Multidisciplinary analysis of Italian Alpine wildflower honey reveals criticalities, diversity and value

Wildflower honeys produced in mountain grasslands are an expression of the biodiversity of these fragile habitats. Despite its importance, the botanical origin of honey is often defined without performing formal analysis. The aim of the study was to characterize six wildflower mountain honeys produced in the Italian Alps with different analytic techniques (SPME–GC–MS, HPLC-Orbitrap, cicatrizing and antioxidant activity) alongside melissopalynological analysis and botanical definition of the production area. Even though the apiaries were in mountain grasslands rich in Alpine herbaceous species, the honey could be defined as rhododendron/raspberry unifloral or raspberry and rhododendron bifloral while the honey produced at the lowest altitude differed due to the presence of linden, heather and chestnut. The non-compliance of the honey could be due to habitat (meadows and pastures) fragmentation, but also to specific compounds involved in the plant–insect relationship, such as kynurenic acid, present in a high quantity in the sample rich in chestnut pollen. 255 volatile compounds were detected as well as some well-known markers of specific botanic essences, in particular chestnut, linden and heather, also responsible for most of the differences in aroma profiling. A high correlation between nicotinaldehyde content and percentage of raspberry pollen (r = 0.853, p < 0.05) was found. Phenolic acid and hydroxy-fatty acid were predominant in the chestnut pollen dominant honey, which presented the highest antioxidant activity and the lowest cicatrizing activity, while the flavonoid fraction was accentuated in one sample (rhododendron pollen prevalent), that was also the one with the highest effect on wound closure, although all samples had similar cicatrizing effects apart from the chestnut pollen dominant honey (lowest cicatrizing activity). Our study highlighted the difficulty of producing mountain wildflower honey and the importance of a thorough characterization of this product, also to encourage its production and valorisation.


Materials and methods
Sampling areas and vegetation analysis. Six bee stations in the mountain areas of high Val Camonica (Southern Alps, Northern Italy, Lombardy region) were chosen for sampling in the spring of 2020 (Fig. 1). Three production areas are inside the Adamello Regional Park and one is in the Stelvio National Park. Phytosociological relevés were performed to collect floristic data about the study area in July, during the blooming season. Later, during the second week of August, beekeepers performed honey extraction to produce the wildflower variety. Finally, honey was sampled from the decanters during the first week of September with the appropriate permission for the collection of samples.
According to the classification of the ecoregions of Italy 23 , high Camonica Valley is within the Central and Eastern Alps section (Alpine Province, Temperate Division) and has a sub-continental climate (average annual rainfall: 900-1100 mm; average annual temperature: 4-10 °C) 24 .
In areas between 1200 and 1700 m a.s.l. there are mostly coniferous forests of Calamagrostio arundinaceae-Piceo excelsae sigmetum 25  In addition, there are mountain meadows of Trisetetum flavescentis association and acidophilus pastures of Siversio-Nardetum 8,25 .
At higher altitudes, especially beyond the treeline (2050-2200 m a.s.l.), there are mainly alpine acidophilus grasslands of Caricetea curvulae class 26 with small shrubs such as Thymus spp., Rhododendron ferrugineum L., Rubus idaeus, Vaccinium myrtillus, V. vitis-idaea L., Calluna vulgaris and Juniperus communis L. 25 . These shrubs (and some trees such as larix and spruce) are colonizing the montane and sub-alpine grasslands abandoned by man 8,9 . Floristic data of the vegetation types close to the hives were collected performing phytosociological relevés in accordance with the method of Braun-Blanquet 27 . In each honey production area ( Fig. 1) 4 relevés were carried out in June-July 2019 on a surface of 100 m 2 (10 × 10 m plot), starting from 5 m away from the apiaries and proceeding towards the four cardinal points. Phanerogam species were identified using "Flora d'Italia" dichotomous keys of Pignatti 28 and their coverage was estimated using the abundance/dominance scale of Braun-Blanquet 27 : r, rare species in the relevé; + , coverage < 1%; 1, coverage 1-5%; 2, coverage > 5-25%; 3, coverage > 25-50%; 4, coverage > 50-75%; 5, coverage > 75-100%. The data of the relevés were arranged in a matrix (relevés x species) where abundance/dominance indexes of the species were converted into the percentage of coverage in accordance with 9 to perform statistical analysis.
