Metabolomic fingerprinting employing DART-TOFMS for authentication of tomatoes and peppers from organic and conventional farming

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.

Food Additives and Contaminants2012, 1–12, iFirst

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.

Keywords: authenticity; chemometrics; statistical analysis; traceability; nutrition; vegetables; organic foods

Introduction

Growing consumer concerns about the quality and

safety of food they eat are reflected in a continuously

increasing demand for organic farming products. They

are believed to be healthier and taste better. To

confirm these expectations, several research studies

have been conducted to compare the composition

of various crops originated from organic and con-

ventional farms (Woese et al. 1997; Carbonaro et al.

2002; Rembialkowska 2007; Lairon 2009; Willer and

Kilcher 2011).Available data show that organic crops typically

have a higher dry weight and, probably on this

account, also a higher content of vitamins and minerals

(Woese et al. 1997; Hajslova et al. 2005;

Rembialkowska 2007; Lairon 2009). The absence/

trace (background) levels of pesticide residues and

lower concentrations of nitrates represent further

benefits associated with organic foods (Woese et al.

1997; Rembialkowska 2007; Lairon 2009). A lot

of attention has been also paid to various natural

toxins, which are considered to be a part of plant

protective systems. In spite of assumptions on their

higher levels in organic crops, no statistically signifi-

cant differences were found (e.g. glycoalkaloids in

tomatoes and potatoes, furanocoumarins in celeriac

and parsnip) (Woese et al. 1997; Bourn and Prescott

2002; Siderer et al. 2005; Pussemier et al. 2006; Winter

and Davis 2006; Rembialkowska 2007; Lairon 2009).

In some cases, increased levels of phenolics were

reported in organic fruits/vegetables such as apple,

peach, pear, tomato, pepper or potatoes (Woese et al.

1997; Carbonaro et al. 2002; Mitchell et al. 2007;

Lairon 2009).Because of lower yields (in average by 20%),

organic products tend to be retailed at a higher price

than their conventional counterparts and thus, not

surprisingly, organic production might be susceptible

to fraud (Rembialkowska 2007). Under these condi-

tions, reliable analytical methods should be available

to verify their authenticity. It should be noted that this

is not an easy task. In spite of intensive research, a

simple authentication strategy applicable in routine

practice has not been developed so far.

*Corresponding author. Email: [email protected]

ISSN 1944–0049 print/ISSN 1944–0057 online

� 2012 Taylor & Francis

http://dx.doi.org/10.1080/19440049.2012.690348

http://www.tandfonline.com

Dow

nloa

ded

by [

Vys

oka

Skol

a C

hem

icko

], [

Han

a N

ovot

na]

at 0

4:59

24

July

201

2

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.

Material and methods

Chemicals and samples

Methanol (HPLC grade) was supplied by Merck

(Darmstadt, Germany). Polyethylene glycol (PEG,average relative molecular weight 600) was purchased

from Sigma-Aldrich (Steinheim, Germany). Water was

purified with the use of a Milli-Q system (Millipore,

Eschborn, Germany).Tomato samples consisted of five varieties (stan-

dard tomatoes: Rumba, Mercury and Akord; cherry

tomatoes: Picolino and Conchita) and pepper samples

of three varieties (Roberta, Spartacus and Berceo).

Both tomato and pepper samples were produced in

plastic tunnels on four different farms, located in the

Mazovia region, Poland. The samples were produced

in the years 2007 and 2008. Two of the farms used

organic and two conventional farming method. The

total number of samples was 40 and 24 for tomatoes

and pepper, respectively. Mazovia region is a flat

lowland and the distance between the farms was

relatively small (on average 15 km), which means they

were subjected to similar climatic conditions in the

individual years.The growing season in 2007 (April–August) was

warmer (average temperature¼ 15.74�C) but with a

lower sum of rainfall (292mm) compared with the

growing season in 2008 (average temperature¼

15.08�C, sum of rainfall¼ 319mm). The average

temperatures and sums of rainfall in each month

during both production seasons are listed in Table 1,

which shows that the largest differences in temperaturewere in May, while the rainfall differed significantly in

