Food Chemistry 121 (2010) 856–862

Contents lists available at ScienceDirect

Food Chemistry

journal homepage: www.elsevier .com/locate / foodchem

Analytical Methods

Botanical discrimination and classification of honey samples applyinggas chromatography/mass spectrometry fingerprinting of headspacevolatile compounds

Konstantinos A. Aliferis a,*, Petros A. Tarantilis b, Paschalis C. Harizanis c, Eleftherios Alissandrakis c

a Department of Plant Science, McGill University, 21111 Lakeshore Rd., Sainte-Anne-de-Bellevue, Quebec, Canada H9X 3V9b Laboratory of Chemistry, Department of Science, Agricultural University of Athens, 75 Iera Odos St., 118 55 Athens, Greecec Laboratory of Sericulture and Apiculture, Department of Crop Science, Agricultural University of Athens, 75 Iera Odos St., 118 55 Athens, Greece

a r t i c l e i n f o

Article history:Received 14 September 2009Received in revised form 3 November 2009Accepted 31 December 2009

Keywords:BiomarkersBotanical discriminationBotanical originChemometricsHeadspace GC/MSHoneySolid-phase microextraction (SPME)Volatiles

0308-8146/$ - see front matter � 2010 Elsevier Ltd. Adoi:10.1016/j.foodchem.2009.12.098

* Corresponding author. Tel.: +1 514 398 7851x071E-mail address: konstantinos.aliferis@mcgill.ca (K.

a b s t r a c t

A validated method for the discrimination and classification of honey samples performing GC/MS finger-printing of headspace volatile compounds was developed. Combined mass spectra of honey samples orig-inated from different plants and geographical regions of Greece were subjected to orthogonal partial leastsquares-discriminant analysis™ (OPLS™-DA), soft independent modelling of class analogy (SIMCA), andOPLS™-hierarchical cluster analysis (OPLS™-HCA). Analyses revealed an excellent separation betweenhoney samples according to their botanical origin with the percentage of misclassification to be as lowas 1.3% applying OPLS™-HCA. Fragments (m/z) responsible for the observed separation were assignedto phenolic, terpenoid, and aliphatic compounds present in the headspace of unifloral honeys. On theother hand, a variable classification for citrus and thyme honeys according to their geographical origincould be achieved. Results suggested that the developed methodology is robust and reliable for thebotanical classification of honey samples, and the study of differences in their chemical composition.

� 2010 Elsevier Ltd. All rights reserved.

1. Introduction

Honey is a very important product because of its nutritional andtherapeutic values (Molan, Mizrahi, & Lensky, 1997; Nasuti, Gab-bianelli, Falcioni, & Cantalamessa, 2006). Because of the high priceof certain honey types, adulteration with low cost and nutritionalvalue substances (Arvanitoyannis, Chalhoub, Gotsiou, Lydakis-Simantiris, & Kefalas, 2005; Cotte et al., 2007) or mislabellingregarding the botanical origin (Alissandrakis, Tarantilis, Harizanis,& Polissiou, 2007a) sometimes occurs. Nonetheless, the discrimina-tion between different types of honey is important for honeys thatpossess discrete aroma and taste (Alissandrakis et al., 2007a) andspecial nutritional properties (Lusby, Coombes, & Wilkinson,2005; Marghitas� et al., 2009; Molan, 1992) which make them pref-erable for consumers. Thus, the development of reliable, rapid andcost effective methodologies for their discrimination and classifica-tion based on their botanical origin is of high importance.

Several analytical platforms such as gas chromatography/massspectrometry (GC/MS) (Tananaki, Thrasyvoulou, Giraudel, & Mon-tury, 2007), nuclear magnetic resonance spectroscopy (NMR) (Lolli,

ll rights reserved.

8; fax: +1 514 398 7897.A. Aliferis).