Multidimensional scaling analysis (MDS) was performed on the matrix of the relevés to highlight floristic differences among the relevés and the honey production areas, and to identify the main ecological gradients that , collection of honey (c) and honey (d) produced in area A. The map was generated using ArcGIS Desktop 10.5 (Esri, Redlands, CA, USA, https:// www. esrii talia. it/ prodo tti/ la-nostra-offer ta/ piatt aforma-esri/ arcgis-10-5). www.nature.com/scientificreports/ determine these differences. MDS was carried out considering the chord distance 29 and was performed using the "vegan" package of R 3.5.2. software 30 . The scientific names of the species mentioned are in accordance with 28 .
Melissopalynological analysis. Melissopalynological analysis was performed according to the techniques proposed by the International Commission for Bee Botany (ICBB) for qualitative melissopalynological analysis 4 opportunely modified as described below. Organoleptic characteristics were also concisely analysed. Slide preparation was performed according to the following method: 12 g of honey were weighted in a 50 mL test tube with conical bottom and dissolved with 40 mL of distilled water. The solution was centrifuged for 15 min at 3000 rpm, and then the liquid was separated from the sediment by pouring or aspiration. To eliminate the sugars present, the sediments were dissolved again in 10 mL of distilled water and centrifuged again for 5 min at 3000 rpm and the supernatant was eliminated by pouring gently. The sediment was carefully dispersed with a Pasteur pipette, transferred to a glass slide, and distributed as evenly as possible over an area of about 1 cm 2 . The glass slide was left to dry at 40 °C. When the sediment was dry, it was included in a drop of glycerol gelatine, previously dissolved in a water bath, and covered with a coverslip.
The microscopic analysis of honey sediment composition provides the percentage of specific pollen observed by microscopic comparison with known pollen grains. Pollen grain counting was performed according to the method described in 31 . If the sediment contained a high percentage of overrepresented pollen, a second count excluding the over-represented pollen was performed in order to determine the relative abundance of other pollen types more precisely. The pollen types present in the honey samples were identified, counted, and classified, according to their percentages, as dominant pollen (more than 45% of the total pollen grains counted), secondary pollen (from 16 to 45%), important minor pollen (from 3 to 15%), minor pollen (less than 3%) or sporadic pollen (less than 1%) 32 .
MDS was performed on the matrix of the melissopalynological data to highlight the differences among the honey samples using R.
Bioactivity. As regards bioactivity, antioxidant (Free Radical Scavenging Activity, FRSA) and cicatrizing properties were examined through DPPH and HaCaT cell culture tests.
The DPPH radical scavenging activity of honey samples was determined according to the method of Brand-Williams 33 modified as in 34 . The DPPH solution in methanol (0.06 mM) was prepared daily, and 2.7 ml of this solution was mixed with 0.3 ml of 50% (w/v) honey solution. The mixture was shaken vigorously and left to stand for 60 min in the dark (until stable absorption values were obtained). The reduction of the DPPH radical was determined by measuring the absorption at 517 nm. The experiment was carried out in triplicate. Radical scavenging activity (RSA) was calculated as a percentage of DPPH discoloration using the equation: where A1 is the absorbance of the sample solution and A0 is the absorbance of the radical blank using distilled water. The honey samples were confronted with a manuka honey sample.
Cicatrizing activity was performed on a HaCaT cell culture. HaCaT cells are immortalized human skin keratinocytes that mimic many properties of normal epidermal keratinocytes. These cells are not invasive and can differentiate under experimental conditions 35 . Cells were cultured in a Dulbecco's modified Eagles' medium (DMEM) supplemented with 10% fetal bovine serum (FBS; Euroclone, Pero, Italy), 1% glutamine, 100 U/ml penicillin and 100 mg/ml streptomycin.
Scratch wounds were made in confluent layers HaCaT cells by using a sterile 2-20 µL pipette tip 36 . After washing away detached cells, cultures were rinsed with complete medium in the presence of honeys (0.5% w/v) for 24 h. After exposures, cells were fixed in 3.7% formaldehyde in PBS for 30 min and stained for 30 min at room temperature with 0.1% toluidine blue.
The space between the wound edges was measured at T0 (at wounding) and at T24 (end of exposure) by using an inverted microscope (Leica Microsystems) equipped with a digital camera.