each month.Organic samples were grown in line with European

Union Regulations 834/2007 and 889/2008, that is,

without the usage of synthetic fertilisers and pesticides

in organic farming. Plant protection against pests was

thus based on herbal extracts. In the conventional

systems, cow manure together with mineral fertilisers

was used. For protection against pests, commercialpesticides were used. The list of all samples used in this

study together with detailed information on farm

localities, soil type, fertilisation and plant protection

is shown in Tables 2A and 2B.The soil types were similar for all farms, as

indicated in Tables 2A and 2B. The organic farms

were converted to this way of farming more than 5

years before the start of this trial, which should be

enough for the stabilisation of the soil qualities

(Zadoks 1989; Oehl et al. 2004). Cow manure was

used as the fertiliser and phacelia as the pre-crop in the

organic system 3 years before the trial. Additionally,

the tomato and pepper plantations were controlled

three times during the vegetation period: in June, at the

end of July and at the end of August (during harvest),

2 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

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).

Year

April May June July August

�C mm �C mm �C mm �C mm �C mm

2007 8.5 18.0 14.9 52.0 18.1 79.0 18.6 111.0 18.6 32.02008 9.0 51.0 12.6 84.0 17.2 33.0 18.1 100.0 18.5 51.0

Food Additives and Contaminants 3

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

2A.Overview

ofthestudiedtomato

samples.Characteristics

ofthelocalities,soiltype,fertiliser

regim

eandplantprotectionusedfororganic

andconventionalcultivation

are

alsopresented.

Farm

Localisation

Varieties

Soiltype

Fertilisers

Plantprotection

system

Fertilisationrange(kgha�1)

NP

KMg

Ca

Microelem

ents

Organic

farm

no.1

Kaszew

skaWola,

51�300N

20�550

E

Rumba,Merkury,

Akord,Picolino,

Conchita

Sandy-loamy,middle

soil,IIIcategory

(35%

floatable

particles)

Green

manure

(35a

or25btha�1),

Patentkali(P:S

30:10),basaltic

flour,Humobak

(mixture

ofbac-

teriaandsoil

fungi)

BioczosBR

(garlic

extract),

Antifung20SL

(Biohumus

mixture)

120a,

140b

25a,b

80a,

90b

50a,

40b

120a,

170b

Yes

Organic

farm

no.2

Radzanow,51�560

N20�050E

Rumba,Merkury,

Akord,Picolino,

Conchita

Sandy-loamy,slight

soil,II

category

(19%

floatable

particles)

Compost

60%,

Biohumus

Grevit200SL

(grapefruit

extract

20%)

130a,

140b

50a,

40b

110a,

150b

45a,

65b

150a,

130b

Yes

Conventional

farm

no.1

Kaszew

skaWola,

51�190N

20�580

E

Rumba,Merkury,

Akord,Picolino,

Conchita

Sandy-loamy,middle

soil,IIIcategory

(32%

floatable

particles)

Cowmanure

(20aor

10btha�1),

Organic

chalk,

calcium

and

magnesium

nitrate

Sumilex

500SC

(procymidone)

160a,

170b

120a,

130b

150a,

170b

70a,

60b

130a,

150b

No

Conventional

farm

no.2

Sew

erynow,51�280

N20�540E

Rumba,Merkury,

Akord,Picolino,

Conchita

Sandy-loamy,slight

soil,II

category

(15%

floatable

particles)

Polidap(N

:P:S

18:46:5),

Grandar(N

:P:K

3:9:19),

Nitrosofosphate

Mospilan20SP

(acetamipiryde),

BetoksanS050

SL

(2-naftoxya-

ceticacid)

160a,

180b

110a,

150b

150a,

165b

50a,b

170a,

190b

Yes

Notes:

aIn

productionyear2007.

bIn

productionyear2008.

4 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

2B.

Overviewofthestudiedpepper

samples.Characteristics

ofthelocalities,soiltype,fertilisersregim

eandplantprotectionusedfororganicandconventionalcultivation

are

alsopresented.

