Introduction

Honey is a higly nutritious natural foodstuff used in the diet of many civilizations since ancient times. It is produced by Apis mellifera honeybees from the nectar of numerous flowers or from honeydew secretions present in conifer trees. It is a supersaturated sugar solution, where sugars are the main constituents accounting for about 95% of dry matter [1]. Water is the second most abundant constituent (12–23%), while low proportions of other compounds such as organic acids, minerals, vitamins, enzymes, proteins, amino acids, Maillard reaction products, volatile compounds, several bioactive substances (phytochemicals), as well as pollen grains have been also reported contributing to the overall composition of honey [2].

A modern trend in the food industry involves the consumption of minimally processed authentic and functional food products, to gain consumer preference and support, due to such products’ documented beneficial health properties. Unifloral honeys often have a higher price in the domestic and international market due to their unique chemical composition and sensory characteristics.

There are indicative studies in the literature, over the past years, using conventional [2] and instrumental methods of analysis including gas chromatography coupled to mass spectrometry [3], high-performance liquid chromatography [4, 5], high-performance anion-exchange chromatography [6], infrared spectroscopy [7], front-phase fluorimetric spectroscopy [8], nuclear magnetic resonance (NMR) [9,10,11,12,13,14,15,16,17,18], inductively coupled plasma optical emission or mass spectrometry [19, 20] or isotope ratio mass spectrometry [21], in combination with supervised statistical techniques, to characterize and accurately determine specific honey chemical markers with respect to honey quality and authenticity.

Several articles have been published, indicating the high efficiency of NMR methods either 1H NMR or 13C NMR or their combination to classify honey samples according to botanical and geographical origin [12, 13, 15, 17, 22]. NMR is a powerful technique for structure determination [14]. It can, thus, provide an in-depth understanding of complex structures in complicated systems, such as food matrices [23].

There are several advantages of NMR spectroscopy compared to conventional methods of analysis, including its non-invasive nature, sample handling/time and cost of analysis, the relative ease and rapidity of data treatment, along with the ability to present information on a broad range of metabolites in a single run [16, 17, 24].

The resonances of sugars in 13C NMR are much more adequately resolved than in 1H NMR and thus, quantification is much easier [9]. However, 13C NMR has much lower sensitivity than 1H NMR and thus minor components are undetectable. In this work, to overcome the difficulties of the 1H NMR spectra complexity, the peaks were deconvoluted. Given the aforementioned higher sensitivity of 1H NMR, minor honey components were also quantified.

Recently, saccharides were determined in Greek honeys using 13C NMR [25]. However, this is the first time that 1H NMR is used for the same purpose.

To the best of our knowledge, the present set of data (fructose, glucose, maltose, sucrose, raffinose, turanose, formic acid, proline, phenylalanine, the sum of the dominant sugars or their ratios, as well as the ratio of dominant sugars to moisture content) has never been reported before, in Greek pine or fir honeys.

Given the increasing interest in consumption of honeydew honey in many countries worldwide [6, 11, 16, 17, 20], knowledge on Greek honeydew honey composition along with its identity could be of considerable interest to consumers, researchers and the food industry at an international level.

Thus, the aim of the present study was to differentiate the most common honeydew honeys produced in Greece using the combination of 1H NMR/HPLC based on metabolites and physicochemical data along with chemometric analysis.

Materials and methods

Sampling plan

70 representative honey samples (N = 70) collected from selected beekeepers and produced in 2011 consisted of pine (N = 39) and fir (N = 31) honeys, were subjected at the earliest to NMR and HPLC analyses. Details on honey sample origin and storage conditions are given in the supplementary material (Supplementary Table 1) [20].

Reagents and solutions

Deuterated water (99.9 atom % D, D2O), tetraethyl-ammonium bromide [(C2H5)4NBr], and maleic acid (C4H4O4, 99.94%) were purchased from Sigma–Aldrich (Germany). Maltose was obtained from Merck (Darmstadt, Germany). Acetonitrile (HPLC grade), methanol (HPLC grade), ammonium hydroxide and ethylenediamine were also obtained from Merck (Darmstadt, Germany).

