This article was downloaded by: [Vysoka Skola Chemicko], [Hana Novotna] On: 24 July 2012, At: 04:59 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK Food Additives & Contaminants: Part A: Chemistry, Analysis, Control, Exposure & Risk Assessment Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/tfac20 Metabolomic fingerprinting employing DART-TOFMS for authentication of tomatoes and peppers from organic and conventional farming H. Novotná a , O. Kmiecik b , M. Gałązka b , V. Krtková a , A. Hurajová a , V. Schulzová a , E. Hallmann b , E. Rembiałkowska b & J. Hajšlová a a Department of Food Analysis and Nutrition, Institute of Chemical Technology, Prague, Czech Republic b Department of Functional Food and Commodities, Warsaw University of Life Sciences, Warsaw, Poland Accepted author version posted online: 04 May 2012. Version of record first published: 19 Jul 2012 To cite this article: H. Novotná, O. Kmiecik, M. Gałązka, V. Krtková, A. Hurajová, V. Schulzová, E. Hallmann, E. Rembiałkowska & J. Hajšlová (2012): Metabolomic fingerprinting employing DART-TOFMS for authentication of tomatoes and peppers from organic and conventional farming, Food Additives & Contaminants: Part A: Chemistry, Analysis, Control, Exposure & Risk Assessment, DOI:10.1080/19440049.2012.690348 To link to this article: http://dx.doi.org/10.1080/19440049.2012.690348 PLEASE SCROLL DOWN FOR ARTICLE Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. The publisher does not give any warranty express or implied or make any representation that the contents will be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses should be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or arising out of the use of this material.
13
Embed
Metabolomic fingerprinting employing DART-TOFMS for authentication of tomatoes and peppers from organic and conventional farming
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
This article was downloaded by: [Vysoka Skola Chemicko], [Hana Novotna]On: 24 July 2012, At: 04:59Publisher: Taylor & FrancisInforma Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,37-41 Mortimer Street, London W1T 3JH, UK
Food Additives & Contaminants: Part A: Chemistry,Analysis, Control, Exposure & Risk AssessmentPublication details, including instructions for authors and subscription information:http://www.tandfonline.com/loi/tfac20
Metabolomic fingerprinting employing DART-TOFMS forauthentication of tomatoes and peppers from organicand conventional farmingH. Novotná a , O. Kmiecik b , M. Gałązka b , V. Krtková a , A. Hurajová a , V. Schulzová a , E.
Hallmann b , E. Rembiałkowska b & J. Hajšlová a
a Department of Food Analysis and Nutrition, Institute of Chemical Technology, Prague,Czech Republicb Department of Functional Food and Commodities, Warsaw University of Life Sciences,Warsaw, Poland
Accepted author version posted online: 04 May 2012. Version of record first published: 19Jul 2012
To cite this article: H. Novotná, O. Kmiecik, M. Gałązka, V. Krtková, A. Hurajová, V. Schulzová, E. Hallmann, E.Rembiałkowska & J. Hajšlová (2012): Metabolomic fingerprinting employing DART-TOFMS for authentication of tomatoesand peppers from organic and conventional farming, Food Additives & Contaminants: Part A: Chemistry, Analysis, Control,Exposure & Risk Assessment, DOI:10.1080/19440049.2012.690348
To link to this article: http://dx.doi.org/10.1080/19440049.2012.690348
PLEASE SCROLL DOWN FOR ARTICLE
Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions
This article may be used for research, teaching, and private study purposes. Any substantial or systematicreproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form toanyone is expressly forbidden.
The publisher does not give any warranty express or implied or make any representation that the contentswill be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses shouldbe independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims,proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly inconnection with or arising out of the use of this material.
