/MeOH (50:50, v/v. containing or not 0.1% formic acid). A working mixed standard solution (1 μg/mL) was prepared by diluting each standard stock solution by a factor of 1000 with water. To minimize degradation of standards, stock and working standard solutions were stored at −20 °C immediately after preparation. Honey Samples. A panel of 76 honeys was purchased by Institut National de la Consommation (INC, which is the French national institute for consumer protection) from different stores in the Paris region (France). These honeys were produced from around the world and from different types of flowers (multifloral, lavender, acacia, mountain, orange tree, and eucalyptus). They were of various prices and quality, ranging from discount, private label, handmade, to luxury, and of different kinds of label (i.e., none, organic farming, red label, and protected environmental indication (IGP)). Sample Preparation. Honey samples were processed by liquid− liquid extraction with acetonitrile solvent, by adapting the protocol developed by Pizzuti et al.34 to extract pesticides from soybeans to honey samples. Each honey sample (5 g) was weighed into a 50 mL polypropylene Falcon tube with a screw cap, and 4 mL of water was added to homogenize the samples of various consistencies (liquid, solid, or creamy). Each sample was then prepared in six replicates. Three of six replicates were spiked with an appropriate volume of working solution (1 μg/mL) to obtain a concentration of 10 ng/g for each targeted pollutant. Samples were then extracted with 10 mL of ACN with 2 g of magnesium sulfate and 2.5 g of sodium acetate added. C
dx.doi.org/10.1021/jf504400c | J. Agric. Food Chem. XXXX, XXX, XXX−XXX
Journal of Agricultural and Food Chemistry
Article
features were quantile normalized.37 Features were then kept if they were detected with the same trend (i.e., present or absent) in the three replicates of 90% of honeys, with a coefficient of variation below 30% and with a sample to blank concentration ratio above a factor of 10. Finally, features were annotated by matching their mass accurate measured mass at ±10 ppm with theoretical ones contained in the KEGG,38 HMDB,39 and Metlin40 databases by using an informatics tool developed in R language. Identification of Chlorinated Xenobiotics in Honey Samples. A script was written in R language to highlight the chlorinated molecules thanks to the specific isotopic patterns provided by chlorine atoms. This script uses the peak table generated by XCMS. It groups two variables with the same retention time (±5 s) and differing by 1.9970 ± 0.0010 m/z, which corresponds to the accurate m/z difference between the two natural chlorine isotope masses (i.e., 35Cl and 37Cl). Statistical Analyses. (a) Multivariate Statistical Analyses. The data resulting from the XCMS process were mean-centered and logarithm-scaled and then analyzed using SIMCA-P11 software (Umetrics, Umea, Sweden) for multivariate analyses using principal component analysis (PCA) and projection to latent structure discriminant analysis (PLS-DA). This software returns a variable importance in projection (VIP) score, which reflects the contribution of the variables to the model. A variable is considered as important for the model when its VIP is above 1. The PLS-DA models were validated using the cross-validation function of SIMCA-P11 and by permutation tests (k = 100). (b) Univariate Statistical Analyses. For each VIP feature, the normality assumption was checked (Shapiro−Wilk test, α = 5%). Then, the Wilcoxon test was used with α = 5% to determine the significance (p value) between the two groups.
triple-quadrupole instruments operated in the MRM mode. In addition, all LOD values determined are below the regulatory values imposed by the European Union, when available (Table 2s). Targeted Analysis of Honey Samples. Seventy-six honey samples from different countries and of different floral origins were selected. The results of the targeted analysis indicate that 74 of 76 honeys are contaminated with at least one molecule of the 83 selected xenobiotics. Figure 1a displays the number of
■
RESULTS AND DISCUSSION Screening of Emerging Pollutants in Honey Samples Using a Targeted Approach. Method Development and Validation. An HRMS-based metabolomic approach was first developed and implemented with honey samples as a proof-ofconcept study. A list of 83 pollutants (55 pesticides and 28 antibiotics) of interest was drawn up (Table 1). A liquid−liquid extraction method and a UHPLC-HRMS method were developed and validated to (i) detect these 83 pollutants, (ii) indicate whether their concentrations in honeys are above or below regulatory limits, and (iii) detect a wide range of chemicals present in honeys. Method development and validation are detailed in the Supporting Information. Briefly, a spectral database including chromatographic retention times and MS information (i.e., protonated, adduct, isotope, and in-source fragment ions for each molecule of interest) was built to select the ions to be extracted from raw data. Table 1s displays retention times and ions used for detection and quantification. Metabolite extraction from honey samples was achieved using acetonitrile supplemented with magnesium sulfate and sodium acetate (please refer to Materials and Methods for additional information), and an LC-HRMS method was optimized to achieve chromatographic separation of the 83 xenobiotics of interest within 30 min. This method was validated by studies of linearity, matrix effect, and intra-assay precision and by determination of the limit of detection (LOD), which corresponds in our case to the lowest concentration for which the coefficient of variation of the areas of triplicates was 0.99 for all compounds. LOD values ranged from 0.1 to 10 μg/kg. Finally, the sensitivity achieved with our method is comparable to12 or even better1,34 than that achieved with
Figure 1. Targeted analysis of xenobiotics in honey samples: (a) histogram representing the number of molecules found in honey; (b) representation of the number of honey samples in which each molecule was found.