NIH ImageJ software was used to analyse the digitalized pictures. Wound closure rate was calculated as the difference between wound width at T0 and at T24. The value of the control was set at 100 percent to obtain wound closure rate %.
Phytochemical analysis. The phenolic and flavonoid profile was obtained through HPLC-Q-Exactive-Orbitrap-MS metabolomic analysis of the six samples. The phenolic compounds were extracted by the following SPE procedure: 2 g of honey was diluted in 5 mL of 0.01 M HCl and 12 μL of IS (4-Fluorobenzoic acid) was added. The SPE cartridges (STRATA X 33) were preconditioned with 3 mL of methanol, followed by 5 mL of water. The honey samples were then introduced into the cartridges at a drop wise flow rate. After loading, the SPE cartridge was left under vacuum for 3 min to completely dry out. The phenolic compounds retained by the cartridges were finally eluted with 3 mL of elution solvent (methanol: acetonitrile = 2:1). The eluent was filtered by 0,45 μm syringe filter and subsequently diluted with 0.1% of formic acid (1:1). 10 μL was introduced into HPLC-Q-Exactive-Orbitrap-HRMS instrumentation. Untargeted metabolomic approach was performed according to the recently developed strategy for propolis 37 . All analyses were performed in triplicate.
Heatmap and Principal Components Analysis (PCA) were performed on the metabolomic data in order to highlight the differences among honey samples.
The volatile profile was obtained through Headspace Solid-Phase Microextraction (HS SPME) coupled to Gas Chromatography-Mass Spectrometry (GC-MS) analysis of Volatile Organic Compounds (VOCs).
All the samples were prepared by weighing exactly 5 g of honey in a 20 mL glass vial along with 100 μL of the IS (4-Methyl-2-pentanone, 2 mg/L in 2-propanol). Each vial was fitted with a cap equipped with a silicon/ ) was exposed to the headspace of the sample for extraction (120 min) by CombiPAL system injector autosampler (CTC analytics, Switzerland). The fibre and the time of extraction used in this study were selected after preliminary study. The best adsorption of analytes was obtained using CAR/PDMS/DVB and 120 min as extraction time. The extraction temperature of 25 °C was selected to prevent possible matrix alterations (oxidation of some compounds, particularly aldehydes and furans). The vials were maintained on a heater plate (CTC Analytics, Zwingen, Switzerland) to keep temperature constant during analysis. GC-MS analysis was performed using a Trace GC Ultra (Thermo-Fisher Scientific, Waltham, MA, USA) Gas Chromatograph coupled to a Trace DSQII quadrupole mass spectrometer (MS) (Thermo-Fisher Scientific, Waltham, MA, USA) and equipped with an Rtx-Wax column (30 m; 0.25 mm i.d.; 0.25 μm film thickness, Restek, USA). The oven temperature program was from 35 °C, hold 8 min, to 60 °C at 4 °C/min, then from 60 to 160 °C at 6 °C/min, and finally from 160 to 200 °C at 20 °C/min, hold 20 min. Carryover and peaks originating from the fibre were regularly assessed by running blank samples. After each analysis, fibres were immediately thermally desorbed in the GC injector for 5 min at 250 °C to prevent contamination. The injections were performed in splitless mode (8 min). The carrier gas was helium at a constant flow of 1 mL/min. The transfer line to the mass spectrometer was maintained at 230 °C, and the ion source temperature was set at 250 °C. Mass spectra were obtained by using a mass selective detector with electronic impact at 70 eV, a multiplier voltage of 1456 V, and by collecting the data at a rate of 1 scan s −1 over the m/z range of 35-350. An alkanes mixture (C8-C22, Sigma R 8769, Saint Louis, MO, USA) was run under the same chromatographic conditions as the samples to calculate the Kovats Retention Indices (RI) of the detected compounds. Compounds were identified by comparing the Kovats retention indices with literature data and through the National Institute of Standards and Technology (NIST) MS spectral database as in previous research [38][39][40] . Volatile compound measurements from each headspace of honey extracts were carried out by peak area normalization (expressed in ppb). All analyses were performed in triplicate.
Heatmap was performed on the VOCs data to highlight the differences among honey samples.

Ethics declaration.
There are no special ethical concerns arising from the use of animals or human subjects.