Metabolomic fingerprinting employing DART-TOFMS for authentication of tomatoes and
peppers from organic and conventional farming
H. Novotnaa, O. Kmiecikb, M. Gala �zkab, V. Krtkovaa, A. Hurajovaa, V. Schulzovaa, E. Hallmannb,E. Rembialkowskab and J. Hajslovaa*
aDepartment of Food Analysis and Nutrition, Institute of Chemical Technology, Prague, Czech Republic;bDepartment of Functional Food and Commodities, Warsaw University of Life Sciences, Warsaw, Poland
(Received 20 January 2012; final version received 29 April 2012)
The rapidly growing demand for organic food requires the availability of analytical tools enabling theirauthentication. Recently, metabolomic fingerprinting/profiling has been demonstrated as a challenging optionfor a comprehensive characterisation of small molecules occurring in plants, since their pattern may reflect theimpact of various external factors. In a two-year pilot study, concerned with the classification of organic versusconventional crops, ambient mass spectrometry consisting of a direct analysis in real time (DART) ion sourceand a time-of-flight mass spectrometer (TOFMS) was employed. This novel methodology was tested on 40tomato and 24 pepper samples grown under specified conditions. To calculate statistical models, the obtaineddata (mass spectra) were processed by the principal component analysis (PCA) followed by linear discriminantanalysis (LDA). The results from the positive ionisation mode enabled better differentiation between organic andconventional samples than the results from the negative mode. In this case, the recognition ability obtained byLDA was 97.5% for tomato and 100% for pepper samples and the prediction abilities were above 80% for bothsample sets. The results suggest that the year of production had stronger influence on the metabolomicfingerprints compared with the type of farming (organic versus conventional). In any case, DART-TOFMS is apromising tool for rapid screening of samples. Establishing comprehensive (multi-sample) long-term databasesmay further help to improve the quality of statistical classification models.
A wide range of analytical approaches has beentested to distinguish organic crops from those grownunder conventional conditions. Screening of typicalcomposition pattern and levels of various specificcomponents such as macronutrients (proteins, sugars,lipids, etc.), minerals or minor components representedby vitamins, antioxidants and other biologically activesubstances is the most common practice (Worthington2001; Siderer et al. 2005; Lairon 2009). However, dueto a high variability of factors influencing cropcomposition, such target analysis of individual com-ponents or their groups does not provide a generalstrategy applicable for unbiased classification of theway of crop farming (Hajslova et al. 2005). Anothertesting strategy is based on the measurement of �15Nvalues in crops or soils on which they were grown.Typically, the mean �15N values for synthetic nitrogenfertilisers are significantly lower compared with thosefor the ‘natural’ fertilisers permitted in organicagriculture. This approach may provide evidencewhether artificial fertilisers were applied to crops;however, it does not say anything about the use ofpesticides or other agrochemicals prohibited in organicfarming (Bateman and Kelly 2007).
Non-target screening of some crops compositionalcharacteristics represents a promising alternative to thetarget analysis. Among the currently very popular‘omics’ fingerprinting strategies, metabolomics is prob-ably the most suitable for the authentication purpose.As a general principle, the aim of this method is todetect the broadest possible range of small moleculesoriginated by metabolomic pathways in a plant matrix.The impact of farming practices is assumed to influ-ence respective cellular processes (Wishart 2008; vanRuth et al. 2011).
Recently, ambient mass spectrometry has emergedas a challenging analytical tool applicable for metabo-lomic fingerprinting of compounds with molecularweight below 1 kDa. This technique requires little or nosample preparation and enables the recording of massspectra without prior separation (Wishart 2008). In therecent years, a large number of novel ambient desorp-tion ionisation techniques, such as desorption electro-spray ionisation (DESI), the atmospheric solidsanalysis probe (ASAP) or direct analysis in real time(DART) have become commercially available (Cookset al. 2006; Venter et al. 2008).
The aim of this pilot study was to investigate thepotential of metabolomic fingerprinting conducted byDART coupled with a time-of-flight mass spectro-meter (TOFMS) to distinguish between organicallyand conventionally grown tomatoes (Solanum lycoper-sicum L.) and sweet bell peppers (Capsicum annuumL.). Various sample preparation approaches and mea-surement conditions were tested. To process theobtained data, advanced chemometric techniqueswere employed.
and no leaf symptoms of nutrient deficiencies wererecognised.