Farm

Localisation

Varieties

Soiltype

Fertilisers

Plantprotection

system

Fertilisationrange(kgha�1)

NP

KMg

Ca

Microelem

ents

Organic

farm

no.1

Kaszew

skaWola,

51�300N

20�550

E

Roberta,

Spartacus,

Berceo

Sandy-loamy,

middle

soil,III

category

(35%

floatable

particles)

Green

manure

(Phacelia

juss)(20aor14bt

ha�1),Patentkali(P:S

30:10),basaltic

flour

BioczosBR

(garlic

extract

25%

),Grevit200SL

(grapefruit

extract

20%

)

150a,

130b

35a,

25b

80a,b

45a,

55b

140a,

160b

Yes

Organic

farm

no.2

Radzanow,51�560

N20�050E

Roberta,

Spartacus,

Berceo

Sandy-loamy,

slightsoil,II

category

(19%

floatable

particles)

Compost

(35aor25b

tha�1)

Grevit200SL

(grapefruit

extract

20%

)

145a,

135b

55a,

45b

145a,

135b

65a,

55b

160a,

140b

Yes

Conventional

farm

no.1

Kaszew

skaWola,

51�190N

20�580E

Roberta,

Spartacus,

Berceo

Sandy-loamy,

middle

soil,III

category

(32%

floatable

particles)

Cow

manure

(10aor20b

tha�1),PolimagS

(N:P:K

:Mg:S

10:8:15:5:4),

Hydrokompleks

(N:P:K

:Mg

12:11:18:25),Polyfeed

(N:P:K

:Mg

15:30:15:2)

Mospilan20SP

(acetami-

piryde),

Sumilex

500S.C.

(procymidone)

180a,

160b

140a,

120b

160a,

140b

75a,

65b

135a,

125b

No

Conventional

farm

no.2

Oronsko,51�180

N20�550E

Roberta,

Spartacus,

Berceo

Sandy-clay,slight

soil,II

category

(20%

floatable

particles)

Patentkali(P:S

30:10),

Polidap(N

:P:S

18:46:5),Grandar

(N:P:K

3:9:19),cal-

cium

nitrate,organic

chalk

Sumilex

500S.C.

(procymidone),

Karate

050CS

(lambda-cyha-

lotryne),

Vertimec

018

EC

(abamek-

tyne),Dithane

NeoTec

75WG

(mankozeb)

170a,

190b

145a,

135b

155a,

165b

50a,

70b

150a,

190b

Yes

Notes:

aIn

productionyear2007.

bIn

productionyear2008.

Food Additives and Contaminants 5

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 3A. Analytical data of selected ions (markers) determined by DART-TOFMS (tomato extract).

m/z Probable ion elemental formula Mass error (ppm) Tentative identification RSDg (%)

(A) DART positive-ion mode84.044 C4H6NO �10.71 24.6985.026 C4H5O2 �31.22 2-Furanonea 21.4797.028 C5H5O2 �11.90 Furfural (degradation product of pentose)a 3.1299.044 C5H7O2 �11.11 3.42103.041 C4H7O3 10.68 �-Ketobutyric acida 11.58111.044 C6H7O2 �9.01 Catechola 9.35113.023 C5H5O3 �5.04 Degradation product of hexosea 9.53115.039 C5H7O3 �1.74 Degradation product of hexosea 6.26127.040 C6H7O3 6.14 Maltol/hydroxymethylfural

(degradation product of fructose)a5.38

129.054 C6H9O3 �9.07 15.02130.050 C5H8NO3 �1.54 Pyroglutamic acida 7.56133.050 C5H9O4 �4.40 12.70145.049 C6H9O4 �8.27 Hexoseb 3.31163.059 C6H11O5 �10.12 Hexosec 2.85180.083 C6H14NO5 �22.76 13.67192.099 C11H14NO2 �19.54 3.86198.099 C6H16NO6 6.74 10.10212.092 C10H14NO4 �2.75 7.22224.091 C11H14NO4 �7.51 4.22230.102 C10H16NO5 �3.26 3.35236.093 C12H14NO4 3.03 8.92238.074 C11H12NO5 9.88 4.39246.099 C10H16NO6 4.62 3.56254.097 C12H16NO5 �22.23 9.37256.081 C11H14NO6 �5.13 7.92268.103 C9H18NO8 �2.40 17.74271.083 C12H15O7 5.24 3.76289.096 C12H17O8 12.99 6.03325.187 C18H29O3S 9.10 7.52