Determination of moisture content

Moisture content (g/100 g) of pine and fir honeys was determined by drying honey until constant weight using a moisture analyzer (KERN, model MLB50-3, Balingen, Germany) [20]. Since water represents the second major constituent of honey, moisture content values were used in the development of the geographical classification model. Reported values are the average of two replicates (n = 2).

Determination of maltose content using HPLC

High-performance liquid chromatography was applied for the determination of maltose since its identification was not possible using 1H NMR. Reported maltose values are the average of three replicates (n = 3).

Preparation of honey samples for HPLC analysis

5 g of honey were weighed into a beaker and dissolved in 40 mL deionized water. Then, 25 mL of methanol was pipetted into a 100-mL volumetric flask and the honey solution was quantitatively transferred into the flask followed by filling to the mark, with deionized water. Finally, the obtained solution was filtered through a 0.45-µm membrane filter (Whatman, UK) prior to HPLC analysis [26]. Each sample was prepared in duplicates (n = 2).

HPLC apparatus operating conditions

HPLC analysis was performed with a SHIMADZU LC unit (Kyoto, Japan), consisting of a quaternary pump (LC-20AD), a thermostated column oven (CTO-10A), a 20-µL loop injector and a SHIMADZU ChemStation software for data analysis. Detection was carried out using a SHIMADZU refractive index detector (RID-10A).

A chromatographic column (Zorbax Rx-SIL, 250 mm × 4.6 mm, i.d. 5 µm, as the stationary phase, Hewlett–Packard, USA) was used. The column temperature was held at 25 °C. The mobile phase for isocratic elution was a mixture of water/acetonitrile (1:2.6 v/v) containing 0.03% (v/v) ethylenediamine as a modifier and ammonium hydroxide (0.05% v/v), used to adjust the pH to 9–10. The flow rate was 1.0 mL/min. Prior to analysis, a mixture of water/acetonitrile (1:2.6 v/v) containing 0.3% (v/v) ethylenediamine was run through the column forming a dynamic coating layer on the silica surface [27].

Quantification analysis

Maltose was quantified by comparison of its chromatographic peak area to that of the standard calibration curve. The calibration curve was prepared in triplicates (n = 3) at five different dilutions (0–10 mg/L). The determination coefficient for the calibration curve of maltose was R2 = 1.000. Limit of detection (LOD) and limit of quantification (LOQ) values were determined using regression parameters from the calibration curve (3 Sy/x/a and 10 Sy/x/a, respectively; where Sy/x is the standard deviation of the residues and a is the slope). The aforementioned values for maltose were: LOD = 0.015 and LOQ = 0.05 mg/kg. The developed HPLC method showed a relative standard deviation < 2%. A typical HPLC-RI chromatogram of fir honey sample no. 9 from Lakonia is given in the supplementary material (Supplementary Fig. 1).

NMR measurements

Sample preparation and calibration

The samples to acquire H1 NMR spectra were prepared by dissolving 50.0 mg of honey in 400 µL of D2O containing the standard. 13C spectra were acquired with samples prepared by dissolving 200 mg of honey in 400 µL of D2O containing the standard.

1H measurements at 300 MHz

All NMR measurements were performed on a Bruker Avance 300 Ultrashield spectrometer (Bruker BioSpin, Rheinstetten, Germany) equipped with a 5-mm SEI probe with Z-gradient coils, 1H NMR spectra were acquired at 300.0 K without sample rotation. 64 scans and 4 prior dummy scans of 32 k points were acquired with a spectral width of 12.0 ppm and 2 s relaxation between each scan. Water suppression was achieved using the NOESY-presaturation pulse sequence (Bruker 1D noesygppr1d pulse sequence) with irradiation at the water frequency (1890.60 Hz) during the recycle and mixing time delays. 13C spectra were acquired with 1024 scans and 32 k points covering 140 ppm and 5 s relaxation between each scan. A standard inverse-gated decoupling pulse sequence from a Bruker library was adopted to avoid the heteronuclear nuclear Overhauser effect derived from proton decoupling. All NMR spectra were phased and baseline-corrected.