Sample preparation
Samples were harvested in full maturity (red-ripe) inamounts of approximately 20 kg per cultivar, systemand farm. To obtain representative samples, thus tocompensate for possible variability in composition,tomatoes/peppers were taken at different parts of thefield and then pooled. Before analysis, every fruit wascleaned, divided into eight parts and four oppositepieces were taken. The samples were freeze-dried using2.5 kg ice per 24 h (freeze drier Labconco, Kansas City,Missouri, USA). After freeze-drying, the samples weremilled into a fine powder (average size of the particlesaround 100 mm) in a laboratory grinder A-11 (IKA,Staufen, Germany) at 19,500 rotations min�1 andstored at �80�C. A methanol–water mixture (1/1, v/v;2ml) was added to 0.2 g of the freeze-dried sample, themixture was intensively shaken and subsequentlycentrifuged (11,000 rpm, 5min). The supernatant wasanalysed using DART-TOFMS.
Instrumentation and experimental conditions
For the experiments, a DART-TOFMS system con-sisting of a DART ion source (IonSense, Saugus, MA,USA), an AccuTOF LP high-resolution time-of-flightmass spectrometer (JEOL (Europe), SAS, Croissy surSeine, France) and a HTC PAL autosamplerAutoDART-96 (Leap Technologies, Carrboro, NC,USA) was used. Helium was used as an ionisation gas.
The operating conditions of the DART ion sourcewere: both positive- and negative-ion mode; heliumflow: 3.5 lmin�1; discharge needle voltage: 3.0 kV;perforated electrode potential: þ150V; and grid elec-trode potentials: þ250 and �350V, for the positive-and negative-ion mode, respectively. Conditions ofTOFMS were: cone voltage: þ20V; monitored massrange: m/z 50–1000; acquisition rate: 5 spectra min�1;and resolving power: approximately 6000 FWHM (fullwidth at half maximum). The distance between theDART gun exit and mass spectrometer inlet was
10mm. Sample insertion was carried out automaticallyusing Dip-itTM samplers (IonSense, Saugus, MA,USA). The sampling glass rod was immersed for 1 sinto the sample hole of a deepwell micro-plate (LifeSystems Design, Merenschwand, Switzerland) andtransferred to the optimised position in front of theDART gun exit. The sample was then desorbed fromthe surface of the glass rod for 30 s by hot helium,creating ions outside the instrument, during which timethe spectral data were recorded. Each sample wasmeasured at least three times. To perform a mass driftcompensation for accurate mass measurements andelemental composition calculations, a polyethyleneglycol 500 mgml�1 solution in methanol was insertedmanually at the end of each run. The relative standarddeviation of the relative ion intensities was typicallyaround 10% (Tables 3A and 3B).
Data processing and chemometric analysis
The Mass Center software version 1.3 (JEOL, Tokyo,Japan) was used for data processing. The data wereobtained by averaging the intensities of the massspectra recorded during the exposure of the sample tothe DART gas beam; background ions were subtractedand the mass drift was corrected. Potential markerswere selected after careful examination of DART-MSspectra; the ions with significant intensities (threshold2% of the ion with the highest intensity in respectivemass spectrum) were chosen and ions with m/zbelonging to 13C isotope of other ion of the samecompound were excluded. All the selected ions areshown in Tables 3A and 3B. The data obtained fromrepeated sample introductions were transformed to aconstant row sum, i.e. the intensity of each ion wasdivided by the sum of intensities of all ions selected forthe given sample, and results of repeated measurementswere averaged.
In the follow-up step, chemometric tools wereemployed for the processing of MS data. The spectralinformation (ions characterised by m/z value andrelative intensity) obtained by instrumental analysiswere firstly processed by principal component analysis(PCA), which allowed the visualisation of multidimen-sional information in the form of a few principal
Table 1. Climatic conditions at the production region during the growing season 2007 and 2008: mean temperatures (�C) andsums of rainfall (mm).
Notes: a[MþH]þ.b[MþH–2H2O]þ.c[MþH–2H2O]þ.d[M�H]�.e[M�H–2H2O]�.f[M�H–H2O]�.gRelative standard deviation (RSD) of peak area. The intensities of detected ions were normalised to constant sum of all ions.
6 H. Novotna et al.
Dow
nloa
ded
by [
Vys
oka
Skol
a C
hem
icko
], [
Han
a N
ovot
na]
at 0
4:59
24
July
201
2
Table 3B. Analytical data of selected ions (markers) determined by DART-TOFMS (pepper extract).
m/z Probable ion elemental formula Mass error (ppm) Tentative identification RSDg (%)
components retaining the maximum possible variabil-ity within the data set.
