Investigation of Volatile Compounds in Combination with Multivariate Analysis for the Characterization of Monoﬂoral Honeys

: Lately there has been a growing demand for monoﬂoral honeys with distinctive properties. Considering the limitations of pollen analysis, the volatile proﬁle of honey has been proposed as a helpful supplementary tool for the conﬁrmation of monoﬂorality; however, research remains regarding the volatile markers that may characterize the monoﬂoral honey types. Therefore, in this study, we tried to expand the research by investigating the aroma proﬁles of ﬁve monoﬂoral honey types (ﬁr, pine, erica, thyme, cotton) and discriminate them through chemometric approach. A purge and trap–gas chromatograph–mass spectrometer system was used for the extraction, separation, and identiﬁcation of volatile and semi-volatile compounds. Thyme honey had the richest quantitatively aroma proﬁle, with 97 volatile compounds, whereas ﬁr and cotton honeys had 65 and 60 volatile compounds, respectively. From a total of 124 compounds, the 38 were detected in all the studied honey types. Thyme honey was distinguished by the presence (or percentage participation) of benzeneacetaldehyde, benzealdehyde, and benzyl nitrile; erica honey of isophorone and furfural; cotton honey of 1-butanol, 2-methyl, 1-pentanol, and 4-methyl-; and honeydew honeys of α -pinene, octane, and nonanal. The discriminant analysis conﬁrmed that the percentage participation of volatile compounds may lead to the discrimination of the studied monoﬂoral honey types.


Introduction
The aroma of honey is a distinguishable characteristic directly related to honey's quality and, in turn, to consumer preferences [1]. This is also evidenced by the fact that more and more food industries are trying to find the ideal conditions for the honey collection and processing to maintain this characteristic unchanged. The aroma of honey consists of a complex mixture of volatile compounds, whose concentration depends on the source area, processing and storage conditions [2][3][4]. Studies report the presence of about 600 different volatile compounds in honey, in low concentrations, but including a variety of hydrocarbons, aldehydes, alcohols, ketones, acetates, benzenes and their derivatives, furan and pyran, norisopreoids, terpenes and their derivatives, sulfates, and cyclic compounds [5][6][7][8][9].
Some monofloral honeys have been found to possess richer aroma profiles compared with polyfloral honeys; thus, they present an increased marketing value [10]. The determination of the botanical origin of various honey species is usually based on pollen analysis [11,12], combined with physicochemical characteristics such as color, electrical conductivity, sugars, pH, mineral content [13][14][15][16], sensory evaluation [17], concentrations of flavonoids and phenolic acids [18,19], contents of nitrogenous components, such as proteins and amino acids [20], as well as the presence of specific organic [21], and inorganic acids [22]. However, with the development of gas chromatography (GC), and especially with the beginning of the use of mass spectrometry for detection, the interest of scientists

Sampling
The honey samples (~1 kg) were sourced by beekeepers all over Greece and harvested in 2019 and 2020 ( Figure 1). The samples were fresh and unprocessed, coming directly from beekeepers in jars, under the guidance of applying the appropriate beekeeping practices in order to ensure the production of monofloral honey samples. Immediately after the harvest, the samples were put into ice boxes, sent through courier to the Laboratory of Apiculture-Sericulture, AUTH, and kept in a freezer until their analysis.

Honey Extraction and Isolation of the Components
For the isolation of the volatile compounds, the trapping device system (Purge and Trap, OI Analytical, model 4560) was used, concentrated in a Tenax 07 column (OI Analytical). The method of Tananaki et al. [32] was applied, with some modifications. Specifically, 10 g (±0.001 g) of honey was diluted with 5 g ultrapure water (Millipore, model Simplicity 185) and 15 μL of internal standard (styrene, Merck, Belgium, >99%), and the solution was transferred in the purge vessels (25 mL). The vessels were stirred in vortex for 30-60 s and put in the Purge and Trap system for the extraction. A steel (No. 316) spiral component of 40 cm length was used to suppress the foam. The vessels were heated at 40 °C for 2 min without importing gas, to reduce the viscosity and facilitate the passage of helium gas from the mass of the solution during the extraction. Afterwards, the vessels were purged with the helium gas (40 mL min −1 ) for 40 min, keeping the temperature of the samples at 40 °C. The volatile and semi-volatile compounds were collected on a Tenax TM TA trap (OI Analytical). The moisture was removed by heating at 100 °C for 2 min, and then the desorption was performed by raising the trap temperature at 180 °C for 6 min with simultaneous passing helium (40 mL min −1 ) and the analytes were transferred through a thermostable transfer line (100 °C) to the gas chromatograph. The trap was cleaned each time by heating at 200 °C for 7 min.