molecules detected in each honey. A bimodal distribution seems to emerge with maximum points corresponding to two and seven molecules per honey. Classifications of the number of molecules per honey were made by country, floral origin, and organic farming, but did not show significant correlation (data not shown). Two honeys, in which no pollutants were found, came from organic farming in Australia and Mexico. Figure 1b displays the number of xenobiotics detected and their presence in honey samples. It indicates that 35 of 83 D
dx.doi.org/10.1021/jf504400c | J. Agric. Food Chem. XXXX, XXX, XXX−XXX
Journal of Agricultural and Food Chemistry
Article
Table 2. Xenobiotics Detected in Honey Samples ID
m/z
RT (min)
interpretation
occurrence in honey samples (floral origin)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
189.9817 236.1407 243.0971 253.0308 256.0598 277.1069 277.2158 284.0834 292.1204 308.1520 325.1255 341.0610 347.1973 357.1520 369.1526 387.1622
7.9 12.5 2.9 11.2 9.6 13.7 12.5 6.0 15.3 16.3 9.4 12.7 16.5 8.5 9.7 9.9
2,6-dichlorobenzamide (C7H5ONCl2) C11H22NO2Cl C5H15N6O3Cl thiacloprid (C10H9N4SCl) imidacloprid (C9H10N5O2Cl) C10H17N4O3Cl C13H29 N4Cl C11H19NO3Cl2 cyproconazole (C15H18N3OCl) tebuconazole (C16H22N3OCl) C17H22N2 Cl2 C11H18N4O2S Cl2 C18H31O4Cl C12H25N4O6Cl U50488b (C19H26N2OCl2) C17H27N4O2SCl
6 (lavender) 70 (miscellaneous) 15 (miscellaneous) 20 (miscellaneous) 1 (orange tree) 1 (multifloral) 13 (miscellaneous) 38 (miscellaneous) 9 (miscellaneous) 30 (miscellaneous) 39 (miscellaneous) 56 (miscellaneous) 48 (miscellaneous) 38 (miscellaneous) 15 (miscellaneous) 47 (miscellaneous)
ID statusa identified a, e a, e identified identified a, e a, e a, e identified identified a, e a, e a, e a, e a, e a, e
(a, b, c)
(a, b, c) (a, b, c)
(a, b, c) (a, b, c)
a
a, elemental composition; b, retention times matching with that of a standard; c, MS/MS matching with that of a standard; d, MS/MS matching with a structural hypothesis; e, MS/MS experiment available. bStructural hypothesis obtained from the Scifinder database (elemental composition).