(B) DART negative-ion mode85.031 C4H5O2 28.75 7.5387.008 C3H3O3 �1.38 Pyruvic acidd 24.4989.024 C3H5O3 0.34 Lactic acidd 42.11101.023 C4H5O3 �4.65 �-Ketobutyric acidd 36.55111.008 C5H3O3 �5.59 Degradation product of hexosea 17.20113.023 C5H5O3 �7.70 Degradation product of hexosea 9.10115.003 C4H3O4 �3.78 Fumaric/maleic acidd/malic acide 3.75117.019 C4H5O4 2.69 Succinic acidd 38.17128.034 C5H6NO3 �7.56 Pyroglutamic acidd 6.87129.020 C5H5O4 6.32 Itaconic acidd 9.85131.034 C5H7O4 �1.03 Glutaric acidd 8.83133.013 C4H5O5 �3.76 Malic acidd 4.66143.034 C6H7O4 �5.84 Hexosee 12.18159.028 C6H7O5 �5.97 8.18161.046 C6H9O5 6.83 Hexosef 7.97173.010 C6H5O6 5.12 Dehydroascorbic acidd 10.19179.055 C6H11O6 �0.92 Hexosed 9.91191.019 C6H7O7 �1.99 Citric/isocitric acidd 14.52205.034 C7H9O7 �2.10 8.90215.033 C12H7O4 �7.14 3.98217.028 4.10254.065 C11H12NO6 �7.34 3.85308.097 C11H18NO9 �2.46 Fructosyl glutamated 11.62309.084 C11H17O10 7.20 7.64313.078 C10H17O11 2.91 19.79

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 (%)

(A) DART positive-ion mode69.034 C4H5O �3.97 Degradation product of hexosea 1.6070.066 C4H8N �1.66 12.7172.081 C4H10N �2.16 6.6682.065 C5H8N �2.92 2.3884.081 C5H10N �3.90 4.3185.029 C4H5O2 �4.68 2-Furanonea 2.8186.097 C5H12N �3.77 8.1097.029 C5H5O2 �1.88 Furfural (degradation product of pentose)a 1.4099.044 C5H7O2 �3.11 3.63100.076 C5H10NO �2.19 27.06101.023 C4H5O3 �4.75 7.30109.029 C6H5O2 �3.62 Degradation product of hexosea 5.78115.039 C5H7O3 �4.26 Degradation product of hexosea 2.84116.071 C5H10NO2 �3.31 Prolinea 17.32118.086 C5H12NO2 �4.18 Valinea 1.34120.081 C8H10N �2.20 4.70127.039 C6H7O3 �4.17 Maltol/hydroxymethylfuran

(degradation product of fructose)a2.86

132.102 C6H14NO2 �5.48 2.89142.122 C8H16NO �5.55 6.26145.050 C6H9O4 �2.96 Hexoseb 6.08146.081 C6H12NO3 �3.01 11.35163.060 C6H11O5 �4.35 Hexosec 18.98166.086 C9H12NO2 �4.30 Phenylalaninea 5.23179.107 C11H15O2 �1.70 12.93180.086 C6H14NO5 �5.16 6.79197.117 C11H17O3 �2.69 14.52198.098 C6H16NO6 �1.08 13.13207.138 C13H19O2 �1.76 11.57325.114 C12H21O10 0.23 28.61342.139 C12H24NO10 �2.12 25.78442.192 C17H32NO12 �1.84 14.91456.207 C18H34NO12 �2.66 40.82