Results and discussion
From the phytosociological relevés performed in each sampling area it is evident that hives were positioned in grasslands rich in Alpine herbaceous species (Table S1). In fact, among the 169 identified species, 85% were herbaceous species common in meadows (of Arrhenatherion elatioris and Triseto flavescentis-Polygonion bistortae phytosociological alliance) and acidophilus pastures (Siversio-Nardetum). 15% of the species were trees and shrubs (not abundant in the floristic relevés of the apiary areas considered), including some of beekeeping interest such as: Rhododendron ferrugineum, Castanea sativa and Rubus idaeus. From the MDS biplot ( Fig. 2) elevation is the main ecological variable that differentiates sampling areas. In particular, the relevés of stations B and F are characterized by a floristic composition which is different from the areas at higher elevation (characterized by www.nature.com/scientificreports/ a higher presence of microthermal alpine species). This is due to the separation between the sub-montane belt and the high mountain belt vegetation on the 1.300 m a.s.l. line in the study area 25 .
Although the beehives were positioned in mountain grasslands, melissopalynological analysis presented a different picture. The pollen of numerous species detected through the floristic relevés were found in the honey samples via melissopalynological analysis, although the latter did not totally overlap with the floristic characterization of the area, in particular from a "quantitative" point of view. In fact, the floristic relevés showed a relative richness of herbaceous species (Table S1) peculiar of mountain grasslands that would seem promising for the production of wildflower honeys. Conversely, in the melissopalynological analysis the species considered interesting but not predominant in the botanical description were relevant ( Fig. 3 and Table S2).
The premises to produce wildflower honey is that the botanical species contributing must be different and sometimes very numerous, without any of them assuming a dominant character. However, this was not fully evident in our research: although it was possible to identify more than seventy species through melissopalynological analysis and even more through the floristic characterization of the areas, most of them were defined as minor or sporadic pollen (Table S2). Even though apiaries were in mountain grasslands, the most relevant role was played by some woody species/shrubs: Rubus (presumably Rubus idaeous L., identified in the floristic relevés) and rhododendron (Rhododendron ferrugineum L.) for the mountain/subalpine belt and Castanea and Ericaceae (heather) in the submountain belt. Following the rules to define ''unifloral honey'' , three of the wildflower honeys could be defined unifloral or bifloral: • Honey F, due to the contribution of pollen from Tilia genus (that was detected only in this sample as an important sporadic pollen, 3.5%) Castanea (96.4% in honey F, but it should be noted that chestnut pollen is an overrepresented pollen) and in the second count Ericaceae (32.45%, that was considered a secondary pollen together with Rubus, with a percentage of 38.59% in honey F) differed from the other honeys (Fig. 3).
Rubus pollen was anyway present in good amounts in all the samples considered, and was a dominant pollen in honey B, a secondary pollen in honeys C, D, E and F and a minor pollen in honey A. Sorbus and Tilia pollens were detected only in honey F, while no rhododendron was detected in honey F. Honey D was characterized by a percentage higher than the "rare pollen" category of some important alpine essences, such as Liliaceae, Centaurea, Campanulaceae, Anthyllis f., Polygonum bistorta, Lotus alpinus and Potentilla/fragaria (Table S2).
Although wildflower honeys are intrinsically characterized by a high variability compared with unifloral honey, this shows the importance of the formal characterization of honey to obtain a product which satisfies consumer expectations, and it was demonstrated that the botanical origin of honey cannot be based on the claims of local beekeepers by considering the predominant flowers surrounding the hive.