Dobson et al.,47 and Peironcely et al.,48 were subsequently published. Peironcely et al., for example, found some discrimination trends between metabolites and nonmetabolites on a PCA score plot on the basis that metabolites tend to have a higher water solubility, lower molecular mass, fewer C, N, and S atoms, and a lower number of cycles and rotatable bonds than nonmetabolites.48 However, from the analysis of these studies, it is difficult to establish rules to discriminate between signals related to either metabolites or xenobiotics in LC-HRMS data sets. The fact that xenobiotics can be metabolized by living organisms or degraded by the environment, and also the difficulty of automatically generating relevant elemental compositions from accurate measured mass despite the availability of chemical and empirical rules,49 further complicates this task. Thus, we decided to focus on chlorine-containing compounds for this proof-of-concept study. As chlorine atoms mainly occur in biological media as sodium chloride, chlorine-containing organic compounds are assumed to be, or at least to come from, xenobiotics produced by human industrial activity. Note that 35% of our 83 targeted xenobiotics contain at least one chlorine atom. In addition, chlorinated compounds can be easily and selectively detected in biological media using HRMS-based instruments thanks to their particular natural isotopic signature consisting of two stable isotopes, 35Cl and 37Cl, the latter having a relative abundance of 33% of the first one and these two isotope masses being separated by 1.9970 u. We therefore developed an algorithm written in R language that highlights features differing by 1.9970 ± 0.0010 u (for singly charged ions) and occurring at the same retention time (±5 s) from peak tables produced by an automatic peak detection and alignment software. First, 76 XCMS peak tables (see Supporting Information for the XCMS parameters, Table 3s) were produced on triplicates of each honey sample to improve the accuracy of RT and measured masses when compared with a single peak table containing 228 samples. These peak tables were then annotated with public databases (see Materials and Methods) and processed by our algorithm. Features highlighted by this algorithm in all 76 peak tables were then merged, leading to a
targeted molecules were detected. The three most frequently recovered compounds were carbendazim, amitraz, and chlorfenvinphos (i.e., 70, 59, and 35 times in 76). Carbendazim is a fungicide belonging to the carbamate family, which is commonly used in cereal and fruit crops. It has been banned in the European Union (EU) since 2009.41 Amitraz is an acaricide belonging to the family of formamidines. It is commonly used by beekeepers to protect their hives against the parasite Varroa destructor.42 Finally, chlorfenvinphos is an insecticide and an acaricide belonging to the family of organophosphates. It was mainly used against fly larvae in farmed chicken or beef.43 Because of its toxicity, it was banned in the EU in 2007. Some beekeepers may still use it to combat Varroa because it is more effective than amitraz.44 However, our results, which provide an order of magnitude of concentration, indicate that no xenobiotics were found at concentrations above the regulatory threshold levels of the EU. Data Mining from LC-HRMS Fingerprints. Unlike data acquired on a triple-quadrupole mass spectrometer operated in the MRM mode, HRMS-based data can be subjected to subsequent data mining procedures to highlight unanticipated information, as already performed with metabolomics in systems biology or for biomarker discovery. Proof-of-concept studies detailed below emphasize the relevance of such an approach for the detection of unexpected xenobiotics and for an improved chemical characterization of honeys. To this end, raw data were processed using XCMS software, a collection of algorithms written in R language for the detection and alignment of peaks. XCMS output is a data matrix that can thus be processed by data mining algorithms and can also be subjected to multivariate statistical analyses. Detection of Unexpected Xenobiotics. How to discriminate between signals related to metabolites (i.e., all organic substances naturally occurring from the metabolism of the living organism studied) or xenobiotics and contained in LCHRMS fingerprints is a key question. It was addressed at the beginning of the 2000s by chemoinformatic analysis that describes metabolites present in the Escherichia coli bacteria listed in the EcoCyc and KEGG databases in terms of fragmentbased fingerprints and descriptors based on physicochemical properties.45 Other studies, such as those of Gupta et al.,46 E
dx.doi.org/10.1021/jf504400c | J. Agric. Food Chem. XXXX, XXX, XXX−XXX
Journal of Agricultural and Food Chemistry
Article
Figure 2. Identification of 2,6-dichlorobenzamide in honey ([M + H]+ = 189.9821, m/z ± 8 ppm).