After PCA, linear discriminant analysis (LDA) wasapplied to reduced data set to establish a predictivemodel for sample classification. Calculation of lineardiscriminant functions, which maximise the ratio ofvariance between respective classes and, at the sametime, minimise the ratio of within-class variance, wasthe basis of this supervised pattern recognition tech-nique. Classification results of LDA are presented interms of recognition and prediction abilities. Therecognition ability represents the percentage of suc-cessfully classified samples in the training set and theprediction ability is the percentage of correctly classi-fied samples in the test set by the model developedduring the training step. For the prediction ability, aleave-one-out cross-validation procedure was used(Berrueta et al. 2007).
PCA and LDA were performed by the software
package statistiXL version 1.8 (statistiXL, Broadway-
Nedlands, Australia).
Results and discussion
As mentioned in the Introduction, the fingerprinting
strategy should enable one to detect as many metabo-
lome components as possible in order to support the
classification of sample categories. In the particular
case, we aimed to obtain comprehensive DART-
TOFMS spectral information on the basis of which
organically grown tomatoes and peppers could be
distinguished from those crops grown conventionally.
For this purpose, an analytical procedure based on
ambient mass spectrometry was optimised.
Table 3B. Continued.
m/z Probable ion elemental formula Mass error (ppm) Tentative identification RSDg (%)
Notes: a[MþH]þ.b[MþH–2H2O]þ.c[MþH–2H2O]þ.d[M�H]�.e[M�H–2H2O]�.f[M�H–H2O]�.gRelative standard deviation (RSD) of peak area. The intensities of detected ions were normalised to constant sum of all ions.
Both positive and negative ions can be generated bythe DART ion source. The type and intensity ofobtained ions depend on various factors like solventtype or helium gas beam temperature (Hajslova et al.2011). Among the tested extraction solvents, a meth-anol–water mixture (1/1, v/v) was chosen for follow-upexperiments since the highest number of ionisablecompounds (in the m/z range of 50–400) occurred inextract prepared from both tomatoes and peppers.Regarding temperature, which is one of the key factorsaffecting experimental results, optimal thermal desorp-tion and ionisation were obtained at 250�C. At highertemperatures, a signal drop of some ions in the highermass region was observed, probably due to thermaldegradation of the respective compounds.
Analysis of tomato samples
Figure 1 shows an example of mass spectra of tomatosamples from organic and conventional farmingsystem. Based on the data, 29 and 25 markers werechosen in the positive and negative mode, respectively.The number of variables was then reduced usingchemometric tools represented by PCA followed byLDA. For each LDA model, the data obtained byanalysis of all samples were used for classification.
Markers obtained in the positive mode providedbetter results when developing a model which couldpredict the production system (organic versus conven-tional). Employing the PCA and LDA chemometrictechniques, the recognition ability of the obtainedmodel was 97.5% and the prediction ability was 82.5%(i.e. 33 samples out of 40 were correctly classified).These results were obtained using 24 principalcomponents. It should be noted that to avoid model
over-fitting the number of initially selected variables
should not exceed the value of (n� g)/3, where n is the
number of objects; and g is the number of classes
(Defernez and Kemsley 1997). In our case, the number
of objects was 40 and the number of classes was 2, so
the calculated maximum number of variables was 13.
Since the number of used variables (24 principal
components) was higher, it can be concluded that the
model for distinguishing between organic and conven-tional tomatoes could not be classified as fully reliable.
However, if only 13 variables were used according to
the above-mentioned criterion, the recognition and
prediction abilities were only 80% and 65%, respec-
tively. A higher number of samples would thus be
necessary for possible improvement of the model.We also tried to distinguish between samples from
the four production localities. The metabolomic
DART-TOFMS fingerprints were processed in the
same way as when distinguishing between the farming
systems. Using 25 principal components obtained from
pre-selected ions on which PCA was applied, the
calculated recognition ability was 100%, but the
prediction ability was ‘only’ 70% (i.e., 28 samples out
of 40 were correctly classified). Although this value
may seem relatively low, it can still be considered as
‘reasonable’ since four different classes (localities) were
classified. Furthermore, different fertilising and plant
protection systems were used in each locality, as shown
in Table 2A. The different treatment regimes in the
individual localities thus had a stronger influence on
the metabolomic profiles than the way of farming.However, as well as the differentiation of organic
and conventional samples, this model was over-fittedtoo, because 25 variables were too many to fulfil the
over-fitting criterion (maximum should be 12).