Bertelli, Plessi, Sabatini, & Restani, 2008), infrared spectroscopy(IR) (Ruoff et al., 2006; Woodcock, Downey, Kelly, & O’Donnell,2007), front face fluorescence spectroscopy (Karoui, Dufour, Bosset,& De Baerdemaeker, 2007), atomic emission spectroscopy (AES),and inductively coupled plasma atomic emission spectrometry(ICP-AES) (Nozal Nalda, Bernal Yaguee, Diego Calva, & Martin Go-mez, 2005) have been used for the chemical analyses of honeysand their classification. Also, microscopic (Dimou, Katsaros, Tzavel-la-Klonari, & Thrasyvoulou, 2006) and physiochemical characteris-tics (Corbella & Cozzolino, 2006; Serrano, Villarejo, Espejo, & Jodral,2004) have been employed for the botanical and geographicaldetermination of honey samples.

Regarding the isolation of honey volatile compounds, severaltechniques have been applied, including the Likens–Nickersonmethodology, dynamic headspace, purge-and-trap systems, ultra-sound-assisted extraction and more recently the solid-phase mic-roextraction (Anklam, 1998; Cuevas-Glorya, Pino, Santiago, &Sauri-Duch, 2007).

The recent developments in chemometrics analyses have en-abled the authentication and classification of food products (Arva-nitoyannis et al., 2005; Tzouros & Arvanitoyannis, 2001). GC/MSfingerprinting has been proved to be a powerful platform forthe discrimination between honey samples (Baroni et al., 2006;

K.A. Aliferis et al. / Food Chemistry 121 (2010) 856–862 857

Senyuva et al., 2009). Therefore, in the present study a new meth-odology for the botanical discrimination and classification of hon-eys was developed, with their classification based on thegeographical origin to be a secondary task. Headspace solid-phasemicroextraction (HS-SPME) was performed for the extraction ofvolatile compounds followed by GC/MS fingerprinting. Chemomet-rics analyses were performed for the combined MS spectra. For theclassification and discrimination between the different types ofhoney and the detection of the corresponding biomarkers, princi-pal components analysis (PCA), orthogonal partial least squares-discriminant analysis™ (OPLS™-DA), soft independent modellingof class analogy (SIMCA), and OPLS™-hierarchical cluster analysis(OPLS™-HCA) were carried out. Results of the analyses highlightthe potential of the use of combined mass spectra for the discrim-ination and classification of honey samples performing GC/MS fin-gerprinting of headspace volatile compounds.

2. Materials and methods

2.1. Chemicals and reagents

Benzophenone was purchased from Fluka Chemika (Buchs,Switzerland) and methanol from Merck (Darmstadt, Germany).

2.2. Botanical and geographical origins of honey samples

A total of 77 unifloral honey samples of different botanical ori-gins namely, chestnut (Castanea sativa, six samples), cotton (Gossy-pium hirsutum, six samples), fir (Abies spp., six samples), heather(Erica manipuliflora, six samples), pine (Pinus spp., six samples),thyme (Corydothymus capitatus, 20 samples) and citrus (Citrusspp., 27 samples), were analysed (Table 1).

Citrus honey samples had been collected from several provincesof north (Arta) and south Greece (Argolida, Crete, Lakonia) andthyme honey samples from the Aegean islands of Calimnos, Cos,and Leros, and Crete. Samples were kept at �20 �C in glass vials.Prior to analyses, samples were left to thaw for 24 h at roomtemperature.

2.3. Isolation and gas chromatography–mass spectrometry analysis ofheadspace volatile honey compounds

The isolation of the aroma compounds was performed using theSPME procedure. An SPME holder (Supelco, Bellefonte, PA, USA)and the divinylbenzene/carboxen/polydimethylsiloxane fibre(Supelco, Bellefonte, PA, USA) were used to extract headspace vol-atiles from honey. This type of fibre provides an excellent sorptioncapacity and the broadest range of extracted volatiles from theheadspace of honey (Cajka, Hajšlová, Cochran, Holadová, & Klimán-ková, 2007). The honey samples (6 mL of deionised water solutionof 3 g honey mL�1) were placed in 15 mL screw-top vials withpolytetrafluoroethylene/silicone septa. Benzophenone was usedas the internal standard and 20 lL of a solution of 10 lg mL�1 in

Table 1Number of honey samples and corresponding period of harvesting.