NMR spectra preprocessing and quantification

The resonances of each spectrum were integrated by line fitting using the MultiSpec NMR software developed in house by Dr. Anastasios Keramidas at the University of Cyprus (a free version of the program can be obtained from Dr. Anastasios Keramidas). For the quantification of each component, the integral of each peak was compared to the integral of the internal standard (C2H5)4NBr (Table 1). The concentration of the internal standard [(C2H5)4NBr] was 0.126 mM. Results were expressed as g/100 g.

Table 1 Chemical shifts (δ, ppm) of major and minor metabolites determined in Greek pine and fir honeys by 1H NMR
Full size table

The concentration of (C2H5)4NBr was calculated by comparing its NMR signals to those of a 0.500 mM solution in D2O of maleic acid internal standard. All samples were prepared in duplicates (n = 2) to reduce to a minimum a possible sample inhomogeneity. The limit of detection (LOD) and limit of quantification (LOQ) calculated from the noise of the spectra were for 1H NMR 0.007 g and 0.023 g glucose per 100 g of honey and for 13C NMR 0.51 g and 1.70 g per 100 g of honey, respectively. The standard error from two measurements of two different samples of the same honey was less than the 0.3%. Reported values are the average of two independent replicates (n = 2). The developed NMR method showed a relative standard deviation < 5%.

Statistical analysis

Multivariate analysis of variance (MANOVA) using the SPSS v.20.0 statistics software was applied to the quantified data set (g/100 g) consisting of 15 variables: fructose, glucose, sucrose, maltose, turanose, raffinose, proline, phenylalanine, formic acid, and moisture, along with the sum of fructose and glucose (F + G), and the ratios of glucose/fructose (F/G), fructose/moisture (F/M), glucose/moisture (G/M), and F + G/M. This step is mandatory in multivariate analysis to investigate which variables (considering the possible effectiveness of all variables simultaneously) are significant (p < 0.05) for the geographical discrimination of pine and fir honeys.

As a follow-up test to MANOVA, linear discriminant analysis (LDA), a supervised statistical technique, was applied only to the significant parameters (dependent variables) (p < 0.05) to determine a linear combination of these groups of subjects which could differentiate pine and fir honeys according to geographical origin (Halkidiki, Thassos, Evia, Samos for pine honeys and Messinia, Karditsa, Lakonia, Arkadia for fir honeys) (independent variables). LDA is a supervised pattern recognition method that has been widely used for authentication purposes in different foodstuffs [20, 28].

Results and discussion

1H NMR profile of Greek pine and fir honeys

The 1H NMR spectra of pine and fir honeys are shown in Fig. 1a. In general, the resonances of the spectra from pine and fir honeys are well resolved although they are broader than resonances of, i.e., blossom honeys (work in preparation). This is probably owed to the presence of resin in pine and fir honeys that did not permit the perfect shimming of the sample. In addition, differences were also observed between the same honey types of different geographical origin. Quantitative analysis obtained by integration of 1H signals of the sugars and minor components of honeys are given in Table 2.

Fig. 1
figure 1figure 1

a 1H NMR spectra and assignments of the resonances in the anomeric region of one fir honey sample no. 8 from Arkadia and two pine honey sample no. 6 from Halkidiki. a Mid-sugars area, b upper-aromatic area and c low-acids area. b 13C NMR spectra and assignments of the resonances in the anomeric region of (a) full spectra of 1 (fir honey sample no. 8 from Arkadia) and 2 (pine honey sample no. 6 from Halkidiki), b expansion of the anomeric region of pine honey sample no. 6 from Halkidiki: F fructose, G glucose, T turanose, FF fructofuranose, FP fructopyranose, GP glucopyranose

Full size image
Table 2 Content (g/100 g) of different metabolites in Greek pine and fir honeys determined by 1H NMR and HPLC according to geographical origin
Full size table

The 1H NMR spectrum of an aqueous honey solution was largely dominated by saccharides, whereas amino and organic acids, characterizing the aliphatic and the aromatic region of the spectrum, were present at very low concentration. The sugar region between 3.49 and 5.22 ppm is the major informative part of the 1H NMR spectrum for the saccharide resonance assignment, containing all the anomeric and part of the side chain protons of saccharides. Although the area between 3.49 and 4.0 ppm is crowded, integration of the 1H NMR resonances by line fitting gave a satisfactory quantification of the sugars and gave results which are comparable to those of 13C NMR. The mid-low-frequency region between δ 4.20 and 3.00 ppm is dominated by very intensive signals of the major monosaccharides (glucose and fructose). Glucose and fructose signals (α and β isomers) were dominant, compared to the other anomeric signals of saccharides. In particular, two disaccharides (sucrose and turanose), and one trisaccharide (raffinose) were also identified (Table 1. The region between 2.00 and 2.50 ppm was the informative part of the 1H NMR spectrum for proline and other organic acids and amino acids [18].