Although honeybees are considered supergeneralists in their foraging choices, there are certain key species or plant groups that are particularly important in honeybee foraging 2 , and many were identified in the botanical was among the frequently found species and tree pollen belonging to Castanea sativa L. as well as, for example, species of Malus, Salix and Quercus spp, was frequently seen. These kinds of preferences could relate to the ease of availability and abundance of the plant, the quality and abundance of the nectar and pollen and/or specific nutrients or trace elements provided by these species or neurological aspects (as will be discussed further). As referred by beekeepers, over the last decades the production of mountain wildflower honey, that often does not meet the characteristics expected and presents flavours that are reminiscent of other kinds of honey such as rhododendron or linden or chestnut, is becoming more and more critical and this was absolutely confirmed by this study. This could be linked to the fragmentation of an important habitat of the Alps-mountain grasslands (meaning pastures and meadows) for anthropic and climatic reasons 8,9 . Honeybees from the same colony forage across areas spanning up to several hundred square kilometres, and at linear distances as far as 9 km from the hive 41 . Onlooker bees are those in charge of finding nectar sources and of giving instructions to the employed bees, the other foraging bees, that communicate the necessity to look for new resources of food to the onlookers through continuous dance communication 42 . Among the onlookers, there is a difference between the bees that scout for different nectar sources or recruit to well known floral resources 43 and there is an optimal ratio of scouts to recruits, for the most effective collective foraging 41 . However, this balance may change based on the structure of the landscape in which the bees forage for food [44][45][46] . Theoretical models 47,48 and empirical tests 49 suggest that when resources are concentrated into a small number of highly rewarding patches, colonies perform best with few scouts and many recruits, while when resource patches are small, evenly distributed, and easy to locate, successful colonies invest more in scouting than in recruitment. This is strictly linked to climate and social changes in the mountains: mountain grasslands are no longer evenly distributed and easily localizable, as they are scattered among expanding areas of shrublands and forests 9 and, for the above-mentioned reasons, it is more efficient for the colony to invest in more recruiters than scouters, as recruiters will identify a small number of highly rewarding patches, such as raspberry or rhododendron shrublands or linden and chestnut woods, that are highly rewarding and very different in quality.
This overlaps with individual and collective honeybee behaviour driven by proximate physiological mechanisms that involve the tryptophan metabolism via kynurenine pathway that is one of main neuroprotective mechanisms. In this research, many of the differences/similarities among the samples might be attributed to metabolic alterations within this pathway, represented by relative amounts of kynurenic acid. However, different quinoline structures have also been identified (Fig. 4). Neurotransmitters play a central role in several of the biological processes that honeybees require to perform activities such as foraging behaviour 50 . A considerable amount of literature highlights the involvement of the neuroprotective kynurenine pathway (KP) final product kynurenic acid (KinA) in the regulation of the stress-related hormone dopamine in the honeybee as well as in other animal species 51,52 . The major known source of dietary KynA are pollen and nectar produced by sweet chestnuts 53,54 and it has been verified that this compound is found in high concentrations in chestnut flowers 55 . This is coherent with the results of this study: chestnut pollen was found in honey F, produced in the lower station where chestnuts also appear in the floristic relevés, and KynA was found to be a dominant compound in honey F. Interestingly, chestnut pollen was found as sporadic pollen in all the other samples, even those produced in the highest apiary stations (Table S2).
Further, KinA may possess positive properties in a number of pathologies of the gastrointestinal tract, especially colitis, colon obstruction or ulceration 56,57 . It has been proposed that KinA may also possess antioxidative properties [56][57][58][59] . This was confirmed by this study, since the wildflower honey with a high component of chestnut pollen was the one with the highest antioxidant properties at the FRSA test (66.61 ± 4.77%), even if lower than manuka honey (84.21 ± 1.04%), a dark honey that is a well-known nutraceutical product and has recently www.nature.com/scientificreports/ attracted attention for its biological properties, especially for its antioxidant and anti-microbial capacities 60 . Honey A showed the lowest power (22.40 ± 0.28%) while the other honeys ranked around 40% (Fig. 5). Interestingly, metabolomic analysis revealed the presence of 3-hydroxyquinaldic acid (Fig. 4), which is a kynurenic acid isomer and, although its function has not been elucidated in detail, a few literature data indicate its role as a precursor of naturally occurring peptide antibiotics from the quinomycin family 61 .
In order to evaluate the ability of honey to induce wound closure, a scratch wound assay was performed (Fig. 6) 62 . Scratch assay creates a gap in confluent keratinocyte monolayer to mimic a wound. It has already been demonstrated that honeys are able to induce wound closure 63 to different extents depending on honey origins and properties.
The six honeys were compared to manuka honey 60 . Results suggested that honeys A-E induce a significant effect on wound closure, allowing a positive action on the wound edge closure after 24 h of exposure. The results were comparable to that obtained with manuka at the same concentration. Only sample F did not induce a relevant effect if compared to control condition (i.e. complete medium).