(B) DART negative-ion mode71.013 C3H3O2 �6.63 Lactatee 18.6387.008 C3H3O3 �6.07 Pyruvic acidd 16.0888.039 C3H6NO2 �6.70 Alanined 10.45101.023 C4H5O3 �7.42 �-Ketobutyric acidd 15.14105.018 C3H5O4 �5.86 Glycerated 10.63113.023 C5H5O3 �5.04 17.89114.055 C5H8NO2 �5.38 Prolined 7.30116.071 C5H10NO2 �4.17 Valined 7.89117.018 C4H5O4 �5.34 Succinic acidd 10.45118.050 C4H8NO3 �4.99 Threonined 10.82119.034 C4H7O4 �5.67 Tetrosed 9.72128.070 C6H10NO2 �6.20 10.56129.018 C5H5O4 �4.92 Itaconic acidd 5.13130.086 C6H12NO2 �4.34 Leucine/isoleucined 6.95131.034 C5H7O4 �4.01 Glutaric acidd 23.75133.013 C4H5O5 �5.04 Malic acidd 6.42133.049 C5H9O4 �6.35 11.11143.034 C6H7O4 �4.16 Hexosee 25.62144.065 C6H10NO3 �5.75 2.33147.029 C5H7O5 �5.37 Citramalic acidd 5.30149.044 C5H9O5 �3.89 Pentosed 7.72159.028 C6H7O5 �5.47 7.74161.044 C6H9O5 �3.23 Hexosef 4.05164.070 C9H10NO2 �5.45 Phenylalanined 4.67173.008 C6H5O6 �5.46 Dehydroascorbic acidd 4.60175.023 C6H7O6 �5.68 Ascorbic acidd 5.88

(continued )

Food Additives and Contaminants 7

Dow

nloa

ded

by [

Vys

oka

Skol

a C

hem

icko

], [

Han

a N

ovot

na]

at 0

4:59

24

July

201

2

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 (%)

176.031 C6H8O6 �6.25 6.70177.040 C6H9O6 �2.17 Gluconolactoned 4.00178.047 C6 H10O6 �2.98 7.57179.046 C8H7N2O3 0.90 5.63179.055 C6H11O6 �5.00 Hexosed 3.40181.059 C5H11NO6 0.01 9.37191.019 C6H7O7 �0.01 Citric/isocitric acidd 26.84191.056 C7H11O6 0.00 Quinic acidd 4.71193.049 C10H9O4 �5.46 Ferulic acidd 26.89205.035 C7H9O7 �0.49 19.97215.032 C10H5N3O3 �4.19 5.18217.029 Unknown 8.18225.061 C7H13O8 �2.20 Glucoheptonic acidd 16.88226.056 Unknown 7.24239.076 C8H15O8 �2.11 15.69242.051 C6H12NO9 �1.81 5.74255.072 C8H15O9 0.51 5.70267.073 C9H15O9 3.59 3.99268.103 C9H18NO8 �0.20 7.46269.088 C9H17O9 2.60 4.38269.212 C16H29O3 2.56 25.25269.248 C17H33O2 �0.50 57.77285.083 C9H17O10 3.10 7.64294.120 C11H20NO8 4.07 8.96296.135 C11H22NO8 2.08 4.73297.083 C10H17O10 1.53 15.95298.114 C10H20NO9 0.77 14.42308.135 C12H22NO8 �0.05 15.02310.151 C12H24NO8 2.02 4.03344.135 C15H22NO8 0.86 8.05359.120 C12H23O12 3.19 13.67435.296 C22H43O8 0.79 15.77

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.

8 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

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.