Botanical origin Number of samples Time of collection

Chestnut 6 JuneCitrus 27 AprilCotton 6 AugustFir 6 JuneHeather 6 NovemberPine 6 OctoberThyme 20 July

methanol were added in the samples prior to extraction. The vialswere maintained in a water bath at 60 �C under constant stirringduring the whole procedure. Equilibration time was set at30 min, followed by 60 min headspace sampling time. The extrac-tion conditions had been previously optimised based on data ofprevious study (Alissandrakis, Tarantilis, Harizanis, & Polissiou,2007b).

The analysis of the isolated compounds was performed using aHewlett Packard 5890 II GC, equipped with a Hewlett Packard 5972MS detector. The column used was an HP-5MS (Crosslinked 5% PHME Siloxane) capillary column (30 m � 0.25 mm i.d., 0.25 lm filmthickness) and the gas carrier was Helium at 1 mL min�1 rate. Theinjector and MS-transfer line temperatures were maintained at220 �C and 290 �C, respectively. Oven temperature was held at40 �C for 3 min, raised to 160 �C at 3 �C min�1 and then to 200 �Cat 10 �C min�1. Electron impact mass spectra were recorded at amass range of 40–500 Da. An electron ionisation system was usedwith ionisation energy of 70 eV.

2.4. Mass spectral data processing

Acquisition of total ion chromatograms, collection of combinedMS spectra, and automated peak deconvolution were performedusing the Hewllett Packard ChemStation G1701AA, VersionA.03.00. The assignment of m/z fragments to the correspondingheadspace volatile compounds was based on fragmentation pat-tern data of our previous works (Alissandrakis et al., 2007a,2007b). Additionally, mass spectra searches were performed usingthe libraries of the National Institute of Standards and Technology(NIST) and on-line databases [The Golm Metabolome Database(http://csbdb.mpimp-golm.mpg.de/csbdb/gmd/msri/gmd_sspq.html)and the Spectral Database for Organic Compounds SDBS (http://riodb01.ibase.aist.go.jp/sdbs/cgi-bin/direct_frame_top.cgi)].

Data were expressed as ratios of the abundance of each m/zfragment against the abundance of the fragment m/z = 182, whichis the molecular ion of the internal standard (benzophenone) thatwas used. Extraction of the ion chromatogram for this fragment(m/z = 182) revealed that benzophenone does not naturally occurin any of the isolated compounds. Data were then exported toMS Excel� and fragments not related to honey volatiles, such asm/z = 44 which corresponds to CO2, were excluded from furtheranalyses. Additionally, major fragments corresponding to naphtha-lene (m/z = 128) and p-dichlorobenzene (m/z = 146 and 148), com-pounds that were formerly used in the common apiculture practise(Bogdanov et al., 2004; Harizanis, Alissandrakis, Tarantilis, & Polis-siou, 2008; Tananaki, Thrasyvoulou, Karazafiris, & Zotou, 2006),were excluded. Finally, after the removal of the fragments non-re-lated to honey headspace volatiles, the data matrix was composedof 77 columns (honey samples) and 154 variables (m/z fragments).Prior to pattern recognition analysis, data were further normalisedby dividing the relative abundances by their sum.

2.5. Multivariate data analyses and visualisation

For chemometrics analyses the pre-processed data were im-ported into the SIMCA-P + 12.0 software (Umetrics, MKS Instru-ments Inc., Sweden). A free demo version of the software isavailable (http://www.umetrics.com/default.asp/pagename/down-loads_software/c/1). Data were Pareto scaled (1/

pSD) which is a

compromise between unit variance scaling (UV) and no scaling(Eriksson, Johansson, Kettaneh-Wold, & Wold, 2001). PCA,OPLS™-DA, and SIMCA were performed for the whole dataset in or-der to discriminate and classify honey samples according to theirbotanical origin and detect corresponding biomarkers. The classifi-cation of honey samples according to their geographical origin wasa secondary task of the research. Performing PCA in a set of