successfully for these 12 compounds. However, only two structures were proposed on the basis of the MS/MS interpretation. The CID spectrum of compound 15 matched with U50488, a compound present in the Scifinder database (Table 2), whereas compound 1 was identified as 2,6dichlorobenzamide by matching its retention time, accurate measured mass, isotopic pattern, and CID spectrum to those of the purchased compound (Figure 2), and it was found in 6 of 11 French lavender honeys (Table 2). Interestingly, this molecule is the metabolite of dichlobenil, a herbicide that was not part on our list of xenobiotics and which was widely used in lavender and lavendin plantations until it was banned in France in 2010. Honey Typing. Besides the detection of chemical contaminants and/or their metabolites, we would also like to show that LC-HRMS and associated data mining could be relevant to improving the chemical characterization of honey samples. Here we focus on the discovery of biomarkers of floral origin that could help to detect fraud and provide a preliminary experiment. To this end, we used the XCMS peak table previously produced on the 76 kinds of honey samples (see Table 4s for XCMS parameters). This peak table contained 121,000 features and 300 samples (i.e., 76 honeys analyzed in triplicate and 72 analytical blanks). It was filtered out (see Materials and Methods for more detail), and 42500 analytically relevant signals were kept for multivariate statistical analyses. The PCA score plot displayed in Figure 3a shows a clear discrimination between single-flower (acacia, lavender and orange) and multifloral (multiple flowers and mountain) honeys, except for eucalyptus honey samples for which the sampling was too small (four honey samples only). It also indicates that acacia honey samples are well separated of orange and lavender honey samples. We then focused on the discrimination between single and multifloral honey samples and used PLS-DA, the score plot of which is shown in Figure 3b, to confirm these results. This PLS-DA model was validated using the cross-validation function of the SIMCA P11 software and by permutation tests (k = 500), which dramatically reduced the performances of the model, thus confirming its validity (Figure 3c).
peak table containing 25000 features. First, it is of note that 97% of the chlorinated molecules previously monitored in the course of the targeted approach were present in this peak table. Then, due to the high number of features, manual inspection was limited to the 1300 most represented features, which were sorted by increasing m/z values. Features corresponding to a unique molecule were then merged in 143 clusters corresponding to couples of potential 35Cl- and 37Cl-containing ions and which were manually inspected. Only 28 of them were confirmed to correspond to chlorinated ions, among which 13 chlorinated compounds were detected in honeys (i.e., the others features were artifacts generated by the automatic peak detection process). The inspection of a fraction of the peak table shows that too many false positives are selected using only the criteria retention time and m/z difference. Additional criteria were then applied: (i) the relative abundance the 37Cl ion had to be between 20 and 120% of the 35Cl, which ensures the detection of ions bearing one to four chlorine atoms; (ii) the retention time had to be above 90 s (to avoid false positives due to the huge number of molecules present in the void volume); (iii) the area of the chromatographic peak related to the monoisotopic ion should be above 106 AU; and (iv) sample (i.e., mean of the three replicates per honey) to blank concentration ratio should be above a factor of 3. By these means, only 4 chlorinated molecules of 35 previously detected in the targeted analysis (i.e., thiacloprid, imidacloprid, cyproconazole, tebuconazole) were detected, but the number of clusters generated was much smaller, thus allowing manual inspection. Moreover, this led to the detection of 18 new chlorinated compounds that are present in at least one honey sample, in addition to the 4 chlorinated compounds that were part of our list. An XCMS peak table including the 228 samples (i.e., 76 honeys analyzed in triplicate) was generated without any specific criteria (see the Supporting Information for the XCMS parameters, Table 4s) to confirm the presence of these new chlorinated compounds. This led to the confirmation of 12 chlorinated molecules of the 18 initial ones, and 10 of them were present in at least 10 kinds of honey samples (Table 2). MS/MS information, elemental composition, and isotope pattern regarding the number of Cl atoms were obtained F
dx.doi.org/10.1021/jf504400c | J. Agric. Food Chem. XXXX, XXX, XXX−XXX
Journal of Agricultural and Food Chemistry
Article
Figure 3. Multivariate statistical analyses of the data set of 76 honey samples. Multifloral (●), mountain (◆), acacia (*), orange tree (□), and lavender (△) honey samples. (a) PCA score plot of honey samples (excluding eucalyptus). Data were log-transformed and centered before multivariate statistical analysis. (b) PLSDA score plot of honey samples. (c) Validation model of PLSDA by permutation tests (k = 500) on the first component. R2(Y) corresponds to the proportion of the variance of the response variable that is explained by the model, and Q2 (cum) expresses the cumulative proportion of the variance of the variables that can be predicted by the model. Permutation of data from multifloral samples related to data from single-flower samples dramatically decreased the performance of the model, thus confirming its validity.
Variables exhibiting a VIP score (please refer to the Materials and Methods for more details) above a value of 2 were then annotated using public databases such as KEGG,38 HMDB,39 and METLIN.40 Six of the most significant annotated variables (VIP > 2) were selected. Their significance was confirmed by the Wilcoxon test with α = 5% (after verification of the normality assumption with the Shapiro−Wilk test; see Materials and Methods for more details), and their box and whisker plots are displayed in Figure 4.