The efficacy of natural honey in wound care has been attributed to its anti-inflammatory activity, for which specific flavonoids and polyphenols, as for example kempherol 64 , are considered to be partly responsible. Honey C was the richest in terms of some specific flavonoids, among which tricin, luteolin, pectolinarigenin, naringenin and kempherol ( Figure S1), together with a specific glycoside (dihydroxyfenchone 6-O-d-glucoside). Majtan et al. 65 discovered two other flavonoids in aqueous extract of honey, one of which, kempherol, suppresses the activity of TNF-α-induced Multiple Medical Problems (MMP)-9 expression in HaCaT and according to Budovsky et al. in 66 in terms of plant phytochemicals, the activity of alkaloids, flavonoids, terpenes, and glycosides in promoting wound healing, has been better researched than other bioactive plant compounds. It is known that the chemical composition of honey predominantly depends on its botanical source 67 . Among all substances in honey, the ones that are the most depending on floral sources are phenolic compounds, that are also important for their antioxidant activity 68 while processing, handling, and storage have a lesser impact on the phenolic profile and composition of honey. Phenolic acids and flavonoids can be then considered important markers for the identification of botanical origin for different honey 69,70 .
Heatmaps, PCAs and pathway molecular networks, very different but complementary visualization techniques, confirmed and complemented one another providing an additional perspective to recognize the results from honey phenolic compounds analysis (the total ion current chromatograms obtained for each honey samples are showed in Figure S3). The heatmap and PCA (Fig. 7) were performed to compare the overall distribution and variation within and between the honey samples. The PCA and hierarchical cluster analysis indicated that the different floral origins caused significant metabolic changes in the honey samples. A clear separation was observed between sample F (chestnut honey) and other samples. Also, C (prevalently rhododendron) and E (rubus and rhododendron) exhibited specific profiles regarding flavonoid composition. On the other hand,  Table S4 and the possible metabolic transformation is presented in Figure S2. The flavonoid fraction was accentuated in C and it corresponds to the heatmap results while phenolic acid and hydroxy-fatty acids were predominant in F. For each type of honey, the most important candidates for differentiation are presented in Figure S1.
Looking at the volatile profiles, analysing the six samples, 255 compounds were detected (Table S3), belonging to the following groups: alcohols, aldehydes, anidrides, aromatics, carboxylic acid, esters, eters, furanoids, hydrocarbons, ketones, nitrogenates, sulphurated, terpenes and other (Table 2). Data analysis showed that honeys A, B, C, and E were very similar while D and F were richer in volatile compounds and presented a high dissimilarity from the other samples. Honeys D and F were the most interesting in terms of volatile fingerprints, showing the Hierarchical cluster analysis (a) and biplot of PCA (b) reflecting the differences between compounds revealed in honey samples (red region in heat map for F present prevalently phenolic acid and hydroxy fatty acids profile while the red region for C stands for reach flavonoids fraction). Capital letters indicate the samples and the number the analytical repetitions. Heat-map (cells, exposed to honeys) was generated using R 3.5.2. software 30 . www.nature.com/scientificreports/ highest quantity in terpenes and terpenoids (both around 45 ppb compared to the other honeys, ranging from 13 to 15 ppb). Honeys D and F also had the highest amount of carboxylic acids ( Table 2). The single compounds were then analysed to identify the compounds significantly different in each sample (Fig. 8). Carboxylic acids are responsible of different scents, depending on carbon chain length. Short chain acids such as acetic acid have for example spicy flavours, while butanoic acid and hexanoic acids determine a rancid aroma 18 . Honey F was the one containing the highest quantity of acetic acid (81.89 ± 4.069 ppb) and the one with a consistent presence of chestnut pollen. Honey F was described having the peculiarly spiced flavour of chestnut together with the vegetable/fruity flavour of wildflower and the bitterness of linden tree and chestnut. Honey F was also the richest in nitrogenated compounds (Fig. 8) such as 2-Methylfuran, detected only in honey F. Hydrate furanoids such as 2-Methyltetrahydro-3-furanone have already been found in chestnut honeys of different countries 71 . Shikimate pathway derivatives such as 2,4-Di-tert-butylphenol, 3-Methylacetophenone and 