858 K.A. Aliferis et al. / Food Chemistry 121 (2010) 856–862

observations representing one or more groups, knowledge relatedto group membership is not used to find the location of the princi-pal components. Thus, for the detection of differences between thehoney samples, OPLS™-DA was chosen because it is a modificationof the PLS-DA that has been designed to handle the variation in Xthat is orthogonal to Y. The partitioning of the X-data applyingOPLS™-DA, provides improved model transparency and interpret-ability, without modifying its predictive power (Trygg & Wold,2002). The detection of the influential fragments (m/z) was basedon scaled and centred PLS regression coefficients (CoeffCS) with95% jack-knifed confidence intervals (Efron & Gong, 1983). Themultivariate analysis was based on the methodology of Aliferisand Chrysayi-Tokousbalides (2006). Additionally, OPLS™-HCAwas applied for selected datasets in order to further evaluate theability of the developed models to classify honey samples accord-ing to their botanical and geographical origins using the Ward’sclustering (Ward, 1963).

2.6. Performance statistics

For the evaluation of the developed models, the cumulativefraction of the total variation of the X’s that could be predictedby the extracted components [predictive ability, Q2

(cum)] and thefraction of the sum of squares of all X’s (R2X) and Y’s (Y2X) ex-plained by the current component (explained variation) were used.

3. Results and discussion

3.1. Chemometrics analysis for the discrimination and classification ofhoney samples according to their botanical origin and the detection ofbiomarkers

3.1.1. Principal components analysis (PCA)The principal task of the present study was the botanical dis-

crimination between honey samples, their classification, and the

Fig. 1. (a) OPLS™-DA PC1/PC2 score plot for the whole dataset. The ellipse representsOPLS™-HCA using the Ward’s method. Honey samples are grouped according to their bothyme ( )].

detection of corresponding biomarkers. For the initial overviewof the dataset and the detection of trends and outliers, PCA wascarried out. Analysis showed no samples being outside the Hotell-ing T2 95% confidence ellipse that could influence the analyses, andhigh values of explained variation (R2X = 0.77) and predictive abil-ity [Q2

(cum) = 0.67, 4 PCs].

3.1.2. Orthogonal partial least squares™-discriminant analysis(OPLS™-DA)

Performing OPLS™-DA a model was developed from the train-ing sets of observations of known membership class. Cross valida-tion was performed for the training sets using the default softwareoptions giving high values for the R2X = 0.82, R2Y = 0.79, andQ2

(cum) = 0.73 (6 PCs). A very satisfactory grouping of honey sam-ples could be achieved according to their botanical origin(Fig. 1a). In order to further examine the predictive ability of themodel and detect the corresponding biomarkers for the observedseparation, OPLS™-DA was performed comparing all possible pairsof groups of honey samples. Analyses revealed a very strong dis-crimination between the different types of honey compared inpairs, as it is indicated by the high values of Q2

(cum) with the excep-tion of the pair fir/pine for which a moderate separation could beachieved [Q2

(cum) = 0.68] (Table S1 in Supplementary data).

3.1.3. Orthogonal partial least squares™-hierarchical cluster analysis(OPLS™-HCA)

In agreement to results of OPLS™-DA, application of OPLS™-HCA for the whole dataset resulted to a very good classificationof the samples according to their botanical origin (Fig. 1b). Out ofthe 77 samples that were simultaneously analysed, only one sam-ple (heather honey) was incorrectly classified as fir honey (Fig. 1b),a percentage of misclassification of 1.3%. The percentage of theright classification of samples based on their botanical origin wasraised from 98.7% to 100% when citrus and thyme groups were ex-cluded from the dataset (Fig. 2a and b).

the Hotelling T2 with 95% confidence, and (b) OPLS™-dendrogram illustrating thetanical origin [chestnut ( ), citrus ( ), cotton ( ), fir ( ), heather ( ), pine ( ), and

Fig. 2. (a) OPLS™-DA PC1/PC2 score plot for the whole dataset excluding citrus and thyme honeys. The ellipse represents the Hotelling T2 with 95% confidence, and (b)OPLS™-dendrogram illustrating the OPLS™-HCA using the Ward’s method. Honey samples are grouped according to their botanical origin [chestnut ( ), cotton ( ), fir ( ),heather ( ), and pine ( )].