Five of these six variables were putatively identified and exhibited MS spectra compatible with the proposed structure; a flavone-like structure was proposed for the last one (Table 3). CID spectra (in HCD mode) are available as Supporting Information. Figure 5 displays CID spectra of m/z 438 (i.e., X84952) and a proposed fragmentation scheme suggesting that the precursor ion could be lunarine, an alkaloid occurring in plants of the Lunaria genus, native to central and southern Europe. A complete fragmentation scheme of lunarine is provided as Supporting Information (Figure 6s). G
dx.doi.org/10.1021/jf504400c | J. Agric. Food Chem. XXXX, XXX, XXX−XXX
Journal of Agricultural and Food Chemistry
Article
Figure 4. Box-plot of the six features of the most significance from PLS-DA. The thick line corresponds to the median. The box corresponds to the first and third quartiles. The whiskers represent 1.5 times the interquartile range. The outliers are marked by dots. Names of molecules putatively identified: X36800, methyl 2-(2-ethoxyacetamido)-4-methylpentanoate; X36809, methyl 2-(2-ethoxyacetamido)-2-methylpentanoate; X42700, N-Dglucosylarylamine; X42932, piscidic acid; X69282, flavone with ramification C18H16O8; X84952, lunarine.
These molecules enable discrimination between multifloral and single-flower honeys and could be used for fraud detection (i.e., selling of multifloral honeys instead of single-flower honeys) after confirmation studies. Indeed, these are preliminary results based on information indicated on labels. Additional data mining procedures should be developed to improve these results, and confirmatory studies have to be undertaken, based on confirmation of floral origins of honey by melissopalinological, sensory, or chemical analyses. In conclusion, this study demonstrates that LC-HRMS and LC-HRMS/MS associated with “metabolomics like” data mining tools are versatile tools allowing both target and global analyses for a better characterization of food matrices by enabling (i) conventional multiplexed and targeted analyses of xenobiotics, (ii) the detection of untargeted and unknown chemical contaminants, and (iii) the detection of metabolites useful for the characterization of food matrices. This was achieved by performing a proof-of-concept study on honey samples. Our results for a panel of 76 honeys showed that at least 1 of the 83 targeted pollutants was detected in 74 of the 76 honeys and that an average of 5 xenobiotics were found per
Table 3. Significant Features for the Discrimination between Single-Flower and Multifloral Honey Samples feature
m/z
RT (min)
X36800
232.15433
5.5
X36809
232.15433
5.7
X42700 X42932 X69282
256.11795 257.0654 361.09179
1.6 3.8 11
X84952
438.23873
7.9
interpretation methyl 2-(2-ethoxyacetamido)2-methylpentanoate or methyl 2-(2-ethoxyacetamido)-4methylpentanoate methyl 2-(2-ethoxyacetamido)2-methylpentanoate or methyl 2-(2-ethoxyacetamido)-4methylpentanoate N-D-glucosylarylamine piscidic acid flavone with ramification C18H16O8 lunarine
ID statusa a, d
a, d
a, d a, d a, d a, d
a
a, m/z matching structure; b, retention times matching with that of a standard; c, CID spectrum matching with that of a standard; d, CID spectrum experiment available.
H
dx.doi.org/10.1021/jf504400c | J. Agric. Food Chem. XXXX, XXX, XXX−XXX
Journal of Agricultural and Food Chemistry
Article
Figure 5. Characterization of lunarine from a honey sample: (a) CID spectrum of protonated variable X84952 annotated as lunarine recorded in HCD mode under (a) 20% and (b) 40% normalized collision energies; (b) summary of the proposed formal mechanism of fragmentations for lunarine (a complete scheme is available as Supporting Information, Figure 6s).
honey, with concentrations below the regulatory limits. Multivariate statistical analyses performed on analytically relevant features discriminated honey samples according to their floral origin. The data mining procedure also highlighted the presence of 12 unknown chlorinated chemicals in honey samples. One of these compounds was present in 4 lavender honeys and was formally identified as 2,6-dichlorobenzamide (a metabolite of dichlobenil, a herbicide that was widely used in lavender plantations before 2010). Such an approach could be applied to any other food matrix to build (i) spectral databases of metabolites and pollutants to improve data set annotation and compound identification and (ii) metabolite fingerprint databases related to any given food matrix that could be coupled with data mining tools to highlight the presence of unexpected compounds. However, this requires the development of standardization and normalization tools to handle batch-to-batch variations of signal intensities and chromatographic retention times and also the analysis of a
large number of samples to build a robust model. Furthermore, and as emphasized by our results, the characterization of untargeted and unknown chemicals by using MS and CID information remains highly challenging, but informatics tools such as in silico fragmenters and software for fragment ion prediction50−53 that could be combined with other physical− chemical properties such as retention time index or octanol− water partitioning coefficients54 could be of value to address this issue.