Food Additives and Contaminants 9

Dow

nloa

ded

by [

Vys

oka

Skol

a C

hem

icko

], [

Han

a N

ovot

na]

at 0

4:59

24

July

201

2

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 (%)

Predictionability (%)

(A) Tomatoes (total 40 samples)Organic (n¼ 20) versus conventional (n¼ 20) 24 PCs 97.5 82.5Four localities (n¼ 10/locality) 25 PCs 100 70Five varieties (n¼ 8/variety) 26 PCs 92.5 45Crop year 2007 (n¼ 20) versus 2008 (n¼ 20) 12 PCs 100 85

(B) Peppers (total 24 samples)Organic (n¼ 12) versus conventional (n¼ 12) 21 PCs 100 87.5Four localities (n¼ 6/locality) 19 PCs 100 45.8Three varieties (n¼ 8/variety) 15 PCs 100 75Crop year 2007 (n¼ 12) versus 2008 (n¼ 12) 7 PCs 95.2 95.2

10 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

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

ambient conditions. Anal Chem. 77:2297–2302.Cooks RG, Ouyang Z, Takats Z, Wiseman JM. 2006.

Ambient mass spectrometry. Science.

311(5767):1566–1570.Defernez M, Kemsley EK. 1997. The use and misuse of

chemometrics for treating classification problems. Trends

Anal Chem. 16(4):216–221.Gomez-Romero M, Segura-Carretero A, Fernandez-

Gutierrez A. 2010. Metabolite profiling and quantification

of phenolic compounds in methanol extracts of tomato

fruit. Phytochemistry. 71(16):1848–1864.

Hajslova J, Cajka T, Vaclavık L. 2011. Challenging appli-

cations offered by direct analysis in real time (DART) in

food-quality and safety analysis. Trends Anal Chem.

30(2):204–218.Hajslova J, Schulzova V, Slanina P, Janne K, Hellenas KE,

Andersson C. 2005. Quality of organically and conven-

tionally grown potatoes: four-year study of micronutri-

ents, metals, secondary metabolites, enzymic browning

and organoleptic properties. Food Addit Contam.

22(6):514–534.Lairon D. 2009. Nutritional quality and safety of organic

food. A review. Agron Sustain Dev. 30(1):33–41.

Mitchell AE, Hong YJ, Koh E, Barrett DM, Bryant DE,

Denison RF, Kaffka S. 2007. Ten-year comparison of the

influence of organic and conventional crop management

practices on the content of flavonoids in tomatoes. J Agric

Food Chem. 55:6154–6159.Oehl F, Sieverding E, Mader P, Dubois D, Ineichen K,

Boller T, Wiemken A. 2004. Impact of long-term conven-

tional and organic farming on the diversity of arbuscular

mycorrhizal fungi. Oecologia. 138:574–583.Pussemier L, Larondelle Y, van Peteghem C,

Huyghebaert A. 2006. Chemical safety of conventionally

and organically produced foodstuffs: a tentative compar-

ison under Belgian conditions. Food Contr. 17:14–21.

Rembialkowska E. 2007. Quality of plant products from

organic agriculture. J Sci Food Agric.

87(15):2757–2762.Siderer Y, Maquet A, Anklam E. 2005. Need for research to

support consumer confidence in the growing organic food

market. Trends Food Sci Technol. 16:332–343.van Ruth S, Alewijn M, Rogers K, Newton-Smith E,

Tena N, Bollen M, Koot A. 2011. Authentication of

organic and conventional eggs by carotenoid profiling.

Food Chem. 126:1299–1305.

Venter A, Nefliu M, Cooks RG. 2008. Ambient desorption

ionization mass spectrometry. Trends Anal Chem.

27(4):284–290.

Willer H, Kilcher L. 2011. The world of organic agriculture.

Statistics and emerging trends 2011. FiBL-IFOAMReport.

Bonn, Germany/Frick, Switzerland: IFOAM and FiBL.Winter CK, Davis SF. 2006. Organic foods. J Food Sci.

71(9):R117–R124.Wishart DS. 2008. Metabolomics: applications to food

science and nutrition research. Trends Food Sci Technol.

19:482–493.Woese K, Lange D, Boess C, Bogl KW. 1997. A comparison

of organically and conventionally grown foods – results of

a review of the relevant literature. J Sci Food Agric.

74:281–293.Worthington V. 2001. Nutritional quality of organic versus

conventional fruits, vegetables, and grains. J Altern

Complement Med. 7(2):161–173.Zadoks JC. 1989. Development of farming systems.

Wageningen (the Netherlands): Pudoc.

12 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