The most characteristic resonances of pine and fir honeys (tree honeys) were those at 6.45 and 8.50 ppm. The resonance at 6.45 ppm was assigned to pyruvic acid, whereas that of 8.50 ppm was assigned to formic acid. Honeydew honeys did not exhibit any other resonances in the aromatic region. Resonances between 7.00 and 7.50 ppm were assigned to phenylalanine and tyrosine, and may be indicative of the mixed feeding of the bees, apart from honeydew secretions, with thyme or flower nectar. Thus, these resonances were used to distinguish between blossom and honeydew honeys (manuscript in preparation). Honey samples that did not contain any pyruvic acid, but only large quantities of phenylalanine and/or tyrosine, were excluded from the present study. However, honeys that contained both pyruvic acid and phenylalanine were included, because this is considered a regional characteristic.

Honey samples were also analyzed by 13C NMR spectroscopy, to evaluate the presence of different sugar isoforms (Fig. 1b). Quantitative analysis obtained by integration of 13C signals for both fructose and glucose, led to the calculation of glucose to fructose ratio (G/F) ratio for pine and fir honey samples. 13C NMR spectra of honey samples revealed the dominant signals of fructose and glucose in their different tautomeric isoforms. In particular, fructose was present in fructofuranose (FF) and fructopyranose (FP) forms, while glucose only in the form of glucopyranose (GP). The quantities of sugars were compared to those of 1H NMR and were found identical.

Sugar profile of Greek pine and fir honeys

The amount and type of sugars vary among honeys of different floral source, comprising, thus, a useful tool for the classification of monofloral honeys [6]. For example, variations in the content of some sugars were used to distinguish between blossom and honeydew honeys, because the latter contain lower levels of monosaccharides and higher values of trisaccharides (mainly melezitose, erlose, raffinose and maltotriose), as well as other higher oligosaccharides [29].The higher the glucose content (higher than 28–30 g/100 g) and melezitose (higher than 10 g/100 g) contents are, the faster the crystallization of honey is [30].

In the present study, slightly higher fructose content (1.89%) was recorded for pine honeys compared to fir honeys (both dark-colored honeys). Fructose content values (g/100 g) of Greek pine and fir honeys are higher (24.30 and 22.86%, respectively) compared to German honeydew honeys [16].

Glucose recorded higher values in pine compared to fir honeys by ca. 5.66%. Present glucose content values (g/100 g) involving Greek pine and fir honeys is considerably higher than those of German honeydew honeys (29.49 and 25.26%, respectively) [16]. To the best of our knowledge, there are very limited studies in the literature that provide quantification data regarding honey major sugars using 1H NMR methodology [16, 18].

Maltose content (g/100 g) was, in general, higher in fir honeys compared to pine honeys (Table 2). An exception to this observation was the high maltose content values in pine honey samples from Samos Island. Such high maltose content values maybe characteristic of Samos pine honey samples (p < 0.05). Ohmenhaeuser et al. [16] and Del Campo et al. [18], using NMR methodologies, did not detect maltose in Spanish and German honeys, respectively. Maltose average content values (g/100 g) obtained for pine (1.12) honeys are in general agreement with those of Spanish forest honeys (1.33) [31], while present average values of fir honeys (2.16) are higher than the average values (1.70) reported for French fir honeys by Cotte et al. [6].

Raffinose content (g/100 g) was higher in fir honeys compared to pine honeys, whereas respective turanose content was higher in pine honeys (Table 2). These sugars may be indicative of certain plant nectar secretions in the investigated regions (p < 0.05). To the best of our knowledge, there are no studies in the literature that provide quantification results regarding honey raffinose and turanose content, using 1H NMR methodology [16,17,18]. Thus, further comparison is not possible.