K.A. Aliferis et al. / Food Chemistry 121 (2010) 856–862 859

3.1.4. Soft independent modelling of class analogy (SIMCA)SIMCA was additionally carried out in order to examine the

hypothesis that following the applied protocols, a simultaneousdiscrimination of citrus and thyme samples could be achieved.For each of those two types of honey, a separate PCA model wascomputed which was used to infer membership class of all treat-ments. The results are summarised in the Coomans’ plot (Fig. 3)where the co-ordinates are the distances to the model values(Moderate outliers, DModX) for the prediction set observationsdown onto the thyme model (X-axis) and the citrus model (Y-axis)(Eriksson et al., 2001). As Fig. 3 shows, there are no samples in thelower left part of the plot which means that there are no honeysamples with fragmentation pattern that fit to both citrus andthyme models. Samples in the lower right and the upper left partbelong to thyme and citrus models, respectively. All other samplesare clustered into the upper right area in which honey samples thatfit to none of the citrus and thyme models can be found. The resultscross-validated those obtained from OPLS™-DA and OPLS™-HCA,indicating an excellent classification and discrimination for citrusand thyme honey samples following the applied protocols.

3.1.5. Detection of biomarkers for the observed discriminationsFurther analyses of the data using scaled and centred PLS

regression coefficients with 95% jack-knifed confidence intervals,unravelled the fragments (m/z) responsible for the observed dis-crimination and classification of honey samples (Table S2 in Sup-plementary data). For the assignment of the fragments to thecorresponding headspace volatiles, results of previous works (Alis-sandrakis et al., 2007a, 2007b) were combined with MS databasesearches. Fragments were assigned to phenolic (m/z 43, 65, 91,and 92), terpenoid (m/z 41, 43, 55, 67, 71, and 93), and aliphaticcompounds such as acids (m/z 74 and 87), esters (m/z 41, 43, 55,56, and 57), alcohols (m/z 41, 43, 55, 56, and 57), and aldehydes

(m/z 41, 43, 55, 56, 57, 69, 70, 81, 82, 84, 95, 96, and 110). Addition-ally, m/z fragments influential for the separation of citrus andthyme honeys were assigned to the major compounds previouslyisolated from the headspace of Greek unifloral citrus (Alissandrakiset al., 2007b) and thyme (Alissandrakis et al., 2007a) honeys(Table S3 in Supplementary data). In Fig. 4 representative chro-matograms of citrus and thyme honey samples and correspondingbiomarkers for their discrimination are presented. Fragments ofthe lilacaldehyde and the 1-p-menthen-9-al isomers, limoneneand methyl anthranilate were influential for the classification ofcitrus honey samples whereas fragments of the phenylacetalde-hyde, 1-phenyl-2,3-butanedione, 3-hydroxy-4-phenyl-2-butanone,3-hydroxy-1-phenyl-2-butanone, and 3-hydroxy-4-phenyl-3-bu-ten-2-one were influential for the classification of thyme honeysamples.

3.2. Discrimination and classification of citrus and thyme honeysaccording to their geographical origin

3.2.1. Geographical discrimination between citrus honey samplesThe application of OPLS™-DA for the citrus dataset revealed

that a variable discrimination could be achieved between samplescollected from different geographical regions. A moderate discrim-ination could be achieved between citrus honey samples collectedfrom Argolida and Chania [Q2

(cum) = 0.74], Argolida and Lakonia[Q2

(cum) = 0.64], Arta and Lakonia [Q2(cum) = 0.88], and Chania and

Lakonia [Q2(cum) = 0.87]. On the other hand, a poor discrimination

was obtained comparing in pairs samples collected from Argolidaand Arta [Q2

(cum) = 0.21], and Arta and Chania [Q2(cum) = 0.20]

(Table S4 in Supplementary data and Fig. 5a). OPLS™-HCA con-firmed the unsatisfactory classification of citrus honeys based ontheir geographical origin, with samples collected from different re-gions to be clustered in the same clusters. For example, samples

Fig. 3. Coomans’ plot with normalised prediction distances (DModX) to citrus (j) model plotted versus distance to thyme (.) model for the whole dataset. The criticaldistances (DCrit) are indicated by solid lines and correspond to 95% tolerance intervals [no class (*)].