■
ASSOCIATED CONTENT
S Supporting Information *
Building of the spectral database, the choice of chromatographic conditions, metabolite extraction, validation of the developed method, and detailed scheme for lunarine fragmentation; table with information about the molecules present in the spectral database (name, formula, detected ions, extracted masses, retention time and log Ko/w), table with I
dx.doi.org/10.1021/jf504400c | J. Agric. Food Chem. XXXX, XXX, XXX−XXX
Journal of Agricultural and Food Chemistry
Article
(13) Lehotay, S. J.; de Kok, A.; Hiemstra, M.; Van, B. P. J. AOAC Int. 2005, 88, 595−614. (14) Schenck, F. J.; Hobbs, J. E. Bull. Environ. Contam. Toxicol. 2004, 73, 24−30. (15) Wang, J.; Chow, W.; Leung, D.; Chang, J. J. Agric. Food Chem. 2012, 60, 12088−12104. (16) Masia, A.; Ibanez, M.; Blasco, C.; Sancho, J. V.; Pico, Y.; Hernandez, F. Anal. Chim. Acta 2013, 761, 117−127. (17) De, D. E.; Commissati, I.; Suman, M. J. Mass Spectrom. 2012, 47, 1232−1241. (18) Kaufmann, A.; Butcher, P.; Maden, K.; Walker, S.; Widmer, M. Anal. Chim. Acta 2011, 700, 86−94. (19) Peters, R. J.; Bolck, Y. J.; Rutgers, P.; Stolker, A. A.; Nielen, M. W. J. Chromatogr., A 2009, 1216, 8206−8216. (20) Petrovic, M.; Gros, M.; Barcelo, D. J. Chromatogr., A 2006, 1124, 68−81. (21) Polgar, L.; Garcia-Reyes, J. F.; Fodor, P.; Gyepes, A.; Dernovics, M.; Abranko, L.; Gilbert-Lopez, B.; Molina-Diaz, A. J. Chromatogr., A 2012, 1249, 83−91. (22) Kaufmann, A.; Butcher, P.; Maden, K.; Walker, S.; Widmer, M. Analyst 2011, 136, 1898−1909. (23) Leon, N.; Roca, M.; Igualada, C.; Martins, C. P.; Pastor, A.; Yusa, V. J. Chromatogr., A 2012, 1258, 55−65. (24) Masia, A.; Campo, J.; Blasco, C.; Pico, Y. J. Chromatogr., A 2014, 1345, 86−97. (25) Farre, M.; Pico, Y.; Barcelo, D. J. Chromatogr., A 2014, 1328, 66−79. (26) Hernandez, F.; Sancho, J. V.; Ibanez, M.; Abad, E.; Portoles, T.; Mattioli, L. Anal. Bioanal. Chem. 2012, 403, 1251−1264. (27) Junot, C.; Madalinski, G.; Tabet, J. C.; Ezan, E. Analyst 2010, 135, 2203−2219. (28) Schymanski, E. L.; Jeon, J.; Gulde, R.; Fenner, K.; Ruff, M.; Singer, H. P.; Hollender, J. Environ. Sci. Technol. 