Comparing raffinose and turanose content values of Greek honeys with those in the literature, in which specific sugars were determined using high-performance liquid chromatography (HPLC), raffinose content values are higher than those reported for Brazilian honeydew honeys [5]. On the other hand, turanose content values of Greek pine and fir honeys are lower than those reported for Brazilian honeydew and French fir honeys, respectively [5, 6]. The same holds for Spanish dark-colored honeys, which recorded higher turanose content values (g/100 g) compared to honeydew honeys of the present study [31].

High sucrose content in honey is related to its botanical origin, honey immaturity, high nectar flux or artificial feeding of bees. In the present study, pine and fir honey samples recorded sucrose content values lower than 5 g/100 g. The aforementioned value is considered the upper limit by European legislation related to honey [32]. It should also be mentioned that significant differences (p < 0.05) were observed in sucrose content values of pine and fir honey samples according to geographical origin.

At this point, it is useful to stress, that the great variability of sucrose, maltose and raffinose content in honey samples from Thassos, Halkidiki, Evia, and Lakonia regions, resulted in higher standard deviation values compared to average values (Table 2). We claim that, in the case of natural products this is very likely to occur and is owed to the biological variability within different honey cultivars/samples and, thus, should be distinguished from analytical variability, to avoid misleading. Besides, this trend in honey minor constituents has been previously reported for Brazilian honeys [5].

Minor constituents profile of Greek pine and fir honeys

Free amino and organic acids comprise minor constituents of honey representing c.a. 0.1–0.3% of total honey weight [4, 33]. Proline is the most abundant free amino acid in honey, ranging from 50–85% of the total amino fraction [33], whereas phenylalanine, alanine, methionine, tryptophan, and serine may also contribute to the free amino acid profile of honey [34].

As the main source of amino acids is pollen, the amino acid profile or specific amino acids may aid to the botanical differentiation of honey at an international level using HPLC or NMR methodology [13, 34]. Likewise, geographical aspects should also be considered since free amino acids are added by bees, and thus, a high degree of variability of free amino acid content for honeys originating from the same floral source is possible to occur [35].

This was the case for pine honeys from Halkidiki, Evia, and Thassos regions, where a high degree of variability among the analyzed pine honeys resulted in higher standard deviation values than the average, for proline and phenylalanine. The same trend was also recorded for fir honey samples from Messinia and Arkadia regions (Table 2). Proline and phenylalanine contents (g/100 g) of Greek pine and fir honeys were higher compared to those of Spanish dark-colored honeys [18].

On the other hand, organic acids contribute to honey acidity, electrical conductivity, and are responsible for honey taste, aroma, color, and preservation, since they inhibit the growth of microorganisms. The principal organic acid in honey is gluconic acid, which results from glucose oxidase activity. Other acids present in honey in smaller amounts are acetic, butyric, lactic, citric, succinic, formic, maleic, malic and oxalic acids [4, 33]. It should also be mentioned that organic acids are produced from nectar and honeydew sugar secretions by the action of enzymes secreted by honeybees during honey ripening and storage (i.e. formic acid) [36]. In that sense, blossom and honeydew honeys may have a different organic acid profile [4].

In the present study, formic acid recorded similar values (g/100 g) in pine and fir honeys (Table 2). Such values are slightly lower than those reported for German honeydew honeys [16].

Physicochemical and 1H NMR ratios of Greek pine and fir honeys

Moisture is a quality parameter related to honey shelf life. Honey moisture content is at normal levels when the beehive cells are totally capped with beeswax. Honey moisture may range between 13 and 25%, the optimum value being 17% [37]. Conversely, honeys with moisture content higher than 18% are prone to fermentation, because sugars’ osmotic pressure is not high enough to avoid the osmophilic (sugar-tolerant) yeasts’ proliferation. The higher the honey moisture content is, the higher the number of yeasts that can cause honey fermentation are [35].

Besides, several properties of honey, such as color, crystallization, viscosity, flavor and density, are also affected by moisture content. As honey is a very hygroscopic product, it is important to avoid environmental moisture uptake during honey processing and packaging. Moisture content of honey may be related to different factors such as the botanical and geographical origin of nectar, edaphic and climatic conditions, harvesting season, intensity of nectar flux, degree of maturation, manipulation by beekeepers during harvesting, as well as extraction, processing and storage conditions [20, 37].