Gas Chromatography-Mass Spectrometry (GC-MS) Conditions
An Agilent 6890 gas chromatograph, model, coupled with an Agilent 5973 mass detector, was used for separation of the extracted components, directly connected via a

Honey Extraction and Isolation of the Components
For the isolation of the volatile compounds, the trapping device system (Purge and Trap, OI Analytical, model 4560) was used, concentrated in a Tenax 07 column (OI Analytical). The method of Tananaki et al. [32] was applied, with some modifications. Specifically, 10 g (±0.001 g) of honey was diluted with 5 g ultrapure water (Millipore, model Simplicity 185) and 15 µL of internal standard (styrene, Merck, Belgium, >99%), and the solution was transferred in the purge vessels (25 mL). The vessels were stirred in vortex for 30-60 s and put in the Purge and Trap system for the extraction. A steel (No. 316) spiral component of 40 cm length was used to suppress the foam. The vessels were heated at 40 • C for 2 min without importing gas, to reduce the viscosity and facilitate the passage of helium gas from the mass of the solution during the extraction. Afterwards, the vessels were purged with the helium gas (40 mL min −1 ) for 40 min, keeping the temperature of the samples at 40 • C. The volatile and semi-volatile compounds were collected on a Tenax TM TA trap (OI Analytical). The moisture was removed by heating at 100 • C for 2 min, and then the desorption was performed by raising the trap temperature at 180 • C for 6 min with simultaneous passing helium (40 mL min −1 ) and the analytes were transferred through a thermostable transfer line (100 • C) to the gas chromatograph. The trap was cleaned each time by heating at 200 • C for 7 min.

Gas Chromatography-Mass Spectrometry (GC-MS) Conditions
An Agilent 6890 gas chromatograph, model, coupled with an Agilent 5973 mass detector, was used for separation of the extracted components, directly connected via a thermostatic transfer line to the extraction system. The gasified mixture was introduced via a split-splitless feeder, while the components were separated on an HP-5MS column (30 × 0.25 mm, df = 0.25 µm). The extraction gas as well as the carrier gas were He of high purity (99.999%), and before entering the analysis system, it was passed through a filter, in order to remove from it any small amounts of oxygen. The temperature program used for the separation was as follows: 40 • C for 5 min, increasing to 55 • C at 1 • C min −1 , to 120 • C at 3 • C min −1 , to 230 • C at 10 • C min −1 and to 280 • C at 20 • C min −1 (hold for 5 min). The operating conditions of the mass spectrometer were as follows: interface temperature, 280 • C; source temperature, 230 • C; quadrupole temperature, 150 • C; ionization, 70 eV. The chromatograms were processed and completed with the MSD ChemStation program, while the peaks were identified using the electronic libraries and tables of retention times and spectra that were kept in the Laboratory of Apiculture-Sericulture, AUTH [41]. The presence of volatile compounds was confirmed by using retention indices (RIs) based on the calculations using the standard mixture of alkanes (Sigma Aldrich, Darmstadt, Germany). Retention indices (RIs) are widely applied for the comparison of results with other studies, as well as to characterize stationary phases.

Statistical Analysis
For statistical processing of the results, the SPSS 19.0 statistical package software for Windows was used. The level of significance was set at α = 0.05. Specifically, a multivariate analysis of variance (MANOVA) was applied for the compounds that were detected in 100% honey samples of each category, to determine the significant volatile compounds that could affect their classification and for those compounds, a linear discriminant analysis (LDA) was followed, in order to determine whether the studied honey types could be further discriminated. Considering the large number of predictors that arose from the LDA, the stepwise method was further applied to select the variables that fit better to the prediction model. Then, samples (test set) were analyzed on their volatile compounds and the possibility of their inclusion to a certain monofloral honey type was examined, to test the accuracy of the prediction model.