2014, 48, 2097−2098. (29) Weiss, J. M.; Simon, E.; Stroomberg, G. J.; de, B. R.; de, B. J.; van der Linden, S. C.; Leonards, P. E.; Lamoree, M. H. Anal. Bioanal. Chem. 2011, 400, 3141−3149. (30) Junot, C.; Fenaille, F.; Colsch, B.; Bécher, F. Mass Spectrom. Rev. 2013, 33, 471−500. (31) Klein, A. M.; Vaissiere, B. E.; Cane, J. H.; Steffan-Dewenter, I.; Cunningham, S. A.; Kremen, C.; Tscharntke, T. Proc. Biol. Sci. 2007, 274, 303−313. (32) Steffan-Dewenter, I.; Potts, S. G.; Packer, L. Trends Ecol. Evol. 2005, 20, 651−652. (33) Regulation (EC) No. 396/2005 of the European Parliament and of the Council of 23 February 2005 on maximum residue levels of pesticides in or on food and feed of plant and animal origin and amending Council Directive 91/414/EEC; http://ec.europa.eu/ sanco_pesticides/public/?event=homepage (accessed April 18, 2014); Of f. J. Eur. Communities 2005, L 70 (1). (34) Pizzutti, I. R.; de Kok, A.; Hiemstra, M.; Wickert, C.; Prestes, O. D. J. Chromatogr., A 2009, 1216, 4539−4552. (35) Kessner, D.; Chambers, M.; Burke, R.; Agus, D.; Mallick, P. Bioinformatics 2008, 24, 2534−2536. (36) Tautenhahn, R.; Bottcher, C.; Neumann, S. BMC Bioinformatics 2008, 9, 504. (37) Lee, J.; Park, J.; Lim, M. S.; Seong, S. J.; Seo, J. J.; Park, S. M.; Lee, H. W.; Yoon, Y. R. Anal. Sci. 2012, 28, 801−805. (38) Kanehisa, M. Novartis Found. Symp. 2002, 247, 91−101. (39) Wishart, D. S.; Jewison, T.; Guo, A. C.; Wilson, M.; Knox, C.; Liu, Y.; Djoumbou, Y.; Mandal, R.; Aziat, F.; Dong, E.; Bouatra, S.; Sinelnikov, I.; Arndt, D.; Xia, J.; Liu, P.; Yallou, F.; Bjorndahl, T.; Perez-Pineiro, R.; Eisner, R.; Allen, F.; Neveu, V.; Greiner, R.; Scalbert, A. Nucleic Acids Res. 2013, 41, D801−D807. (40) Smith, C. A.; O’Maille, G.; Want, E. J.; Qin, C.; Trauger, S. A.; Brandon, T. R.; Custodio, D. E.; Abagyan, R.; Siuzdak, G. Ther. Drug Monit. 2005, 27, 747−751. (41) Regulation (EC) No. 1107/2009 of the European Parliament and of the Council of 21 October 2009 concerning the placing of plant
information about the validation parameters (i.e., recovery (%), RSD (%), LOD (μg/kg)) and EU MLR (μg/kg), two tables that list the two sets of XCMS parameters used for the detection of features in honey samples, and MS2 spectra of the six features that discriminate between floral origins. This material is available free of charge via the Internet at http:// pubs.acs.org.
■
AUTHOR INFORMATION
Corresponding Author
*(C.J.) Phone: +33 1 69 08 43 66. Fax: +33 1 69 08 59 07. Email:
[email protected]. Funding
This work was funded by the French National Institutes of Consumers (INC) and by Bpi France as part of the Agrifood GPS collaborative project. J.C. is supported by a grant from Association Nationale de Recherche et de Technologie. Notes
The authors declare no competing financial interest.