3,4-Dimethylacetophenone, and some specific monoterpenes such as z-Rose oxide, α-terpinene and cymene, in higher quantities in honey F, were previously associated to linden honey or in general to the genus Tilia 71 , that includes different species generally called linden trees for the European species. These species contribute to the production of linden tree honey, which has a taste which is sweet, bitter, medicinal, floral, woody and hay-like 72 . Honey F is the one produced at the lower station, where the linden and chestnut botanical belt starts, and is more definable organoleptically as a mix of Tilia and Castanea honey than a mountain wildflower honey. Honey F is the only one in which Tilia pollen was detected. Tilia was not detected by the floristic relevé but is present in the production area of honey F; this could mean that bees are willing to travel far to find this essence. 1-Phenylethanol, 2-Phenylethanol, 2-aminoacetophenone and acetophenone content distinguished honey F with a high content of chestnut pollen coherently with a previous report 73 . Considered as a product related to the shikimic acid pathway as well 74 , acetophenone is formed during phenylpropane metabolism by enzymatic reactions from hydroxy-substituted aromatic acids 75 . According to 73 , 1-Phenylethanol can be associated to a floral odour and may contribute to the characteristic floral aroma of chestnut honey in synergy with other compounds, such as phenylacetaldehyde or 2-phenylethanol. 2-aminoacetophenone is probably not issued by the shikimate pathway but might be the result of tryptophane degradation 76 . In this study tryptophane was not detected by GC-MS but by HPLC Orbitrap. 2-aminoacetophenone is a strong scent 73 that can also contribute to the flavour of chestnut honey. Moreover, honey F had the highest concentrations of products produced by the non-enzymatic browning of sugars such as 2-acetylfuran, 3-Methylfuran, 2-Ethylfuran, menthofuran, 4,7-Dimethyl-benzofuran and 2,3-Dihydro-benzofuran. Only 4,5-Dimethyl-2-formylfuran was highest in honey E. These results fit in with the large amounts of furan and methylfuran previously measured in 77 with a headspace dynamic method in the same origin. These furan derivatives cannot be considered specific floral markers because they are associated to specific www.nature.com/scientificreports/ thermal treatment and storage conditions 77 but they can be useful to authenticate honey of chestnut origin that is characterized by a bitter, sweet, burnt caramel and woody flavour 78 . Considering instead benzene derivatives, benzoic acid is considered a marker of nectar from the genera Erica and Calluna and was found predominantly in honey F (besides honey D, 10 ppb, against less than 1 ppb in the other honeys). Honey F, as well as being chestnut pollen prevalent, it is also the only honey with a dominant presence of Ericaceae pollen. Nectar from the genera Erica and Calluna contribute to the production of heather honey 79 , its flavour being characterized by sweet and candy-like notes 80 . Benzaldehyde was also found in a quantity higher than in the other honeys in honey F (29.16 ± 6.36 ppb) and has been previously described as a chestnut honey marker 71 , while in honeys B, D and E it was around 10 ppb and in the remaining less than 2 ppb. Different compounds that were responsible for the dissimilar profile in honey D were a category that are considered both "non-specific" (deriving from the degradation of carotenoid precursors producing different C9-norisoprenoids) and markers of some specific essences such as Ericaeae (heather honey) and were coherently found in higher quantities also in honey F (Fig. 8). For example, significantly higher quantities of α-Isophorone and 4-Oxoisophorone were found only in honeys D and F, while in the other samples they were present only in traces. Other non-specific compounds found in higher quantities in honeys D and F were numerous carboxylic acids (Hexa-2,4-Dienoic acid isomer, Hexanoic acid, Butanoic acid and Acetic acid) and the terpenes Linalool, trans-Linalool oxide and cis-Linalool oxide, Lilac Aldheide and its isomers. These kinds of compound are among the most common in honey, and many VOCs considered nonspecific such as linalool and linalool oxide have been seen to be involved in communication between flowers of fruit crops and their pollinators 81 . Moreover, honeys D and F were the ones containing higher quantities of Hotrienol, a regular monoterpene derived from geranyl pyrophosphate (GPP), among the most common honey terpenes. β-Damascenone, defined as a Rhododendron honey and Ericacee in general honey marker 71,75,82 was found in a higher quantity in honey D compared to the other honeys (2.49 ppb while in the other honeys it was always lower than 1 ppb) (Fig. 8).