In all pine and fir honey samples analyzed, average moisture content (g/100 g) was within the range set by European quality standards [32]. In general, pine honeys recorded lower moisture content values than fir honeys (Table 2). Such moisture content values for Greek pine honeys are lower than those reported for Italian honeydew honeys, whereas respective moisture content values of Greek fir honeys are higher than those of Italian honeydew honeys [38].

Apart from individual sugar values, ratios between specific sugars have been proposed as suitable indicators of honey botanical origin [31]. For example, blossom honeys showed a fructose–glucose ratio (F/G) about 1.0, while in honeydew honeys the ratio ranged between 1.5 and 2.0 [39]. In the present study, fir honeys recorded higher average F/G values than pine honeys (1.75 vs. 1.23) (Table 2). The F/G values of Greek pine and fir honeys are lower than those reported previously for dark-colored Spanish honeys (2.65) [18].

In addition, such F/G values for Greek pine and fir honeys are in general agreement with the F/G values reported for French fir honeys, even though the F/G ratio was determined using HPLC [6].

The F/G ratio is also of great interest for the quality control of honey since it affects the degree of crystallization during aging [9]. Honey samples with F/G ratio higher than 1.30 (i.e. acacia and sage, among others) and glucose/moisture (G/M) ratio lower than 1.70 tend to crystallize slowly while in samples with ratios of F/G lower than 1.0 (i.e. dandelion, rape and sunflower, among others) and G/M higher than 2.0, crystallization is faster as reported for Nigerian honeys [40]. This hypothesis does not totally hold for Greek pine and fir honeys (Table 2). Discrepancies may be attributed to honey production country, or beekeepers’ manipulations regarding honey maturity in the beehive.

Finally, with respect to fructose-moisture ratio (F/M), glucose-moisture ratio (G/M), and sum of fructose and glucose-moisture ratio (F + G/M), some interesting data were collected. Significant variations (p < 0.05) on such ratios depending on geographical origin of the investigated honeys were recorded, and thus, these parameters could be used as honey geographical origin indicators (Table 2).

Discrimination of Greek pine honeys according to geographical origin based on major and minor honey components

MANOVA analysis was applied to all determined parameters (dependent variables), to determine those which are significant for pine honey geographical discrimination. Geographical origin was taken as the independent variable.

Pillai’s Trace = 2.145 (F = 4.827, p = 0.001) and Wilks’ Lambda = 0.016 (F = 5.301, p = 0.001) index values showed that there was a significant multivariable effect of the determined parameters on geographical origin of pine honey. Nine physicochemical parameters were found to be significant (p < 0.05) for the geographical discrimination of pine honeys. Thus, these nine physicochemical markers including: fructose, maltose, raffinose, phenylalanine, proline, moisture content, sum of fructose and glucose, fructose/moisture ratio, and sum of fructose and glucose to moisture ratio) were subjected to LDA.

Results showed that two statistically significant discriminant functions were formed: Wilks’ Lambda = 0.035, X2 = 105.488, df = 27, p = 0.001 for the first and Wilks’ Lambda = 0.166, X 2 = 56.259, df = 17, p = 0.001, for the second. The uniformity of variability, by application of Box’ M index (Box’s M index = 497.241, F = 2.325, p = 0.053), was insignificant at the 95% confidence level showing the existence of uniformity of sample variability for each geographical origin.

The first discriminant function accounted for 66.3% of total variance, while the second accounted for 22.7%. Both accounted for 89% of total variance, which is a satisfactory rate. In Fig. 2a, it is shown that pine honeys from Samos are clearly separated. In particular, the first discriminant function fully differentiates Samos from the other regions, while the second discriminant function differentiates the Thassos region. The overall correct classification rate was 87.2% for the original and 76.9% for the cross validation method, considered satisfactory especially for the second method. The physicochemical markers (independent variables) that contributed to the first and second discriminant function are given in Table 2.