Results and Discussion
In total, 124 chemical compounds were found. Peaks with a low correlation of their mass spectra were not identified, but were rendered "unknown" and their respective fragments were described, whereas 7 compounds were identified as isomers. The percentage participation (%) of compounds (average and standard deviation), their retention time (R.T.), retention indices (RIs), and mass fractions (m/z) (the underlined fractions are the main for each compound) are given in Table 1, whereas the representative chromatogram for each kind is presented in Figure 2.
The percentage participation of different categories of organic compounds is shown in Figure 3, where the presence of hydrocarbons, aldehydes, and alcohols is highlighted. Indeed, the group of hydrocarbons stood out (40.3%), followed by ketones, alcohols, and aldehydes with 15.3%, 14.5%, and 12.9%, respectively. To a lesser extent, esters (2.4%) and sulfur compounds (1.7%) were found, whereas a small percentage of organic compounds was not identified, named as "unknown" (5.6%). Another group, which accounted for 7.3%, included volatile compounds such as heterocyclics, which were rarely detectable in honey samples, without justifying their grouping into a separate class of compounds.  The percentage participation of different categories of organic compounds is shown in Figure 3, where the presence of hydrocarbons, aldehydes, and alcohols is highlighted. Indeed, the group of hydrocarbons stood out (40.3%), followed by ketones, alcohols, and aldehydes with 15.3%, 14.5%, and 12.9%, respectively. To a lesser extent, esters (2.4%) and sulfur compounds (1.7%) were found, whereas a small percentage of organic compounds was not identified, named as "unknown" (5.6%). Another group, which accounted for 7.3%, included volatile compounds such as heterocyclics, which were rarely detectable in honey samples, without justifying their grouping into a separate class of compounds.  Thyme honey quantitatively had the richest aroma profile compared with the other studied honey types: in total, 97 volatile compounds were detected, followed by erica with 83, and pine with 81 compounds. Cotton and fir honeys had the fewest volatile compounds detected, with 65 and 63, respectively ( Table 1). The rich volatile fraction of thyme honey has been discussed in the literature [31,84,85], whereas for erica and cotton honey, the references are inadequate [86].
The use of P&T linked to GC-MS is widely used for the identification of volatile compounds in food science [92,93]. The high-resolution power and sensitivity of gas chromatography (GC), together with the structural information provided by mass spectrometry (MS), has made the coupling GC-MS the technique of choice for the qualitative and quantitative analysis of food volatiles [93,94]. However, the preliminary steps of fractionation from major non-volatile components of the honey matrix (sugars and water) and preconcentration are still necessary for the chromatographic separation of honey volatiles. In the purge-and-trap (P&T) technique, volatiles swept by a flow of inert gas are trapped on an adsorbent, while further thermal/solvent desorption of the trap allows volatiles to enter the chromatographic system for separation. Furthermore, in P&T, the reduction in or elimination of organic solvents is achieved, offering a green alterative to sample preparation, combined with simultaneous multiclass compound extraction and reproducibility associated with a totally automated system [25].
The MANOVA that was applied for the compounds found in 100% of honey samples of each honey type showed significant differences among the honeys for some characteristic compounds, such as 1-octene (C20), benzaldehyde (C40), and nonanal (C80) ( Table 1). In turn, a linear discriminant analysis through the stepwise method (SLDA) was used in order to determine whether the significant volatile compounds could discriminate the examined monofloral honey types. The first canonical variable corresponded to 58.9%, whereas the second corresponded to 25.5% of variation (Table 2). Additionally, regarding the group centroids of the first function, fir honeys had a mean of −10.623, pine honeys a mean of −2.250, erica honeys a mean of 21.424, thyme honeys a mean of 0.918, and cotton honeys a mean of −19.655. All the monofloral honey types were discriminated, and this was more obvious for erica and cotton honeys because they were located in the most distant places (Figure 4), which was also confirmed by the different chromatograms that these two honey types presented, compared with the other studied monofloral honey types ( Figure 2). Samples of the same honey type were located at the same quadrant, concentrated around the centroid of the respective honey type, showing the large correlation inside the same group ( Figure 4).
Additionally, the stepwise method was applied to determine the compounds that will be used in the prediction model. The original group cases were correctly classified at a 94.2% rate, whereas the cross-validated group cases were correctly classified at a 92.3% rate (Table 3).
Five F scores were calculated (Table 4) and used for the classification in the group membership of five samples (test set) whose botanical origin was known ("reference samples") for verification of the predictive model. The samples were assigned to the group for which the classification function had the largest F score (Table 4). For all the reference samples, the prediction was correct.
The combination of chemometric approaches, such as principal component analysis (PCA), cluster analysis (CA), or discriminant analysis (DA), with the analysis of volatile compounds for the characterization of other monofloral honey types studied by other authors had also led to promising results for the correct classification of honey samples [32,95,96]. However, due to the complexity of volatile compounds' structure that makes them unstable along with the various methods of analysis used for their determination, a larger number of honey samples could optimize the prediction model, and thus, the honey authentication. Additionally, the stepwise method was applied to determine the compounds that will be used in the prediction model. The original group cases were correctly classified at a 94.2% rate, whereas the cross-validated group cases were correctly classified at a 92.3% rate (Table 3). Table 3. Classification results (original and cross-validated) extracted from the prediction model after applying SLDA analysis. Five F scores were calculated (Table 4) and used for the classification in the group membership of five samples (test set) whose botanical origin was known ("reference samples") for verification of the predictive model. The samples were assigned to the group for

Conclusions
A Purge and Trap GC-MS system was used for the extraction, separation, and identification of characteristic volatile and semi-volatile compounds of five distinguished monofloral honey types-thyme, pine, fir, erica, and cotton-with the latter two remaining less well explored regarding their aroma profile. In total, 124 compounds were found. Thyme honey, whose profile, regarding the presence of volatiles, was the richest compared with the other honey types of this study, was characterized by the presence of benzeneacetaldehyde, benzealdehyde, and benzyl nitrile. Isophorone and furfural seem to characterize erica honey, 1-butanol, 2-methyl-, and 1-pentanol, 4-methyl-the cotton honey and alpha.-pinene, octane and nonanal the honeydew honeys. The application of multivariate statistical analysis to P&T GC-MS data showed that the use of volatile profiles could lead to discrimination among the different monofloral honey types. Further analysis of more monofloral honey types could be a helpful tool for the confirmation of honey authentication through volatile profile analysis.