Fig. 4. Representative total ion chromatograms of thyme (a) and citrus (b) honey samples extracted performing headspace solid-phase microextraction. Compounds thatserve as biomarkers for their discrimination are indicated.

860 K.A. Aliferis et al. / Food Chemistry 121 (2010) 856–862

collected from Argolida, Arta, and Chania are clustered in clustersII, III and IV in Fig. 5b. On the other hand, all samples collected fromLakonia were cluster in cluster I (Fig. 5b).

3.2.2. Geographical discrimination between thyme honey samples.For the thyme honey dataset, the application of OPLS™-DA re-

vealed a satisfactory discrimination between samples collected

Fig. 5. (a) OPLS™-DA PC1/PC2 score plot for citrus honeys collected in Argolida (j), Arta ( ), Chania ( ), and Lakonia ( ). The ellipse represents the Hotelling T2 with 95%confidence, and (b) OPLS™-dendrogram illustrating the OPLS™-HCA using the Ward’s method.

Fig. 6. (a) OPLS™-DA PC1/PC2 score plot for thyme honeys collected in Calymnos ( ), Cos ( ), Crete ( ), and Leros ( ). The ellipse represents the Hotelling T2 with 95%confidence, and (b) OPLS™-dendrogram illustrating the OPLS™-HCA using the Ward’s method.

K.A. Aliferis et al. / Food Chemistry 121 (2010) 856–862 861

862 K.A. Aliferis et al. / Food Chemistry 121 (2010) 856–862

from Calimnos and Cos [Q2(cum) = 0.76], and Crete and Calimnos

[Q2(cum) = 0.73]. A moderate discrimination was observed between

samples collected from Crete and Cos [Q2(cum) = 0.51], and Leros

and Calimnos [Q2(cum) = 0.51], whereas a poor discrimination was

observed between honey samples collected from Crete and Leros[Q2

(cum) = 0.21], and Leros and Cos [Q2(cum) = 0.05] (Table S5 in Sup-

plementary data and Fig. 6a). OPLS™-HCA confirmed the results ofthe OPLS™-DA (Fig. 6b). The only satisfactory classification couldbe achieved between honey samples collected from broader geo-graphical regions [i.e. between Crete/Leros (clusters I–IV) andCos/Calymnos (clusters V–VII)].

4. Conclusion

The developed chemometrics model based on combined MSspectra of headspace volatiles of 77 samples of honey from sevenof the most common botanical origins provided a strong discrimi-nation applying OPLS™-DA and a percentage of correct classifica-tion of samples higher than 98% performing OPLS™-HCA.Additionally, compounds belonging to several chemical groupssuch as phenolics, terpenoids, and aliphatics (i.e. acids, esters, alco-hols, and aldehydes) were detected as biomarkers for the observeddiscrimination. The results clearly showed that HS-SPME–GC/MSfingerprinting of honey volatiles combined with state-of-the-artchemometrics analyses is a robust, reliable, rapid, and of high po-tential method for the discrimination and classification of honeysamples based on their botanical origin. Therefore, it can be consid-ered as a non-time consuming and cost effective methodology suit-able for routine analyses of honeys for their botanical classification.

On the other hand, the developed models showed a variable dis-criminative ability for citrus and thyme honey samples accordingto their geographical origin. This is not surprising since the regionsfrom which the samples had been collected share a very similar cli-mate and composition of plant populations.

To our knowledge, the present work is the first report highlight-ing the use and potential of combined MS spectra in GC/MS finger-printing studies. The use of combined MS spectra provides asignificant advantage to such approach since the identification ofthe isolated compounds is not a priori required. The results re-vealed the applicability of this approach in targeted chemometricsstudies of volatile compounds of honey and possibly other foodproducts and/or biological samples.