Figure 1. Examples of DART-TOFMS mass spectra of tomato extracts (variety Picolino from crop year 2008): (a) conventionalproduction in negative-ion mode, (b) organic production in negative-ion mode, (c) conventional production in positive-ion mode,and (d) organic production in positive-ion mode.
To distinguish between tomato samples from twoconsecutive crop years (2007 and 2008), the modelbased on the negative-ion mode provided better results.The obtained recognition ability was 100% and theprediction ability was 85%. The model was calculatedusing 12 principal components, which means that theover-fitting criterion was met, i.e. the model could beclassified as reliable.
Since especially acidic compounds ionise in thenegative mode, it can be assumed that the differencesin the climatic conditions during the growing seasonbetween the two crop years probably influenced thecontent of organic acids. In any case, the distinguishingof samples from different crop years was to someextent influenced by the ripeness of the crop at the timeof its harvest, as it is rather difficult to get completelyidentical physiological status of samples in two con-secutive years.
When distinguishing between tomato samples offive different varieties (using data from all the sam-ples), we obtained better results using variables fromthe positive mode. The recognition ability was 92.5%and the prediction ability was 45%. These results wereobtained using 26 principal components. The overviewof all these models is shown in Table 4.
Analysis of pepper samples
An example of mass spectra of pepper samples fromorganic and conventional farming system is shown inFigure 2. From the obtained data, 32 and 58 markerswere chosen in the positive and negative mode,respectively. The strategy employed for data processingwas the same as that described for the tomato samples.The number of variables was then reduced usingchemometric tools PCA and LDA. For each LDAmodel, the data obtained by analysis of all sampleswere used for classification. To distinguish betweensamples from the organic and conventional productionsystem, as in the case of tomatoes, markers obtained in
the positive mode showed better results. Employing thePCA and LDA techniques, the achieved recognitionability was 100% and the prediction ability was 87.5%.To obtain these results, 21 principal components wereused. According to the criterion (n� g)/3, the numberof variables in this case should not exceed seven. Thiscriterion was not met due to the fact that we had only24 samples available for this study. If only sevenvariables were used according to the criterion, therecognition and prediction abilities were 66.7% and50%, respectively.
As regards distinguishing between pepper samplesfrom four different localities, we achieved worse resultscompared with the tomato samples. The recognitionability was 100%, but the prediction ability was only45.8% (these results were achieved using 19 principalcomponents).
The negative ionisation mode provided, again as inthe case of tomatoes, better results for distinguishingbetween pepper samples from the two crop years. Weobtained both the recognition and prediction ability of95.2% using only seven variables. This means that theearlier mentioned over-fitting criterion was met and themodel may be considered as reliable.
Three different varieties of peppers were repre-sented in our set of samples and the positive modeenabled better differentiation between these varieties.The achieved recognition ability was 100% and theprediction ability was 75%. These results wereachieved using 15 principal components. The overviewof all these models is shown in Table 4.
Tentative identification of marker metabolites
All the above-mentioned chemometric models werebased on the metabolomic fingerprinting approach. Inother words, we used non-target analyses of a widerange of metabolites without previous identification.However, the high-resolution TOF-MS detector pro-vides accurate mass spectra, so the calculation of
Table 4. Overall summary of PCA-LDA models.