Fig. 2
figure 2

a Geographical discrimination of pine honeys using the combination of 1H NMR and HPLC based on metabolites along with physicochemical data. Functions 1 and 2 are canonical discriminant functions. b Geographical discrimination of fir honeys using the combination of 1H NMR and HPLC based on metabolites along with physicochemical data. Functions 1 and 2 are canonical discriminant functions

Full size image

Discrimination of Greek fir honeys according to geographical origin based on major and minor honey components

As in the case of pine honeys, Pillai’s Trace = 2.445 (F = 5.759, p = 0.001) and Wilks’ Lambda = 0.006 (F = 5.504, p = 0.001) index values showed that there was a significant multivariable effect of 11 determined parameter values (glucose, sucrose, turanose, proline, formic acid, sum of fructose and glucose, moisture, fructose/moisture ratio, glucose/moisture ratio, sum of fructose and glucose/moisture ratio, and maltose) on geographical origin of fir honeys (p < 0.05). Thus, these 11 physicochemical markers were subjected to LDA.

Results showed that two statistically significant discriminant functions were formed: Wilks’ Lambda = 0.010, X 2 = 104.943, df = 30, p = 0.001 for the first, and Wilks’ Lambda = 0.078, X 2 = 58.693, df = 18, p = 0.001, for the second. The uniformity of variability, by application of Box’s M index, was insignificant at the 95% confidence level (Box’s M index = 59.952, F = 2.216, p = 0.054) showing the existence of uniformity of sample variability for each geographical origin.

The first discriminant function accounted for 54.9% of total variance, while the second accounted for 28.9%. Both accounted for 83.8% of total variance, being a satisfactory rate. In Fig. 2b, it is shown that fir honeys are well separated. In particular, the first discriminant function differentiates Karditsa and Messinia regions, while the second discriminant function differentiates Lakonia region.

The overall correct classification rate was 96.8% for the original and 80.6% for the cross-validation method, considered satisfactory especially for the cross-validation method. The independent variables that contributed to the first and second discriminant function are given in Table 3.

Table 3 1H NMR and HPLC metabolites along with physicochemical data used for the geographical discrimination of pine honeys and canonical discriminant function coefficients
Full size table

Summary regarding the correct prediction rates using 1H NMR metabolites and physicochemical data

The application of chemometric methods to metabolites determined using 1H NMR and HPLC along with physicochemical data resulted to a satisfactory geographical discrimination of Greek pine and fir honeys (p < 0.05).

Tables 3 and 4 list the standardized canonical discriminant function coefficients obtained in the selected models for every different origin parameter, with respect to the geographical origin of pine and fir honeys. It should also be pointed that, the higher the absolute value of a standardized canonical coefficient, the more significant the variable is.

Table 4 1H NMR and HPLC metabolites along with physicochemical data used for the geographical discrimination of fir honeys and canonical discriminant function coefficients
Full size table

The first discriminant function was the one that best differentiated between honey groups, given that it represented the highest variability. The variables that most contributed to the geographical discrimination of honeys were: (1) maltose, moisture, raffinose, fructose-moisture ratio, and phenylalanine in the case of pine honeys and (2) moisture, sum of fructose and glucose/moisture ratio, proline, glucose/moisture ratio, fructose/moisture ratio, sucrose, maltose, and turanose in the case of fir honeys.

Present differentiation rates (> 75%), using a model validation technique, support previous work in the literature on a similar topic involving Italian honeys [15]. It should also be stressed, that this specific set of data has not been previously published, this constituting the novelty of the present work. Finally, the geographical origin differentiation of Greek honeydew honey proved to be a more difficult procedure than that of botanical origin identification [20].

Finally, authenticity of honey has become a major subject of investigation at an international level, as shown in a recent and exhaustive study of Spiteri et al. [17] using 1400 honey samples of different botanical and geographical origins. Above authors reported that future research on the quantification of minor honey components using 1H NMR methodologies will contribute to the discrimination of monofloral honeys, in agreement with the philosophy of the present study.

Conclusion

The combined use of instrumental (1H NMR/HPLC) and conventional methods has the potential to provide specific physicochemical fingerprints of honey. Major and minor constituents of Greek pine and fir honeys, as assessed by multivariate analyses, may provide useful information on geographical origin and could aid to the accurate labeling of monofloral honeys.