Appendix A. Supplementary data

Supplementary data associated with this article can be found, inthe online version, at doi:10.1016/j.foodchem.2009.12.098.

References

Aliferis, K. A., & Chrysayi-Tokousbalides, M. (2006). Metabonomic strategy for theinvestigation of the mode of action of the phytotoxin (5S, 8R, 13S, 16R)-(�)-pyrenophorol using 1H nuclear magnetic resonance fingerprinting. Journal ofAgricultural and Food Chemistry, 54, 1687–1692.

Alissandrakis, E., Tarantilis, P. A., Harizanis, P. C., & Polissiou, M. (2007a).Comparison of the volatile composition in thyme honeys from several originsin Greece. Journal of Agricultural and Food Chemistry, 55, 8152–8157.

Alissandrakis, E., Tarantilis, P. A., Harizanis, P. C., & Polissiou, M. (2007b). Aromainvestigation of unifloral Greek citrus honey using solid-phase microextractioncoupled to gas chromatography–mass spectrometry. Food Chemistry, 100,396–404.

Anklam, E. (1998). A review of the analytical methods to determine thegeographical and botanical origin of honey. Food Chemistry, 63, 549–562.

Arvanitoyannis, I. S., Chalhoub, C., Gotsiou, P., Lydakis-Simantiris, N., & Kefalas, P.(2005). Novel quality control methods in conjunction with chemometrics(multivariate analysis) for detecting honey authenticity. Critical Reviews in FoodScience and Nutrition, 45, 193–203.

Baroni, M. V., Nores, M. L., Díaz, M. P., Chiabrando, G. A., Fassano, J. P., Costa, C., et al.(2006). Determination of volatile organic compound patterns characteristic of

five unifloral honey by solid-phase microextraction–gas chromatography–massspectrometry coupled to chemometrics. Journal of Agricultural and FoodChemistry, 54, 7235–7241.

Bogdanov, S., Kichenmann, V., Seiler, K., Pfefferli, H., Frey, T., Roux, B., et al. (2004).Residues of p-dichlorobenzene in honey and beeswax. Journal of ApiculturalResearch, 43, 14–16.

Cajka, T., Hajšlová, J., Cochran, J., Holadová, K., & Klimánková, E. (2007). Solid phasemicroextraction-comprehensive two-dimensional gas chromatography–time-of-flight mass spectrometry for the analysis of honey volatiles. Journal ofSeparation Science, 30, 534–546.

Corbella, E., & Cozzolino, D. (2006). Classification of the floral origin of Uruguayanhoneys by chemical and physical characteristics combined with chemometrics.LWT-Food Science and Technology, 39, 534–539.

Cotte, J. F., Casabianca, H., Lhéritier, J., Perrucchietti, C., Sanglar, C., Waton, H., et al.(2007). Study and validity of 13C stable carbon isotopic ratio analysis by massspectrometry and 2H site-specific natural isotopic fractionation by nuclearmagnetic resonance isotopic measurements to characterize and control theauthenticity of honey. Analytica Chimica Acta, 582, 125–136.

Cuevas-Glorya, L. F., Pino, J. A., Santiago, L. S., & Sauri-Duch, E. (2007). A review ofvolatile analytical methods for determining the botanical origin of honey. FoodChemistry, 103, 1032–1043.

Dimou, M., Katsaros, J., Tzavella-Klonari, K., & Thrasyvoulou, A. (2006).Discriminating pine and fir honeydew honeys by microscopic characteristics.Journal of Apicultural Research, 45, 16–21.

Efron, B., & Gong, G. (1983). A leisurely look at the bootstrap, the jackknife, andcross-validation. The American Statistician, 37, 36–48.

Eriksson, L., Johansson, E., Kettaneh-Wold, N., & Wold, S. (2001). Multi- andmegavariate data analysis. Principles and applications. Umeå, Sweden: UmetricsAcademy.

Harizanis, P. C., Alissandrakis, E., Tarantilis, P. A., & Polissiou, M. (2008). Solid-phasemicroextraction/gas-chromatographic/mass spectrometric analysis of p-dichlorobenzene and naphthalene in honey. Food Additives and Contaminants,25, 1272–1277.