A significant amount of nicotinaldehyde was found in honeys B (6.65 ppb), D (2.6 ppb), E (7.88) and F (3.01 ppb), while this compound was not detected in honeys A (almost pure rhododendron) and C where it was found in lower quantities (0.16 ppb) (Table S3). This is remarkably coherent with the melissopalynological analysis where Rubus pollen was found mainly in honeys B (67.12%), D (34.74%), E (44.25%) and F (38.59%), while it was found in far lower amounts in honeys A (11.83%) and C (17.24%). The high correlation between nicotinaldehyde content and percentage of Rubus pollen (r = 0.853, p < 0.05) qualifies this molecule as a promising marker for Rubus pollen occurrence. A lower content in nicotinaldehyde (0.16 ppb) was found in sample C that was characterized by 17.24% of Rubus pollen, whereas it was not detected in sample A that was characterized by 11% of Rubus pollen. Therefore, cut-off values for nicotinaldehyde as a Rubus pollen marker can preliminarily be set at 17%. Raspberry honey is quite rare and only few references exist concerning volatile compounds of this honey produced in Estonia and Slovakia 71 , although without mention of the compound nicotinaldehyde. However, nicotinaldehyde was reported in the chemical composition of volatiles released by the flowers and fruits of the red raspberry (Rubus idaeus) by 83 . Nicotinaldehydes could be involved in the synthesis of quinolones, the class of naturally occurring compounds that was abundantly present in samples according to our HPLC Orbitrap analysis. It is clear that there is a need for further studies to ascertain the existence of a common volatile profile for this type of honey 71 . Quinoline alkaloids, sought by bees in both chestnut and raspberry, could therefore have an important role in insect-plant communication. It is important to consider the volatile composition of honey as these substances are the main factors responsible for aroma which, together with other factors such as taste and physical factors, contribute to flavour 17 . Nevertheless, the volatile fingerprint of honey as a recognition method for floral origin is a debated topic because it is rather difficult to find reliable chemical markers for the discrimination of honey collected from different floral sources. Therefore, scientific publications may report different floral markers for honey of the same floral origin.
The organoleptic properties of honey (flavour, colour, aroma, and texture) are important factors in consumer's choice of together with the new concept of functional food. Some kinds of honey can have higher market values because of peculiar sensorial properties or for the content in compounds with antioxidant and antimicrobial activity 67,84,85 . Therefore, the characterisation of different honeys could increase their commercial value. Mountain wildflower honeys, for example, can be considered particularly valuable, as they are the expression of meadows and pastures biodiversity. In fact, they are included in the presidia to protect following the rules of the Slow Food Foundation for Biodiversity 10 . When transhumance was widespread and regular in the Alps, bees (wild and domesticated) also benefited: the pastures were cared for and cleaned to the benefit not only of the animals, but also of the vegetation, and then the bees 10 .
As exposed, the chemical composition and the consequent properties of honey depend largely on its floral source which is very dependent on the geographical origin of honey 14 . As well, potential contamination sources can characterise different honey production areas 86 , also in the case of organic honey 87 . The production of honey in uncontaminated areas as mountain areas, can be a way of promoting this product so difficult to obtain, together with the promotion of bioactive properties of a sustainable product from marginal territories.
Nevertheless, the honey of the study area, although produced in a limited territory, produced very different honeys according to the different apiary stations and, most importantly, the honeys were not totally definable as wildflower. This highlights the importance of formal analysis (at least melissopalynological analysis if phytochemical is not possible) to characterize honey, that could often reveal food products even more interesting than expected. A better definition of honeys, in addition, could give important cues for beekeeping management, as there are certain key species or plant groups that are particularly important 2 and a valuable area of further research is to discover why these particular species are important and how landscapes affect bee foraging in different environments. An understanding of the reasons why honeybees target certain plants could help to provide guidance on what constitutes a balanced honeybee diet 2 . The characterization of honeys, furthermore, could lead www.nature.com/scientificreports/ to the discovery of important compounds both for bees, in particular considering bee behaviour and bee-plant communication, and for human nutrition.

Conclusion
Our study highlights the difficulty of producing a mountain wildflower honey in the Alps taking into consideration expected aroma and flavour characteristics. This could be attributed to environmental and behavioural features, also linked to the presence of particular neuroactive compounds in some essences. Thorough characterization of the product is essential in order to reveal unexpected characteristics also from a nutraceutical point of view. It is to be hoped that this work will contribute to the valorisation of wildflower honey, this important product of the Alps, and give new impetus to its production conditional on bee behaviour which in turn depends on the anthropic and climatic changes affecting mountain territories.