Model (number of samples) Variables usedRecognitionability (%)
elemental composition of ions is possible. Forconfirmation, the evaluation of isotopic patterns ofrespective ions can be used. A wide range of com-pounds can be ionised under DART ion sourceconditions. In the positive mode, protonated ions[MþH]þ, electron ions Mþ� or adduct ions[MþNH4]
þ typically occur in the mass spectra. Inthe spectra from the negative mode, deprotonatedmolecules [M�H]�, electron ions M�� or adduct ions[MþCl]� could be present (Cody et al. 2005; Hajslovaet al. 2011). The overview of tentatively identifiedmetabolites in tomato and pepper samples is shown inTables 3A and 3B. In both cases, more compoundscould be identified in the negative mode. Thesecompounds were mainly polar primary metabolites(amino acids and organic acids). In peppers, alsophenolic acids, typical secondary metabolites, could befound. It should be noted that for the purpose of thisstudy, the degradation of metabolome components wasnot a problem since the ion intensities are proportionalto the amount of their precursors, so distinguishing ofsamples differing in their content is not hampered. Asfar as a detailed identification of metabolome compo-nents is required, then the LC-ESI-TOFMS (liquidchromatography-electrospray ionisation-time of flightmass spectrometry) technique might be a better option.In addition to retention times, which might be helpfulfor identification purpose, the commonly usedelectrospray provides more efficient ionisation ofsample components compared to DART. In this ionsource, only those ionisable compounds which can bethermo-desorbed might be detected (this is not possiblefor many polar compounds, which can be extracted by
the methanol–water mixture). For instance, in a recentstudy (Gomez-Romero et al. 2010) concerned withmetabolite profiling of tomato samples, 135 com-pounds were identified by the use of the LC-ESI-TOFMS technique.
Conclusions
The knowledge obtained in this two-year pilot studyconcerned with the classification of organic versusconventional tomato and pepper samples can besummarised as follows:
. The DART-TOFMS technique is applicablefor a very rapid examination of crop samples.Using the methanol/water mixture for extrac-tion, mainly polar metabolome componentswere isolated and their ions recorded in bothpositive and negative ionisation mode.
. Based on the obtained fingerprints (massspectra), suitable marker ions were selectedfor statistical processing by PCA and subse-quently LDA to obtain statistical models forsample classification.
. The differentiation between samples fromdifferent farming systems (organic and con-ventional) was reasonable. The recognitionabilities were 97.5% and 100% and predictionabilities were 82.5% and 87.5% for the tomatoand pepper samples, respectively. However,the year of production had a stronger impacton the measured metabolomic fingerprintscompared with the farming system.
Figure 2. Examples of DART-TOFMS mass spectra of pepper extracts (variety Roberta from crop year 2008): (a) conventionalproduction in negative-ion mode, (b) organic production in negative-ion mode, (c) conventional production in positive-ion mode,and (d) organic production in positive-ion mode.
Food Additives and Contaminants 11
Dow
nloa
ded
by [
Vys
oka
Skol
a C
hem
icko
], [
Han
a N
ovot
na]
at 0
4:59
24
July
201
2
. Most of the obtained models were not reliableenough, thus a larger number of samplesshould be employed for future improvementof statistical models for classification. Themodel employed for differentiation of samplesfrom the 2 years of production was the onlyone that could be considered as reliable.
. Using the high-resolution TOF-MS detector,it was possible to estimate the elementalcomposition of detected ions and tentativelyidentify 25 and 38 compounds in the tomatoand pepper extracts, respectively.
Acknowledgements
This study was carried out with financial support from theMinistry of Education, Youth and Sports, Czech Republic(MSM 6046137305) and specific university research (MSMTNo. 21/2012); the Polish team also wants to thank the PolishMinistry of Agriculture and Rural Development for financialsupport.
References
Bateman A, Kelly S. 2007. Fertilizer nitrogen isotope
signatures. Isot Environ Hlth S. 43(3):237–247.Berrueta LA, Alonso-Salces RM, Heberger K. 2007.
Supervised pattern recognition in food analysis. J
Chromatogr A. 1158:196–214.Bourn D, Prescott J. 2002. A comparison of the nutritional
value, sensory qualities, and food safety of organically and
conventionally produced foods. Crit Rev Food Sci Nutr.
42(1):1–34.Carbonaro M, Mattera M, Nicoli S, Bergamo P,
Cappelloni M. 2002. Modulation of antioxidant com-
pounds in organic vs conventional fruit (peach, Prunus
persica L., and pear, Pyrus communis L.). J Agric Food
Chem. 50(19):5458–5462.Cody RB, Laramee JA, Durst HD. 2005. Versatile new ion
source for the analysis of materials in open air under