Karoui, R., Dufour, E., Bosset, J.-O., & De Baerdemaeker, J. (2007). The use of frontface fluorescence spectroscopy to classify the botanical origin of honey samplesproduced in Switzerland. Food Chemistry, 101, 314–323.

Lolli, M., Bertelli, D., Plessi, M., Sabatini, A. G., & Restani, C. (2008). Classification ofItalian honeys by 2D HR-NMR. Journal of Agricultural and Food Chemistry, 56,1298–1304.

Lusby, P. E., Coombes, A. L., & Wilkinson, J. M. (2005). Bactericidal activity ofdifferent honeys against pathogenic bacteria. Archives of Medical Research, 36,464–467.

Marghitas�, L. A., Dezmirean, D., Moise, A., Bobis, O., Laslo, L., & Bogdanov, S. (2009).Physico-chemical and bioactive properties of different floral origin honeys fromRomania. Food Chemistry, 112, 863–867.

Molan, P. C. (1992). The antibacterial activity of honey. 2. Variation in the potency ofthe antibacterial activity. Bee World, 73, 59–76.

Molan, P. C., Mizrahi, A., & Lensky, Y. (1997). Honey as an antimicrobial agent. In A.Mizrahi & Y. Lensky (Eds.), Bee products: Properties, applications and apitherapy(pp. 27–37). New York: Plenum Press.

Nasuti, C., Gabbianelli, R., Falcioni, G., & Cantalamessa, F. (2006). Antioxidative andgastroprotective activities of anti-inflammatory formulations derived fromchestnut honey in rats. Nutrition Research, 26, 130–137.

Nozal Nalda, M. J., Bernal Yaguee, J. L., Diego Calva, J. C., & Martin Gomez, M. T.(2005). Classifying honeys from the Soria province of Spain via multivariateanalysis. Analytical and Bioanalytical Chemistry, 382, 311–319.

Ruoff, K., Luginbuehl, W., Kuenzli, R., Iglesias, M. T., Bogdanov, S., Bosset, J. O., et al.(2006). Authentication of the botanical and geographical origin of honey bymid-infrared spectroscopy. Journal of Agricultural and Food Chemistry, 54,6873–6880.

Senyuva, H. Z., Gilbert, J., Silici, S., Charlton, A., Dal, C., Cürel, N., et al. (2009).Profiling Turkish honeys to determine authenticity using physical and chemicalcharacteristics. Journal of Agricultural and Food Chemistry, 57, 3911–3919.

Serrano, S., Villarejo, M., Espejo, R., & Jodral, M. (2004). Chemical and physicalparameters of Andalusian honey: classification of citrus and eucalyptus honeysby discriminant analysis. Food Chemistry, 87, 619–625.

Tananaki, C., Thrasyvoulou, A., Giraudel, J. L., & Montury, M. (2007). Determinationof volatile characteristics of Greek and Turkish pine honey samples and theirclassification by using Kohonen self organising maps. Food Chemistry, 101,1687–1693.

Tananaki, C., Thrasyvoulou, A., Karazafiris, E., & Zotou, A. (2006). Contamination ofhoney by chemicals applied to protect honeybee combs from wax-moth(Galleria mellonela L.). Food Additives and Contaminants, 23, 159–163.

Trygg, J., & Wold, S. (2002). Orthogonal projections to latent structures (OPLS).Journal of Chemometrics, 16, 119–128.

Tzouros, N. E., & Arvanitoyannis, I. S. (2001). Agricultural produces: Synopsis ofemployed Quality Control methods for the authentication of foods for theclassification of foods according to their variety of geographical origin. CriticalReviews in Food Science and Nutrition, 41, 287–319.

Ward, J. H. (1963). Hierarchical grouping to optimize an objective function. Journalof the American Statistical Association, 58, 236–244.

Woodcock, T., Downey, G., Kelly, J. D., & O’Donnell, C. (2007). Geographicalclassification of honey samples by near-infrared spectroscopy: A feasibilitystudy. Journal of Agricultural and Food Chemistry, 55, 9128–9134.