■ ■
ACKNOWLEDGMENTS We thank Dr. François Fenaille for helpful discussion. ABBREVIATIONS USED EU MRL, European maximal residue level; HCD, higher collision-induced dissociation; LC-HRMS, liquid chromatography coupled to high-resolution mass spectrometry; LOD, limit of detection; MRM, multiple reaction monitoring; PCA, principal component analysis; PLS-DA, projection on latent structures-discriminant analysis; QqQ, triple quadrupole; QuEChERS, quick, easy, cheap, effective, rugged, and safe; RSD, relative standard deviation; VIP, variable importance in projections
■
REFERENCES
(1) Gomez-Perez, M. L.; Plaza-Bolanos, P.; Romero-Gonzalez, R.; Martinez-Vidal, J. L.; Garrido-Frenich, A. J. Chromatogr., A 2012, 1248, 130−138. (2) Jackson, R. J.; Minjares, R.; Naumoff, K. S.; Shrimali, B. P.; Martin, L. K. J. Hunger Environ. Nutr. 2009, 4, 393−408. (3) Bohm, D. A.; Stachel, C. S.; Gowik, P. Anal. Bioanal. Chem. 2012, 403, 2943−2953. (4) Cherta, L.; Portoles, T.; Beltran, J.; Pitarch, E.; Mol, J. G. J.; Hernandez, F. J. Chromatogr., A 2013, 1314, 224−240. (5) Feo, M. L.; Eljarrat, E.; Barcelo, D. Rapid Commun. Mass Spectrom. 2011, 25, 869−876. (6) Gros, M.; Petrovic, M.; Barcelo, D. Anal. Bioanal. Chem. 2006, 386, 941−952. (7) Herrmann, A.; Rosen, J.; Jansson, D.; Hellenas, K. E. J. Chromatogr., A 2012, 1235, 115−124. (8) Ho, Y. B.; Zakaria, M. P.; Latif, P. A.; Saari, N. J. Chromatogr., A 2012, 1262, 160−168. (9) Wiest, L.; Bulete, A.; Giroud, B.; Fratta, C.; Amic, S.; Lambert, O.; Pouliquen, H.; Arnaudguilhem, C. J. Chromatogr., A 2011, 1218, 5743−5756. (10) Cannon, W. R.; Jarman, K. H.; Webb-Robertson, B. J.; Baxter, D. J.; Oehmen, C. S.; Jarman, K. D.; Heredia-Langner, A.; Auberry, K. J.; Anderson, G. A. J. Proteome Res. 2005, 4, 1687−1698. (11) Kapp, E. A.; Schutz, F.; Connolly, L. M.; Chakel, J. A.; Meza, J. E.; Miller, C. A.; Fenyo, D.; Eng, J. K.; Adkins, J. N.; Omenn, G. S.; Simpson, R. J. Proteomics 2005, 5, 3475−3490. (12) Kmellar, B.; Fodor, P.; Pareja, L.; Ferrer, C.; Martinez-Uroz, M. A.; Valverde, A.; Fernandez-Alba, A. R. J. Chromatogr., A 2008, 1215, 37−50. J
dx.doi.org/10.1021/jf504400c | J. Agric. Food Chem. XXXX, XXX, XXX−XXX
Journal of Agricultural and Food Chemistry
Article
protection products on the market and repealing Council Directives 79/117/EEC and 91/414/EEC; Of f. J. Eur. Communities 2009. (42) The European Agency for the Evaluation of Medicinal Products (EMEA). Commitee for veterinary medicinal products: Amitraz (Bees); EMEA/MRL/572/99-FINAL, 1999. (43) Agency for Toxic Substances and Disease Registry (ATSDR). Toxicological Profile for Chlorfenvinphos; U.S. Department of Health and Human Services: Public Health Service: Washington, DC, USA, 1997. (44) Higes, M.; Martin-Hernandez, R.; Martinez-Salvador, A.; Garrido-Bailon, E.; Gonzalez-Porto, A. V.; Meana, A.; Bernal, J. L.; Del Nozal, M. J.; Bernal, J. Environ. Microbiol. Rep. 2010, 2, 243−250. (45) Nobeli, I.; Ponstingl, H.; Krissinel, E. B.; Thornton, J. M. J. Mol. Biol. 2003, 334, 697−719. (46) Gupta, S.; ires-de-Sousa, J. Mol. Divers. 2007, 11, 23−36. (47) Dobson, P. D.; Patel, Y.; Kell, D. B. Drug Discovery Today 2009, 14, 31−40. (48) Peironcely, J. E.; Reijmers, T.; Coulier, L.; Bender, A.; Hankemeier, T. PLoS One 2011, 6, No. e28966. (49) Kind, T.; Fiehn, O. BMC Bioinformatics 2007, 8, 105. (50) Wolf, S.; Schmidt, S.; Muller-Hannemann, M.; Neumann, S. BMC Bioinformatics 2010, 11, 148. (51) Gerlich, M.; Neumann, S. J. Mass Spectrom. 2013, 48, 291−298. (52) Allen, F.; Pon, A.; Wilson, M.; Greiner, R.; Wishart, D. Nucleic Acids Res. 2014, 42, W94−W99. (53) Wang, Y.; Kora, G.; Bowen, B. P.; Pan, C. Anal. Chem. 2014, 86, 9496−9503. (54) Schymanski, E. L.; Meringer, M.; Brack, W. Anal. Chem. 2011, 83, 903−912.
K
dx.doi.org/10.1021/jf504400c | J. Agric. Food Chem. XXXX, XXX, XXX−XXX
